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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 391 - (show annotations) (download)
Tue Mar 17 21:16:01 2009 UTC (5 years, 7 months ago) by ph10
File MIME type: text/plain
File size: 207878 byte(s)
Add support for UTF-8 in EBCDIC environments.

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) 1997-2009 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 DEBUG is defined, we need the pcre_printint() function, which is also
57 used by pcretest. DEBUG is not defined when building a production library. */
58
59 #ifdef DEBUG
60 #include "pcre_printint.src"
61 #endif
62
63
64 /* Macro for setting individual bits in class bitmaps. */
65
66 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
67
68 /* Maximum length value to check against when making sure that the integer that
69 holds the compiled pattern length does not overflow. We make it a bit less than
70 INT_MAX to allow for adding in group terminating bytes, so that we don't have
71 to check them every time. */
72
73 #define OFLOW_MAX (INT_MAX - 20)
74
75
76 /*************************************************
77 * Code parameters and static tables *
78 *************************************************/
79
80 /* This value specifies the size of stack workspace that is used during the
81 first pre-compile phase that determines how much memory is required. The regex
82 is partly compiled into this space, but the compiled parts are discarded as
83 soon as they can be, so that hopefully there will never be an overrun. The code
84 does, however, check for an overrun. The largest amount I've seen used is 218,
85 so this number is very generous.
86
87 The same workspace is used during the second, actual compile phase for
88 remembering forward references to groups so that they can be filled in at the
89 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
90 is 4 there is plenty of room. */
91
92 #define COMPILE_WORK_SIZE (4096)
93
94
95 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
96 are simple data values; negative values are for special things like \d and so
97 on. Zero means further processing is needed (for things like \x), or the escape
98 is invalid. */
99
100 #ifndef EBCDIC
101
102 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
103 in UTF-8 mode. */
104
105 static const short int escapes[] = {
106 0, 0,
107 0, 0,
108 0, 0,
109 0, 0,
110 0, 0,
111 CHAR_COLON, CHAR_SEMICOLON,
112 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
113 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
114 CHAR_COMMERCIAL_AT, -ESC_A,
115 -ESC_B, -ESC_C,
116 -ESC_D, -ESC_E,
117 0, -ESC_G,
118 -ESC_H, 0,
119 0, -ESC_K,
120 0, 0,
121 0, 0,
122 -ESC_P, -ESC_Q,
123 -ESC_R, -ESC_S,
124 0, 0,
125 -ESC_V, -ESC_W,
126 -ESC_X, 0,
127 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
128 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
129 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
130 CHAR_GRAVE_ACCENT, 7,
131 -ESC_b, 0,
132 -ESC_d, ESC_e,
133 ESC_f, 0,
134 -ESC_h, 0,
135 0, -ESC_k,
136 0, 0,
137 ESC_n, 0,
138 -ESC_p, 0,
139 ESC_r, -ESC_s,
140 ESC_tee, 0,
141 -ESC_v, -ESC_w,
142 0, 0,
143 -ESC_z
144 };
145
146 #else
147
148 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
149
150 static const short int escapes[] = {
151 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
152 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
153 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
154 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
155 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
156 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
157 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
158 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
159 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
160 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
161 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
162 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
163 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
164 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
165 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
166 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
167 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
168 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
169 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
170 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
171 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
172 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
173 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
174 };
175 #endif
176
177
178 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
179 searched linearly. Put all the names into a single string, in order to reduce
180 the number of relocations when a shared library is dynamically linked. The
181 string is built from string macros so that it works in UTF-8 mode on EBCDIC
182 platforms. */
183
184 typedef struct verbitem {
185 int len;
186 int op;
187 } verbitem;
188
189 static const char verbnames[] =
190 STRING_ACCEPT0
191 STRING_COMMIT0
192 STRING_F0
193 STRING_FAIL0
194 STRING_PRUNE0
195 STRING_SKIP0
196 STRING_THEN;
197
198 static const verbitem verbs[] = {
199 { 6, OP_ACCEPT },
200 { 6, OP_COMMIT },
201 { 1, OP_FAIL },
202 { 4, OP_FAIL },
203 { 5, OP_PRUNE },
204 { 4, OP_SKIP },
205 { 4, OP_THEN }
206 };
207
208 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
209
210
211 /* Tables of names of POSIX character classes and their lengths. The names are
212 now all in a single string, to reduce the number of relocations when a shared
213 library is dynamically loaded. The list of lengths is terminated by a zero
214 length entry. The first three must be alpha, lower, upper, as this is assumed
215 for handling case independence. */
216
217 static const char posix_names[] =
218 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
219 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
220 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
221 STRING_word0 STRING_xdigit;
222
223 static const uschar posix_name_lengths[] = {
224 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
225
226 /* Table of class bit maps for each POSIX class. Each class is formed from a
227 base map, with an optional addition or removal of another map. Then, for some
228 classes, there is some additional tweaking: for [:blank:] the vertical space
229 characters are removed, and for [:alpha:] and [:alnum:] the underscore
230 character is removed. The triples in the table consist of the base map offset,
231 second map offset or -1 if no second map, and a non-negative value for map
232 addition or a negative value for map subtraction (if there are two maps). The
233 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
234 remove vertical space characters, 2 => remove underscore. */
235
236 static const int posix_class_maps[] = {
237 cbit_word, cbit_digit, -2, /* alpha */
238 cbit_lower, -1, 0, /* lower */
239 cbit_upper, -1, 0, /* upper */
240 cbit_word, -1, 2, /* alnum - word without underscore */
241 cbit_print, cbit_cntrl, 0, /* ascii */
242 cbit_space, -1, 1, /* blank - a GNU extension */
243 cbit_cntrl, -1, 0, /* cntrl */
244 cbit_digit, -1, 0, /* digit */
245 cbit_graph, -1, 0, /* graph */
246 cbit_print, -1, 0, /* print */
247 cbit_punct, -1, 0, /* punct */
248 cbit_space, -1, 0, /* space */
249 cbit_word, -1, 0, /* word - a Perl extension */
250 cbit_xdigit,-1, 0 /* xdigit */
251 };
252
253
254 #define STRING(a) # a
255 #define XSTRING(s) STRING(s)
256
257 /* The texts of compile-time error messages. These are "char *" because they
258 are passed to the outside world. Do not ever re-use any error number, because
259 they are documented. Always add a new error instead. Messages marked DEAD below
260 are no longer used. This used to be a table of strings, but in order to reduce
261 the number of relocations needed when a shared library is loaded dynamically,
262 it is now one long string. We cannot use a table of offsets, because the
263 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
264 simply count through to the one we want - this isn't a performance issue
265 because these strings are used only when there is a compilation error. */
266
267 static const char error_texts[] =
268 "no error\0"
269 "\\ at end of pattern\0"
270 "\\c at end of pattern\0"
271 "unrecognized character follows \\\0"
272 "numbers out of order in {} quantifier\0"
273 /* 5 */
274 "number too big in {} quantifier\0"
275 "missing terminating ] for character class\0"
276 "invalid escape sequence in character class\0"
277 "range out of order in character class\0"
278 "nothing to repeat\0"
279 /* 10 */
280 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
281 "internal error: unexpected repeat\0"
282 "unrecognized character after (? or (?-\0"
283 "POSIX named classes are supported only within a class\0"
284 "missing )\0"
285 /* 15 */
286 "reference to non-existent subpattern\0"
287 "erroffset passed as NULL\0"
288 "unknown option bit(s) set\0"
289 "missing ) after comment\0"
290 "parentheses nested too deeply\0" /** DEAD **/
291 /* 20 */
292 "regular expression is too large\0"
293 "failed to get memory\0"
294 "unmatched parentheses\0"
295 "internal error: code overflow\0"
296 "unrecognized character after (?<\0"
297 /* 25 */
298 "lookbehind assertion is not fixed length\0"
299 "malformed number or name after (?(\0"
300 "conditional group contains more than two branches\0"
301 "assertion expected after (?(\0"
302 "(?R or (?[+-]digits must be followed by )\0"
303 /* 30 */
304 "unknown POSIX class name\0"
305 "POSIX collating elements are not supported\0"
306 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
307 "spare error\0" /** DEAD **/
308 "character value in \\x{...} sequence is too large\0"
309 /* 35 */
310 "invalid condition (?(0)\0"
311 "\\C not allowed in lookbehind assertion\0"
312 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
313 "number after (?C is > 255\0"
314 "closing ) for (?C expected\0"
315 /* 40 */
316 "recursive call could loop indefinitely\0"
317 "unrecognized character after (?P\0"
318 "syntax error in subpattern name (missing terminator)\0"
319 "two named subpatterns have the same name\0"
320 "invalid UTF-8 string\0"
321 /* 45 */
322 "support for \\P, \\p, and \\X has not been compiled\0"
323 "malformed \\P or \\p sequence\0"
324 "unknown property name after \\P or \\p\0"
325 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
326 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
327 /* 50 */
328 "repeated subpattern is too long\0" /** DEAD **/
329 "octal value is greater than \\377 (not in UTF-8 mode)\0"
330 "internal error: overran compiling workspace\0"
331 "internal error: previously-checked referenced subpattern not found\0"
332 "DEFINE group contains more than one branch\0"
333 /* 55 */
334 "repeating a DEFINE group is not allowed\0"
335 "inconsistent NEWLINE options\0"
336 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
337 "a numbered reference must not be zero\0"
338 "(*VERB) with an argument is not supported\0"
339 /* 60 */
340 "(*VERB) not recognized\0"
341 "number is too big\0"
342 "subpattern name expected\0"
343 "digit expected after (?+\0"
344 "] is an invalid data character in JavaScript compatibility mode";
345
346
347 /* Table to identify digits and hex digits. This is used when compiling
348 patterns. Note that the tables in chartables are dependent on the locale, and
349 may mark arbitrary characters as digits - but the PCRE compiling code expects
350 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
351 a private table here. It costs 256 bytes, but it is a lot faster than doing
352 character value tests (at least in some simple cases I timed), and in some
353 applications one wants PCRE to compile efficiently as well as match
354 efficiently.
355
356 For convenience, we use the same bit definitions as in chartables:
357
358 0x04 decimal digit
359 0x08 hexadecimal digit
360
361 Then we can use ctype_digit and ctype_xdigit in the code. */
362
363 #ifndef EBCDIC
364
365 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
366 UTF-8 mode. */
367
368 static const unsigned char digitab[] =
369 {
370 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
371 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
372 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
373 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
374 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
375 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
376 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
377 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
378 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
379 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
382 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
383 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
384 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
387 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
388 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
394 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
395 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
396 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
402
403 #else
404
405 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
406
407 static const unsigned char digitab[] =
408 {
409 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
410 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
411 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
412 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
413 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
414 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
420 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
421 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
422 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
423 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
424 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
425 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
426 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
431 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
433 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
439 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
440 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
441
442 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
443 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
444 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
445 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
447 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
451 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
452 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
454 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
456 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
459 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
460 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
461 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
462 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
463 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
464 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
465 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
466 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
467 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
468 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
469 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
470 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
471 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
472 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
473 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
474 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
475 #endif
476
477
478 /* Definition to allow mutual recursion */
479
480 static BOOL
481 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
482 int *, int *, branch_chain *, compile_data *, int *);
483
484
485
486 /*************************************************
487 * Find an error text *
488 *************************************************/
489
490 /* The error texts are now all in one long string, to save on relocations. As
491 some of the text is of unknown length, we can't use a table of offsets.
492 Instead, just count through the strings. This is not a performance issue
493 because it happens only when there has been a compilation error.
494
495 Argument: the error number
496 Returns: pointer to the error string
497 */
498
499 static const char *
500 find_error_text(int n)
501 {
502 const char *s = error_texts;
503 for (; n > 0; n--) while (*s++ != 0) {};
504 return s;
505 }
506
507
508 /*************************************************
509 * Handle escapes *
510 *************************************************/
511
512 /* This function is called when a \ has been encountered. It either returns a
513 positive value for a simple escape such as \n, or a negative value which
514 encodes one of the more complicated things such as \d. A backreference to group
515 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
516 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
517 ptr is pointing at the \. On exit, it is on the final character of the escape
518 sequence.
519
520 Arguments:
521 ptrptr points to the pattern position pointer
522 errorcodeptr points to the errorcode variable
523 bracount number of previous extracting brackets
524 options the options bits
525 isclass TRUE if inside a character class
526
527 Returns: zero or positive => a data character
528 negative => a special escape sequence
529 on error, errorcodeptr is set
530 */
531
532 static int
533 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
534 int options, BOOL isclass)
535 {
536 BOOL utf8 = (options & PCRE_UTF8) != 0;
537 const uschar *ptr = *ptrptr + 1;
538 int c, i;
539
540 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
541 ptr--; /* Set pointer back to the last byte */
542
543 /* If backslash is at the end of the pattern, it's an error. */
544
545 if (c == 0) *errorcodeptr = ERR1;
546
547 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
548 in a table. A non-zero result is something that can be returned immediately.
549 Otherwise further processing may be required. */
550
551 #ifndef EBCDIC /* ASCII/UTF-8 coding */
552 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
553 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
554
555 #else /* EBCDIC coding */
556 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
557 else if ((i = escapes[c - 0x48]) != 0) c = i;
558 #endif
559
560 /* Escapes that need further processing, or are illegal. */
561
562 else
563 {
564 const uschar *oldptr;
565 BOOL braced, negated;
566
567 switch (c)
568 {
569 /* A number of Perl escapes are not handled by PCRE. We give an explicit
570 error. */
571
572 case CHAR_l:
573 case CHAR_L:
574 case CHAR_N:
575 case CHAR_u:
576 case CHAR_U:
577 *errorcodeptr = ERR37;
578 break;
579
580 /* \g must be followed by one of a number of specific things:
581
582 (1) A number, either plain or braced. If positive, it is an absolute
583 backreference. If negative, it is a relative backreference. This is a Perl
584 5.10 feature.
585
586 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
587 is part of Perl's movement towards a unified syntax for back references. As
588 this is synonymous with \k{name}, we fudge it up by pretending it really
589 was \k.
590
591 (3) For Oniguruma compatibility we also support \g followed by a name or a
592 number either in angle brackets or in single quotes. However, these are
593 (possibly recursive) subroutine calls, _not_ backreferences. Just return
594 the -ESC_g code (cf \k). */
595
596 case CHAR_g:
597 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
598 {
599 c = -ESC_g;
600 break;
601 }
602
603 /* Handle the Perl-compatible cases */
604
605 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
606 {
607 const uschar *p;
608 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
609 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
610 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
611 {
612 c = -ESC_k;
613 break;
614 }
615 braced = TRUE;
616 ptr++;
617 }
618 else braced = FALSE;
619
620 if (ptr[1] == CHAR_MINUS)
621 {
622 negated = TRUE;
623 ptr++;
624 }
625 else negated = FALSE;
626
627 c = 0;
628 while ((digitab[ptr[1]] & ctype_digit) != 0)
629 c = c * 10 + *(++ptr) - CHAR_0;
630
631 if (c < 0) /* Integer overflow */
632 {
633 *errorcodeptr = ERR61;
634 break;
635 }
636
637 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
638 {
639 *errorcodeptr = ERR57;
640 break;
641 }
642
643 if (c == 0)
644 {
645 *errorcodeptr = ERR58;
646 break;
647 }
648
649 if (negated)
650 {
651 if (c > bracount)
652 {
653 *errorcodeptr = ERR15;
654 break;
655 }
656 c = bracount - (c - 1);
657 }
658
659 c = -(ESC_REF + c);
660 break;
661
662 /* The handling of escape sequences consisting of a string of digits
663 starting with one that is not zero is not straightforward. By experiment,
664 the way Perl works seems to be as follows:
665
666 Outside a character class, the digits are read as a decimal number. If the
667 number is less than 10, or if there are that many previous extracting
668 left brackets, then it is a back reference. Otherwise, up to three octal
669 digits are read to form an escaped byte. Thus \123 is likely to be octal
670 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
671 value is greater than 377, the least significant 8 bits are taken. Inside a
672 character class, \ followed by a digit is always an octal number. */
673
674 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
675 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
676
677 if (!isclass)
678 {
679 oldptr = ptr;
680 c -= CHAR_0;
681 while ((digitab[ptr[1]] & ctype_digit) != 0)
682 c = c * 10 + *(++ptr) - CHAR_0;
683 if (c < 0) /* Integer overflow */
684 {
685 *errorcodeptr = ERR61;
686 break;
687 }
688 if (c < 10 || c <= bracount)
689 {
690 c = -(ESC_REF + c);
691 break;
692 }
693 ptr = oldptr; /* Put the pointer back and fall through */
694 }
695
696 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
697 generates a binary zero byte and treats the digit as a following literal.
698 Thus we have to pull back the pointer by one. */
699
700 if ((c = *ptr) >= CHAR_8)
701 {
702 ptr--;
703 c = 0;
704 break;
705 }
706
707 /* \0 always starts an octal number, but we may drop through to here with a
708 larger first octal digit. The original code used just to take the least
709 significant 8 bits of octal numbers (I think this is what early Perls used
710 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
711 than 3 octal digits. */
712
713 case CHAR_0:
714 c -= CHAR_0;
715 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
716 c = c * 8 + *(++ptr) - CHAR_0;
717 if (!utf8 && c > 255) *errorcodeptr = ERR51;
718 break;
719
720 /* \x is complicated. \x{ddd} is a character number which can be greater
721 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
722 treated as a data character. */
723
724 case CHAR_x:
725 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
726 {
727 const uschar *pt = ptr + 2;
728 int count = 0;
729
730 c = 0;
731 while ((digitab[*pt] & ctype_xdigit) != 0)
732 {
733 register int cc = *pt++;
734 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
735 count++;
736
737 #ifndef EBCDIC /* ASCII/UTF-8 coding */
738 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
739 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
740 #else /* EBCDIC coding */
741 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
742 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
743 #endif
744 }
745
746 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
747 {
748 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
749 ptr = pt;
750 break;
751 }
752
753 /* If the sequence of hex digits does not end with '}', then we don't
754 recognize this construct; fall through to the normal \x handling. */
755 }
756
757 /* Read just a single-byte hex-defined char */
758
759 c = 0;
760 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
761 {
762 int cc; /* Some compilers don't like */
763 cc = *(++ptr); /* ++ in initializers */
764 #ifndef EBCDIC /* ASCII/UTF-8 coding */
765 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
766 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
767 #else /* EBCDIC coding */
768 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
769 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
770 #endif
771 }
772 break;
773
774 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
775 This coding is ASCII-specific, but then the whole concept of \cx is
776 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
777
778 case CHAR_c:
779 c = *(++ptr);
780 if (c == 0)
781 {
782 *errorcodeptr = ERR2;
783 break;
784 }
785
786 #ifndef EBCDIC /* ASCII/UTF-8 coding */
787 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
788 c ^= 0x40;
789 #else /* EBCDIC coding */
790 if (c >= CHAR_a && c <= CHAR_z) c += 64;
791 c ^= 0xC0;
792 #endif
793 break;
794
795 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
796 other alphanumeric following \ is an error if PCRE_EXTRA was set;
797 otherwise, for Perl compatibility, it is a literal. This code looks a bit
798 odd, but there used to be some cases other than the default, and there may
799 be again in future, so I haven't "optimized" it. */
800
801 default:
802 if ((options & PCRE_EXTRA) != 0) switch(c)
803 {
804 default:
805 *errorcodeptr = ERR3;
806 break;
807 }
808 break;
809 }
810 }
811
812 *ptrptr = ptr;
813 return c;
814 }
815
816
817
818 #ifdef SUPPORT_UCP
819 /*************************************************
820 * Handle \P and \p *
821 *************************************************/
822
823 /* This function is called after \P or \p has been encountered, provided that
824 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
825 pointing at the P or p. On exit, it is pointing at the final character of the
826 escape sequence.
827
828 Argument:
829 ptrptr points to the pattern position pointer
830 negptr points to a boolean that is set TRUE for negation else FALSE
831 dptr points to an int that is set to the detailed property value
832 errorcodeptr points to the error code variable
833
834 Returns: type value from ucp_type_table, or -1 for an invalid type
835 */
836
837 static int
838 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
839 {
840 int c, i, bot, top;
841 const uschar *ptr = *ptrptr;
842 char name[32];
843
844 c = *(++ptr);
845 if (c == 0) goto ERROR_RETURN;
846
847 *negptr = FALSE;
848
849 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
850 negation. */
851
852 if (c == CHAR_LEFT_CURLY_BRACKET)
853 {
854 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
855 {
856 *negptr = TRUE;
857 ptr++;
858 }
859 for (i = 0; i < (int)sizeof(name) - 1; i++)
860 {
861 c = *(++ptr);
862 if (c == 0) goto ERROR_RETURN;
863 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
864 name[i] = c;
865 }
866 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
867 name[i] = 0;
868 }
869
870 /* Otherwise there is just one following character */
871
872 else
873 {
874 name[0] = c;
875 name[1] = 0;
876 }
877
878 *ptrptr = ptr;
879
880 /* Search for a recognized property name using binary chop */
881
882 bot = 0;
883 top = _pcre_utt_size;
884
885 while (bot < top)
886 {
887 i = (bot + top) >> 1;
888 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
889 if (c == 0)
890 {
891 *dptr = _pcre_utt[i].value;
892 return _pcre_utt[i].type;
893 }
894 if (c > 0) bot = i + 1; else top = i;
895 }
896
897 *errorcodeptr = ERR47;
898 *ptrptr = ptr;
899 return -1;
900
901 ERROR_RETURN:
902 *errorcodeptr = ERR46;
903 *ptrptr = ptr;
904 return -1;
905 }
906 #endif
907
908
909
910
911 /*************************************************
912 * Check for counted repeat *
913 *************************************************/
914
915 /* This function is called when a '{' is encountered in a place where it might
916 start a quantifier. It looks ahead to see if it really is a quantifier or not.
917 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
918 where the ddds are digits.
919
920 Arguments:
921 p pointer to the first char after '{'
922
923 Returns: TRUE or FALSE
924 */
925
926 static BOOL
927 is_counted_repeat(const uschar *p)
928 {
929 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
930 while ((digitab[*p] & ctype_digit) != 0) p++;
931 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
932
933 if (*p++ != CHAR_COMMA) return FALSE;
934 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
935
936 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
937 while ((digitab[*p] & ctype_digit) != 0) p++;
938
939 return (*p == CHAR_RIGHT_CURLY_BRACKET);
940 }
941
942
943
944 /*************************************************
945 * Read repeat counts *
946 *************************************************/
947
948 /* Read an item of the form {n,m} and return the values. This is called only
949 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
950 so the syntax is guaranteed to be correct, but we need to check the values.
951
952 Arguments:
953 p pointer to first char after '{'
954 minp pointer to int for min
955 maxp pointer to int for max
956 returned as -1 if no max
957 errorcodeptr points to error code variable
958
959 Returns: pointer to '}' on success;
960 current ptr on error, with errorcodeptr set non-zero
961 */
962
963 static const uschar *
964 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
965 {
966 int min = 0;
967 int max = -1;
968
969 /* Read the minimum value and do a paranoid check: a negative value indicates
970 an integer overflow. */
971
972 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
973 if (min < 0 || min > 65535)
974 {
975 *errorcodeptr = ERR5;
976 return p;
977 }
978
979 /* Read the maximum value if there is one, and again do a paranoid on its size.
980 Also, max must not be less than min. */
981
982 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
983 {
984 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
985 {
986 max = 0;
987 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
988 if (max < 0 || max > 65535)
989 {
990 *errorcodeptr = ERR5;
991 return p;
992 }
993 if (max < min)
994 {
995 *errorcodeptr = ERR4;
996 return p;
997 }
998 }
999 }
1000
1001 /* Fill in the required variables, and pass back the pointer to the terminating
1002 '}'. */
1003
1004 *minp = min;
1005 *maxp = max;
1006 return p;
1007 }
1008
1009
1010
1011 /*************************************************
1012 * Find forward referenced subpattern *
1013 *************************************************/
1014
1015 /* This function scans along a pattern's text looking for capturing
1016 subpatterns, and counting them. If it finds a named pattern that matches the
1017 name it is given, it returns its number. Alternatively, if the name is NULL, it
1018 returns when it reaches a given numbered subpattern. This is used for forward
1019 references to subpatterns. We know that if (?P< is encountered, the name will
1020 be terminated by '>' because that is checked in the first pass.
1021
1022 Arguments:
1023 ptr current position in the pattern
1024 cd compile background data
1025 name name to seek, or NULL if seeking a numbered subpattern
1026 lorn name length, or subpattern number if name is NULL
1027 xmode TRUE if we are in /x mode
1028
1029 Returns: the number of the named subpattern, or -1 if not found
1030 */
1031
1032 static int
1033 find_parens(const uschar *ptr, compile_data *cd, const uschar *name, int lorn,
1034 BOOL xmode)
1035 {
1036 const uschar *thisname;
1037 int count = cd->bracount;
1038
1039 for (; *ptr != 0; ptr++)
1040 {
1041 int term;
1042
1043 /* Skip over backslashed characters and also entire \Q...\E */
1044
1045 if (*ptr == CHAR_BACKSLASH)
1046 {
1047 if (*(++ptr) == 0) return -1;
1048 if (*ptr == CHAR_Q) for (;;)
1049 {
1050 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1051 if (*ptr == 0) return -1;
1052 if (*(++ptr) == CHAR_E) break;
1053 }
1054 continue;
1055 }
1056
1057 /* Skip over character classes; this logic must be similar to the way they
1058 are handled for real. If the first character is '^', skip it. Also, if the
1059 first few characters (either before or after ^) are \Q\E or \E we skip them
1060 too. This makes for compatibility with Perl. Note the use of STR macros to
1061 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1062
1063 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1064 {
1065 BOOL negate_class = FALSE;
1066 for (;;)
1067 {
1068 int c = *(++ptr);
1069 if (c == CHAR_BACKSLASH)
1070 {
1071 if (ptr[1] == CHAR_E)
1072 ptr++;
1073 else if (strncmp((const char *)ptr+1,
1074 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1075 ptr += 3;
1076 else
1077 break;
1078 }
1079 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
1080 negate_class = TRUE;
1081 else break;
1082 }
1083
1084 /* If the next character is ']', it is a data character that must be
1085 skipped, except in JavaScript compatibility mode. */
1086
1087 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1088 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1089 ptr++;
1090
1091 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1092 {
1093 if (*ptr == 0) return -1;
1094 if (*ptr == CHAR_BACKSLASH)
1095 {
1096 if (*(++ptr) == 0) return -1;
1097 if (*ptr == CHAR_Q) for (;;)
1098 {
1099 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1100 if (*ptr == 0) return -1;
1101 if (*(++ptr) == CHAR_E) break;
1102 }
1103 continue;
1104 }
1105 }
1106 continue;
1107 }
1108
1109 /* Skip comments in /x mode */
1110
1111 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1112 {
1113 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1114 if (*ptr == 0) return -1;
1115 continue;
1116 }
1117
1118 /* An opening parens must now be a real metacharacter */
1119
1120 if (*ptr != CHAR_LEFT_PARENTHESIS) continue;
1121 if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1122 {
1123 count++;
1124 if (name == NULL && count == lorn) return count;
1125 continue;
1126 }
1127
1128 ptr += 2;
1129 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1130
1131 /* We have to disambiguate (?<! and (?<= from (?<name> */
1132
1133 if ((*ptr != CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_EXCLAMATION_MARK ||
1134 ptr[1] == CHAR_EQUALS_SIGN) && *ptr != CHAR_APOSTROPHE)
1135 continue;
1136
1137 count++;
1138
1139 if (name == NULL && count == lorn) return count;
1140 term = *ptr++;
1141 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1142 thisname = ptr;
1143 while (*ptr != term) ptr++;
1144 if (name != NULL && lorn == ptr - thisname &&
1145 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1146 return count;
1147 }
1148
1149 return -1;
1150 }
1151
1152
1153
1154 /*************************************************
1155 * Find first significant op code *
1156 *************************************************/
1157
1158 /* This is called by several functions that scan a compiled expression looking
1159 for a fixed first character, or an anchoring op code etc. It skips over things
1160 that do not influence this. For some calls, a change of option is important.
1161 For some calls, it makes sense to skip negative forward and all backward
1162 assertions, and also the \b assertion; for others it does not.
1163
1164 Arguments:
1165 code pointer to the start of the group
1166 options pointer to external options
1167 optbit the option bit whose changing is significant, or
1168 zero if none are
1169 skipassert TRUE if certain assertions are to be skipped
1170
1171 Returns: pointer to the first significant opcode
1172 */
1173
1174 static const uschar*
1175 first_significant_code(const uschar *code, int *options, int optbit,
1176 BOOL skipassert)
1177 {
1178 for (;;)
1179 {
1180 switch ((int)*code)
1181 {
1182 case OP_OPT:
1183 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1184 *options = (int)code[1];
1185 code += 2;
1186 break;
1187
1188 case OP_ASSERT_NOT:
1189 case OP_ASSERTBACK:
1190 case OP_ASSERTBACK_NOT:
1191 if (!skipassert) return code;
1192 do code += GET(code, 1); while (*code == OP_ALT);
1193 code += _pcre_OP_lengths[*code];
1194 break;
1195
1196 case OP_WORD_BOUNDARY:
1197 case OP_NOT_WORD_BOUNDARY:
1198 if (!skipassert) return code;
1199 /* Fall through */
1200
1201 case OP_CALLOUT:
1202 case OP_CREF:
1203 case OP_RREF:
1204 case OP_DEF:
1205 code += _pcre_OP_lengths[*code];
1206 break;
1207
1208 default:
1209 return code;
1210 }
1211 }
1212 /* Control never reaches here */
1213 }
1214
1215
1216
1217
1218 /*************************************************
1219 * Find the fixed length of a pattern *
1220 *************************************************/
1221
1222 /* Scan a pattern and compute the fixed length of subject that will match it,
1223 if the length is fixed. This is needed for dealing with backward assertions.
1224 In UTF8 mode, the result is in characters rather than bytes.
1225
1226 Arguments:
1227 code points to the start of the pattern (the bracket)
1228 options the compiling options
1229
1230 Returns: the fixed length, or -1 if there is no fixed length,
1231 or -2 if \C was encountered
1232 */
1233
1234 static int
1235 find_fixedlength(uschar *code, int options)
1236 {
1237 int length = -1;
1238
1239 register int branchlength = 0;
1240 register uschar *cc = code + 1 + LINK_SIZE;
1241
1242 /* Scan along the opcodes for this branch. If we get to the end of the
1243 branch, check the length against that of the other branches. */
1244
1245 for (;;)
1246 {
1247 int d;
1248 register int op = *cc;
1249 switch (op)
1250 {
1251 case OP_CBRA:
1252 case OP_BRA:
1253 case OP_ONCE:
1254 case OP_COND:
1255 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1256 if (d < 0) return d;
1257 branchlength += d;
1258 do cc += GET(cc, 1); while (*cc == OP_ALT);
1259 cc += 1 + LINK_SIZE;
1260 break;
1261
1262 /* Reached end of a branch; if it's a ket it is the end of a nested
1263 call. If it's ALT it is an alternation in a nested call. If it is
1264 END it's the end of the outer call. All can be handled by the same code. */
1265
1266 case OP_ALT:
1267 case OP_KET:
1268 case OP_KETRMAX:
1269 case OP_KETRMIN:
1270 case OP_END:
1271 if (length < 0) length = branchlength;
1272 else if (length != branchlength) return -1;
1273 if (*cc != OP_ALT) return length;
1274 cc += 1 + LINK_SIZE;
1275 branchlength = 0;
1276 break;
1277
1278 /* Skip over assertive subpatterns */
1279
1280 case OP_ASSERT:
1281 case OP_ASSERT_NOT:
1282 case OP_ASSERTBACK:
1283 case OP_ASSERTBACK_NOT:
1284 do cc += GET(cc, 1); while (*cc == OP_ALT);
1285 /* Fall through */
1286
1287 /* Skip over things that don't match chars */
1288
1289 case OP_REVERSE:
1290 case OP_CREF:
1291 case OP_RREF:
1292 case OP_DEF:
1293 case OP_OPT:
1294 case OP_CALLOUT:
1295 case OP_SOD:
1296 case OP_SOM:
1297 case OP_EOD:
1298 case OP_EODN:
1299 case OP_CIRC:
1300 case OP_DOLL:
1301 case OP_NOT_WORD_BOUNDARY:
1302 case OP_WORD_BOUNDARY:
1303 cc += _pcre_OP_lengths[*cc];
1304 break;
1305
1306 /* Handle literal characters */
1307
1308 case OP_CHAR:
1309 case OP_CHARNC:
1310 case OP_NOT:
1311 branchlength++;
1312 cc += 2;
1313 #ifdef SUPPORT_UTF8
1314 if ((options & PCRE_UTF8) != 0)
1315 {
1316 while ((*cc & 0xc0) == 0x80) cc++;
1317 }
1318 #endif
1319 break;
1320
1321 /* Handle exact repetitions. The count is already in characters, but we
1322 need to skip over a multibyte character in UTF8 mode. */
1323
1324 case OP_EXACT:
1325 branchlength += GET2(cc,1);
1326 cc += 4;
1327 #ifdef SUPPORT_UTF8
1328 if ((options & PCRE_UTF8) != 0)
1329 {
1330 while((*cc & 0x80) == 0x80) cc++;
1331 }
1332 #endif
1333 break;
1334
1335 case OP_TYPEEXACT:
1336 branchlength += GET2(cc,1);
1337 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1338 cc += 4;
1339 break;
1340
1341 /* Handle single-char matchers */
1342
1343 case OP_PROP:
1344 case OP_NOTPROP:
1345 cc += 2;
1346 /* Fall through */
1347
1348 case OP_NOT_DIGIT:
1349 case OP_DIGIT:
1350 case OP_NOT_WHITESPACE:
1351 case OP_WHITESPACE:
1352 case OP_NOT_WORDCHAR:
1353 case OP_WORDCHAR:
1354 case OP_ANY:
1355 case OP_ALLANY:
1356 branchlength++;
1357 cc++;
1358 break;
1359
1360 /* The single-byte matcher isn't allowed */
1361
1362 case OP_ANYBYTE:
1363 return -2;
1364
1365 /* Check a class for variable quantification */
1366
1367 #ifdef SUPPORT_UTF8
1368 case OP_XCLASS:
1369 cc += GET(cc, 1) - 33;
1370 /* Fall through */
1371 #endif
1372
1373 case OP_CLASS:
1374 case OP_NCLASS:
1375 cc += 33;
1376
1377 switch (*cc)
1378 {
1379 case OP_CRSTAR:
1380 case OP_CRMINSTAR:
1381 case OP_CRQUERY:
1382 case OP_CRMINQUERY:
1383 return -1;
1384
1385 case OP_CRRANGE:
1386 case OP_CRMINRANGE:
1387 if (GET2(cc,1) != GET2(cc,3)) return -1;
1388 branchlength += GET2(cc,1);
1389 cc += 5;
1390 break;
1391
1392 default:
1393 branchlength++;
1394 }
1395 break;
1396
1397 /* Anything else is variable length */
1398
1399 default:
1400 return -1;
1401 }
1402 }
1403 /* Control never gets here */
1404 }
1405
1406
1407
1408
1409 /*************************************************
1410 * Scan compiled regex for numbered bracket *
1411 *************************************************/
1412
1413 /* This little function scans through a compiled pattern until it finds a
1414 capturing bracket with the given number.
1415
1416 Arguments:
1417 code points to start of expression
1418 utf8 TRUE in UTF-8 mode
1419 number the required bracket number
1420
1421 Returns: pointer to the opcode for the bracket, or NULL if not found
1422 */
1423
1424 static const uschar *
1425 find_bracket(const uschar *code, BOOL utf8, int number)
1426 {
1427 for (;;)
1428 {
1429 register int c = *code;
1430 if (c == OP_END) return NULL;
1431
1432 /* XCLASS is used for classes that cannot be represented just by a bit
1433 map. This includes negated single high-valued characters. The length in
1434 the table is zero; the actual length is stored in the compiled code. */
1435
1436 if (c == OP_XCLASS) code += GET(code, 1);
1437
1438 /* Handle capturing bracket */
1439
1440 else if (c == OP_CBRA)
1441 {
1442 int n = GET2(code, 1+LINK_SIZE);
1443 if (n == number) return (uschar *)code;
1444 code += _pcre_OP_lengths[c];
1445 }
1446
1447 /* Otherwise, we can get the item's length from the table, except that for
1448 repeated character types, we have to test for \p and \P, which have an extra
1449 two bytes of parameters. */
1450
1451 else
1452 {
1453 switch(c)
1454 {
1455 case OP_TYPESTAR:
1456 case OP_TYPEMINSTAR:
1457 case OP_TYPEPLUS:
1458 case OP_TYPEMINPLUS:
1459 case OP_TYPEQUERY:
1460 case OP_TYPEMINQUERY:
1461 case OP_TYPEPOSSTAR:
1462 case OP_TYPEPOSPLUS:
1463 case OP_TYPEPOSQUERY:
1464 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1465 break;
1466
1467 case OP_TYPEUPTO:
1468 case OP_TYPEMINUPTO:
1469 case OP_TYPEEXACT:
1470 case OP_TYPEPOSUPTO:
1471 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1472 break;
1473 }
1474
1475 /* Add in the fixed length from the table */
1476
1477 code += _pcre_OP_lengths[c];
1478
1479 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1480 a multi-byte character. The length in the table is a minimum, so we have to
1481 arrange to skip the extra bytes. */
1482
1483 #ifdef SUPPORT_UTF8
1484 if (utf8) switch(c)
1485 {
1486 case OP_CHAR:
1487 case OP_CHARNC:
1488 case OP_EXACT:
1489 case OP_UPTO:
1490 case OP_MINUPTO:
1491 case OP_POSUPTO:
1492 case OP_STAR:
1493 case OP_MINSTAR:
1494 case OP_POSSTAR:
1495 case OP_PLUS:
1496 case OP_MINPLUS:
1497 case OP_POSPLUS:
1498 case OP_QUERY:
1499 case OP_MINQUERY:
1500 case OP_POSQUERY:
1501 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1502 break;
1503 }
1504 #else
1505 (void)(utf8); /* Keep compiler happy by referencing function argument */
1506 #endif
1507 }
1508 }
1509 }
1510
1511
1512
1513 /*************************************************
1514 * Scan compiled regex for recursion reference *
1515 *************************************************/
1516
1517 /* This little function scans through a compiled pattern until it finds an
1518 instance of OP_RECURSE.
1519
1520 Arguments:
1521 code points to start of expression
1522 utf8 TRUE in UTF-8 mode
1523
1524 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1525 */
1526
1527 static const uschar *
1528 find_recurse(const uschar *code, BOOL utf8)
1529 {
1530 for (;;)
1531 {
1532 register int c = *code;
1533 if (c == OP_END) return NULL;
1534 if (c == OP_RECURSE) return code;
1535
1536 /* XCLASS is used for classes that cannot be represented just by a bit
1537 map. This includes negated single high-valued characters. The length in
1538 the table is zero; the actual length is stored in the compiled code. */
1539
1540 if (c == OP_XCLASS) code += GET(code, 1);
1541
1542 /* Otherwise, we can get the item's length from the table, except that for
1543 repeated character types, we have to test for \p and \P, which have an extra
1544 two bytes of parameters. */
1545
1546 else
1547 {
1548 switch(c)
1549 {
1550 case OP_TYPESTAR:
1551 case OP_TYPEMINSTAR:
1552 case OP_TYPEPLUS:
1553 case OP_TYPEMINPLUS:
1554 case OP_TYPEQUERY:
1555 case OP_TYPEMINQUERY:
1556 case OP_TYPEPOSSTAR:
1557 case OP_TYPEPOSPLUS:
1558 case OP_TYPEPOSQUERY:
1559 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1560 break;
1561
1562 case OP_TYPEPOSUPTO:
1563 case OP_TYPEUPTO:
1564 case OP_TYPEMINUPTO:
1565 case OP_TYPEEXACT:
1566 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1567 break;
1568 }
1569
1570 /* Add in the fixed length from the table */
1571
1572 code += _pcre_OP_lengths[c];
1573
1574 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1575 by a multi-byte character. The length in the table is a minimum, so we have
1576 to arrange to skip the extra bytes. */
1577
1578 #ifdef SUPPORT_UTF8
1579 if (utf8) switch(c)
1580 {
1581 case OP_CHAR:
1582 case OP_CHARNC:
1583 case OP_EXACT:
1584 case OP_UPTO:
1585 case OP_MINUPTO:
1586 case OP_POSUPTO:
1587 case OP_STAR:
1588 case OP_MINSTAR:
1589 case OP_POSSTAR:
1590 case OP_PLUS:
1591 case OP_MINPLUS:
1592 case OP_POSPLUS:
1593 case OP_QUERY:
1594 case OP_MINQUERY:
1595 case OP_POSQUERY:
1596 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1597 break;
1598 }
1599 #else
1600 (void)(utf8); /* Keep compiler happy by referencing function argument */
1601 #endif
1602 }
1603 }
1604 }
1605
1606
1607
1608 /*************************************************
1609 * Scan compiled branch for non-emptiness *
1610 *************************************************/
1611
1612 /* This function scans through a branch of a compiled pattern to see whether it
1613 can match the empty string or not. It is called from could_be_empty()
1614 below and from compile_branch() when checking for an unlimited repeat of a
1615 group that can match nothing. Note that first_significant_code() skips over
1616 backward and negative forward assertions when its final argument is TRUE. If we
1617 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1618 bracket whose current branch will already have been scanned.
1619
1620 Arguments:
1621 code points to start of search
1622 endcode points to where to stop
1623 utf8 TRUE if in UTF8 mode
1624
1625 Returns: TRUE if what is matched could be empty
1626 */
1627
1628 static BOOL
1629 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1630 {
1631 register int c;
1632 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1633 code < endcode;
1634 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1635 {
1636 const uschar *ccode;
1637
1638 c = *code;
1639
1640 /* Skip over forward assertions; the other assertions are skipped by
1641 first_significant_code() with a TRUE final argument. */
1642
1643 if (c == OP_ASSERT)
1644 {
1645 do code += GET(code, 1); while (*code == OP_ALT);
1646 c = *code;
1647 continue;
1648 }
1649
1650 /* Groups with zero repeats can of course be empty; skip them. */
1651
1652 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1653 {
1654 code += _pcre_OP_lengths[c];
1655 do code += GET(code, 1); while (*code == OP_ALT);
1656 c = *code;
1657 continue;
1658 }
1659
1660 /* For other groups, scan the branches. */
1661
1662 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1663 {
1664 BOOL empty_branch;
1665 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1666
1667 /* Scan a closed bracket */
1668
1669 empty_branch = FALSE;
1670 do
1671 {
1672 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1673 empty_branch = TRUE;
1674 code += GET(code, 1);
1675 }
1676 while (*code == OP_ALT);
1677 if (!empty_branch) return FALSE; /* All branches are non-empty */
1678 c = *code;
1679 continue;
1680 }
1681
1682 /* Handle the other opcodes */
1683
1684 switch (c)
1685 {
1686 /* Check for quantifiers after a class. XCLASS is used for classes that
1687 cannot be represented just by a bit map. This includes negated single
1688 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1689 actual length is stored in the compiled code, so we must update "code"
1690 here. */
1691
1692 #ifdef SUPPORT_UTF8
1693 case OP_XCLASS:
1694 ccode = code += GET(code, 1);
1695 goto CHECK_CLASS_REPEAT;
1696 #endif
1697
1698 case OP_CLASS:
1699 case OP_NCLASS:
1700 ccode = code + 33;
1701
1702 #ifdef SUPPORT_UTF8
1703 CHECK_CLASS_REPEAT:
1704 #endif
1705
1706 switch (*ccode)
1707 {
1708 case OP_CRSTAR: /* These could be empty; continue */
1709 case OP_CRMINSTAR:
1710 case OP_CRQUERY:
1711 case OP_CRMINQUERY:
1712 break;
1713
1714 default: /* Non-repeat => class must match */
1715 case OP_CRPLUS: /* These repeats aren't empty */
1716 case OP_CRMINPLUS:
1717 return FALSE;
1718
1719 case OP_CRRANGE:
1720 case OP_CRMINRANGE:
1721 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1722 break;
1723 }
1724 break;
1725
1726 /* Opcodes that must match a character */
1727
1728 case OP_PROP:
1729 case OP_NOTPROP:
1730 case OP_EXTUNI:
1731 case OP_NOT_DIGIT:
1732 case OP_DIGIT:
1733 case OP_NOT_WHITESPACE:
1734 case OP_WHITESPACE:
1735 case OP_NOT_WORDCHAR:
1736 case OP_WORDCHAR:
1737 case OP_ANY:
1738 case OP_ALLANY:
1739 case OP_ANYBYTE:
1740 case OP_CHAR:
1741 case OP_CHARNC:
1742 case OP_NOT:
1743 case OP_PLUS:
1744 case OP_MINPLUS:
1745 case OP_POSPLUS:
1746 case OP_EXACT:
1747 case OP_NOTPLUS:
1748 case OP_NOTMINPLUS:
1749 case OP_NOTPOSPLUS:
1750 case OP_NOTEXACT:
1751 case OP_TYPEPLUS:
1752 case OP_TYPEMINPLUS:
1753 case OP_TYPEPOSPLUS:
1754 case OP_TYPEEXACT:
1755 return FALSE;
1756
1757 /* These are going to continue, as they may be empty, but we have to
1758 fudge the length for the \p and \P cases. */
1759
1760 case OP_TYPESTAR:
1761 case OP_TYPEMINSTAR:
1762 case OP_TYPEPOSSTAR:
1763 case OP_TYPEQUERY:
1764 case OP_TYPEMINQUERY:
1765 case OP_TYPEPOSQUERY:
1766 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1767 break;
1768
1769 /* Same for these */
1770
1771 case OP_TYPEUPTO:
1772 case OP_TYPEMINUPTO:
1773 case OP_TYPEPOSUPTO:
1774 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1775 break;
1776
1777 /* End of branch */
1778
1779 case OP_KET:
1780 case OP_KETRMAX:
1781 case OP_KETRMIN:
1782 case OP_ALT:
1783 return TRUE;
1784
1785 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1786 MINUPTO, and POSUPTO may be followed by a multibyte character */
1787
1788 #ifdef SUPPORT_UTF8
1789 case OP_STAR:
1790 case OP_MINSTAR:
1791 case OP_POSSTAR:
1792 case OP_QUERY:
1793 case OP_MINQUERY:
1794 case OP_POSQUERY:
1795 case OP_UPTO:
1796 case OP_MINUPTO:
1797 case OP_POSUPTO:
1798 if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
1799 break;
1800 #endif
1801 }
1802 }
1803
1804 return TRUE;
1805 }
1806
1807
1808
1809 /*************************************************
1810 * Scan compiled regex for non-emptiness *
1811 *************************************************/
1812
1813 /* This function is called to check for left recursive calls. We want to check
1814 the current branch of the current pattern to see if it could match the empty
1815 string. If it could, we must look outwards for branches at other levels,
1816 stopping when we pass beyond the bracket which is the subject of the recursion.
1817
1818 Arguments:
1819 code points to start of the recursion
1820 endcode points to where to stop (current RECURSE item)
1821 bcptr points to the chain of current (unclosed) branch starts
1822 utf8 TRUE if in UTF-8 mode
1823
1824 Returns: TRUE if what is matched could be empty
1825 */
1826
1827 static BOOL
1828 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1829 BOOL utf8)
1830 {
1831 while (bcptr != NULL && bcptr->current >= code)
1832 {
1833 if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1834 bcptr = bcptr->outer;
1835 }
1836 return TRUE;
1837 }
1838
1839
1840
1841 /*************************************************
1842 * Check for POSIX class syntax *
1843 *************************************************/
1844
1845 /* This function is called when the sequence "[:" or "[." or "[=" is
1846 encountered in a character class. It checks whether this is followed by a
1847 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1848 reach an unescaped ']' without the special preceding character, return FALSE.
1849
1850 Originally, this function only recognized a sequence of letters between the
1851 terminators, but it seems that Perl recognizes any sequence of characters,
1852 though of course unknown POSIX names are subsequently rejected. Perl gives an
1853 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1854 didn't consider this to be a POSIX class. Likewise for [:1234:].
1855
1856 The problem in trying to be exactly like Perl is in the handling of escapes. We
1857 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1858 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1859 below handles the special case of \], but does not try to do any other escape
1860 processing. This makes it different from Perl for cases such as [:l\ower:]
1861 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1862 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1863 I think.
1864
1865 Arguments:
1866 ptr pointer to the initial [
1867 endptr where to return the end pointer
1868
1869 Returns: TRUE or FALSE
1870 */
1871
1872 static BOOL
1873 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1874 {
1875 int terminator; /* Don't combine these lines; the Solaris cc */
1876 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1877 for (++ptr; *ptr != 0; ptr++)
1878 {
1879 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
1880 {
1881 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
1882 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
1883 {
1884 *endptr = ptr;
1885 return TRUE;
1886 }
1887 }
1888 }
1889 return FALSE;
1890 }
1891
1892
1893
1894
1895 /*************************************************
1896 * Check POSIX class name *
1897 *************************************************/
1898
1899 /* This function is called to check the name given in a POSIX-style class entry
1900 such as [:alnum:].
1901
1902 Arguments:
1903 ptr points to the first letter
1904 len the length of the name
1905
1906 Returns: a value representing the name, or -1 if unknown
1907 */
1908
1909 static int
1910 check_posix_name(const uschar *ptr, int len)
1911 {
1912 const char *pn = posix_names;
1913 register int yield = 0;
1914 while (posix_name_lengths[yield] != 0)
1915 {
1916 if (len == posix_name_lengths[yield] &&
1917 strncmp((const char *)ptr, pn, len) == 0) return yield;
1918 pn += posix_name_lengths[yield] + 1;
1919 yield++;
1920 }
1921 return -1;
1922 }
1923
1924
1925 /*************************************************
1926 * Adjust OP_RECURSE items in repeated group *
1927 *************************************************/
1928
1929 /* OP_RECURSE items contain an offset from the start of the regex to the group
1930 that is referenced. This means that groups can be replicated for fixed
1931 repetition simply by copying (because the recursion is allowed to refer to
1932 earlier groups that are outside the current group). However, when a group is
1933 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
1934 inserted before it, after it has been compiled. This means that any OP_RECURSE
1935 items within it that refer to the group itself or any contained groups have to
1936 have their offsets adjusted. That one of the jobs of this function. Before it
1937 is called, the partially compiled regex must be temporarily terminated with
1938 OP_END.
1939
1940 This function has been extended with the possibility of forward references for
1941 recursions and subroutine calls. It must also check the list of such references
1942 for the group we are dealing with. If it finds that one of the recursions in
1943 the current group is on this list, it adjusts the offset in the list, not the
1944 value in the reference (which is a group number).
1945
1946 Arguments:
1947 group points to the start of the group
1948 adjust the amount by which the group is to be moved
1949 utf8 TRUE in UTF-8 mode
1950 cd contains pointers to tables etc.
1951 save_hwm the hwm forward reference pointer at the start of the group
1952
1953 Returns: nothing
1954 */
1955
1956 static void
1957 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
1958 uschar *save_hwm)
1959 {
1960 uschar *ptr = group;
1961
1962 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
1963 {
1964 int offset;
1965 uschar *hc;
1966
1967 /* See if this recursion is on the forward reference list. If so, adjust the
1968 reference. */
1969
1970 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
1971 {
1972 offset = GET(hc, 0);
1973 if (cd->start_code + offset == ptr + 1)
1974 {
1975 PUT(hc, 0, offset + adjust);
1976 break;
1977 }
1978 }
1979
1980 /* Otherwise, adjust the recursion offset if it's after the start of this
1981 group. */
1982
1983 if (hc >= cd->hwm)
1984 {
1985 offset = GET(ptr, 1);
1986 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
1987 }
1988
1989 ptr += 1 + LINK_SIZE;
1990 }
1991 }
1992
1993
1994
1995 /*************************************************
1996 * Insert an automatic callout point *
1997 *************************************************/
1998
1999 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2000 callout points before each pattern item.
2001
2002 Arguments:
2003 code current code pointer
2004 ptr current pattern pointer
2005 cd pointers to tables etc
2006
2007 Returns: new code pointer
2008 */
2009
2010 static uschar *
2011 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2012 {
2013 *code++ = OP_CALLOUT;
2014 *code++ = 255;
2015 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2016 PUT(code, LINK_SIZE, 0); /* Default length */
2017 return code + 2*LINK_SIZE;
2018 }
2019
2020
2021
2022 /*************************************************
2023 * Complete a callout item *
2024 *************************************************/
2025
2026 /* A callout item contains the length of the next item in the pattern, which
2027 we can't fill in till after we have reached the relevant point. This is used
2028 for both automatic and manual callouts.
2029
2030 Arguments:
2031 previous_callout points to previous callout item
2032 ptr current pattern pointer
2033 cd pointers to tables etc
2034
2035 Returns: nothing
2036 */
2037
2038 static void
2039 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2040 {
2041 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2042 PUT(previous_callout, 2 + LINK_SIZE, length);
2043 }
2044
2045
2046
2047 #ifdef SUPPORT_UCP
2048 /*************************************************
2049 * Get othercase range *
2050 *************************************************/
2051
2052 /* This function is passed the start and end of a class range, in UTF-8 mode
2053 with UCP support. It searches up the characters, looking for internal ranges of
2054 characters in the "other" case. Each call returns the next one, updating the
2055 start address.
2056
2057 Arguments:
2058 cptr points to starting character value; updated
2059 d end value
2060 ocptr where to put start of othercase range
2061 odptr where to put end of othercase range
2062
2063 Yield: TRUE when range returned; FALSE when no more
2064 */
2065
2066 static BOOL
2067 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2068 unsigned int *odptr)
2069 {
2070 unsigned int c, othercase, next;
2071
2072 for (c = *cptr; c <= d; c++)
2073 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2074
2075 if (c > d) return FALSE;
2076
2077 *ocptr = othercase;
2078 next = othercase + 1;
2079
2080 for (++c; c <= d; c++)
2081 {
2082 if (UCD_OTHERCASE(c) != next) break;
2083 next++;
2084 }
2085
2086 *odptr = next - 1;
2087 *cptr = c;
2088
2089 return TRUE;
2090 }
2091 #endif /* SUPPORT_UCP */
2092
2093
2094
2095 /*************************************************
2096 * Check if auto-possessifying is possible *
2097 *************************************************/
2098
2099 /* This function is called for unlimited repeats of certain items, to see
2100 whether the next thing could possibly match the repeated item. If not, it makes
2101 sense to automatically possessify the repeated item.
2102
2103 Arguments:
2104 op_code the repeated op code
2105 this data for this item, depends on the opcode
2106 utf8 TRUE in UTF-8 mode
2107 utf8_char used for utf8 character bytes, NULL if not relevant
2108 ptr next character in pattern
2109 options options bits
2110 cd contains pointers to tables etc.
2111
2112 Returns: TRUE if possessifying is wanted
2113 */
2114
2115 static BOOL
2116 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2117 const uschar *ptr, int options, compile_data *cd)
2118 {
2119 int next;
2120
2121 /* Skip whitespace and comments in extended mode */
2122
2123 if ((options & PCRE_EXTENDED) != 0)
2124 {
2125 for (;;)
2126 {
2127 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2128 if (*ptr == CHAR_NUMBER_SIGN)
2129 {
2130 while (*(++ptr) != 0)
2131 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2132 }
2133 else break;
2134 }
2135 }
2136
2137 /* If the next item is one that we can handle, get its value. A non-negative
2138 value is a character, a negative value is an escape value. */
2139
2140 if (*ptr == CHAR_BACKSLASH)
2141 {
2142 int temperrorcode = 0;
2143 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2144 if (temperrorcode != 0) return FALSE;
2145 ptr++; /* Point after the escape sequence */
2146 }
2147
2148 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2149 {
2150 #ifdef SUPPORT_UTF8
2151 if (utf8) { GETCHARINC(next, ptr); } else
2152 #endif
2153 next = *ptr++;
2154 }
2155
2156 else return FALSE;
2157
2158 /* Skip whitespace and comments in extended mode */
2159
2160 if ((options & PCRE_EXTENDED) != 0)
2161 {
2162 for (;;)
2163 {
2164 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2165 if (*ptr == CHAR_NUMBER_SIGN)
2166 {
2167 while (*(++ptr) != 0)
2168 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2169 }
2170 else break;
2171 }
2172 }
2173
2174 /* If the next thing is itself optional, we have to give up. */
2175
2176 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2177 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2178 return FALSE;
2179
2180 /* Now compare the next item with the previous opcode. If the previous is a
2181 positive single character match, "item" either contains the character or, if
2182 "item" is greater than 127 in utf8 mode, the character's bytes are in
2183 utf8_char. */
2184
2185
2186 /* Handle cases when the next item is a character. */
2187
2188 if (next >= 0) switch(op_code)
2189 {
2190 case OP_CHAR:
2191 #ifdef SUPPORT_UTF8
2192 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2193 #else
2194 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2195 #endif
2196 return item != next;
2197
2198 /* For CHARNC (caseless character) we must check the other case. If we have
2199 Unicode property support, we can use it to test the other case of
2200 high-valued characters. */
2201
2202 case OP_CHARNC:
2203 #ifdef SUPPORT_UTF8
2204 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2205 #endif
2206 if (item == next) return FALSE;
2207 #ifdef SUPPORT_UTF8
2208 if (utf8)
2209 {
2210 unsigned int othercase;
2211 if (next < 128) othercase = cd->fcc[next]; else
2212 #ifdef SUPPORT_UCP
2213 othercase = UCD_OTHERCASE((unsigned int)next);
2214 #else
2215 othercase = NOTACHAR;
2216 #endif
2217 return (unsigned int)item != othercase;
2218 }
2219 else
2220 #endif /* SUPPORT_UTF8 */
2221 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2222
2223 /* For OP_NOT, "item" must be a single-byte character. */
2224
2225 case OP_NOT:
2226 if (item == next) return TRUE;
2227 if ((options & PCRE_CASELESS) == 0) return FALSE;
2228 #ifdef SUPPORT_UTF8
2229 if (utf8)
2230 {
2231 unsigned int othercase;
2232 if (next < 128) othercase = cd->fcc[next]; else
2233 #ifdef SUPPORT_UCP
2234 othercase = UCD_OTHERCASE(next);
2235 #else
2236 othercase = NOTACHAR;
2237 #endif
2238 return (unsigned int)item == othercase;
2239 }
2240 else
2241 #endif /* SUPPORT_UTF8 */
2242 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2243
2244 case OP_DIGIT:
2245 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2246
2247 case OP_NOT_DIGIT:
2248 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2249
2250 case OP_WHITESPACE:
2251 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2252
2253 case OP_NOT_WHITESPACE:
2254 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2255
2256 case OP_WORDCHAR:
2257 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2258
2259 case OP_NOT_WORDCHAR:
2260 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2261
2262 case OP_HSPACE:
2263 case OP_NOT_HSPACE:
2264 switch(next)
2265 {
2266 case 0x09:
2267 case 0x20:
2268 case 0xa0:
2269 case 0x1680:
2270 case 0x180e:
2271 case 0x2000:
2272 case 0x2001:
2273 case 0x2002:
2274 case 0x2003:
2275 case 0x2004:
2276 case 0x2005:
2277 case 0x2006:
2278 case 0x2007:
2279 case 0x2008:
2280 case 0x2009:
2281 case 0x200A:
2282 case 0x202f:
2283 case 0x205f:
2284 case 0x3000:
2285 return op_code != OP_HSPACE;
2286 default:
2287 return op_code == OP_HSPACE;
2288 }
2289
2290 case OP_VSPACE:
2291 case OP_NOT_VSPACE:
2292 switch(next)
2293 {
2294 case 0x0a:
2295 case 0x0b:
2296 case 0x0c:
2297 case 0x0d:
2298 case 0x85:
2299 case 0x2028:
2300 case 0x2029:
2301 return op_code != OP_VSPACE;
2302 default:
2303 return op_code == OP_VSPACE;
2304 }
2305
2306 default:
2307 return FALSE;
2308 }
2309
2310
2311 /* Handle the case when the next item is \d, \s, etc. */
2312
2313 switch(op_code)
2314 {
2315 case OP_CHAR:
2316 case OP_CHARNC:
2317 #ifdef SUPPORT_UTF8
2318 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2319 #endif
2320 switch(-next)
2321 {
2322 case ESC_d:
2323 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2324
2325 case ESC_D:
2326 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2327
2328 case ESC_s:
2329 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2330
2331 case ESC_S:
2332 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2333
2334 case ESC_w:
2335 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2336
2337 case ESC_W:
2338 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2339
2340 case ESC_h:
2341 case ESC_H:
2342 switch(item)
2343 {
2344 case 0x09:
2345 case 0x20:
2346 case 0xa0:
2347 case 0x1680:
2348 case 0x180e:
2349 case 0x2000:
2350 case 0x2001:
2351 case 0x2002:
2352 case 0x2003:
2353 case 0x2004:
2354 case 0x2005:
2355 case 0x2006:
2356 case 0x2007:
2357 case 0x2008:
2358 case 0x2009:
2359 case 0x200A:
2360 case 0x202f:
2361 case 0x205f:
2362 case 0x3000:
2363 return -next != ESC_h;
2364 default:
2365 return -next == ESC_h;
2366 }
2367
2368 case ESC_v:
2369 case ESC_V:
2370 switch(item)
2371 {
2372 case 0x0a:
2373 case 0x0b:
2374 case 0x0c:
2375 case 0x0d:
2376 case 0x85:
2377 case 0x2028:
2378 case 0x2029:
2379 return -next != ESC_v;
2380 default:
2381 return -next == ESC_v;
2382 }
2383
2384 default:
2385 return FALSE;
2386 }
2387
2388 case OP_DIGIT:
2389 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2390 next == -ESC_h || next == -ESC_v;
2391
2392 case OP_NOT_DIGIT:
2393 return next == -ESC_d;
2394
2395 case OP_WHITESPACE:
2396 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2397
2398 case OP_NOT_WHITESPACE:
2399 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2400
2401 case OP_HSPACE:
2402 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2403
2404 case OP_NOT_HSPACE:
2405 return next == -ESC_h;
2406
2407 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2408 case OP_VSPACE:
2409 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2410
2411 case OP_NOT_VSPACE:
2412 return next == -ESC_v;
2413
2414 case OP_WORDCHAR:
2415 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2416
2417 case OP_NOT_WORDCHAR:
2418 return next == -ESC_w || next == -ESC_d;
2419
2420 default:
2421 return FALSE;
2422 }
2423
2424 /* Control does not reach here */
2425 }
2426
2427
2428
2429 /*************************************************
2430 * Compile one branch *
2431 *************************************************/
2432
2433 /* Scan the pattern, compiling it into the a vector. If the options are
2434 changed during the branch, the pointer is used to change the external options
2435 bits. This function is used during the pre-compile phase when we are trying
2436 to find out the amount of memory needed, as well as during the real compile
2437 phase. The value of lengthptr distinguishes the two phases.
2438
2439 Arguments:
2440 optionsptr pointer to the option bits
2441 codeptr points to the pointer to the current code point
2442 ptrptr points to the current pattern pointer
2443 errorcodeptr points to error code variable
2444 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2445 reqbyteptr set to the last literal character required, else < 0
2446 bcptr points to current branch chain
2447 cd contains pointers to tables etc.
2448 lengthptr NULL during the real compile phase
2449 points to length accumulator during pre-compile phase
2450
2451 Returns: TRUE on success
2452 FALSE, with *errorcodeptr set non-zero on error
2453 */
2454
2455 static BOOL
2456 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2457 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2458 compile_data *cd, int *lengthptr)
2459 {
2460 int repeat_type, op_type;
2461 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2462 int bravalue = 0;
2463 int greedy_default, greedy_non_default;
2464 int firstbyte, reqbyte;
2465 int zeroreqbyte, zerofirstbyte;
2466 int req_caseopt, reqvary, tempreqvary;
2467 int options = *optionsptr;
2468 int after_manual_callout = 0;
2469 int length_prevgroup = 0;
2470 register int c;
2471 register uschar *code = *codeptr;
2472 uschar *last_code = code;
2473 uschar *orig_code = code;
2474 uschar *tempcode;
2475 BOOL inescq = FALSE;
2476 BOOL groupsetfirstbyte = FALSE;
2477 const uschar *ptr = *ptrptr;
2478 const uschar *tempptr;
2479 uschar *previous = NULL;
2480 uschar *previous_callout = NULL;
2481 uschar *save_hwm = NULL;
2482 uschar classbits[32];
2483
2484 #ifdef SUPPORT_UTF8
2485 BOOL class_utf8;
2486 BOOL utf8 = (options & PCRE_UTF8) != 0;
2487 uschar *class_utf8data;
2488 uschar *class_utf8data_base;
2489 uschar utf8_char[6];
2490 #else
2491 BOOL utf8 = FALSE;
2492 uschar *utf8_char = NULL;
2493 #endif
2494
2495 #ifdef DEBUG
2496 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2497 #endif
2498
2499 /* Set up the default and non-default settings for greediness */
2500
2501 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2502 greedy_non_default = greedy_default ^ 1;
2503
2504 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2505 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2506 matches a non-fixed char first char; reqbyte just remains unset if we never
2507 find one.
2508
2509 When we hit a repeat whose minimum is zero, we may have to adjust these values
2510 to take the zero repeat into account. This is implemented by setting them to
2511 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2512 item types that can be repeated set these backoff variables appropriately. */
2513
2514 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2515
2516 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2517 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2518 value > 255. It is added into the firstbyte or reqbyte variables to record the
2519 case status of the value. This is used only for ASCII characters. */
2520
2521 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2522
2523 /* Switch on next character until the end of the branch */
2524
2525 for (;; ptr++)
2526 {
2527 BOOL negate_class;
2528 BOOL should_flip_negation;
2529 BOOL possessive_quantifier;
2530 BOOL is_quantifier;
2531 BOOL is_recurse;
2532 BOOL reset_bracount;
2533 int class_charcount;
2534 int class_lastchar;
2535 int newoptions;
2536 int recno;
2537 int refsign;
2538 int skipbytes;
2539 int subreqbyte;
2540 int subfirstbyte;
2541 int terminator;
2542 int mclength;
2543 uschar mcbuffer[8];
2544
2545 /* Get next byte in the pattern */
2546
2547 c = *ptr;
2548
2549 /* If we are in the pre-compile phase, accumulate the length used for the
2550 previous cycle of this loop. */
2551
2552 if (lengthptr != NULL)
2553 {
2554 #ifdef DEBUG
2555 if (code > cd->hwm) cd->hwm = code; /* High water info */
2556 #endif
2557 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2558 {
2559 *errorcodeptr = ERR52;
2560 goto FAILED;
2561 }
2562
2563 /* There is at least one situation where code goes backwards: this is the
2564 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2565 the class is simply eliminated. However, it is created first, so we have to
2566 allow memory for it. Therefore, don't ever reduce the length at this point.
2567 */
2568
2569 if (code < last_code) code = last_code;
2570
2571 /* Paranoid check for integer overflow */
2572
2573 if (OFLOW_MAX - *lengthptr < code - last_code)
2574 {
2575 *errorcodeptr = ERR20;
2576 goto FAILED;
2577 }
2578
2579 *lengthptr += code - last_code;
2580 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2581
2582 /* If "previous" is set and it is not at the start of the work space, move
2583 it back to there, in order to avoid filling up the work space. Otherwise,
2584 if "previous" is NULL, reset the current code pointer to the start. */
2585
2586 if (previous != NULL)
2587 {
2588 if (previous > orig_code)
2589 {
2590 memmove(orig_code, previous, code - previous);
2591 code -= previous - orig_code;
2592 previous = orig_code;
2593 }
2594 }
2595 else code = orig_code;
2596
2597 /* Remember where this code item starts so we can pick up the length
2598 next time round. */
2599
2600 last_code = code;
2601 }
2602
2603 /* In the real compile phase, just check the workspace used by the forward
2604 reference list. */
2605
2606 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2607 {
2608 *errorcodeptr = ERR52;
2609 goto FAILED;
2610 }
2611
2612 /* If in \Q...\E, check for the end; if not, we have a literal */
2613
2614 if (inescq && c != 0)
2615 {
2616 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2617 {
2618 inescq = FALSE;
2619 ptr++;
2620 continue;
2621 }
2622 else
2623 {
2624 if (previous_callout != NULL)
2625 {
2626 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2627 complete_callout(previous_callout, ptr, cd);
2628 previous_callout = NULL;
2629 }
2630 if ((options & PCRE_AUTO_CALLOUT) != 0)
2631 {
2632 previous_callout = code;
2633 code = auto_callout(code, ptr, cd);
2634 }
2635 goto NORMAL_CHAR;
2636 }
2637 }
2638
2639 /* Fill in length of a previous callout, except when the next thing is
2640 a quantifier. */
2641
2642 is_quantifier =
2643 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2644 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2645
2646 if (!is_quantifier && previous_callout != NULL &&
2647 after_manual_callout-- <= 0)
2648 {
2649 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2650 complete_callout(previous_callout, ptr, cd);
2651 previous_callout = NULL;
2652 }
2653
2654 /* In extended mode, skip white space and comments */
2655
2656 if ((options & PCRE_EXTENDED) != 0)
2657 {
2658 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2659 if (c == CHAR_NUMBER_SIGN)
2660 {
2661 while (*(++ptr) != 0)
2662 {
2663 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2664 }
2665 if (*ptr != 0) continue;
2666
2667 /* Else fall through to handle end of string */
2668 c = 0;
2669 }
2670 }
2671
2672 /* No auto callout for quantifiers. */
2673
2674 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2675 {
2676 previous_callout = code;
2677 code = auto_callout(code, ptr, cd);
2678 }
2679
2680 switch(c)
2681 {
2682 /* ===================================================================*/
2683 case 0: /* The branch terminates at string end */
2684 case CHAR_VERTICAL_LINE: /* or | or ) */
2685 case CHAR_RIGHT_PARENTHESIS:
2686 *firstbyteptr = firstbyte;
2687 *reqbyteptr = reqbyte;
2688 *codeptr = code;
2689 *ptrptr = ptr;
2690 if (lengthptr != NULL)
2691 {
2692 if (OFLOW_MAX - *lengthptr < code - last_code)
2693 {
2694 *errorcodeptr = ERR20;
2695 goto FAILED;
2696 }
2697 *lengthptr += code - last_code; /* To include callout length */
2698 DPRINTF((">> end branch\n"));
2699 }
2700 return TRUE;
2701
2702
2703 /* ===================================================================*/
2704 /* Handle single-character metacharacters. In multiline mode, ^ disables
2705 the setting of any following char as a first character. */
2706
2707 case CHAR_CIRCUMFLEX_ACCENT:
2708 if ((options & PCRE_MULTILINE) != 0)
2709 {
2710 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2711 }
2712 previous = NULL;
2713 *code++ = OP_CIRC;
2714 break;
2715
2716 case CHAR_DOLLAR_SIGN:
2717 previous = NULL;
2718 *code++ = OP_DOLL;
2719 break;
2720
2721 /* There can never be a first char if '.' is first, whatever happens about
2722 repeats. The value of reqbyte doesn't change either. */
2723
2724 case CHAR_DOT:
2725 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2726 zerofirstbyte = firstbyte;
2727 zeroreqbyte = reqbyte;
2728 previous = code;
2729 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2730 break;
2731
2732
2733 /* ===================================================================*/
2734 /* Character classes. If the included characters are all < 256, we build a
2735 32-byte bitmap of the permitted characters, except in the special case
2736 where there is only one such character. For negated classes, we build the
2737 map as usual, then invert it at the end. However, we use a different opcode
2738 so that data characters > 255 can be handled correctly.
2739
2740 If the class contains characters outside the 0-255 range, a different
2741 opcode is compiled. It may optionally have a bit map for characters < 256,
2742 but those above are are explicitly listed afterwards. A flag byte tells
2743 whether the bitmap is present, and whether this is a negated class or not.
2744
2745 In JavaScript compatibility mode, an isolated ']' causes an error. In
2746 default (Perl) mode, it is treated as a data character. */
2747
2748 case CHAR_RIGHT_SQUARE_BRACKET:
2749 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2750 {
2751 *errorcodeptr = ERR64;
2752 goto FAILED;
2753 }
2754 goto NORMAL_CHAR;
2755
2756 case CHAR_LEFT_SQUARE_BRACKET:
2757 previous = code;
2758
2759 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2760 they are encountered at the top level, so we'll do that too. */
2761
2762 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2763 ptr[1] == CHAR_EQUALS_SIGN) &&
2764 check_posix_syntax(ptr, &tempptr))
2765 {
2766 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2767 goto FAILED;
2768 }
2769
2770 /* If the first character is '^', set the negation flag and skip it. Also,
2771 if the first few characters (either before or after ^) are \Q\E or \E we
2772 skip them too. This makes for compatibility with Perl. */
2773
2774 negate_class = FALSE;
2775 for (;;)
2776 {
2777 c = *(++ptr);
2778 if (c == CHAR_BACKSLASH)
2779 {
2780 if (ptr[1] == CHAR_E)
2781 ptr++;
2782 else if (strncmp((const char *)ptr+1,
2783 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2784 ptr += 3;
2785 else
2786 break;
2787 }
2788 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2789 negate_class = TRUE;
2790 else break;
2791 }
2792
2793 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2794 an initial ']' is taken as a data character -- the code below handles
2795 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2796 [^] must match any character, so generate OP_ALLANY. */
2797
2798 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2799 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2800 {
2801 *code++ = negate_class? OP_ALLANY : OP_FAIL;
2802 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2803 zerofirstbyte = firstbyte;
2804 break;
2805 }
2806
2807 /* If a class contains a negative special such as \S, we need to flip the
2808 negation flag at the end, so that support for characters > 255 works
2809 correctly (they are all included in the class). */
2810
2811 should_flip_negation = FALSE;
2812
2813 /* Keep a count of chars with values < 256 so that we can optimize the case
2814 of just a single character (as long as it's < 256). However, For higher
2815 valued UTF-8 characters, we don't yet do any optimization. */
2816
2817 class_charcount = 0;
2818 class_lastchar = -1;
2819
2820 /* Initialize the 32-char bit map to all zeros. We build the map in a
2821 temporary bit of memory, in case the class contains only 1 character (less
2822 than 256), because in that case the compiled code doesn't use the bit map.
2823 */
2824
2825 memset(classbits, 0, 32 * sizeof(uschar));
2826
2827 #ifdef SUPPORT_UTF8
2828 class_utf8 = FALSE; /* No chars >= 256 */
2829 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2830 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2831 #endif
2832
2833 /* Process characters until ] is reached. By writing this as a "do" it
2834 means that an initial ] is taken as a data character. At the start of the
2835 loop, c contains the first byte of the character. */
2836
2837 if (c != 0) do
2838 {
2839 const uschar *oldptr;
2840
2841 #ifdef SUPPORT_UTF8
2842 if (utf8 && c > 127)
2843 { /* Braces are required because the */
2844 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2845 }
2846
2847 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2848 data and reset the pointer. This is so that very large classes that
2849 contain a zillion UTF-8 characters no longer overwrite the work space
2850 (which is on the stack). */
2851
2852 if (lengthptr != NULL)
2853 {
2854 *lengthptr += class_utf8data - class_utf8data_base;
2855 class_utf8data = class_utf8data_base;
2856 }
2857
2858 #endif
2859
2860 /* Inside \Q...\E everything is literal except \E */
2861
2862 if (inescq)
2863 {
2864 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
2865 {
2866 inescq = FALSE; /* Reset literal state */
2867 ptr++; /* Skip the 'E' */
2868 continue; /* Carry on with next */
2869 }
2870 goto CHECK_RANGE; /* Could be range if \E follows */
2871 }
2872
2873 /* Handle POSIX class names. Perl allows a negation extension of the
2874 form [:^name:]. A square bracket that doesn't match the syntax is
2875 treated as a literal. We also recognize the POSIX constructions
2876 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2877 5.6 and 5.8 do. */
2878
2879 if (c == CHAR_LEFT_SQUARE_BRACKET &&
2880 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2881 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
2882 {
2883 BOOL local_negate = FALSE;
2884 int posix_class, taboffset, tabopt;
2885 register const uschar *cbits = cd->cbits;
2886 uschar pbits[32];
2887
2888 if (ptr[1] != CHAR_COLON)
2889 {
2890 *errorcodeptr = ERR31;
2891 goto FAILED;
2892 }
2893
2894 ptr += 2;
2895 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
2896 {
2897 local_negate = TRUE;
2898 should_flip_negation = TRUE; /* Note negative special */
2899 ptr++;
2900 }
2901
2902 posix_class = check_posix_name(ptr, tempptr - ptr);
2903 if (posix_class < 0)
2904 {
2905 *errorcodeptr = ERR30;
2906 goto FAILED;
2907 }
2908
2909 /* If matching is caseless, upper and lower are converted to
2910 alpha. This relies on the fact that the class table starts with
2911 alpha, lower, upper as the first 3 entries. */
2912
2913 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
2914 posix_class = 0;
2915
2916 /* We build the bit map for the POSIX class in a chunk of local store
2917 because we may be adding and subtracting from it, and we don't want to
2918 subtract bits that may be in the main map already. At the end we or the
2919 result into the bit map that is being built. */
2920
2921 posix_class *= 3;
2922
2923 /* Copy in the first table (always present) */
2924
2925 memcpy(pbits, cbits + posix_class_maps[posix_class],
2926 32 * sizeof(uschar));
2927
2928 /* If there is a second table, add or remove it as required. */
2929
2930 taboffset = posix_class_maps[posix_class + 1];
2931 tabopt = posix_class_maps[posix_class + 2];
2932
2933 if (taboffset >= 0)
2934 {
2935 if (tabopt >= 0)
2936 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
2937 else
2938 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
2939 }
2940
2941 /* Not see if we need to remove any special characters. An option
2942 value of 1 removes vertical space and 2 removes underscore. */
2943
2944 if (tabopt < 0) tabopt = -tabopt;
2945 if (tabopt == 1) pbits[1] &= ~0x3c;
2946 else if (tabopt == 2) pbits[11] &= 0x7f;
2947
2948 /* Add the POSIX table or its complement into the main table that is
2949 being built and we are done. */
2950
2951 if (local_negate)
2952 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
2953 else
2954 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
2955
2956 ptr = tempptr + 1;
2957 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
2958 continue; /* End of POSIX syntax handling */
2959 }
2960
2961 /* Backslash may introduce a single character, or it may introduce one
2962 of the specials, which just set a flag. The sequence \b is a special
2963 case. Inside a class (and only there) it is treated as backspace.
2964 Elsewhere it marks a word boundary. Other escapes have preset maps ready
2965 to 'or' into the one we are building. We assume they have more than one
2966 character in them, so set class_charcount bigger than one. */
2967
2968 if (c == CHAR_BACKSLASH)
2969 {
2970 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
2971 if (*errorcodeptr != 0) goto FAILED;
2972
2973 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
2974 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
2975 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
2976 else if (-c == ESC_Q) /* Handle start of quoted string */
2977 {
2978 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
2979 {
2980 ptr += 2; /* avoid empty string */
2981 }
2982 else inescq = TRUE;
2983 continue;
2984 }
2985 else if (-c == ESC_E) continue; /* Ignore orphan \E */
2986
2987 if (c < 0)
2988 {
2989 register const uschar *cbits = cd->cbits;
2990 class_charcount += 2; /* Greater than 1 is what matters */
2991
2992 /* Save time by not doing this in the pre-compile phase. */
2993
2994 if (lengthptr == NULL) switch (-c)
2995 {
2996 case ESC_d:
2997 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
2998 continue;
2999
3000 case ESC_D:
3001 should_flip_negation = TRUE;
3002 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3003 continue;
3004
3005 case ESC_w:
3006 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3007 continue;
3008
3009 case ESC_W:
3010 should_flip_negation = TRUE;
3011 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3012 continue;
3013
3014 case ESC_s:
3015 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3016 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3017 continue;
3018
3019 case ESC_S:
3020 should_flip_negation = TRUE;
3021 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3022 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3023 continue;
3024
3025 default: /* Not recognized; fall through */
3026 break; /* Need "default" setting to stop compiler warning. */
3027 }
3028
3029 /* In the pre-compile phase, just do the recognition. */
3030
3031 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3032 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3033
3034 /* We need to deal with \H, \h, \V, and \v in both phases because
3035 they use extra memory. */
3036
3037 if (-c == ESC_h)
3038 {
3039 SETBIT(classbits, 0x09); /* VT */
3040 SETBIT(classbits, 0x20); /* SPACE */
3041 SETBIT(classbits, 0xa0); /* NSBP */
3042 #ifdef SUPPORT_UTF8
3043 if (utf8)
3044 {
3045 class_utf8 = TRUE;
3046 *class_utf8data++ = XCL_SINGLE;
3047 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3048 *class_utf8data++ = XCL_SINGLE;
3049 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3050 *class_utf8data++ = XCL_RANGE;
3051 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3052 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3053 *class_utf8data++ = XCL_SINGLE;
3054 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3055 *class_utf8data++ = XCL_SINGLE;
3056 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3057 *class_utf8data++ = XCL_SINGLE;
3058 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3059 }
3060 #endif
3061 continue;
3062 }
3063
3064 if (-c == ESC_H)
3065 {
3066 for (c = 0; c < 32; c++)
3067 {
3068 int x = 0xff;
3069 switch (c)
3070 {
3071 case 0x09/8: x ^= 1 << (0x09%8); break;
3072 case 0x20/8: x ^= 1 << (0x20%8); break;
3073 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3074 default: break;
3075 }
3076 classbits[c] |= x;
3077 }
3078
3079 #ifdef SUPPORT_UTF8
3080 if (utf8)
3081 {
3082 class_utf8 = TRUE;
3083 *class_utf8data++ = XCL_RANGE;
3084 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3085 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3086 *class_utf8data++ = XCL_RANGE;
3087 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3088 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3089 *class_utf8data++ = XCL_RANGE;
3090 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3091 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3092 *class_utf8data++ = XCL_RANGE;
3093 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3094 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3095 *class_utf8data++ = XCL_RANGE;
3096 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3097 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3098 *class_utf8data++ = XCL_RANGE;
3099 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3100 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3101 *class_utf8data++ = XCL_RANGE;
3102 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3103 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3104 }
3105 #endif
3106 continue;
3107 }
3108
3109 if (-c == ESC_v)
3110 {
3111 SETBIT(classbits, 0x0a); /* LF */
3112 SETBIT(classbits, 0x0b); /* VT */
3113 SETBIT(classbits, 0x0c); /* FF */
3114 SETBIT(classbits, 0x0d); /* CR */
3115 SETBIT(classbits, 0x85); /* NEL */
3116 #ifdef SUPPORT_UTF8
3117 if (utf8)
3118 {
3119 class_utf8 = TRUE;
3120 *class_utf8data++ = XCL_RANGE;
3121 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3122 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3123 }
3124 #endif
3125 continue;
3126 }
3127
3128 if (-c == ESC_V)
3129 {
3130 for (c = 0; c < 32; c++)
3131 {
3132 int x = 0xff;
3133 switch (c)
3134 {
3135 case 0x0a/8: x ^= 1 << (0x0a%8);
3136 x ^= 1 << (0x0b%8);
3137 x ^= 1 << (0x0c%8);
3138 x ^= 1 << (0x0d%8);
3139 break;
3140 case 0x85/8: x ^= 1 << (0x85%8); break;
3141 default: break;
3142 }
3143 classbits[c] |= x;
3144 }
3145
3146 #ifdef SUPPORT_UTF8
3147 if (utf8)
3148 {
3149 class_utf8 = TRUE;
3150 *class_utf8data++ = XCL_RANGE;
3151 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3152 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3153 *class_utf8data++ = XCL_RANGE;
3154 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3155 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3156 }
3157 #endif
3158 continue;
3159 }
3160
3161 /* We need to deal with \P and \p in both phases. */
3162
3163 #ifdef SUPPORT_UCP
3164 if (-c == ESC_p || -c == ESC_P)
3165 {
3166 BOOL negated;
3167 int pdata;
3168 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3169 if (ptype < 0) goto FAILED;
3170 class_utf8 = TRUE;
3171 *class_utf8data++ = ((-c == ESC_p) != negated)?
3172 XCL_PROP : XCL_NOTPROP;
3173 *class_utf8data++ = ptype;
3174 *class_utf8data++ = pdata;
3175 class_charcount -= 2; /* Not a < 256 character */
3176 continue;
3177 }
3178 #endif
3179 /* Unrecognized escapes are faulted if PCRE is running in its
3180 strict mode. By default, for compatibility with Perl, they are
3181 treated as literals. */
3182
3183 if ((options & PCRE_EXTRA) != 0)
3184 {
3185 *errorcodeptr = ERR7;
3186 goto FAILED;
3187 }
3188
3189 class_charcount -= 2; /* Undo the default count from above */
3190 c = *ptr; /* Get the final character and fall through */
3191 }
3192
3193 /* Fall through if we have a single character (c >= 0). This may be
3194 greater than 256 in UTF-8 mode. */
3195
3196 } /* End of backslash handling */
3197
3198 /* A single character may be followed by '-' to form a range. However,
3199 Perl does not permit ']' to be the end of the range. A '-' character
3200 at the end is treated as a literal. Perl ignores orphaned \E sequences
3201 entirely. The code for handling \Q and \E is messy. */
3202
3203 CHECK_RANGE:
3204 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3205 {
3206 inescq = FALSE;
3207 ptr += 2;
3208 }
3209
3210 oldptr = ptr;
3211
3212 /* Remember \r or \n */
3213
3214 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3215
3216 /* Check for range */
3217
3218 if (!inescq && ptr[1] == CHAR_MINUS)
3219 {
3220 int d;
3221 ptr += 2;
3222 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3223
3224 /* If we hit \Q (not followed by \E) at this point, go into escaped
3225 mode. */
3226
3227 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3228 {
3229 ptr += 2;
3230 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3231 { ptr += 2; continue; }
3232 inescq = TRUE;
3233 break;
3234 }
3235
3236 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3237 {
3238 ptr = oldptr;
3239 goto LONE_SINGLE_CHARACTER;
3240 }
3241
3242 #ifdef SUPPORT_UTF8
3243 if (utf8)
3244 { /* Braces are required because the */
3245 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3246 }
3247 else
3248 #endif
3249 d = *ptr; /* Not UTF-8 mode */
3250
3251 /* The second part of a range can be a single-character escape, but
3252 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3253 in such circumstances. */
3254
3255 if (!inescq && d == CHAR_BACKSLASH)
3256 {
3257 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3258 if (*errorcodeptr != 0) goto FAILED;
3259
3260 /* \b is backspace; \X is literal X; \R is literal R; any other
3261 special means the '-' was literal */
3262
3263 if (d < 0)
3264 {
3265 if (d == -ESC_b) d = CHAR_BS;
3266 else if (d == -ESC_X) d = CHAR_X;
3267 else if (d == -ESC_R) d = CHAR_R; else
3268 {
3269 ptr = oldptr;
3270 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3271 }
3272 }
3273 }
3274
3275 /* Check that the two values are in the correct order. Optimize
3276 one-character ranges */
3277
3278 if (d < c)
3279 {
3280 *errorcodeptr = ERR8;
3281 goto FAILED;
3282 }
3283
3284 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3285
3286 /* Remember \r or \n */
3287
3288 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3289
3290 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3291 matching, we have to use an XCLASS with extra data items. Caseless
3292 matching for characters > 127 is available only if UCP support is
3293 available. */
3294
3295 #ifdef SUPPORT_UTF8
3296 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3297 {
3298 class_utf8 = TRUE;
3299
3300 /* With UCP support, we can find the other case equivalents of
3301 the relevant characters. There may be several ranges. Optimize how
3302 they fit with the basic range. */
3303
3304 #ifdef SUPPORT_UCP
3305 if ((options & PCRE_CASELESS) != 0)
3306 {
3307 unsigned int occ, ocd;
3308 unsigned int cc = c;
3309 unsigned int origd = d;
3310 while (get_othercase_range(&cc, origd, &occ, &ocd))
3311 {
3312 if (occ >= (unsigned int)c &&
3313 ocd <= (unsigned int)d)
3314 continue; /* Skip embedded ranges */
3315
3316 if (occ < (unsigned int)c &&
3317 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3318 { /* if there is overlap, */
3319 c = occ; /* noting that if occ < c */
3320 continue; /* we can't have ocd > d */
3321 } /* because a subrange is */
3322 if (ocd > (unsigned int)d &&
3323 occ <= (unsigned int)d + 1) /* always shorter than */
3324 { /* the basic range. */
3325 d = ocd;
3326 continue;
3327 }
3328
3329 if (occ == ocd)
3330 {
3331 *class_utf8data++ = XCL_SINGLE;
3332 }
3333 else
3334 {
3335 *class_utf8data++ = XCL_RANGE;
3336 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3337 }
3338 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3339 }
3340 }
3341 #endif /* SUPPORT_UCP */
3342
3343 /* Now record the original range, possibly modified for UCP caseless
3344 overlapping ranges. */
3345
3346 *class_utf8data++ = XCL_RANGE;
3347 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3348 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3349
3350 /* With UCP support, we are done. Without UCP support, there is no
3351 caseless matching for UTF-8 characters > 127; we can use the bit map
3352 for the smaller ones. */
3353
3354 #ifdef SUPPORT_UCP
3355 continue; /* With next character in the class */
3356 #else
3357 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3358
3359 /* Adjust upper limit and fall through to set up the map */
3360
3361 d = 127;
3362
3363 #endif /* SUPPORT_UCP */
3364 }
3365 #endif /* SUPPORT_UTF8 */
3366
3367 /* We use the bit map for all cases when not in UTF-8 mode; else
3368 ranges that lie entirely within 0-127 when there is UCP support; else
3369 for partial ranges without UCP support. */
3370
3371 class_charcount += d - c + 1;
3372 class_lastchar = d;
3373
3374 /* We can save a bit of time by skipping this in the pre-compile. */
3375
3376 if (lengthptr == NULL) for (; c <= d; c++)
3377 {
3378 classbits[c/8] |= (1 << (c&7));
3379 if ((options & PCRE_CASELESS) != 0)
3380 {
3381 int uc = cd->fcc[c]; /* flip case */
3382 classbits[uc/8] |= (1 << (uc&7));
3383 }
3384 }
3385
3386 continue; /* Go get the next char in the class */
3387 }
3388
3389 /* Handle a lone single character - we can get here for a normal
3390 non-escape char, or after \ that introduces a single character or for an
3391 apparent range that isn't. */
3392
3393 LONE_SINGLE_CHARACTER:
3394
3395 /* Handle a character that cannot go in the bit map */
3396
3397 #ifdef SUPPORT_UTF8
3398 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3399 {
3400 class_utf8 = TRUE;
3401 *class_utf8data++ = XCL_SINGLE;
3402 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3403
3404 #ifdef SUPPORT_UCP
3405 if ((options & PCRE_CASELESS) != 0)
3406 {
3407 unsigned int othercase;
3408 if ((othercase = UCD_OTHERCASE(c)) != c)
3409 {
3410 *class_utf8data++ = XCL_SINGLE;
3411 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3412 }
3413 }
3414 #endif /* SUPPORT_UCP */
3415
3416 }
3417 else
3418 #endif /* SUPPORT_UTF8 */
3419
3420 /* Handle a single-byte character */
3421 {
3422 classbits[c/8] |= (1 << (c&7));
3423 if ((options & PCRE_CASELESS) != 0)
3424 {
3425 c = cd->fcc[c]; /* flip case */
3426 classbits[c/8] |= (1 << (c&7));
3427 }
3428 class_charcount++;
3429 class_lastchar = c;
3430 }
3431 }
3432
3433 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3434
3435 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3436
3437 if (c == 0) /* Missing terminating ']' */
3438 {
3439 *errorcodeptr = ERR6;
3440 goto FAILED;
3441 }
3442
3443
3444 /* This code has been disabled because it would mean that \s counts as
3445 an explicit \r or \n reference, and that's not really what is wanted. Now
3446 we set the flag only if there is a literal "\r" or "\n" in the class. */
3447
3448 #if 0
3449 /* Remember whether \r or \n are in this class */
3450
3451 if (negate_class)
3452 {
3453 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3454 }
3455 else
3456 {
3457 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3458 }
3459 #endif
3460
3461
3462 /* If class_charcount is 1, we saw precisely one character whose value is
3463 less than 256. As long as there were no characters >= 128 and there was no
3464 use of \p or \P, in other words, no use of any XCLASS features, we can
3465 optimize.
3466
3467 In UTF-8 mode, we can optimize the negative case only if there were no
3468 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3469 operate on single-bytes only. This is an historical hangover. Maybe one day
3470 we can tidy these opcodes to handle multi-byte characters.
3471
3472 The optimization throws away the bit map. We turn the item into a
3473 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3474 that OP_NOT does not support multibyte characters. In the positive case, it
3475 can cause firstbyte to be set. Otherwise, there can be no first char if
3476 this item is first, whatever repeat count may follow. In the case of
3477 reqbyte, save the previous value for reinstating. */
3478
3479 #ifdef SUPPORT_UTF8
3480 if (class_charcount == 1 && !class_utf8 &&
3481 (!utf8 || !negate_class || class_lastchar < 128))
3482 #else
3483 if (class_charcount == 1)
3484 #endif
3485 {
3486 zeroreqbyte = reqbyte;
3487
3488 /* The OP_NOT opcode works on one-byte characters only. */
3489
3490 if (negate_class)
3491 {
3492 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3493 zerofirstbyte = firstbyte;
3494 *code++ = OP_NOT;
3495 *code++ = class_lastchar;
3496 break;
3497 }
3498
3499 /* For a single, positive character, get the value into mcbuffer, and
3500 then we can handle this with the normal one-character code. */
3501
3502 #ifdef SUPPORT_UTF8
3503 if (utf8 && class_lastchar > 127)
3504 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3505 else
3506 #endif
3507 {
3508 mcbuffer[0] = class_lastchar;
3509 mclength = 1;
3510 }
3511 goto ONE_CHAR;
3512 } /* End of 1-char optimization */
3513
3514 /* The general case - not the one-char optimization. If this is the first
3515 thing in the branch, there can be no first char setting, whatever the
3516 repeat count. Any reqbyte setting must remain unchanged after any kind of
3517 repeat. */
3518
3519 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3520 zerofirstbyte = firstbyte;
3521 zeroreqbyte = reqbyte;
3522
3523 /* If there are characters with values > 255, we have to compile an
3524 extended class, with its own opcode, unless there was a negated special
3525 such as \S in the class, because in that case all characters > 255 are in
3526 the class, so any that were explicitly given as well can be ignored. If
3527 (when there are explicit characters > 255 that must be listed) there are no
3528 characters < 256, we can omit the bitmap in the actual compiled code. */
3529
3530 #ifdef SUPPORT_UTF8
3531 if (class_utf8 && !should_flip_negation)
3532 {
3533 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3534 *code++ = OP_XCLASS;
3535 code += LINK_SIZE;
3536 *code = negate_class? XCL_NOT : 0;
3537
3538 /* If the map is required, move up the extra data to make room for it;
3539 otherwise just move the code pointer to the end of the extra data. */
3540
3541 if (class_charcount > 0)
3542 {
3543 *code++ |= XCL_MAP;
3544 memmove(code + 32, code, class_utf8data - code);
3545 memcpy(code, classbits, 32);
3546 code = class_utf8data + 32;
3547 }
3548 else code = class_utf8data;
3549
3550 /* Now fill in the complete length of the item */
3551
3552 PUT(previous, 1, code - previous);
3553 break; /* End of class handling */
3554 }
3555 #endif
3556
3557 /* If there are no characters > 255, set the opcode to OP_CLASS or
3558 OP_NCLASS, depending on whether the whole class was negated and whether
3559 there were negative specials such as \S in the class. Then copy the 32-byte
3560 map into the code vector, negating it if necessary. */
3561
3562 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3563 if (negate_class)
3564 {
3565 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3566 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3567 }
3568 else
3569 {
3570 memcpy(code, classbits, 32);
3571 }
3572 code += 32;
3573 break;
3574
3575
3576 /* ===================================================================*/
3577 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3578 has been tested above. */
3579
3580 case CHAR_LEFT_CURLY_BRACKET:
3581 if (!is_quantifier) goto NORMAL_CHAR;
3582 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3583 if (*errorcodeptr != 0) goto FAILED;
3584 goto REPEAT;
3585
3586 case CHAR_ASTERISK:
3587 repeat_min = 0;
3588 repeat_max = -1;
3589 goto REPEAT;
3590
3591 case CHAR_PLUS:
3592 repeat_min = 1;
3593 repeat_max = -1;
3594 goto REPEAT;
3595
3596 case CHAR_QUESTION_MARK:
3597 repeat_min = 0;
3598 repeat_max = 1;
3599
3600 REPEAT:
3601 if (previous == NULL)
3602 {
3603 *errorcodeptr = ERR9;
3604 goto FAILED;
3605 }
3606
3607 if (repeat_min == 0)
3608 {
3609 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3610 reqbyte = zeroreqbyte; /* Ditto */
3611 }
3612
3613 /* Remember whether this is a variable length repeat */
3614
3615 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3616
3617 op_type = 0; /* Default single-char op codes */
3618 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3619
3620 /* Save start of previous item, in case we have to move it up to make space
3621 for an inserted OP_ONCE for the additional '+' extension. */
3622
3623 tempcode = previous;
3624
3625 /* If the next character is '+', we have a possessive quantifier. This
3626 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3627 If the next character is '?' this is a minimizing repeat, by default,
3628 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3629 repeat type to the non-default. */
3630
3631 if (ptr[1] == CHAR_PLUS)
3632 {
3633 repeat_type = 0; /* Force greedy */
3634 possessive_quantifier = TRUE;
3635 ptr++;
3636 }
3637 else if (ptr[1] == CHAR_QUESTION_MARK)
3638 {
3639 repeat_type = greedy_non_default;
3640 ptr++;
3641 }
3642 else repeat_type = greedy_default;
3643
3644 /* If previous was a character match, abolish the item and generate a
3645 repeat item instead. If a char item has a minumum of more than one, ensure
3646 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3647 the first thing in a branch because the x will have gone into firstbyte
3648 instead. */
3649
3650 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3651 {
3652 /* Deal with UTF-8 characters that take up more than one byte. It's
3653 easier to write this out separately than try to macrify it. Use c to
3654 hold the length of the character in bytes, plus 0x80 to flag that it's a
3655 length rather than a small character. */
3656
3657 #ifdef SUPPORT_UTF8
3658 if (utf8 && (code[-1] & 0x80) != 0)
3659 {
3660 uschar *lastchar = code - 1;
3661 while((*lastchar & 0xc0) == 0x80) lastchar--;
3662 c = code - lastchar; /* Length of UTF-8 character */
3663 memcpy(utf8_char, lastchar, c); /* Save the char */
3664 c |= 0x80; /* Flag c as a length */
3665 }
3666 else
3667 #endif
3668
3669 /* Handle the case of a single byte - either with no UTF8 support, or
3670 with UTF-8 disabled, or for a UTF-8 character < 128. */
3671
3672 {
3673 c = code[-1];
3674 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3675 }
3676
3677 /* If the repetition is unlimited, it pays to see if the next thing on
3678 the line is something that cannot possibly match this character. If so,
3679 automatically possessifying this item gains some performance in the case
3680 where the match fails. */
3681
3682 if (!possessive_quantifier &&
3683 repeat_max < 0 &&
3684 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3685 options, cd))
3686 {
3687 repeat_type = 0; /* Force greedy */
3688 possessive_quantifier = TRUE;
3689 }
3690
3691 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3692 }
3693
3694 /* If previous was a single negated character ([^a] or similar), we use
3695 one of the special opcodes, replacing it. The code is shared with single-
3696 character repeats by setting opt_type to add a suitable offset into
3697 repeat_type. We can also test for auto-possessification. OP_NOT is
3698 currently used only for single-byte chars. */
3699
3700 else if (*previous == OP_NOT)
3701 {
3702 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3703 c = previous[1];
3704 if (!possessive_quantifier &&
3705 repeat_max < 0 &&
3706 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3707 {
3708 repeat_type = 0; /* Force greedy */
3709 possessive_quantifier = TRUE;
3710 }
3711 goto OUTPUT_SINGLE_REPEAT;
3712 }
3713
3714 /* If previous was a character type match (\d or similar), abolish it and
3715 create a suitable repeat item. The code is shared with single-character
3716 repeats by setting op_type to add a suitable offset into repeat_type. Note
3717 the the Unicode property types will be present only when SUPPORT_UCP is
3718 defined, but we don't wrap the little bits of code here because it just
3719 makes it horribly messy. */
3720
3721 else if (*previous < OP_EODN)
3722 {
3723 uschar *oldcode;
3724 int prop_type, prop_value;
3725 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3726 c = *previous;
3727
3728 if (!possessive_quantifier &&
3729 repeat_max < 0 &&
3730 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3731 {
3732 repeat_type = 0; /* Force greedy */
3733 possessive_quantifier = TRUE;
3734 }
3735
3736 OUTPUT_SINGLE_REPEAT:
3737 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3738 {
3739 prop_type = previous[1];
3740 prop_value = previous[2];
3741 }
3742 else prop_type = prop_value = -1;
3743
3744 oldcode = code;
3745 code = previous; /* Usually overwrite previous item */
3746
3747 /* If the maximum is zero then the minimum must also be zero; Perl allows
3748 this case, so we do too - by simply omitting the item altogether. */
3749
3750 if (repeat_max == 0) goto END_REPEAT;
3751
3752 /* All real repeats make it impossible to handle partial matching (maybe
3753 one day we will be able to remove this restriction). */
3754
3755 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3756
3757 /* Combine the op_type with the repeat_type */
3758
3759 repeat_type += op_type;
3760
3761 /* A minimum of zero is handled either as the special case * or ?, or as
3762 an UPTO, with the maximum given. */
3763
3764 if (repeat_min == 0)
3765 {
3766 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3767 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3768 else
3769 {
3770 *code++ = OP_UPTO + repeat_type;
3771 PUT2INC(code, 0, repeat_max);
3772 }
3773 }
3774
3775 /* A repeat minimum of 1 is optimized into some special cases. If the
3776 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3777 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3778 one less than the maximum. */
3779
3780 else if (repeat_min == 1)
3781 {
3782 if (repeat_max == -1)
3783 *code++ = OP_PLUS + repeat_type;
3784 else
3785 {
3786 code = oldcode; /* leave previous item in place */
3787 if (repeat_max == 1) goto END_REPEAT;
3788 *code++ = OP_UPTO + repeat_type;
3789 PUT2INC(code, 0, repeat_max - 1);
3790 }
3791 }
3792
3793 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3794 handled as an EXACT followed by an UPTO. */
3795
3796 else
3797 {
3798 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3799 PUT2INC(code, 0, repeat_min);
3800
3801 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3802 we have to insert the character for the previous code. For a repeated
3803 Unicode property match, there are two extra bytes that define the
3804 required property. In UTF-8 mode, long characters have their length in
3805 c, with the 0x80 bit as a flag. */
3806
3807 if (repeat_max < 0)
3808 {
3809 #ifdef SUPPORT_UTF8
3810 if (utf8 && c >= 128)
3811 {
3812 memcpy(code, utf8_char, c & 7);
3813 code += c & 7;
3814 }
3815 else
3816 #endif
3817 {
3818 *code++ = c;
3819 if (prop_type >= 0)
3820 {
3821 *code++ = prop_type;
3822 *code++ = prop_value;
3823 }
3824 }
3825 *code++ = OP_STAR + repeat_type;
3826 }
3827
3828 /* Else insert an UPTO if the max is greater than the min, again
3829 preceded by the character, for the previously inserted code. If the
3830 UPTO is just for 1 instance, we can use QUERY instead. */
3831
3832 else if (repeat_max != repeat_min)
3833 {
3834 #ifdef SUPPORT_UTF8
3835 if (utf8 && c >= 128)
3836 {
3837 memcpy(code, utf8_char, c & 7);
3838 code += c & 7;
3839 }
3840 else
3841 #endif
3842 *code++ = c;
3843 if (prop_type >= 0)
3844 {
3845 *code++ = prop_type;
3846 *code++ = prop_value;
3847 }
3848 repeat_max -= repeat_min;
3849
3850 if (repeat_max == 1)
3851 {
3852 *code++ = OP_QUERY + repeat_type;
3853 }
3854 else
3855 {
3856 *code++ = OP_UPTO + repeat_type;
3857 PUT2INC(code, 0, repeat_max);
3858 }
3859 }
3860 }
3861
3862 /* The character or character type itself comes last in all cases. */
3863
3864 #ifdef SUPPORT_UTF8
3865 if (utf8 && c >= 128)
3866 {
3867 memcpy(code, utf8_char, c & 7);
3868 code += c & 7;
3869 }
3870 else
3871 #endif
3872 *code++ = c;
3873
3874 /* For a repeated Unicode property match, there are two extra bytes that
3875 define the required property. */
3876
3877 #ifdef SUPPORT_UCP
3878 if (prop_type >= 0)
3879 {
3880 *code++ = prop_type;
3881 *code++ = prop_value;
3882 }
3883 #endif
3884 }
3885
3886 /* If previous was a character class or a back reference, we put the repeat
3887 stuff after it, but just skip the item if the repeat was {0,0}. */
3888
3889 else if (*previous == OP_CLASS ||
3890 *previous == OP_NCLASS ||
3891 #ifdef SUPPORT_UTF8
3892 *previous == OP_XCLASS ||
3893 #endif
3894 *previous == OP_REF)
3895 {
3896 if (repeat_max == 0)
3897 {
3898 code = previous;
3899 goto END_REPEAT;
3900 }
3901
3902 /* All real repeats make it impossible to handle partial matching (maybe
3903 one day we will be able to remove this restriction). */
3904
3905 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3906
3907 if (repeat_min == 0 && repeat_max == -1)
3908 *code++ = OP_CRSTAR + repeat_type;
3909 else if (repeat_min == 1 && repeat_max == -1)
3910 *code++ = OP_CRPLUS + repeat_type;
3911 else if (repeat_min == 0 && repeat_max == 1)
3912 *code++ = OP_CRQUERY + repeat_type;
3913 else
3914 {
3915 *code++ = OP_CRRANGE + repeat_type;
3916 PUT2INC(code, 0, repeat_min);
3917 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
3918 PUT2INC(code, 0, repeat_max);
3919 }
3920 }
3921
3922 /* If previous was a bracket group, we may have to replicate it in certain
3923 cases. */
3924
3925 else if (*previous == OP_BRA || *previous == OP_CBRA ||
3926 *previous == OP_ONCE || *previous == OP_COND)
3927 {
3928 register int i;
3929 int ketoffset = 0;
3930 int len = code - previous;
3931 uschar *bralink = NULL;
3932
3933 /* Repeating a DEFINE group is pointless */
3934
3935 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
3936 {
3937 *errorcodeptr = ERR55;
3938 goto FAILED;
3939 }
3940
3941 /* If the maximum repeat count is unlimited, find the end of the bracket
3942 by scanning through from the start, and compute the offset back to it
3943 from the current code pointer. There may be an OP_OPT setting following
3944 the final KET, so we can't find the end just by going back from the code
3945 pointer. */
3946
3947 if (repeat_max == -1)
3948 {
3949 register uschar *ket = previous;
3950 do ket += GET(ket, 1); while (*ket != OP_KET);
3951 ketoffset = code - ket;
3952 }
3953
3954 /* The case of a zero minimum is special because of the need to stick
3955 OP_BRAZERO in front of it, and because the group appears once in the
3956 data, whereas in other cases it appears the minimum number of times. For
3957 this reason, it is simplest to treat this case separately, as otherwise
3958 the code gets far too messy. There are several special subcases when the
3959 minimum is zero. */
3960
3961 if (repeat_min == 0)
3962 {
3963 /* If the maximum is also zero, we used to just omit the group from the
3964 output altogether, like this:
3965
3966 ** if (repeat_max == 0)
3967 ** {
3968 ** code = previous;
3969 ** goto END_REPEAT;
3970 ** }
3971
3972 However, that fails when a group is referenced as a subroutine from
3973 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
3974 so that it is skipped on execution. As we don't have a list of which
3975 groups are referenced, we cannot do this selectively.
3976
3977 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
3978 and do no more at this point. However, we do need to adjust any
3979 OP_RECURSE calls inside the group that refer to the group itself or any
3980 internal or forward referenced group, because the offset is from the
3981 start of the whole regex. Temporarily terminate the pattern while doing
3982 this. */
3983
3984 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
3985 {
3986 *code = OP_END;
3987 adjust_recurse(previous, 1, utf8, cd, save_hwm);
3988 memmove(previous+1, previous, len);
3989 code++;
3990 if (repeat_max == 0)
3991 {
3992 *previous++ = OP_SKIPZERO;
3993 goto END_REPEAT;
3994 }
3995 *previous++ = OP_BRAZERO + repeat_type;
3996 }
3997
3998 /* If the maximum is greater than 1 and limited, we have to replicate
3999 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4000 The first one has to be handled carefully because it's the original
4001 copy, which has to be moved up. The remainder can be handled by code
4002 that is common with the non-zero minimum case below. We have to
4003 adjust the value or repeat_max, since one less copy is required. Once
4004 again, we may have to adjust any OP_RECURSE calls inside the group. */
4005
4006 else
4007 {
4008 int offset;
4009 *code = OP_END;
4010 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4011 memmove(previous + 2 + LINK_SIZE, previous, len);
4012 code += 2 + LINK_SIZE;
4013 *previous++ = OP_BRAZERO + repeat_type;
4014 *previous++ = OP_BRA;
4015
4016 /* We chain together the bracket offset fields that have to be
4017 filled in later when the ends of the brackets are reached. */
4018
4019 offset = (bralink == NULL)? 0 : previous - bralink;
4020 bralink = previous;
4021 PUTINC(previous, 0, offset);
4022 }
4023
4024 repeat_max--;
4025 }
4026
4027 /* If the minimum is greater than zero, replicate the group as many
4028 times as necessary, and adjust the maximum to the number of subsequent
4029 copies that we need. If we set a first char from the group, and didn't
4030 set a required char, copy the latter from the former. If there are any
4031 forward reference subroutine calls in the group, there will be entries on
4032 the workspace list; replicate these with an appropriate increment. */
4033
4034 else
4035 {
4036 if (repeat_min > 1)
4037 {
4038 /* In the pre-compile phase, we don't actually do the replication. We
4039 just adjust the length as if we had. Do some paranoid checks for
4040 potential integer overflow. */
4041
4042 if (lengthptr != NULL)
4043 {
4044 int delta = (repeat_min - 1)*length_prevgroup;
4045 if ((double)(repeat_min - 1)*(double)length_prevgroup >
4046 (double)INT_MAX ||
4047 OFLOW_MAX - *lengthptr < delta)
4048 {
4049 *errorcodeptr = ERR20;
4050 goto FAILED;
4051 }
4052 *lengthptr += delta;
4053 }
4054
4055 /* This is compiling for real */
4056
4057 else
4058 {
4059 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4060 for (i = 1; i < repeat_min; i++)
4061 {
4062 uschar *hc;
4063 uschar *this_hwm = cd->hwm;
4064 memcpy(code, previous, len);
4065 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4066 {
4067 PUT(cd->hwm, 0, GET(hc, 0) + len);
4068 cd->hwm += LINK_SIZE;
4069 }
4070 save_hwm = this_hwm;
4071 code += len;
4072 }
4073 }
4074 }
4075
4076 if (repeat_max > 0) repeat_max -= repeat_min;
4077 }
4078
4079 /* This code is common to both the zero and non-zero minimum cases. If
4080 the maximum is limited, it replicates the group in a nested fashion,
4081 remembering the bracket starts on a stack. In the case of a zero minimum,
4082 the first one was set up above. In all cases the repeat_max now specifies
4083 the number of additional copies needed. Again, we must remember to
4084 replicate entries on the forward reference list. */
4085
4086 if (repeat_max >= 0)
4087 {
4088 /* In the pre-compile phase, we don't actually do the replication. We
4089 just adjust the length as if we had. For each repetition we must add 1
4090 to the length for BRAZERO and for all but the last repetition we must
4091 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4092 paranoid checks to avoid integer overflow. */
4093
4094 if (lengthptr != NULL && repeat_max > 0)
4095 {
4096 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4097 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4098 if ((double)repeat_max *
4099 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4100 > (double)INT_MAX ||
4101 OFLOW_MAX - *lengthptr < delta)
4102 {
4103 *errorcodeptr = ERR20;
4104 goto FAILED;
4105 }
4106 *lengthptr += delta;
4107 }
4108
4109 /* This is compiling for real */
4110
4111 else for (i = repeat_max - 1; i >= 0; i--)
4112 {
4113 uschar *hc;
4114 uschar *this_hwm = cd->hwm;
4115
4116 *code++ = OP_BRAZERO + repeat_type;
4117
4118 /* All but the final copy start a new nesting, maintaining the
4119 chain of brackets outstanding. */
4120
4121 if (i != 0)
4122 {
4123 int offset;
4124 *code++ = OP_BRA;
4125 offset = (bralink == NULL)? 0 : code - bralink;
4126 bralink = code;
4127 PUTINC(code, 0, offset);
4128 }
4129
4130 memcpy(code, previous, len);
4131 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4132 {
4133 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4134 cd->hwm += LINK_SIZE;
4135 }
4136 save_hwm = this_hwm;
4137 code += len;
4138 }
4139
4140 /* Now chain through the pending brackets, and fill in their length
4141 fields (which are holding the chain links pro tem). */
4142
4143 while (bralink != NULL)
4144 {
4145 int oldlinkoffset;
4146 int offset = code - bralink + 1;
4147 uschar *bra = code - offset;
4148 oldlinkoffset = GET(bra, 1);
4149 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4150 *code++ = OP_KET;
4151 PUTINC(code, 0, offset);
4152 PUT(bra, 1, offset);
4153 }
4154 }
4155
4156 /* If the maximum is unlimited, set a repeater in the final copy. We
4157 can't just offset backwards from the current code point, because we
4158 don't know if there's been an options resetting after the ket. The
4159 correct offset was computed above.
4160
4161 Then, when we are doing the actual compile phase, check to see whether
4162 this group is a non-atomic one that could match an empty string. If so,
4163 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4164 that runtime checking can be done. [This check is also applied to
4165 atomic groups at runtime, but in a different way.] */
4166
4167 else
4168 {
4169 uschar *ketcode = code - ketoffset;
4170 uschar *bracode = ketcode - GET(ketcode, 1);
4171 *ketcode = OP_KETRMAX + repeat_type;
4172 if (lengthptr == NULL && *bracode != OP_ONCE)
4173 {
4174 uschar *scode = bracode;
4175 do
4176 {
4177 if (could_be_empty_branch(scode, ketcode, utf8))
4178 {
4179 *bracode += OP_SBRA - OP_BRA;
4180 break;
4181 }
4182 scode += GET(scode, 1);
4183 }
4184 while (*scode == OP_ALT);
4185 }
4186 }
4187 }
4188
4189 /* If previous is OP_FAIL, it was generated by an empty class [] in
4190 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4191 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4192 error above. We can just ignore the repeat in JS case. */
4193
4194 else if (*previous == OP_FAIL) goto END_REPEAT;
4195
4196 /* Else there's some kind of shambles */
4197
4198 else
4199 {
4200 *errorcodeptr = ERR11;
4201 goto FAILED;
4202 }
4203
4204 /* If the character following a repeat is '+', or if certain optimization
4205 tests above succeeded, possessive_quantifier is TRUE. For some of the
4206 simpler opcodes, there is an special alternative opcode for this. For
4207 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4208 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4209 but the special opcodes can optimize it a bit. The repeated item starts at
4210 tempcode, not at previous, which might be the first part of a string whose
4211 (former) last char we repeated.
4212
4213 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4214 an 'upto' may follow. We skip over an 'exact' item, and then test the
4215 length of what remains before proceeding. */
4216
4217 if (possessive_quantifier)
4218 {
4219 int len;
4220 if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
4221 *tempcode == OP_NOTEXACT)
4222 tempcode += _pcre_OP_lengths[*tempcode] +
4223 ((*tempcode == OP_TYPEEXACT &&
4224 (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);
4225 len = code - tempcode;
4226 if (len > 0) switch (*tempcode)
4227 {
4228 case OP_STAR: *tempcode = OP_POSSTAR; break;
4229 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4230 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4231 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4232
4233 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4234 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4235 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4236 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4237
4238 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4239 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4240 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4241 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4242
4243 default:
4244 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4245 code += 1 + LINK_SIZE;
4246 len += 1 + LINK_SIZE;
4247 tempcode[0] = OP_ONCE;
4248 *code++ = OP_KET;
4249 PUTINC(code, 0, len);
4250 PUT(tempcode, 1, len);
4251 break;
4252 }
4253 }
4254
4255 /* In all case we no longer have a previous item. We also set the
4256 "follows varying string" flag for subsequently encountered reqbytes if
4257 it isn't already set and we have just passed a varying length item. */
4258
4259 END_REPEAT:
4260 previous = NULL;
4261 cd->req_varyopt |= reqvary;
4262 break;
4263
4264
4265 /* ===================================================================*/
4266 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4267 lookbehind or option setting or condition or all the other extended
4268 parenthesis forms. */
4269
4270 case CHAR_LEFT_PARENTHESIS:
4271 newoptions = options;
4272 skipbytes = 0;
4273 bravalue = OP_CBRA;
4274 save_hwm = cd->hwm;
4275 reset_bracount = FALSE;
4276
4277 /* First deal with various "verbs" that can be introduced by '*'. */
4278
4279 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4280 {
4281 int i, namelen;
4282 const char *vn = verbnames;
4283 const uschar *name = ++ptr;
4284 previous = NULL;
4285 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4286 if (*ptr == CHAR_COLON)
4287 {
4288 *errorcodeptr = ERR59; /* Not supported */
4289 goto FAILED;
4290 }
4291 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4292 {
4293 *errorcodeptr = ERR60;
4294 goto FAILED;
4295 }
4296 namelen = ptr - name;
4297 for (i = 0; i < verbcount; i++)
4298 {
4299 if (namelen == verbs[i].len &&
4300 strncmp((char *)name, vn, namelen) == 0)
4301 {
4302 *code = verbs[i].op;
4303 if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
4304 break;
4305 }
4306 vn += verbs[i].len + 1;
4307 }
4308 if (i < verbcount) continue;
4309 *errorcodeptr = ERR60;
4310 goto FAILED;
4311 }
4312
4313 /* Deal with the extended parentheses; all are introduced by '?', and the
4314 appearance of any of them means that this is not a capturing group. */
4315
4316 else if (*ptr == CHAR_QUESTION_MARK)
4317 {
4318 int i, set, unset, namelen;
4319 int *optset;
4320 const uschar *name;
4321 uschar *slot;
4322
4323 switch (*(++ptr))
4324 {
4325 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4326 ptr++;
4327 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4328 if (*ptr == 0)
4329 {
4330 *errorcodeptr = ERR18;
4331 goto FAILED;
4332 }
4333 continue;
4334
4335
4336 /* ------------------------------------------------------------ */
4337 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4338 reset_bracount = TRUE;
4339 /* Fall through */
4340
4341 /* ------------------------------------------------------------ */
4342 case CHAR_COLON: /* Non-capturing bracket */
4343 bravalue = OP_BRA;
4344 ptr++;
4345 break;
4346
4347
4348 /* ------------------------------------------------------------ */
4349 case CHAR_LEFT_PARENTHESIS:
4350 bravalue = OP_COND; /* Conditional group */
4351
4352 /* A condition can be an assertion, a number (referring to a numbered
4353 group), a name (referring to a named group), or 'R', referring to
4354 recursion. R<digits> and R&name are also permitted for recursion tests.
4355
4356 There are several syntaxes for testing a named group: (?(name)) is used
4357 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4358
4359 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4360 be the recursive thing or the name 'R' (and similarly for 'R' followed
4361 by digits), and (b) a number could be a name that consists of digits.
4362 In both cases, we look for a name first; if not found, we try the other
4363 cases. */
4364
4365 /* For conditions that are assertions, check the syntax, and then exit
4366 the switch. This will take control down to where bracketed groups,
4367 including assertions, are processed. */
4368
4369 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4370 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4371 break;
4372
4373 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4374 below), and all need to skip 3 bytes at the start of the group. */
4375
4376 code[1+LINK_SIZE] = OP_CREF;
4377 skipbytes = 3;
4378 refsign = -1;
4379
4380 /* Check for a test for recursion in a named group. */
4381
4382 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4383 {
4384 terminator = -1;
4385 ptr += 2;
4386 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4387 }
4388
4389 /* Check for a test for a named group's having been set, using the Perl
4390 syntax (?(<name>) or (?('name') */
4391
4392 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4393 {
4394 terminator = CHAR_GREATER_THAN_SIGN;
4395 ptr++;
4396 }
4397 else if (ptr[1] == CHAR_APOSTROPHE)
4398 {
4399 terminator = CHAR_APOSTROPHE;
4400 ptr++;
4401 }
4402 else
4403 {
4404 terminator = 0;
4405 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4406 }
4407
4408 /* We now expect to read a name; any thing else is an error */
4409
4410 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4411 {
4412 ptr += 1; /* To get the right offset */
4413 *errorcodeptr = ERR28;
4414 goto FAILED;
4415 }
4416
4417 /* Read the name, but also get it as a number if it's all digits */
4418
4419 recno = 0;
4420 name = ++ptr;
4421 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4422 {
4423 if (recno >= 0)
4424 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4425 recno * 10 + *ptr - CHAR_0 : -1;
4426 ptr++;
4427 }
4428 namelen = ptr - name;
4429
4430 if ((terminator > 0 && *ptr++ != terminator) ||
4431 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4432 {
4433 ptr--; /* Error offset */
4434 *errorcodeptr = ERR26;
4435 goto FAILED;
4436 }
4437
4438 /* Do no further checking in the pre-compile phase. */
4439
4440 if (lengthptr != NULL) break;
4441
4442 /* In the real compile we do the work of looking for the actual
4443 reference. If the string started with "+" or "-" we require the rest to
4444 be digits, in which case recno will be set. */
4445
4446 if (refsign > 0)
4447 {
4448 if (recno <= 0)
4449 {
4450 *errorcodeptr = ERR58;
4451 goto FAILED;
4452 }
4453 recno = (refsign == CHAR_MINUS)?
4454 cd->bracount - recno + 1 : recno +cd->bracount;
4455 if (recno <= 0 || recno > cd->final_bracount)
4456 {
4457 *errorcodeptr = ERR15;
4458 goto FAILED;
4459 }
4460 PUT2(code, 2+LINK_SIZE, recno);
4461 break;
4462 }
4463
4464 /* Otherwise (did not start with "+" or "-"), start by looking for the
4465 name. */
4466
4467 slot = cd->name_table;
4468 for (i = 0; i < cd->names_found; i++)
4469 {
4470 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4471 slot += cd->name_entry_size;
4472 }
4473
4474 /* Found a previous named subpattern */
4475
4476 if (i < cd->names_found)
4477 {
4478 recno = GET2(slot, 0);
4479 PUT2(code, 2+LINK_SIZE, recno);
4480 }
4481
4482 /* Search the pattern for a forward reference */
4483
4484 else if ((i = find_parens(ptr, cd, name, namelen,
4485 (options & PCRE_EXTENDED) != 0)) > 0)
4486 {
4487 PUT2(code, 2+LINK_SIZE, i);
4488 }
4489
4490 /* If terminator == 0 it means that the name followed directly after
4491 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4492 some further alternatives to try. For the cases where terminator != 0
4493 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4494 now checked all the possibilities, so give an error. */
4495
4496 else if (terminator != 0)
4497 {
4498 *errorcodeptr = ERR15;
4499 goto FAILED;
4500 }
4501
4502 /* Check for (?(R) for recursion. Allow digits after R to specify a
4503 specific group number. */
4504
4505 else if (*name == CHAR_R)
4506 {
4507 recno = 0;
4508 for (i = 1; i < namelen; i++)
4509 {
4510 if ((digitab[name[i]] & ctype_digit) == 0)
4511 {
4512 *errorcodeptr = ERR15;
4513 goto FAILED;
4514 }
4515 recno = recno * 10 + name[i] - CHAR_0;
4516 }
4517 if (recno == 0) recno = RREF_ANY;
4518 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4519 PUT2(code, 2+LINK_SIZE, recno);
4520 }
4521
4522 /* Similarly, check for the (?(DEFINE) "condition", which is always
4523 false. */
4524
4525 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4526 {
4527 code[1+LINK_SIZE] = OP_DEF;
4528 skipbytes = 1;
4529 }
4530
4531 /* Check for the "name" actually being a subpattern number. We are
4532 in the second pass here, so final_bracount is set. */
4533
4534 else if (recno > 0 && recno <= cd->final_bracount)
4535 {
4536 PUT2(code, 2+LINK_SIZE, recno);
4537 }
4538
4539 /* Either an unidentified subpattern, or a reference to (?(0) */
4540
4541 else
4542 {
4543 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4544 goto FAILED;
4545 }
4546 break;
4547
4548
4549 /* ------------------------------------------------------------ */
4550 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4551 bravalue = OP_ASSERT;
4552 ptr++;
4553 break;
4554
4555
4556 /* ------------------------------------------------------------ */
4557 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4558 ptr++;
4559 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4560 {
4561 *code++ = OP_FAIL;
4562 previous = NULL;
4563 continue;
4564 }
4565 bravalue = OP_ASSERT_NOT;
4566 break;
4567
4568
4569 /* ------------------------------------------------------------ */
4570 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4571 switch (ptr[1])
4572 {
4573 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4574 bravalue = OP_ASSERTBACK;
4575 ptr += 2;
4576 break;
4577
4578 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4579 bravalue = OP_ASSERTBACK_NOT;
4580 ptr += 2;
4581 break;
4582
4583 default: /* Could be name define, else bad */
4584 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4585 ptr++; /* Correct offset for error */
4586 *errorcodeptr = ERR24;
4587 goto FAILED;
4588 }
4589 break;
4590
4591
4592 /* ------------------------------------------------------------ */
4593 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4594 bravalue = OP_ONCE;
4595 ptr++;
4596 break;
4597
4598
4599 /* ------------------------------------------------------------ */
4600 case CHAR_C: /* Callout - may be followed by digits; */
4601 previous_callout = code; /* Save for later completion */
4602 after_manual_callout = 1; /* Skip one item before completing */
4603 *code++ = OP_CALLOUT;
4604 {
4605 int n = 0;
4606 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4607 n = n * 10 + *ptr - CHAR_0;
4608 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4609 {
4610 *errorcodeptr = ERR39;
4611 goto FAILED;
4612 }
4613 if (n > 255)
4614 {
4615 *errorcodeptr = ERR38;
4616 goto FAILED;
4617 }
4618 *code++ = n;
4619 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4620 PUT(code, LINK_SIZE, 0); /* Default length */
4621 code += 2 * LINK_SIZE;
4622 }
4623 previous = NULL;
4624 continue;
4625
4626
4627 /* ------------------------------------------------------------ */
4628 case CHAR_P: /* Python-style named subpattern handling */
4629 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4630 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4631 {
4632 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4633 terminator = CHAR_RIGHT_PARENTHESIS;
4634 goto NAMED_REF_OR_RECURSE;
4635 }
4636 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4637 {
4638 *errorcodeptr = ERR41;
4639 goto FAILED;
4640 }
4641 /* Fall through to handle (?P< as (?< is handled */
4642
4643
4644 /* ------------------------------------------------------------ */
4645 DEFINE_NAME: /* Come here from (?< handling */
4646 case CHAR_APOSTROPHE:
4647 {
4648 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4649 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4650 name = ++ptr;
4651
4652 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4653 namelen = ptr - name;
4654
4655 /* In the pre-compile phase, just do a syntax check. */
4656
4657 if (lengthptr != NULL)
4658 {
4659 if (*ptr != terminator)
4660 {
4661 *errorcodeptr = ERR42;
4662 goto FAILED;
4663 }
4664 if (cd->names_found >= MAX_NAME_COUNT)
4665 {
4666 *errorcodeptr = ERR49;
4667 goto FAILED;
4668 }
4669 if (namelen + 3 > cd->name_entry_size)
4670 {
4671 cd->name_entry_size = namelen + 3;
4672 if (namelen > MAX_NAME_SIZE)
4673 {
4674 *errorcodeptr = ERR48;
4675 goto FAILED;
4676 }
4677 }
4678 }
4679
4680 /* In the real compile, create the entry in the table */
4681
4682 else
4683 {
4684 slot = cd->name_table;
4685 for (i = 0; i < cd->names_found; i++)
4686 {
4687 int crc = memcmp(name, slot+2, namelen);
4688 if (crc == 0)
4689 {
4690 if (slot[2+namelen] == 0)
4691 {
4692 if ((options & PCRE_DUPNAMES) == 0)
4693 {
4694 *errorcodeptr = ERR43;
4695 goto FAILED;
4696 }
4697 }
4698 else crc = -1; /* Current name is substring */
4699 }
4700 if (crc < 0)
4701 {
4702 memmove(slot + cd->name_entry_size, slot,
4703 (cd->names_found - i) * cd->name_entry_size);
4704 break;
4705 }
4706 slot += cd->name_entry_size;
4707 }
4708
4709 PUT2(slot, 0, cd->bracount + 1);
4710 memcpy(slot + 2, name, namelen);
4711 slot[2+namelen] = 0;
4712 }
4713 }
4714
4715 /* In both cases, count the number of names we've encountered. */
4716
4717 ptr++; /* Move past > or ' */
4718 cd->names_found++;
4719 goto NUMBERED_GROUP;
4720
4721
4722 /* ------------------------------------------------------------ */
4723 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
4724 terminator = CHAR_RIGHT_PARENTHESIS;
4725 is_recurse = TRUE;
4726 /* Fall through */
4727
4728 /* We come here from the Python syntax above that handles both
4729 references (?P=name) and recursion (?P>name), as well as falling
4730 through from the Perl recursion syntax (?&name). We also come here from
4731 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4732 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
4733
4734 NAMED_REF_OR_RECURSE:
4735 name = ++ptr;
4736 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4737 namelen = ptr - name;
4738
4739 /* In the pre-compile phase, do a syntax check and set a dummy
4740 reference number. */
4741
4742 if (lengthptr != NULL)
4743 {
4744 if (namelen == 0)
4745 {
4746 *errorcodeptr = ERR62;
4747 goto FAILED;
4748 }
4749 if (*ptr != terminator)
4750 {
4751 *errorcodeptr = ERR42;
4752 goto FAILED;
4753 }
4754 if (namelen > MAX_NAME_SIZE)
4755 {
4756 *errorcodeptr = ERR48;
4757 goto FAILED;
4758 }
4759 recno = 0;
4760 }
4761
4762 /* In the real compile, seek the name in the table. We check the name
4763 first, and then check that we have reached the end of the name in the
4764 table. That way, if the name that is longer than any in the table,
4765 the comparison will fail without reading beyond the table entry. */
4766
4767 else
4768 {
4769 slot = cd->name_table;
4770 for (i = 0; i < cd->names_found; i++)
4771 {
4772 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
4773 slot[2+namelen] == 0)
4774 break;
4775 slot += cd->name_entry_size;
4776 }
4777
4778 if (i < cd->names_found) /* Back reference */
4779 {
4780 recno = GET2(slot, 0);
4781 }
4782 else if ((recno = /* Forward back reference */
4783 find_parens(ptr, cd, name, namelen,
4784 (options & PCRE_EXTENDED) != 0)) <= 0)
4785 {
4786 *errorcodeptr = ERR15;
4787 goto FAILED;
4788 }
4789 }
4790
4791 /* In both phases, we can now go to the code than handles numerical
4792 recursion or backreferences. */
4793
4794 if (is_recurse) goto HANDLE_RECURSION;
4795 else goto HANDLE_REFERENCE;
4796
4797
4798 /* ------------------------------------------------------------ */
4799 case CHAR_R: /* Recursion */
4800 ptr++; /* Same as (?0) */
4801 /* Fall through */
4802
4803
4804 /* ------------------------------------------------------------ */
4805 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
4806 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
4807 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
4808 {
4809 const uschar *called;
4810 terminator = CHAR_RIGHT_PARENTHESIS;
4811
4812 /* Come here from the \g<...> and \g'...' code (Oniguruma
4813 compatibility). However, the syntax has been checked to ensure that
4814 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
4815 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
4816 ever be taken. */
4817
4818 HANDLE_NUMERICAL_RECURSION:
4819
4820 if ((refsign = *ptr) == CHAR_PLUS)
4821 {
4822 ptr++;
4823 if ((digitab[*ptr] & ctype_digit) == 0)
4824 {
4825 *errorcodeptr = ERR63;
4826 goto FAILED;
4827 }
4828 }
4829 else if (refsign == CHAR_MINUS)
4830 {
4831 if ((digitab[ptr[1]] & ctype_digit) == 0)
4832 goto OTHER_CHAR_AFTER_QUERY;
4833 ptr++;
4834 }
4835
4836 recno = 0;
4837 while((digitab[*ptr] & ctype_digit) != 0)
4838 recno = recno * 10 + *ptr++ - CHAR_0;
4839
4840 if (*ptr != terminator)
4841 {
4842 *errorcodeptr = ERR29;
4843 goto FAILED;
4844 }
4845
4846 if (refsign == CHAR_MINUS)
4847 {
4848 if (recno == 0)
4849 {
4850 *errorcodeptr = ERR58;
4851 goto FAILED;
4852 }
4853 recno = cd->bracount - recno + 1;
4854 if (recno <= 0)
4855 {
4856 *errorcodeptr = ERR15;
4857 goto FAILED;
4858 }
4859 }
4860 else if (refsign == CHAR_PLUS)
4861 {
4862 if (recno == 0)
4863 {
4864 *errorcodeptr = ERR58;
4865 goto FAILED;
4866 }
4867 recno += cd->bracount;
4868 }
4869
4870 /* Come here from code above that handles a named recursion */
4871
4872 HANDLE_RECURSION:
4873
4874 previous = code;
4875 called = cd->start_code;
4876
4877 /* When we are actually compiling, find the bracket that is being
4878 referenced. Temporarily end the regex in case it doesn't exist before
4879 this point. If we end up with a forward reference, first check that
4880 the bracket does occur later so we can give the error (and position)
4881 now. Then remember this forward reference in the workspace so it can
4882 be filled in at the end. */
4883
4884 if (lengthptr == NULL)
4885 {
4886 *code = OP_END;
4887 if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
4888
4889 /* Forward reference */
4890
4891 if (called == NULL)
4892 {
4893 if (find_parens(ptr, cd, NULL, recno,
4894 (options & PCRE_EXTENDED) != 0) < 0)
4895 {
4896 *errorcodeptr = ERR15;
4897 goto FAILED;
4898 }
4899 called = cd->start_code + recno;
4900 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
4901 }
4902
4903 /* If not a forward reference, and the subpattern is still open,
4904 this is a recursive call. We check to see if this is a left
4905 recursion that could loop for ever, and diagnose that case. */
4906
4907 else if (GET(called, 1) == 0 &&
4908 could_be_empty(called, code, bcptr, utf8))
4909 {
4910 *errorcodeptr = ERR40;
4911 goto FAILED;
4912 }
4913 }
4914
4915 /* Insert the recursion/subroutine item, automatically wrapped inside
4916 "once" brackets. Set up a "previous group" length so that a
4917 subsequent quantifier will work. */
4918
4919 *code = OP_ONCE;
4920 PUT(code, 1, 2 + 2*LINK_SIZE);
4921 code += 1 + LINK_SIZE;
4922
4923 *code = OP_RECURSE;
4924 PUT(code, 1, called - cd->start_code);
4925 code += 1 + LINK_SIZE;
4926
4927 *code = OP_KET;
4928 PUT(code, 1, 2 + 2*LINK_SIZE);
4929 code += 1 + LINK_SIZE;
4930
4931 length_prevgroup = 3 + 3*LINK_SIZE;
4932 }
4933
4934 /* Can't determine a first byte now */
4935
4936 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4937 continue;
4938
4939
4940 /* ------------------------------------------------------------ */
4941 default: /* Other characters: check option setting */
4942 OTHER_CHAR_AFTER_QUERY:
4943 set = unset = 0;
4944 optset = &set;
4945
4946 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
4947 {
4948 switch (*ptr++)
4949 {
4950 case CHAR_MINUS: optset = &unset; break;
4951
4952 case CHAR_J: /* Record that it changed in the external options */
4953 *optset |= PCRE_DUPNAMES;
4954 cd->external_flags |= PCRE_JCHANGED;
4955 break;
4956
4957 case CHAR_i: *optset |= PCRE_CASELESS; break;
4958 case CHAR_m: *optset |= PCRE_MULTILINE; break;
4959 case CHAR_s: *optset |= PCRE_DOTALL; break;
4960 case CHAR_x: *optset |= PCRE_EXTENDED; break;
4961 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
4962 case CHAR_X: *optset |= PCRE_EXTRA; break;
4963
4964 default: *errorcodeptr = ERR12;
4965 ptr--; /* Correct the offset */
4966 goto FAILED;
4967 }
4968 }
4969
4970 /* Set up the changed option bits, but don't change anything yet. */
4971
4972 newoptions = (options | set) & (~unset);
4973
4974 /* If the options ended with ')' this is not the start of a nested
4975 group with option changes, so the options change at this level. If this
4976 item is right at the start of the pattern, the options can be
4977 abstracted and made external in the pre-compile phase, and ignored in
4978 the compile phase. This can be helpful when matching -- for instance in
4979 caseless checking of required bytes.
4980
4981 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
4982 definitely *not* at the start of the pattern because something has been
4983 compiled. In the pre-compile phase, however, the code pointer can have
4984 that value after the start, because it gets reset as code is discarded
4985 during the pre-compile. However, this can happen only at top level - if
4986 we are within parentheses, the starting BRA will still be present. At
4987 any parenthesis level, the length value can be used to test if anything
4988 has been compiled at that level. Thus, a test for both these conditions
4989 is necessary to ensure we correctly detect the start of the pattern in
4990 both phases.
4991
4992 If we are not at the pattern start, compile code to change the ims
4993 options if this setting actually changes any of them, and reset the
4994 greedy defaults and the case value for firstbyte and reqbyte. */
4995
4996 if (*ptr == CHAR_RIGHT_PARENTHESIS)
4997 {
4998 if (code == cd->start_code + 1 + LINK_SIZE &&
4999 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5000 {
5001 cd->external_options = newoptions;
5002 }
5003 else
5004 {
5005 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5006 {
5007 *code++ = OP_OPT;
5008 *code++ = newoptions & PCRE_IMS;
5009 }
5010 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5011 greedy_non_default = greedy_default ^ 1;
5012 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5013 }
5014
5015 /* Change options at this level, and pass them back for use
5016 in subsequent branches. When not at the start of the pattern, this
5017 information is also necessary so that a resetting item can be
5018 compiled at the end of a group (if we are in a group). */
5019
5020 *optionsptr = options = newoptions;
5021 previous = NULL; /* This item can't be repeated */
5022 continue; /* It is complete */
5023 }
5024
5025 /* If the options ended with ':' we are heading into a nested group
5026 with possible change of options. Such groups are non-capturing and are
5027 not assertions of any kind. All we need to do is skip over the ':';
5028 the newoptions value is handled below. */
5029
5030 bravalue = OP_BRA;
5031 ptr++;
5032 } /* End of switch for character following (? */
5033 } /* End of (? handling */
5034
5035 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5036 all unadorned brackets become non-capturing and behave like (?:...)
5037 brackets. */
5038
5039 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5040 {
5041 bravalue = OP_BRA;
5042 }
5043
5044 /* Else we have a capturing group. */
5045
5046 else
5047 {
5048 NUMBERED_GROUP:
5049 cd->bracount += 1;
5050 PUT2(code, 1+LINK_SIZE, cd->bracount);
5051 skipbytes = 2;
5052 }
5053
5054 /* Process nested bracketed regex. Assertions may not be repeated, but
5055 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5056 non-register variable in order to be able to pass its address because some
5057 compilers complain otherwise. Pass in a new setting for the ims options if
5058 they have changed. */
5059
5060 previous = (bravalue >= OP_ONCE)? code : NULL;
5061 *code = bravalue;
5062 tempcode = code;
5063 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5064 length_prevgroup = 0; /* Initialize for pre-compile phase */
5065
5066 if (!compile_regex(
5067 newoptions, /* The complete new option state */
5068 options & PCRE_IMS, /* The previous ims option state */
5069 &tempcode, /* Where to put code (updated) */
5070 &ptr, /* Input pointer (updated) */
5071 errorcodeptr, /* Where to put an error message */
5072 (bravalue == OP_ASSERTBACK ||
5073 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5074 reset_bracount, /* True if (?| group */
5075 skipbytes, /* Skip over bracket number */
5076 &subfirstbyte, /* For possible first char */
5077 &subreqbyte, /* For possible last char */
5078 bcptr, /* Current branch chain */
5079 cd, /* Tables block */
5080 (lengthptr == NULL)? NULL : /* Actual compile phase */
5081 &length_prevgroup /* Pre-compile phase */
5082 ))
5083 goto FAILED;
5084
5085 /* At the end of compiling, code is still pointing to the start of the
5086 group, while tempcode has been updated to point past the end of the group
5087 and any option resetting that may follow it. The pattern pointer (ptr)
5088 is on the bracket. */
5089
5090 /* If this is a conditional bracket, check that there are no more than
5091 two branches in the group, or just one if it's a DEFINE group. We do this
5092 in the real compile phase, not in the pre-pass, where the whole group may
5093 not be available. */
5094
5095 if (bravalue == OP_COND && lengthptr == NULL)
5096 {
5097 uschar *tc = code;
5098 int condcount = 0;
5099
5100 do {
5101 condcount++;
5102 tc += GET(tc,1);
5103 }
5104 while (*tc != OP_KET);
5105
5106 /* A DEFINE group is never obeyed inline (the "condition" is always
5107 false). It must have only one branch. */
5108
5109 if (code[LINK_SIZE+1] == OP_DEF)
5110 {
5111 if (condcount > 1)
5112 {
5113 *errorcodeptr = ERR54;
5114 goto FAILED;
5115 }
5116 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5117 }
5118
5119 /* A "normal" conditional group. If there is just one branch, we must not
5120 make use of its firstbyte or reqbyte, because this is equivalent to an
5121 empty second branch. */
5122
5123 else
5124 {
5125 if (condcount > 2)
5126 {
5127 *errorcodeptr = ERR27;
5128 goto FAILED;
5129 }
5130 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5131 }
5132 }
5133
5134 /* Error if hit end of pattern */
5135
5136 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5137 {
5138 *errorcodeptr = ERR14;
5139 goto FAILED;
5140 }
5141
5142 /* In the pre-compile phase, update the length by the length of the group,
5143 less the brackets at either end. Then reduce the compiled code to just a
5144 set of non-capturing brackets so that it doesn't use much memory if it is
5145 duplicated by a quantifier.*/
5146
5147 if (lengthptr != NULL)
5148 {
5149 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5150 {
5151 *errorcodeptr = ERR20;
5152 goto FAILED;
5153 }
5154 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5155 *code++ = OP_BRA;
5156 PUTINC(code, 0, 1 + LINK_SIZE);
5157 *code++ = OP_KET;
5158 PUTINC(code, 0, 1 + LINK_SIZE);
5159 break; /* No need to waste time with special character handling */
5160 }
5161
5162 /* Otherwise update the main code pointer to the end of the group. */
5163
5164 code = tempcode;
5165
5166 /* For a DEFINE group, required and first character settings are not
5167 relevant. */
5168
5169 if (bravalue == OP_DEF) break;
5170
5171 /* Handle updating of the required and first characters for other types of
5172 group. Update for normal brackets of all kinds, and conditions with two
5173 branches (see code above). If the bracket is followed by a quantifier with
5174 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5175 zerofirstbyte outside the main loop so that they can be accessed for the
5176 back off. */
5177
5178 zeroreqbyte = reqbyte;
5179 zerofirstbyte = firstbyte;
5180 groupsetfirstbyte = FALSE;
5181
5182 if (bravalue >= OP_ONCE)
5183 {
5184 /* If we have not yet set a firstbyte in this branch, take it from the
5185 subpattern, remembering that it was set here so that a repeat of more
5186 than one can replicate it as reqbyte if necessary. If the subpattern has
5187 no firstbyte, set "none" for the whole branch. In both cases, a zero
5188 repeat forces firstbyte to "none". */
5189
5190 if (firstbyte == REQ_UNSET)
5191 {
5192 if (subfirstbyte >= 0)
5193 {
5194 firstbyte = subfirstbyte;
5195 groupsetfirstbyte = TRUE;
5196 }
5197 else firstbyte = REQ_NONE;
5198 zerofirstbyte = REQ_NONE;
5199 }
5200
5201 /* If firstbyte was previously set, convert the subpattern's firstbyte
5202 into reqbyte if there wasn't one, using the vary flag that was in
5203 existence beforehand. */
5204
5205 else if (subfirstbyte >= 0 && subreqbyte < 0)
5206 subreqbyte = subfirstbyte | tempreqvary;
5207
5208 /* If the subpattern set a required byte (or set a first byte that isn't
5209 really the first byte - see above), set it. */
5210
5211 if (subreqbyte >= 0) reqbyte = subreqbyte;
5212 }
5213
5214 /* For a forward assertion, we take the reqbyte, if set. This can be
5215 helpful if the pattern that follows the assertion doesn't set a different
5216 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5217 for an assertion, however because it leads to incorrect effect for patterns
5218 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5219 of a firstbyte. This is overcome by a scan at the end if there's no
5220 firstbyte, looking for an asserted first char. */
5221
5222 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5223 break; /* End of processing '(' */
5224
5225
5226 /* ===================================================================*/
5227 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5228 are arranged to be the negation of the corresponding OP_values. For the
5229 back references, the values are ESC_REF plus the reference number. Only
5230 back references and those types that consume a character may be repeated.
5231 We can test for values between ESC_b and ESC_Z for the latter; this may
5232 have to change if any new ones are ever created. */
5233
5234 case CHAR_BACKSLASH:
5235 tempptr = ptr;
5236 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5237 if (*errorcodeptr != 0) goto FAILED;
5238
5239 if (c < 0)
5240 {
5241 if (-c == ESC_Q) /* Handle start of quoted string */
5242 {
5243 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5244 ptr += 2; /* avoid empty string */
5245 else inescq = TRUE;
5246 continue;
5247 }
5248
5249 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5250
5251 /* For metasequences that actually match a character, we disable the
5252 setting of a first character if it hasn't already been set. */
5253
5254 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5255 firstbyte = REQ_NONE;
5256
5257 /* Set values to reset to if this is followed by a zero repeat. */
5258
5259 zerofirstbyte = firstbyte;
5260 zeroreqbyte = reqbyte;
5261
5262 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5263 is a subroutine call by number (Oniguruma syntax). In fact, the value
5264 -ESC_g is returned only for these cases. So we don't need to check for <
5265 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5266 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5267 that is a synonym for a named back reference). */
5268
5269 if (-c == ESC_g)
5270 {
5271 const uschar *p;
5272 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5273 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5274 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5275
5276 /* These two statements stop the compiler for warning about possibly
5277 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5278 fact, because we actually check for a number below, the paths that
5279 would actually be in error are never taken. */
5280
5281 skipbytes = 0;
5282 reset_bracount = FALSE;
5283
5284 /* Test for a name */
5285
5286 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5287 {
5288 BOOL isnumber = TRUE;
5289 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5290 {
5291 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5292 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5293 }
5294 if (*p != terminator)
5295 {
5296 *errorcodeptr = ERR57;
5297 break;
5298 }
5299 if (isnumber)
5300 {
5301 ptr++;
5302 goto HANDLE_NUMERICAL_RECURSION;
5303 }
5304 is_recurse = TRUE;
5305 goto NAMED_REF_OR_RECURSE;
5306 }
5307
5308 /* Test a signed number in angle brackets or quotes. */
5309
5310 p = ptr + 2;
5311 while ((digitab[*p] & ctype_digit) != 0) p++;
5312 if (*p != terminator)
5313 {
5314 *errorcodeptr = ERR57;
5315 break;
5316 }
5317 ptr++;
5318 goto HANDLE_NUMERICAL_RECURSION;
5319 }
5320
5321 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5322 We also support \k{name} (.NET syntax) */
5323
5324 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5325 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5326 {
5327 is_recurse = FALSE;
5328 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5329 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5330 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5331 goto NAMED_REF_OR_RECURSE;
5332 }
5333
5334 /* Back references are handled specially; must disable firstbyte if
5335 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5336 ':' later. */
5337
5338 if (-c >= ESC_REF)
5339 {
5340 recno = -c - ESC_REF;
5341
5342 HANDLE_REFERENCE: /* Come here from named backref handling */
5343 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5344 previous = code;
5345 *code++ = OP_REF;
5346 PUT2INC(code, 0, recno);
5347 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5348 if (recno > cd->top_backref) cd->top_backref = recno;
5349 }
5350
5351 /* So are Unicode property matches, if supported. */
5352
5353 #ifdef SUPPORT_UCP
5354 else if (-c == ESC_P || -c == ESC_p)
5355 {
5356 BOOL negated;
5357 int pdata;
5358 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5359 if (ptype < 0) goto FAILED;
5360 previous = code;
5361 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5362 *code++ = ptype;
5363 *code++ = pdata;
5364 }
5365 #else
5366
5367 /* If Unicode properties are not supported, \X, \P, and \p are not
5368 allowed. */
5369
5370 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5371 {
5372 *errorcodeptr = ERR45;
5373 goto FAILED;
5374 }
5375 #endif
5376
5377 /* For the rest (including \X when Unicode properties are supported), we
5378 can obtain the OP value by negating the escape value. */
5379
5380 else
5381 {
5382 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5383 *code++ = -c;
5384 }
5385 continue;
5386 }
5387
5388 /* We have a data character whose value is in c. In UTF-8 mode it may have
5389 a value > 127. We set its representation in the length/buffer, and then
5390 handle it as a data character. */
5391
5392 #ifdef SUPPORT_UTF8
5393 if (utf8 && c > 127)
5394 mclength = _pcre_ord2utf8(c, mcbuffer);
5395 else
5396 #endif
5397
5398 {
5399 mcbuffer[0] = c;
5400 mclength = 1;
5401 }
5402 goto ONE_CHAR;
5403
5404
5405 /* ===================================================================*/
5406 /* Handle a literal character. It is guaranteed not to be whitespace or #
5407 when the extended flag is set. If we are in UTF-8 mode, it may be a
5408 multi-byte literal character. */
5409
5410 default:
5411 NORMAL_CHAR:
5412 mclength = 1;
5413 mcbuffer[0] = c;
5414
5415 #ifdef SUPPORT_UTF8
5416 if (utf8 && c >= 0xc0)
5417 {
5418 while ((ptr[1] & 0xc0) == 0x80)
5419 mcbuffer[mclength++] = *(++ptr);
5420 }
5421 #endif
5422
5423 /* At this point we have the character's bytes in mcbuffer, and the length
5424 in mclength. When not in UTF-8 mode, the length is always 1. */
5425
5426 ONE_CHAR:
5427 previous = code;
5428 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5429 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5430
5431 /* Remember if \r or \n were seen */
5432
5433 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5434 cd->external_flags |= PCRE_HASCRORLF;
5435
5436 /* Set the first and required bytes appropriately. If no previous first
5437 byte, set it from this character, but revert to none on a zero repeat.
5438 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5439 repeat. */
5440
5441 if (firstbyte == REQ_UNSET)
5442 {
5443 zerofirstbyte = REQ_NONE;
5444 zeroreqbyte = reqbyte;
5445
5446 /* If the character is more than one byte long, we can set firstbyte
5447 only if it is not to be matched caselessly. */
5448
5449 if (mclength == 1 || req_caseopt == 0)
5450 {
5451 firstbyte = mcbuffer[0] | req_caseopt;
5452 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5453 }
5454 else firstbyte = reqbyte = REQ_NONE;
5455 }
5456
5457 /* firstbyte was previously set; we can set reqbyte only the length is
5458 1 or the matching is caseful. */
5459
5460 else
5461 {
5462 zerofirstbyte = firstbyte;
5463 zeroreqbyte = reqbyte;
5464 if (mclength == 1 || req_caseopt == 0)
5465 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5466 }
5467
5468 break; /* End of literal character handling */
5469 }
5470 } /* end of big loop */
5471
5472
5473 /* Control never reaches here by falling through, only by a goto for all the
5474 error states. Pass back the position in the pattern so that it can be displayed
5475 to the user for diagnosing the error. */
5476
5477 FAILED:
5478 *ptrptr = ptr;
5479 return FALSE;
5480 }
5481
5482
5483
5484
5485 /*************************************************
5486 * Compile sequence of alternatives *
5487 *************************************************/
5488
5489 /* On entry, ptr is pointing past the bracket character, but on return it
5490 points to the closing bracket, or vertical bar, or end of string. The code
5491 variable is pointing at the byte into which the BRA operator has been stored.
5492 If the ims options are changed at the start (for a (?ims: group) or during any
5493 branch, we need to insert an OP_OPT item at the start of every following branch
5494 to ensure they get set correctly at run time, and also pass the new options
5495 into every subsequent branch compile.
5496
5497 This function is used during the pre-compile phase when we are trying to find
5498 out the amount of memory needed, as well as during the real compile phase. The
5499 value of lengthptr distinguishes the two phases.
5500
5501 Arguments:
5502 options option bits, including any changes for this subpattern
5503 oldims previous settings of ims option bits
5504 codeptr -> the address of the current code pointer
5505 ptrptr -> the address of the current pattern pointer
5506 errorcodeptr -> pointer to error code variable
5507 lookbehind TRUE if this is a lookbehind assertion
5508 reset_bracount TRUE to reset the count for each branch
5509 skipbytes skip this many bytes at start (for brackets and OP_COND)
5510 firstbyteptr place to put the first required character, or a negative number
5511 reqbyteptr place to put the last required character, or a negative number
5512 bcptr pointer to the chain of currently open branches
5513 cd points to the data block with tables pointers etc.
5514 lengthptr NULL during the real compile phase
5515 points to length accumulator during pre-compile phase
5516
5517 Returns: TRUE on success
5518 */
5519
5520 static BOOL
5521 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5522 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5523 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5524 int *lengthptr)
5525 {
5526 const uschar *ptr = *ptrptr;
5527 uschar *code = *codeptr;
5528 uschar *last_branch = code;
5529 uschar *start_bracket = code;
5530 uschar *reverse_count = NULL;
5531 int firstbyte, reqbyte;
5532 int branchfirstbyte, branchreqbyte;
5533 int length;
5534 int orig_bracount;
5535 int max_bracount;
5536 branch_chain bc;
5537
5538 bc.outer = bcptr;
5539 bc.current = code;
5540
5541 firstbyte = reqbyte = REQ_UNSET;
5542
5543 /* Accumulate the length for use in the pre-compile phase. Start with the
5544 length of the BRA and KET and any extra bytes that are required at the
5545 beginning. We accumulate in a local variable to save frequent testing of
5546 lenthptr for NULL. We cannot do this by looking at the value of code at the
5547 start and end of each alternative, because compiled items are discarded during
5548 the pre-compile phase so that the work space is not exceeded. */
5549
5550 length = 2 + 2*LINK_SIZE + skipbytes;
5551
5552 /* WARNING: If the above line is changed for any reason, you must also change
5553 the code that abstracts option settings at the start of the pattern and makes
5554 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5555 pre-compile phase to find out whether anything has yet been compiled or not. */
5556
5557 /* Offset is set zero to mark that this bracket is still open */
5558
5559 PUT(code, 1, 0);
5560 code += 1 + LINK_SIZE + skipbytes;
5561
5562 /* Loop for each alternative branch */
5563
5564 orig_bracount = max_bracount = cd->bracount;
5565 for (;;)
5566 {
5567 /* For a (?| group, reset the capturing bracket count so that each branch
5568 uses the same numbers. */
5569
5570 if (reset_bracount) cd->bracount = orig_bracount;
5571
5572 /* Handle a change of ims options at the start of the branch */
5573
5574 if ((options & PCRE_IMS) != oldims)
5575 {
5576 *code++ = OP_OPT;
5577 *code++ = options & PCRE_IMS;
5578 length += 2;
5579 }
5580
5581 /* Set up dummy OP_REVERSE if lookbehind assertion */
5582
5583 if (lookbehind)
5584 {
5585 *code++ = OP_REVERSE;
5586 reverse_count = code;
5587 PUTINC(code, 0, 0);
5588 length += 1 + LINK_SIZE;
5589 }
5590
5591 /* Now compile the branch; in the pre-compile phase its length gets added
5592 into the length. */
5593
5594 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5595 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5596 {
5597 *ptrptr = ptr;
5598 return FALSE;
5599 }
5600
5601 /* Keep the highest bracket count in case (?| was used and some branch
5602 has fewer than the rest. */
5603
5604 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5605
5606 /* In the real compile phase, there is some post-processing to be done. */
5607
5608 if (lengthptr == NULL)
5609 {
5610 /* If this is the first branch, the firstbyte and reqbyte values for the
5611 branch become the values for the regex. */
5612
5613 if (*last_branch != OP_ALT)
5614 {
5615 firstbyte = branchfirstbyte;
5616 reqbyte = branchreqbyte;
5617 }
5618
5619 /* If this is not the first branch, the first char and reqbyte have to
5620 match the values from all the previous branches, except that if the
5621 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5622 and we set REQ_VARY for the regex. */
5623
5624 else
5625 {
5626 /* If we previously had a firstbyte, but it doesn't match the new branch,
5627 we have to abandon the firstbyte for the regex, but if there was
5628 previously no reqbyte, it takes on the value of the old firstbyte. */
5629
5630 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5631 {
5632 if (reqbyte < 0) reqbyte = firstbyte;
5633 firstbyte = REQ_NONE;
5634 }
5635
5636 /* If we (now or from before) have no firstbyte, a firstbyte from the
5637 branch becomes a reqbyte if there isn't a branch reqbyte. */
5638
5639 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5640 branchreqbyte = branchfirstbyte;
5641
5642 /* Now ensure that the reqbytes match */
5643
5644 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5645 reqbyte = REQ_NONE;
5646 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5647 }
5648
5649 /* If lookbehind, check that this branch matches a fixed-length string, and
5650 put the length into the OP_REVERSE item. Temporarily mark the end of the
5651 branch with OP_END. */
5652
5653 if (lookbehind)
5654 {
5655 int fixed_length;
5656 *code = OP_END;
5657 fixed_length = find_fixedlength(last_branch, options);
5658 DPRINTF(("fixed length = %d\n", fixed_length));
5659 if (fixed_length < 0)
5660 {
5661 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5662 *ptrptr = ptr;
5663 return FALSE;
5664 }
5665 PUT(reverse_count, 0, fixed_length);
5666 }
5667 }
5668
5669 /* Reached end of expression, either ')' or end of pattern. In the real
5670 compile phase, go back through the alternative branches and reverse the chain
5671 of offsets, with the field in the BRA item now becoming an offset to the
5672 first alternative. If there are no alternatives, it points to the end of the
5673 group. The length in the terminating ket is always the length of the whole
5674 bracketed item. If any of the ims options were changed inside the group,
5675 compile a resetting op-code following, except at the very end of the pattern.
5676 Return leaving the pointer at the terminating char. */
5677
5678 if (*ptr != CHAR_VERTICAL_LINE)
5679 {
5680 if (lengthptr == NULL)
5681 {
5682 int branch_length = code - last_branch;
5683 do
5684 {
5685 int prev_length = GET(last_branch, 1);
5686 PUT(last_branch, 1, branch_length);
5687 branch_length = prev_length;
5688 last_branch -= branch_length;
5689 }
5690 while (branch_length > 0);
5691 }
5692
5693 /* Fill in the ket */
5694
5695 *code = OP_KET;
5696 PUT(code, 1, code - start_bracket);
5697 code += 1 + LINK_SIZE;
5698
5699 /* Resetting option if needed */
5700
5701 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
5702 {
5703 *code++ = OP_OPT;
5704 *code++ = oldims;
5705 length += 2;
5706 }
5707
5708 /* Retain the highest bracket number, in case resetting was used. */
5709
5710 cd->bracount = max_bracount;
5711
5712 /* Set values to pass back */
5713
5714 *codeptr = code;
5715 *ptrptr = ptr;
5716 *firstbyteptr = firstbyte;
5717 *reqbyteptr = reqbyte;
5718 if (lengthptr != NULL)
5719 {
5720 if (OFLOW_MAX - *lengthptr < length)
5721 {
5722 *errorcodeptr = ERR20;
5723 return FALSE;
5724 }
5725 *lengthptr += length;
5726 }
5727 return TRUE;
5728 }
5729
5730 /* Another branch follows. In the pre-compile phase, we can move the code
5731 pointer back to where it was for the start of the first branch. (That is,
5732 pretend that each branch is the only one.)
5733
5734 In the real compile phase, insert an ALT node. Its length field points back
5735 to the previous branch while the bracket remains open. At the end the chain
5736 is reversed. It's done like this so that the start of the bracket has a
5737 zero offset until it is closed, making it possible to detect recursion. */
5738
5739 if (lengthptr != NULL)
5740 {
5741 code = *codeptr + 1 + LINK_SIZE + skipbytes;
5742 length += 1 + LINK_SIZE;
5743 }
5744 else
5745 {
5746 *code = OP_ALT;
5747 PUT(code, 1, code - last_branch);
5748 bc.current = last_branch = code;
5749 code += 1 + LINK_SIZE;
5750 }
5751
5752 ptr++;
5753 }
5754 /* Control never reaches here */
5755 }
5756
5757
5758
5759
5760 /*************************************************
5761 * Check for anchored expression *
5762 *************************************************/
5763
5764 /* Try to find out if this is an anchored regular expression. Consider each
5765 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
5766 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
5767 it's anchored. However, if this is a multiline pattern, then only OP_SOD
5768 counts, since OP_CIRC can match in the middle.
5769
5770 We can also consider a regex to be anchored if OP_SOM starts all its branches.
5771 This is the code for \G, which means "match at start of match position, taking
5772 into account the match offset".
5773
5774 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
5775 because that will try the rest of the pattern at all possible matching points,
5776 so there is no point trying again.... er ....
5777
5778 .... except when the .* appears inside capturing parentheses, and there is a
5779 subsequent back reference to those parentheses. We haven't enough information
5780 to catch that case precisely.
5781
5782 At first, the best we could do was to detect when .* was in capturing brackets
5783 and the highest back reference was greater than or equal to that level.
5784 However, by keeping a bitmap of the first 31 back references, we can catch some
5785 of the more common cases more precisely.
5786
5787 Arguments:
5788 code points to start of expression (the bracket)
5789 options points to the options setting
5790 bracket_map a bitmap of which brackets we are inside while testing; this
5791 handles up to substring 31; after that we just have to take
5792 the less precise approach
5793 backref_map the back reference bitmap
5794
5795 Returns: TRUE or FALSE
5796 */
5797
5798 static BOOL
5799 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
5800 unsigned int backref_map)
5801 {
5802 do {
5803 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5804 options, PCRE_MULTILINE, FALSE);
5805 register int op = *scode;
5806
5807 /* Non-capturing brackets */
5808
5809 if (op == OP_BRA)
5810 {
5811 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5812 }
5813
5814 /* Capturing brackets */
5815
5816 else if (op == OP_CBRA)
5817 {
5818 int n = GET2(scode, 1+LINK_SIZE);
5819 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5820 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
5821 }
5822
5823 /* Other brackets */
5824
5825 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
5826 {
5827 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5828 }
5829
5830 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
5831 it isn't in brackets that are or may be referenced. */
5832
5833 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
5834 op == OP_TYPEPOSSTAR))
5835 {
5836 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
5837 return FALSE;
5838 }
5839
5840 /* Check for explicit anchoring */
5841
5842 else if (op != OP_SOD && op != OP_SOM &&
5843 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
5844 return FALSE;
5845 code += GET(code, 1);
5846 }
5847 while (*code == OP_ALT); /* Loop for each alternative */
5848 return TRUE;
5849 }
5850
5851
5852
5853 /*************************************************
5854 * Check for starting with ^ or .* *
5855 *************************************************/
5856
5857 /* This is called to find out if every branch starts with ^ or .* so that
5858 "first char" processing can be done to speed things up in multiline
5859 matching and for non-DOTALL patterns that start with .* (which must start at
5860 the beginning or after \n). As in the case of is_anchored() (see above), we
5861 have to take account of back references to capturing brackets that contain .*
5862 because in that case we can't make the assumption.
5863
5864 Arguments:
5865 code points to start of expression (the bracket)
5866 bracket_map a bitmap of which brackets we are inside while testing; this
5867 handles up to substring 31; after that we just have to take
5868 the less precise approach
5869 backref_map the back reference bitmap
5870
5871 Returns: TRUE or FALSE
5872 */
5873
5874 static BOOL
5875 is_startline(const uschar *code, unsigned int bracket_map,
5876 unsigned int backref_map)
5877 {
5878 do {
5879 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5880 NULL, 0, FALSE);
5881 register int op = *scode;
5882
5883 /* If we are at the start of a conditional assertion group, *both* the
5884 conditional assertion *and* what follows the condition must satisfy the test
5885 for start of line. Other kinds of condition fail. Note that there may be an
5886 auto-callout at the start of a condition. */
5887
5888 if (op == OP_COND)
5889 {
5890 scode += 1 + LINK_SIZE;
5891 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
5892 switch (*scode)
5893 {
5894 case OP_CREF:
5895 case OP_RREF:
5896 case OP_DEF:
5897 return FALSE;
5898
5899 default: /* Assertion */
5900 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5901 do scode += GET(scode, 1); while (*scode == OP_ALT);
5902 scode += 1 + LINK_SIZE;
5903 break;
5904 }
5905 scode = first_significant_code(scode, NULL, 0, FALSE);
5906 op = *scode;
5907 }
5908
5909 /* Non-capturing brackets */
5910
5911 if (op == OP_BRA)
5912 {
5913 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5914 }
5915
5916 /* Capturing brackets */
5917
5918 else if (op == OP_CBRA)
5919 {
5920 int n = GET2(scode, 1+LINK_SIZE);
5921 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5922 if (!is_startline(scode, new_map, backref_map)) return FALSE;
5923 }
5924
5925 /* Other brackets */
5926
5927 else if (op == OP_ASSERT || op == OP_ONCE)
5928 {
5929 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5930 }
5931
5932 /* .* means "start at start or after \n" if it isn't in brackets that
5933 may be referenced. */
5934
5935 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
5936 {
5937 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
5938 }
5939
5940 /* Check for explicit circumflex */
5941
5942 else if (op != OP_CIRC) return FALSE;
5943
5944 /* Move on to the next alternative */
5945
5946 code += GET(code, 1);
5947 }
5948 while (*code == OP_ALT); /* Loop for each alternative */
5949 return TRUE;
5950 }
5951
5952
5953
5954 /*************************************************
5955 * Check for asserted fixed first char *
5956 *************************************************/
5957
5958 /* During compilation, the "first char" settings from forward assertions are
5959 discarded, because they can cause conflicts with actual literals that follow.
5960 However, if we end up without a first char setting for an unanchored pattern,
5961 it is worth scanning the regex to see if there is an initial asserted first
5962 char. If all branches start with the same asserted char, or with a bracket all
5963 of whose alternatives start with the same asserted char (recurse ad lib), then
5964 we return that char, otherwise -1.
5965
5966 Arguments:
5967 code points to start of expression (the bracket)
5968 options pointer to the options (used to check casing changes)
5969 inassert TRUE if in an assertion
5970
5971 Returns: -1 or the fixed first char
5972 */
5973
5974 static int
5975 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
5976 {
5977 register int c = -1;
5978 do {
5979 int d;
5980 const uschar *scode =
5981 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
5982 register int op = *scode;
5983
5984 switch(op)
5985 {
5986 default:
5987 return -1;
5988
5989 case OP_BRA:
5990 case OP_CBRA:
5991 case OP_ASSERT:
5992 case OP_ONCE:
5993 case OP_COND:
5994 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
5995 return -1;
5996 if (c < 0) c = d; else if (c != d) return -1;
5997 break;
5998
5999 case OP_EXACT: /* Fall through */
6000 scode += 2;
6001
6002 case OP_CHAR:
6003 case OP_CHARNC:
6004 case OP_PLUS:
6005 case OP_MINPLUS:
6006 case OP_POSPLUS:
6007 if (!inassert) return -1;
6008 if (c < 0)
6009 {
6010 c = scode[1];
6011 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6012 }
6013 else if (c != scode[1]) return -1;
6014 break;
6015 }
6016
6017 code += GET(code, 1);
6018 }
6019 while (*code == OP_ALT);
6020 return c;
6021 }
6022
6023
6024
6025 /*************************************************
6026 * Compile a Regular Expression *
6027 *************************************************/
6028
6029 /* This function takes a string and returns a pointer to a block of store
6030 holding a compiled version of the expression. The original API for this
6031 function had no error code return variable; it is retained for backwards
6032 compatibility. The new function is given a new name.
6033
6034 Arguments:
6035 pattern the regular expression
6036 options various option bits
6037 errorcodeptr pointer to error code variable (pcre_compile2() only)
6038 can be NULL if you don't want a code value
6039 errorptr pointer to pointer to error text
6040 erroroffset ptr offset in pattern where error was detected
6041 tables pointer to character tables or NULL
6042
6043 Returns: pointer to compiled data block, or NULL on error,
6044 with errorptr and erroroffset set
6045 */
6046
6047 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6048 pcre_compile(const char *pattern, int options, const char **errorptr,
6049 int *erroroffset, const unsigned char *tables)
6050 {
6051 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6052 }
6053
6054
6055 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6056 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6057 const char **errorptr, int *erroroffset, const unsigned char *tables)
6058 {
6059 real_pcre *re;
6060 int length = 1; /* For final END opcode */
6061 int firstbyte, reqbyte, newline;
6062 int errorcode = 0;
6063 int skipatstart = 0;
6064 #ifdef SUPPORT_UTF8
6065 BOOL utf8;
6066 #endif
6067 size_t size;
6068 uschar *code;
6069 const uschar *codestart;
6070 const uschar *ptr;
6071 compile_data compile_block;
6072 compile_data *cd = &compile_block;
6073
6074 /* This space is used for "compiling" into during the first phase, when we are
6075 computing the amount of memory that is needed. Compiled items are thrown away
6076 as soon as possible, so that a fairly large buffer should be sufficient for
6077 this purpose. The same space is used in the second phase for remembering where
6078 to fill in forward references to subpatterns. */
6079
6080 uschar cworkspace[COMPILE_WORK_SIZE];
6081
6082 /* Set this early so that early errors get offset 0. */
6083
6084 ptr = (const uschar *)pattern;
6085
6086 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6087 can do is just return NULL, but we can set a code value if there is a code
6088 pointer. */
6089
6090 if (errorptr == NULL)
6091 {
6092 if (errorcodeptr != NULL) *errorcodeptr = 99;
6093 return NULL;
6094 }
6095
6096 *errorptr = NULL;
6097 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6098
6099 /* However, we can give a message for this error */
6100
6101 if (erroroffset == NULL)
6102 {
6103 errorcode = ERR16;
6104 goto PCRE_EARLY_ERROR_RETURN2;
6105 }
6106
6107 *erroroffset = 0;
6108
6109 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6110
6111 #ifdef SUPPORT_UTF8
6112 utf8 = (options & PCRE_UTF8) != 0;
6113 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6114 (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
6115 {
6116 errorcode = ERR44;
6117 goto PCRE_EARLY_ERROR_RETURN2;
6118 }
6119 #else
6120 if ((options & PCRE_UTF8) != 0)
6121 {
6122 errorcode = ERR32;
6123 goto PCRE_EARLY_ERROR_RETURN;
6124 }
6125 #endif
6126
6127 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6128 {
6129 errorcode = ERR17;
6130 goto PCRE_EARLY_ERROR_RETURN;
6131 }
6132
6133 /* Set up pointers to the individual character tables */
6134
6135 if (tables == NULL) tables = _pcre_default_tables;
6136 cd->lcc = tables + lcc_offset;
6137 cd->fcc = tables + fcc_offset;
6138 cd->cbits = tables + cbits_offset;
6139 cd->ctypes = tables + ctypes_offset;
6140
6141 /* Check for global one-time settings at the start of the pattern, and remember
6142 the offset for later. */
6143
6144 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6145 ptr[skipatstart+1] == CHAR_ASTERISK)
6146 {
6147 int newnl = 0;
6148 int newbsr = 0;
6149
6150 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6151 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6152 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6153 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6154 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6155 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6156 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6157 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6158 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6159 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6160
6161 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6162 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6163 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6164 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6165
6166 if (newnl != 0)
6167 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6168 else if (newbsr != 0)
6169 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6170 else break;
6171 }
6172
6173 /* Check validity of \R options. */
6174
6175 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6176 {
6177 case 0:
6178 case PCRE_BSR_ANYCRLF:
6179 case PCRE_BSR_UNICODE:
6180 break;
6181 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6182 }
6183
6184 /* Handle different types of newline. The three bits give seven cases. The
6185 current code allows for fixed one- or two-byte sequences, plus "any" and
6186 "anycrlf". */
6187
6188 switch (options & PCRE_NEWLINE_BITS)
6189 {
6190 case 0: newline = NEWLINE; break; /* Build-time default */
6191 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6192 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6193 case PCRE_NEWLINE_CR+
6194 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6195 case PCRE_NEWLINE_ANY: newline = -1; break;
6196 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6197 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6198 }
6199
6200 if (newline == -2)
6201 {
6202 cd->nltype = NLTYPE_ANYCRLF;
6203 }
6204 else if (newline < 0)
6205 {
6206 cd->nltype = NLTYPE_ANY;
6207 }
6208 else
6209 {
6210 cd->nltype = NLTYPE_FIXED;
6211 if (newline > 255)
6212 {
6213 cd->nllen = 2;
6214 cd->nl[0] = (newline >> 8) & 255;
6215 cd->nl[1] = newline & 255;
6216 }
6217 else
6218 {
6219 cd->nllen = 1;
6220 cd->nl[0] = newline;
6221 }
6222 }
6223
6224 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6225 references to help in deciding whether (.*) can be treated as anchored or not.
6226 */
6227
6228 cd->top_backref = 0;
6229 cd->backref_map = 0;
6230
6231 /* Reflect pattern for debugging output */
6232
6233 DPRINTF(("------------------------------------------------------------------\n"));
6234 DPRINTF(("%s\n", pattern));
6235
6236 /* Pretend to compile the pattern while actually just accumulating the length
6237 of memory required. This behaviour is triggered by passing a non-NULL final
6238 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6239 to compile parts of the pattern into; the compiled code is discarded when it is
6240 no longer needed, so hopefully this workspace will never overflow, though there
6241 is a test for its doing so. */
6242
6243 cd->bracount = cd->final_bracount = 0;
6244 cd->names_found = 0;
6245 cd->name_entry_size = 0;
6246 cd->name_table = NULL;
6247 cd->start_workspace = cworkspace;
6248 cd->start_code = cworkspace;
6249 cd->hwm = cworkspace;
6250 cd->start_pattern = (const uschar *)pattern;
6251 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6252 cd->req_varyopt = 0;
6253 cd->external_options = options;
6254 cd->external_flags = 0;
6255
6256 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6257 don't need to look at the result of the function here. The initial options have
6258 been put into the cd block so that they can be changed if an option setting is
6259 found within the regex right at the beginning. Bringing initial option settings
6260 outside can help speed up starting point checks. */
6261
6262 ptr += skipatstart;
6263 code = cworkspace;
6264 *code = OP_BRA;
6265 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6266 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6267 &length);
6268 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6269
6270 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6271 cd->hwm - cworkspace));
6272
6273 if (length > MAX_PATTERN_SIZE)
6274 {
6275 errorcode = ERR20;
6276 goto PCRE_EARLY_ERROR_RETURN;
6277 }
6278
6279 /* Compute the size of data block needed and get it, either from malloc or
6280 externally provided function. Integer overflow should no longer be possible
6281 because nowadays we limit the maximum value of cd->names_found and
6282 cd->name_entry_size. */
6283
6284 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6285 re = (real_pcre *)(pcre_malloc)(size);
6286
6287 if (re == NULL)
6288 {
6289 errorcode = ERR21;
6290 goto PCRE_EARLY_ERROR_RETURN;
6291 }
6292
6293 /* Put in the magic number, and save the sizes, initial options, internal
6294 flags, and character table pointer. NULL is used for the default character
6295 tables. The nullpad field is at the end; it's there to help in the case when a
6296 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6297 pointers. */
6298
6299 re->magic_number = MAGIC_NUMBER;
6300 re->size = size;
6301 re->options = cd->external_options;
6302 re->flags = cd->external_flags;
6303 re->dummy1 = 0;
6304 re->first_byte = 0;
6305 re->req_byte = 0;
6306 re->name_table_offset = sizeof(real_pcre);
6307 re->name_entry_size = cd->name_entry_size;
6308 re->name_count = cd->names_found;
6309 re->ref_count = 0;
6310 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6311 re->nullpad = NULL;
6312
6313 /* The starting points of the name/number translation table and of the code are
6314 passed around in the compile data block. The start/end pattern and initial
6315 options are already set from the pre-compile phase, as is the name_entry_size
6316 field. Reset the bracket count and the names_found field. Also reset the hwm
6317 field; this time it's used for remembering forward references to subpatterns.
6318 */
6319
6320 cd->final_bracount = cd->bracount; /* Save for checking forward references */
6321 cd->bracount = 0;
6322 cd->names_found = 0;
6323 cd->name_table = (uschar *)re + re->name_table_offset;
6324 codestart = cd->name_table + re->name_entry_size * re->name_count;
6325 cd->start_code = codestart;
6326 cd->hwm = cworkspace;
6327 cd->req_varyopt = 0;
6328 cd->had_accept = FALSE;
6329
6330 /* Set up a starting, non-extracting bracket, then compile the expression. On
6331 error, errorcode will be set non-zero, so we don't need to look at the result
6332 of the function here. */
6333
6334 ptr = (const uschar *)pattern + skipatstart;
6335 code = (uschar *)codestart;
6336 *code = OP_BRA;
6337 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
6338 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
6339 re->top_bracket = cd->bracount;
6340 re->top_backref = cd->top_backref;
6341 re->flags = cd->external_flags;
6342
6343 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
6344
6345 /* If not reached end of pattern on success, there's an excess bracket. */
6346
6347 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
6348
6349 /* Fill in the terminating state and check for disastrous overflow, but
6350 if debugging, leave the test till after things are printed out. */
6351
6352 *code++ = OP_END;
6353
6354 #ifndef DEBUG
6355 if (code - codestart > length) errorcode = ERR23;
6356 #endif
6357
6358 /* Fill in any forward references that are required. */
6359
6360 while (errorcode == 0 && cd->hwm > cworkspace)
6361 {
6362 int offset, recno;
6363 const uschar *groupptr;
6364 cd->hwm -= LINK_SIZE;
6365 offset = GET(cd->hwm, 0);
6366 recno = GET(codestart, offset);
6367 groupptr = find_bracket(codestart, (re->options & PCRE_UTF8) != 0, recno);
6368 if (groupptr == NULL) errorcode = ERR53;
6369 else PUT(((uschar *)codestart), offset, groupptr - codestart);
6370 }
6371
6372 /* Give an error if there's back reference to a non-existent capturing
6373 subpattern. */
6374
6375 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
6376
6377 /* Failed to compile, or error while post-processing */
6378
6379 if (errorcode != 0)
6380 {
6381 (pcre_free)(re);
6382 PCRE_EARLY_ERROR_RETURN:
6383 *erroroffset = ptr - (const uschar *)pattern;
6384 PCRE_EARLY_ERROR_RETURN2:
6385 *errorptr = find_error_text(errorcode);
6386 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
6387 return NULL;
6388 }
6389
6390 /* If the anchored option was not passed, set the flag if we can determine that
6391 the pattern is anchored by virtue of ^ characters or \A or anything else (such
6392 as starting with .* when DOTALL is set).
6393
6394 Otherwise, if we know what the first byte has to be, save it, because that
6395 speeds up unanchored matches no end. If not, see if we can set the
6396 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
6397 start with ^. and also when all branches start with .* for non-DOTALL matches.
6398 */
6399
6400 if ((re->options & PCRE_ANCHORED) == 0)
6401 {
6402 int temp_options = re->options; /* May get changed during these scans */
6403 if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
6404 re->options |= PCRE_ANCHORED;
6405 else
6406 {
6407 if (firstbyte < 0)
6408 firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
6409 if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */
6410 {
6411 int ch = firstbyte & 255;
6412 re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
6413 cd->fcc[ch] == ch)? ch : firstbyte;
6414 re->flags |= PCRE_FIRSTSET;
6415 }
6416 else if (is_startline(codestart, 0, cd->backref_map))
6417 re->flags |= PCRE_STARTLINE;
6418 }
6419 }
6420
6421 /* For an anchored pattern, we use the "required byte" only if it follows a
6422 variable length item in the regex. Remove the caseless flag for non-caseable
6423 bytes. */
6424
6425 if (reqbyte >= 0 &&
6426 ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0))
6427 {
6428 int ch = reqbyte & 255;
6429 re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
6430 cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
6431 re->flags |= PCRE_REQCHSET;
6432 }
6433
6434 /* Print out the compiled data if debugging is enabled. This is never the
6435 case when building a production library. */
6436
6437 #ifdef DEBUG
6438
6439 printf("Length = %d top_bracket = %d top_backref = %d\n",
6440 length, re->top_bracket, re->top_backref);
6441
6442 printf("Options=%08x\n", re->options);
6443
6444 if ((re->flags & PCRE_FIRSTSET) != 0)
6445 {
6446 int ch = re->first_byte & 255;
6447 const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
6448 "" : " (caseless)";
6449 if (isprint(ch)) printf("First char = %c%s\n", ch, caseless);
6450 else printf("First char = \\x%02x%s\n", ch, caseless);
6451 }
6452
6453 if ((re->flags & PCRE_REQCHSET) != 0)
6454 {
6455 int ch = re->req_byte & 255;
6456 const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
6457 "" : " (caseless)";
6458 if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless);
6459 else printf("Req char = \\x%02x%s\n", ch, caseless);
6460 }
6461
6462 pcre_printint(re, stdout, TRUE);
6463
6464 /* This check is done here in the debugging case so that the code that
6465 was compiled can be seen. */
6466
6467 if (code - codestart > length)
6468 {
6469 (pcre_free)(re);
6470 *errorptr = find_error_text(ERR23);
6471 *erroroffset = ptr - (uschar *)pattern;
6472 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
6473 return NULL;
6474 }
6475 #endif /* DEBUG */
6476
6477 return (pcre *)re;
6478 }
6479
6480 /* End of pcre_compile.c */

Properties

Name Value
svn:eol-style native
svn:keywords "Author Date Id Revision Url"

webmaster@exim.org
ViewVC Help
Powered by ViewVC 1.1.12