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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 345 - (show annotations) (download)
Mon Apr 28 15:10:02 2008 UTC (6 years, 4 months ago) by ph10
File MIME type: text/plain
File size: 202682 byte(s)
Tidies for the 7.7-RC1 distribution.

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