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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 269 - (show annotations) (download)
Fri Nov 16 16:22:24 2007 UTC (6 years, 8 months ago) by ph10
File MIME type: text/plain
File size: 195027 byte(s)
Improve error messages for (?+-a) and (?-+a).

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