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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 395 - (show annotations) (download)
Fri Mar 20 11:22:42 2009 UTC (5 years, 1 month ago) by ph10
File MIME type: text/plain
File size: 208090 byte(s)
Fix looping bug by recognizing that a conditional with only one branch may 
match an empty string.

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2009 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When DEBUG is defined, we need the pcre_printint() function, which is also
57 used by pcretest. DEBUG is not defined when building a production library. */
58
59 #ifdef DEBUG
60 #include "pcre_printint.src"
61 #endif
62
63
64 /* Macro for setting individual bits in class bitmaps. */
65
66 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
67
68 /* Maximum length value to check against when making sure that the integer that
69 holds the compiled pattern length does not overflow. We make it a bit less than
70 INT_MAX to allow for adding in group terminating bytes, so that we don't have
71 to check them every time. */
72
73 #define OFLOW_MAX (INT_MAX - 20)
74
75
76 /*************************************************
77 * Code parameters and static tables *
78 *************************************************/
79
80 /* This value specifies the size of stack workspace that is used during the
81 first pre-compile phase that determines how much memory is required. The regex
82 is partly compiled into this space, but the compiled parts are discarded as
83 soon as they can be, so that hopefully there will never be an overrun. The code
84 does, however, check for an overrun. The largest amount I've seen used is 218,
85 so this number is very generous.
86
87 The same workspace is used during the second, actual compile phase for
88 remembering forward references to groups so that they can be filled in at the
89 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
90 is 4 there is plenty of room. */
91
92 #define COMPILE_WORK_SIZE (4096)
93
94
95 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
96 are simple data values; negative values are for special things like \d and so
97 on. Zero means further processing is needed (for things like \x), or the escape
98 is invalid. */
99
100 #ifndef EBCDIC
101
102 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
103 in UTF-8 mode. */
104
105 static const short int escapes[] = {
106 0, 0,
107 0, 0,
108 0, 0,
109 0, 0,
110 0, 0,
111 CHAR_COLON, CHAR_SEMICOLON,
112 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
113 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
114 CHAR_COMMERCIAL_AT, -ESC_A,
115 -ESC_B, -ESC_C,
116 -ESC_D, -ESC_E,
117 0, -ESC_G,
118 -ESC_H, 0,
119 0, -ESC_K,
120 0, 0,
121 0, 0,
122 -ESC_P, -ESC_Q,
123 -ESC_R, -ESC_S,
124 0, 0,
125 -ESC_V, -ESC_W,
126 -ESC_X, 0,
127 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
128 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
129 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
130 CHAR_GRAVE_ACCENT, 7,
131 -ESC_b, 0,
132 -ESC_d, ESC_e,
133 ESC_f, 0,
134 -ESC_h, 0,
135 0, -ESC_k,
136 0, 0,
137 ESC_n, 0,
138 -ESC_p, 0,
139 ESC_r, -ESC_s,
140 ESC_tee, 0,
141 -ESC_v, -ESC_w,
142 0, 0,
143 -ESC_z
144 };
145
146 #else
147
148 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
149
150 static const short int escapes[] = {
151 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
152 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
153 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
154 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
155 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
156 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
157 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
158 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
159 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
160 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
161 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
162 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
163 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
164 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
165 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
166 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
167 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
168 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
169 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
170 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
171 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
172 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
173 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
174 };
175 #endif
176
177
178 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
179 searched linearly. Put all the names into a single string, in order to reduce
180 the number of relocations when a shared library is dynamically linked. The
181 string is built from string macros so that it works in UTF-8 mode on EBCDIC
182 platforms. */
183
184 typedef struct verbitem {
185 int len;
186 int op;
187 } verbitem;
188
189 static const char verbnames[] =
190 STRING_ACCEPT0
191 STRING_COMMIT0
192 STRING_F0
193 STRING_FAIL0
194 STRING_PRUNE0
195 STRING_SKIP0
196 STRING_THEN;
197
198 static const verbitem verbs[] = {
199 { 6, OP_ACCEPT },
200 { 6, OP_COMMIT },
201 { 1, OP_FAIL },
202 { 4, OP_FAIL },
203 { 5, OP_PRUNE },
204 { 4, OP_SKIP },
205 { 4, OP_THEN }
206 };
207
208 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
209
210
211 /* Tables of names of POSIX character classes and their lengths. The names are
212 now all in a single string, to reduce the number of relocations when a shared
213 library is dynamically loaded. The list of lengths is terminated by a zero
214 length entry. The first three must be alpha, lower, upper, as this is assumed
215 for handling case independence. */
216
217 static const char posix_names[] =
218 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
219 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
220 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
221 STRING_word0 STRING_xdigit;
222
223 static const uschar posix_name_lengths[] = {
224 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
225
226 /* Table of class bit maps for each POSIX class. Each class is formed from a
227 base map, with an optional addition or removal of another map. Then, for some
228 classes, there is some additional tweaking: for [:blank:] the vertical space
229 characters are removed, and for [:alpha:] and [:alnum:] the underscore
230 character is removed. The triples in the table consist of the base map offset,
231 second map offset or -1 if no second map, and a non-negative value for map
232 addition or a negative value for map subtraction (if there are two maps). The
233 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
234 remove vertical space characters, 2 => remove underscore. */
235
236 static const int posix_class_maps[] = {
237 cbit_word, cbit_digit, -2, /* alpha */
238 cbit_lower, -1, 0, /* lower */
239 cbit_upper, -1, 0, /* upper */
240 cbit_word, -1, 2, /* alnum - word without underscore */
241 cbit_print, cbit_cntrl, 0, /* ascii */
242 cbit_space, -1, 1, /* blank - a GNU extension */
243 cbit_cntrl, -1, 0, /* cntrl */
244 cbit_digit, -1, 0, /* digit */
245 cbit_graph, -1, 0, /* graph */
246 cbit_print, -1, 0, /* print */
247 cbit_punct, -1, 0, /* punct */
248 cbit_space, -1, 0, /* space */
249 cbit_word, -1, 0, /* word - a Perl extension */
250 cbit_xdigit,-1, 0 /* xdigit */
251 };
252
253
254 #define STRING(a) # a
255 #define XSTRING(s) STRING(s)
256
257 /* The texts of compile-time error messages. These are "char *" because they
258 are passed to the outside world. Do not ever re-use any error number, because
259 they are documented. Always add a new error instead. Messages marked DEAD below
260 are no longer used. This used to be a table of strings, but in order to reduce
261 the number of relocations needed when a shared library is loaded dynamically,
262 it is now one long string. We cannot use a table of offsets, because the
263 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
264 simply count through to the one we want - this isn't a performance issue
265 because these strings are used only when there is a compilation error. */
266
267 static const char error_texts[] =
268 "no error\0"
269 "\\ at end of pattern\0"
270 "\\c at end of pattern\0"
271 "unrecognized character follows \\\0"
272 "numbers out of order in {} quantifier\0"
273 /* 5 */
274 "number too big in {} quantifier\0"
275 "missing terminating ] for character class\0"
276 "invalid escape sequence in character class\0"
277 "range out of order in character class\0"
278 "nothing to repeat\0"
279 /* 10 */
280 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
281 "internal error: unexpected repeat\0"
282 "unrecognized character after (? or (?-\0"
283 "POSIX named classes are supported only within a class\0"
284 "missing )\0"
285 /* 15 */
286 "reference to non-existent subpattern\0"
287 "erroffset passed as NULL\0"
288 "unknown option bit(s) set\0"
289 "missing ) after comment\0"
290 "parentheses nested too deeply\0" /** DEAD **/
291 /* 20 */
292 "regular expression is too large\0"
293 "failed to get memory\0"
294 "unmatched parentheses\0"
295 "internal error: code overflow\0"
296 "unrecognized character after (?<\0"
297 /* 25 */
298 "lookbehind assertion is not fixed length\0"
299 "malformed number or name after (?(\0"
300 "conditional group contains more than two branches\0"
301 "assertion expected after (?(\0"
302 "(?R or (?[+-]digits must be followed by )\0"
303 /* 30 */
304 "unknown POSIX class name\0"
305 "POSIX collating elements are not supported\0"
306 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
307 "spare error\0" /** DEAD **/
308 "character value in \\x{...} sequence is too large\0"
309 /* 35 */
310 "invalid condition (?(0)\0"
311 "\\C not allowed in lookbehind assertion\0"
312 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
313 "number after (?C is > 255\0"
314 "closing ) for (?C expected\0"
315 /* 40 */
316 "recursive call could loop indefinitely\0"
317 "unrecognized character after (?P\0"
318 "syntax error in subpattern name (missing terminator)\0"
319 "two named subpatterns have the same name\0"
320 "invalid UTF-8 string\0"
321 /* 45 */
322 "support for \\P, \\p, and \\X has not been compiled\0"
323 "malformed \\P or \\p sequence\0"
324 "unknown property name after \\P or \\p\0"
325 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
326 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
327 /* 50 */
328 "repeated subpattern is too long\0" /** DEAD **/
329 "octal value is greater than \\377 (not in UTF-8 mode)\0"
330 "internal error: overran compiling workspace\0"
331 "internal error: previously-checked referenced subpattern not found\0"
332 "DEFINE group contains more than one branch\0"
333 /* 55 */
334 "repeating a DEFINE group is not allowed\0"
335 "inconsistent NEWLINE options\0"
336 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
337 "a numbered reference must not be zero\0"
338 "(*VERB) with an argument is not supported\0"
339 /* 60 */
340 "(*VERB) not recognized\0"
341 "number is too big\0"
342 "subpattern name expected\0"
343 "digit expected after (?+\0"
344 "] is an invalid data character in JavaScript compatibility mode";
345
346
347 /* Table to identify digits and hex digits. This is used when compiling
348 patterns. Note that the tables in chartables are dependent on the locale, and
349 may mark arbitrary characters as digits - but the PCRE compiling code expects
350 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
351 a private table here. It costs 256 bytes, but it is a lot faster than doing
352 character value tests (at least in some simple cases I timed), and in some
353 applications one wants PCRE to compile efficiently as well as match
354 efficiently.
355
356 For convenience, we use the same bit definitions as in chartables:
357
358 0x04 decimal digit
359 0x08 hexadecimal digit
360
361 Then we can use ctype_digit and ctype_xdigit in the code. */
362
363 #ifndef EBCDIC
364
365 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
366 UTF-8 mode. */
367
368 static const unsigned char digitab[] =
369 {
370 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
371 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
372 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
373 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
374 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
375 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
376 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
377 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
378 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
379 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
382 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
383 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
384 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
387 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
388 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
394 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
395 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
396 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
402
403 #else
404
405 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
406
407 static const unsigned char digitab[] =
408 {
409 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
410 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
411 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
412 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
413 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
414 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
420 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
421 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
422 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
423 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
424 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
425 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
426 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
431 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
433 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
439 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
440 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
441
442 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
443 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
444 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
445 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
447 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
451 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
452 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
454 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
456 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
459 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
460 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
461 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
462 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
463 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
464 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
465 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
466 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
467 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
468 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
469 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
470 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
471 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
472 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
473 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
474 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
475 #endif
476
477
478 /* Definition to allow mutual recursion */
479
480 static BOOL
481 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
482 int *, int *, branch_chain *, compile_data *, int *);
483
484
485
486 /*************************************************
487 * Find an error text *
488 *************************************************/
489
490 /* The error texts are now all in one long string, to save on relocations. As
491 some of the text is of unknown length, we can't use a table of offsets.
492 Instead, just count through the strings. This is not a performance issue
493 because it happens only when there has been a compilation error.
494
495 Argument: the error number
496 Returns: pointer to the error string
497 */
498
499 static const char *
500 find_error_text(int n)
501 {
502 const char *s = error_texts;
503 for (; n > 0; n--) while (*s++ != 0) {};
504 return s;
505 }
506
507
508 /*************************************************
509 * Handle escapes *
510 *************************************************/
511
512 /* This function is called when a \ has been encountered. It either returns a
513 positive value for a simple escape such as \n, or a negative value which
514 encodes one of the more complicated things such as \d. A backreference to group
515 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
516 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
517 ptr is pointing at the \. On exit, it is on the final character of the escape
518 sequence.
519
520 Arguments:
521 ptrptr points to the pattern position pointer
522 errorcodeptr points to the errorcode variable
523 bracount number of previous extracting brackets
524 options the options bits
525 isclass TRUE if inside a character class
526
527 Returns: zero or positive => a data character
528 negative => a special escape sequence
529 on error, errorcodeptr is set
530 */
531
532 static int
533 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
534 int options, BOOL isclass)
535 {
536 BOOL utf8 = (options & PCRE_UTF8) != 0;
537 const uschar *ptr = *ptrptr + 1;
538 int c, i;
539
540 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
541 ptr--; /* Set pointer back to the last byte */
542
543 /* If backslash is at the end of the pattern, it's an error. */
544
545 if (c == 0) *errorcodeptr = ERR1;
546
547 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
548 in a table. A non-zero result is something that can be returned immediately.
549 Otherwise further processing may be required. */
550
551 #ifndef EBCDIC /* ASCII/UTF-8 coding */
552 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
553 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
554
555 #else /* EBCDIC coding */
556 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
557 else if ((i = escapes[c - 0x48]) != 0) c = i;
558 #endif
559
560 /* Escapes that need further processing, or are illegal. */
561
562 else
563 {
564 const uschar *oldptr;
565 BOOL braced, negated;
566
567 switch (c)
568 {
569 /* A number of Perl escapes are not handled by PCRE. We give an explicit
570 error. */
571
572 case CHAR_l:
573 case CHAR_L:
574 case CHAR_N:
575 case CHAR_u:
576 case CHAR_U:
577 *errorcodeptr = ERR37;
578 break;
579
580 /* \g must be followed by one of a number of specific things:
581
582 (1) A number, either plain or braced. If positive, it is an absolute
583 backreference. If negative, it is a relative backreference. This is a Perl
584 5.10 feature.
585
586 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
587 is part of Perl's movement towards a unified syntax for back references. As
588 this is synonymous with \k{name}, we fudge it up by pretending it really
589 was \k.
590
591 (3) For Oniguruma compatibility we also support \g followed by a name or a
592 number either in angle brackets or in single quotes. However, these are
593 (possibly recursive) subroutine calls, _not_ backreferences. Just return
594 the -ESC_g code (cf \k). */
595
596 case CHAR_g:
597 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
598 {
599 c = -ESC_g;
600 break;
601 }
602
603 /* Handle the Perl-compatible cases */
604
605 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
606 {
607 const uschar *p;
608 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
609 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
610 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
611 {
612 c = -ESC_k;
613 break;
614 }
615 braced = TRUE;
616 ptr++;
617 }
618 else braced = FALSE;
619
620 if (ptr[1] == CHAR_MINUS)
621 {
622 negated = TRUE;
623 ptr++;
624 }
625 else negated = FALSE;
626
627 c = 0;
628 while ((digitab[ptr[1]] & ctype_digit) != 0)
629 c = c * 10 + *(++ptr) - CHAR_0;
630
631 if (c < 0) /* Integer overflow */
632 {
633 *errorcodeptr = ERR61;
634 break;
635 }
636
637 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
638 {
639 *errorcodeptr = ERR57;
640 break;
641 }
642
643 if (c == 0)
644 {
645 *errorcodeptr = ERR58;
646 break;
647 }
648
649 if (negated)
650 {
651 if (c > bracount)
652 {
653 *errorcodeptr = ERR15;
654 break;
655 }
656 c = bracount - (c - 1);
657 }
658
659 c = -(ESC_REF + c);
660 break;
661
662 /* The handling of escape sequences consisting of a string of digits
663 starting with one that is not zero is not straightforward. By experiment,
664 the way Perl works seems to be as follows:
665
666 Outside a character class, the digits are read as a decimal number. If the
667 number is less than 10, or if there are that many previous extracting
668 left brackets, then it is a back reference. Otherwise, up to three octal
669 digits are read to form an escaped byte. Thus \123 is likely to be octal
670 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
671 value is greater than 377, the least significant 8 bits are taken. Inside a
672 character class, \ followed by a digit is always an octal number. */
673
674 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
675 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
676
677 if (!isclass)
678 {
679 oldptr = ptr;
680 c -= CHAR_0;
681 while ((digitab[ptr[1]] & ctype_digit) != 0)
682 c = c * 10 + *(++ptr) - CHAR_0;
683 if (c < 0) /* Integer overflow */
684 {
685 *errorcodeptr = ERR61;
686 break;
687 }
688 if (c < 10 || c <= bracount)
689 {
690 c = -(ESC_REF + c);
691 break;
692 }
693 ptr = oldptr; /* Put the pointer back and fall through */
694 }
695
696 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
697 generates a binary zero byte and treats the digit as a following literal.
698 Thus we have to pull back the pointer by one. */
699
700 if ((c = *ptr) >= CHAR_8)
701 {
702 ptr--;
703 c = 0;
704 break;
705 }
706
707 /* \0 always starts an octal number, but we may drop through to here with a
708 larger first octal digit. The original code used just to take the least
709 significant 8 bits of octal numbers (I think this is what early Perls used
710 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
711 than 3 octal digits. */
712
713 case CHAR_0:
714 c -= CHAR_0;
715 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
716 c = c * 8 + *(++ptr) - CHAR_0;
717 if (!utf8 && c > 255) *errorcodeptr = ERR51;
718 break;
719
720 /* \x is complicated. \x{ddd} is a character number which can be greater
721 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
722 treated as a data character. */
723
724 case CHAR_x:
725 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
726 {
727 const uschar *pt = ptr + 2;
728 int count = 0;
729
730 c = 0;
731 while ((digitab[*pt] & ctype_xdigit) != 0)
732 {
733 register int cc = *pt++;
734 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
735 count++;
736
737 #ifndef EBCDIC /* ASCII/UTF-8 coding */
738 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
739 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
740 #else /* EBCDIC coding */
741 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
742 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
743 #endif
744 }
745
746 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
747 {
748 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
749 ptr = pt;
750 break;
751 }
752
753 /* If the sequence of hex digits does not end with '}', then we don't
754 recognize this construct; fall through to the normal \x handling. */
755 }
756
757 /* Read just a single-byte hex-defined char */
758
759 c = 0;
760 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
761 {
762 int cc; /* Some compilers don't like */
763 cc = *(++ptr); /* ++ in initializers */
764 #ifndef EBCDIC /* ASCII/UTF-8 coding */
765 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
766 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
767 #else /* EBCDIC coding */
768 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
769 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
770 #endif
771 }
772 break;
773
774 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
775 This coding is ASCII-specific, but then the whole concept of \cx is
776 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
777
778 case CHAR_c:
779 c = *(++ptr);
780 if (c == 0)
781 {
782 *errorcodeptr = ERR2;
783 break;
784 }
785
786 #ifndef EBCDIC /* ASCII/UTF-8 coding */
787 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
788 c ^= 0x40;
789 #else /* EBCDIC coding */
790 if (c >= CHAR_a && c <= CHAR_z) c += 64;
791 c ^= 0xC0;
792 #endif
793 break;
794
795 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
796 other alphanumeric following \ is an error if PCRE_EXTRA was set;
797 otherwise, for Perl compatibility, it is a literal. This code looks a bit
798 odd, but there used to be some cases other than the default, and there may
799 be again in future, so I haven't "optimized" it. */
800
801 default:
802 if ((options & PCRE_EXTRA) != 0) switch(c)
803 {
804 default:
805 *errorcodeptr = ERR3;
806 break;
807 }
808 break;
809 }
810 }
811
812 *ptrptr = ptr;
813 return c;
814 }
815
816
817
818 #ifdef SUPPORT_UCP
819 /*************************************************
820 * Handle \P and \p *
821 *************************************************/
822
823 /* This function is called after \P or \p has been encountered, provided that
824 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
825 pointing at the P or p. On exit, it is pointing at the final character of the
826 escape sequence.
827
828 Argument:
829 ptrptr points to the pattern position pointer
830 negptr points to a boolean that is set TRUE for negation else FALSE
831 dptr points to an int that is set to the detailed property value
832 errorcodeptr points to the error code variable
833
834 Returns: type value from ucp_type_table, or -1 for an invalid type
835 */
836
837 static int
838 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
839 {
840 int c, i, bot, top;
841 const uschar *ptr = *ptrptr;
842 char name[32];
843
844 c = *(++ptr);
845 if (c == 0) goto ERROR_RETURN;
846
847 *negptr = FALSE;
848
849 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
850 negation. */
851
852 if (c == CHAR_LEFT_CURLY_BRACKET)
853 {
854 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
855 {
856 *negptr = TRUE;
857 ptr++;
858 }
859 for (i = 0; i < (int)sizeof(name) - 1; i++)
860 {
861 c = *(++ptr);
862 if (c == 0) goto ERROR_RETURN;
863 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
864 name[i] = c;
865 }
866 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
867 name[i] = 0;
868 }
869
870 /* Otherwise there is just one following character */
871
872 else
873 {
874 name[0] = c;
875 name[1] = 0;
876 }
877
878 *ptrptr = ptr;
879
880 /* Search for a recognized property name using binary chop */
881
882 bot = 0;
883 top = _pcre_utt_size;
884
885 while (bot < top)
886 {
887 i = (bot + top) >> 1;
888 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
889 if (c == 0)
890 {
891 *dptr = _pcre_utt[i].value;
892 return _pcre_utt[i].type;
893 }
894 if (c > 0) bot = i + 1; else top = i;
895 }
896
897 *errorcodeptr = ERR47;
898 *ptrptr = ptr;
899 return -1;
900
901 ERROR_RETURN:
902 *errorcodeptr = ERR46;
903 *ptrptr = ptr;
904 return -1;
905 }
906 #endif
907
908
909
910
911 /*************************************************
912 * Check for counted repeat *
913 *************************************************/
914
915 /* This function is called when a '{' is encountered in a place where it might
916 start a quantifier. It looks ahead to see if it really is a quantifier or not.
917 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
918 where the ddds are digits.
919
920 Arguments:
921 p pointer to the first char after '{'
922
923 Returns: TRUE or FALSE
924 */
925
926 static BOOL
927 is_counted_repeat(const uschar *p)
928 {
929 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
930 while ((digitab[*p] & ctype_digit) != 0) p++;
931 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
932
933 if (*p++ != CHAR_COMMA) return FALSE;
934 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
935
936 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
937 while ((digitab[*p] & ctype_digit) != 0) p++;
938
939 return (*p == CHAR_RIGHT_CURLY_BRACKET);
940 }
941
942
943
944 /*************************************************
945 * Read repeat counts *
946 *************************************************/
947
948 /* Read an item of the form {n,m} and return the values. This is called only
949 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
950 so the syntax is guaranteed to be correct, but we need to check the values.
951
952 Arguments:
953 p pointer to first char after '{'
954 minp pointer to int for min
955 maxp pointer to int for max
956 returned as -1 if no max
957 errorcodeptr points to error code variable
958
959 Returns: pointer to '}' on success;
960 current ptr on error, with errorcodeptr set non-zero
961 */
962
963 static const uschar *
964 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
965 {
966 int min = 0;
967 int max = -1;
968
969 /* Read the minimum value and do a paranoid check: a negative value indicates
970 an integer overflow. */
971
972 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
973 if (min < 0 || min > 65535)
974 {
975 *errorcodeptr = ERR5;
976 return p;
977 }
978
979 /* Read the maximum value if there is one, and again do a paranoid on its size.
980 Also, max must not be less than min. */
981
982 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
983 {
984 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
985 {
986 max = 0;
987 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
988 if (max < 0 || max > 65535)
989 {
990 *errorcodeptr = ERR5;
991 return p;
992 }
993 if (max < min)
994 {
995 *errorcodeptr = ERR4;
996 return p;
997 }
998 }
999 }
1000
1001 /* Fill in the required variables, and pass back the pointer to the terminating
1002 '}'. */
1003
1004 *minp = min;
1005 *maxp = max;
1006 return p;
1007 }
1008
1009
1010
1011 /*************************************************
1012 * Find forward referenced subpattern *
1013 *************************************************/
1014
1015 /* This function scans along a pattern's text looking for capturing
1016 subpatterns, and counting them. If it finds a named pattern that matches the
1017 name it is given, it returns its number. Alternatively, if the name is NULL, it
1018 returns when it reaches a given numbered subpattern. This is used for forward
1019 references to subpatterns. We know that if (?P< is encountered, the name will
1020 be terminated by '>' because that is checked in the first pass.
1021
1022 Arguments:
1023 ptr current position in the pattern
1024 cd compile background data
1025 name name to seek, or NULL if seeking a numbered subpattern
1026 lorn name length, or subpattern number if name is NULL
1027 xmode TRUE if we are in /x mode
1028
1029 Returns: the number of the named subpattern, or -1 if not found
1030 */
1031
1032 static int
1033 find_parens(const uschar *ptr, compile_data *cd, const uschar *name, int lorn,
1034 BOOL xmode)
1035 {
1036 const uschar *thisname;
1037 int count = cd->bracount;
1038
1039 for (; *ptr != 0; ptr++)
1040 {
1041 int term;
1042
1043 /* Skip over backslashed characters and also entire \Q...\E */
1044
1045 if (*ptr == CHAR_BACKSLASH)
1046 {
1047 if (*(++ptr) == 0) return -1;
1048 if (*ptr == CHAR_Q) for (;;)
1049 {
1050 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1051 if (*ptr == 0) return -1;
1052 if (*(++ptr) == CHAR_E) break;
1053 }
1054 continue;
1055 }
1056
1057 /* Skip over character classes; this logic must be similar to the way they
1058 are handled for real. If the first character is '^', skip it. Also, if the
1059 first few characters (either before or after ^) are \Q\E or \E we skip them
1060 too. This makes for compatibility with Perl. Note the use of STR macros to
1061 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1062
1063 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1064 {
1065 BOOL negate_class = FALSE;
1066 for (;;)
1067 {
1068 int c = *(++ptr);
1069 if (c == CHAR_BACKSLASH)
1070 {
1071 if (ptr[1] == CHAR_E)
1072 ptr++;
1073 else if (strncmp((const char *)ptr+1,
1074 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1075 ptr += 3;
1076 else
1077 break;
1078 }
1079 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
1080 negate_class = TRUE;
1081 else break;
1082 }
1083
1084 /* If the next character is ']', it is a data character that must be
1085 skipped, except in JavaScript compatibility mode. */
1086
1087 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1088 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1089 ptr++;
1090
1091 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1092 {
1093 if (*ptr == 0) return -1;
1094 if (*ptr == CHAR_BACKSLASH)
1095 {
1096 if (*(++ptr) == 0) return -1;
1097 if (*ptr == CHAR_Q) for (;;)
1098 {
1099 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1100 if (*ptr == 0) return -1;
1101 if (*(++ptr) == CHAR_E) break;
1102 }
1103 continue;
1104 }
1105 }
1106 continue;
1107 }
1108
1109 /* Skip comments in /x mode */
1110
1111 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1112 {
1113 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1114 if (*ptr == 0) return -1;
1115 continue;
1116 }
1117
1118 /* An opening parens must now be a real metacharacter */
1119
1120 if (*ptr != CHAR_LEFT_PARENTHESIS) continue;
1121 if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1122 {
1123 count++;
1124 if (name == NULL && count == lorn) return count;
1125 continue;
1126 }
1127
1128 ptr += 2;
1129 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1130
1131 /* We have to disambiguate (?<! and (?<= from (?<name> */
1132
1133 if ((*ptr != CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_EXCLAMATION_MARK ||
1134 ptr[1] == CHAR_EQUALS_SIGN) && *ptr != CHAR_APOSTROPHE)
1135 continue;
1136
1137 count++;
1138
1139 if (name == NULL && count == lorn) return count;
1140 term = *ptr++;
1141 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1142 thisname = ptr;
1143 while (*ptr != term) ptr++;
1144 if (name != NULL && lorn == ptr - thisname &&
1145 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1146 return count;
1147 }
1148
1149 return -1;
1150 }
1151
1152
1153
1154 /*************************************************
1155 * Find first significant op code *
1156 *************************************************/
1157
1158 /* This is called by several functions that scan a compiled expression looking
1159 for a fixed first character, or an anchoring op code etc. It skips over things
1160 that do not influence this. For some calls, a change of option is important.
1161 For some calls, it makes sense to skip negative forward and all backward
1162 assertions, and also the \b assertion; for others it does not.
1163
1164 Arguments:
1165 code pointer to the start of the group
1166 options pointer to external options
1167 optbit the option bit whose changing is significant, or
1168 zero if none are
1169 skipassert TRUE if certain assertions are to be skipped
1170
1171 Returns: pointer to the first significant opcode
1172 */
1173
1174 static const uschar*
1175 first_significant_code(const uschar *code, int *options, int optbit,
1176 BOOL skipassert)
1177 {
1178 for (;;)
1179 {
1180 switch ((int)*code)
1181 {
1182 case OP_OPT:
1183 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1184 *options = (int)code[1];
1185 code += 2;
1186 break;
1187
1188 case OP_ASSERT_NOT:
1189 case OP_ASSERTBACK:
1190 case OP_ASSERTBACK_NOT:
1191 if (!skipassert) return code;
1192 do code += GET(code, 1); while (*code == OP_ALT);
1193 code += _pcre_OP_lengths[*code];
1194 break;
1195
1196 case OP_WORD_BOUNDARY:
1197 case OP_NOT_WORD_BOUNDARY:
1198 if (!skipassert) return code;
1199 /* Fall through */
1200
1201 case OP_CALLOUT:
1202 case OP_CREF:
1203 case OP_RREF:
1204 case OP_DEF:
1205 code += _pcre_OP_lengths[*code];
1206 break;
1207
1208 default:
1209 return code;
1210 }
1211 }
1212 /* Control never reaches here */
1213 }
1214
1215
1216
1217
1218 /*************************************************
1219 * Find the fixed length of a pattern *
1220 *************************************************/
1221
1222 /* Scan a pattern and compute the fixed length of subject that will match it,
1223 if the length is fixed. This is needed for dealing with backward assertions.
1224 In UTF8 mode, the result is in characters rather than bytes.
1225
1226 Arguments:
1227 code points to the start of the pattern (the bracket)
1228 options the compiling options
1229
1230 Returns: the fixed length, or -1 if there is no fixed length,
1231 or -2 if \C was encountered
1232 */
1233
1234 static int
1235 find_fixedlength(uschar *code, int options)
1236 {
1237 int length = -1;
1238
1239 register int branchlength = 0;
1240 register uschar *cc = code + 1 + LINK_SIZE;
1241
1242 /* Scan along the opcodes for this branch. If we get to the end of the
1243 branch, check the length against that of the other branches. */
1244
1245 for (;;)
1246 {
1247 int d;
1248 register int op = *cc;
1249 switch (op)
1250 {
1251 case OP_CBRA:
1252 case OP_BRA:
1253 case OP_ONCE:
1254 case OP_COND:
1255 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1256 if (d < 0) return d;
1257 branchlength += d;
1258 do cc += GET(cc, 1); while (*cc == OP_ALT);
1259 cc += 1 + LINK_SIZE;
1260 break;
1261
1262 /* Reached end of a branch; if it's a ket it is the end of a nested
1263 call. If it's ALT it is an alternation in a nested call. If it is
1264 END it's the end of the outer call. All can be handled by the same code. */
1265
1266 case OP_ALT:
1267 case OP_KET:
1268 case OP_KETRMAX:
1269 case OP_KETRMIN:
1270 case OP_END:
1271 if (length < 0) length = branchlength;
1272 else if (length != branchlength) return -1;
1273 if (*cc != OP_ALT) return length;
1274 cc += 1 + LINK_SIZE;
1275 branchlength = 0;
1276 break;
1277
1278 /* Skip over assertive subpatterns */
1279
1280 case OP_ASSERT:
1281 case OP_ASSERT_NOT:
1282 case OP_ASSERTBACK:
1283 case OP_ASSERTBACK_NOT:
1284 do cc += GET(cc, 1); while (*cc == OP_ALT);
1285 /* Fall through */
1286
1287 /* Skip over things that don't match chars */
1288
1289 case OP_REVERSE:
1290 case OP_CREF:
1291 case OP_RREF:
1292 case OP_DEF:
1293 case OP_OPT:
1294 case OP_CALLOUT:
1295 case OP_SOD:
1296 case OP_SOM:
1297 case OP_EOD:
1298 case OP_EODN:
1299 case OP_CIRC:
1300 case OP_DOLL:
1301 case OP_NOT_WORD_BOUNDARY:
1302 case OP_WORD_BOUNDARY:
1303 cc += _pcre_OP_lengths[*cc];
1304 break;
1305
1306 /* Handle literal characters */
1307
1308 case OP_CHAR:
1309 case OP_CHARNC:
1310 case OP_NOT:
1311 branchlength++;
1312 cc += 2;
1313 #ifdef SUPPORT_UTF8
1314 if ((options & PCRE_UTF8) != 0)
1315 {
1316 while ((*cc & 0xc0) == 0x80) cc++;
1317 }
1318 #endif
1319 break;
1320
1321 /* Handle exact repetitions. The count is already in characters, but we
1322 need to skip over a multibyte character in UTF8 mode. */
1323
1324 case OP_EXACT:
1325 branchlength += GET2(cc,1);
1326 cc += 4;
1327 #ifdef SUPPORT_UTF8
1328 if ((options & PCRE_UTF8) != 0)
1329 {
1330 while((*cc & 0x80) == 0x80) cc++;
1331 }
1332 #endif
1333 break;
1334
1335 case OP_TYPEEXACT:
1336 branchlength += GET2(cc,1);
1337 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1338 cc += 4;
1339 break;
1340
1341 /* Handle single-char matchers */
1342
1343 case OP_PROP:
1344 case OP_NOTPROP:
1345 cc += 2;
1346 /* Fall through */
1347
1348 case OP_NOT_DIGIT:
1349 case OP_DIGIT:
1350 case OP_NOT_WHITESPACE:
1351 case OP_WHITESPACE:
1352 case OP_NOT_WORDCHAR:
1353 case OP_WORDCHAR:
1354 case OP_ANY:
1355 case OP_ALLANY:
1356 branchlength++;
1357 cc++;
1358 break;
1359
1360 /* The single-byte matcher isn't allowed */
1361
1362 case OP_ANYBYTE:
1363 return -2;
1364
1365 /* Check a class for variable quantification */
1366
1367 #ifdef SUPPORT_UTF8
1368 case OP_XCLASS:
1369 cc += GET(cc, 1) - 33;
1370 /* Fall through */
1371 #endif
1372
1373 case OP_CLASS:
1374 case OP_NCLASS:
1375 cc += 33;
1376
1377 switch (*cc)
1378 {
1379 case OP_CRSTAR:
1380 case OP_CRMINSTAR:
1381 case OP_CRQUERY:
1382 case OP_CRMINQUERY:
1383 return -1;
1384
1385 case OP_CRRANGE:
1386 case OP_CRMINRANGE:
1387 if (GET2(cc,1) != GET2(cc,3)) return -1;
1388 branchlength += GET2(cc,1);
1389 cc += 5;
1390 break;
1391
1392 default:
1393 branchlength++;
1394 }
1395 break;
1396
1397 /* Anything else is variable length */
1398
1399 default:
1400 return -1;
1401 }
1402 }
1403 /* Control never gets here */
1404 }
1405
1406
1407
1408
1409 /*************************************************
1410 * Scan compiled regex for numbered bracket *
1411 *************************************************/
1412
1413 /* This little function scans through a compiled pattern until it finds a
1414 capturing bracket with the given number.
1415
1416 Arguments:
1417 code points to start of expression
1418 utf8 TRUE in UTF-8 mode
1419 number the required bracket number
1420
1421 Returns: pointer to the opcode for the bracket, or NULL if not found
1422 */
1423
1424 static const uschar *
1425 find_bracket(const uschar *code, BOOL utf8, int number)
1426 {
1427 for (;;)
1428 {
1429 register int c = *code;
1430 if (c == OP_END) return NULL;
1431
1432 /* XCLASS is used for classes that cannot be represented just by a bit
1433 map. This includes negated single high-valued characters. The length in
1434 the table is zero; the actual length is stored in the compiled code. */
1435
1436 if (c == OP_XCLASS) code += GET(code, 1);
1437
1438 /* Handle capturing bracket */
1439
1440 else if (c == OP_CBRA)
1441 {
1442 int n = GET2(code, 1+LINK_SIZE);
1443 if (n == number) return (uschar *)code;
1444 code += _pcre_OP_lengths[c];
1445 }
1446
1447 /* Otherwise, we can get the item's length from the table, except that for
1448 repeated character types, we have to test for \p and \P, which have an extra
1449 two bytes of parameters. */
1450
1451 else
1452 {
1453 switch(c)
1454 {
1455 case OP_TYPESTAR:
1456 case OP_TYPEMINSTAR:
1457 case OP_TYPEPLUS:
1458 case OP_TYPEMINPLUS:
1459 case OP_TYPEQUERY:
1460 case OP_TYPEMINQUERY:
1461 case OP_TYPEPOSSTAR:
1462 case OP_TYPEPOSPLUS:
1463 case OP_TYPEPOSQUERY:
1464 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1465 break;
1466
1467 case OP_TYPEUPTO:
1468 case OP_TYPEMINUPTO:
1469 case OP_TYPEEXACT:
1470 case OP_TYPEPOSUPTO:
1471 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1472 break;
1473 }
1474
1475 /* Add in the fixed length from the table */
1476
1477 code += _pcre_OP_lengths[c];
1478
1479 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1480 a multi-byte character. The length in the table is a minimum, so we have to
1481 arrange to skip the extra bytes. */
1482
1483 #ifdef SUPPORT_UTF8
1484 if (utf8) switch(c)
1485 {
1486 case OP_CHAR:
1487 case OP_CHARNC:
1488 case OP_EXACT:
1489 case OP_UPTO:
1490 case OP_MINUPTO:
1491 case OP_POSUPTO:
1492 case OP_STAR:
1493 case OP_MINSTAR:
1494 case OP_POSSTAR:
1495 case OP_PLUS:
1496 case OP_MINPLUS:
1497 case OP_POSPLUS:
1498 case OP_QUERY:
1499 case OP_MINQUERY:
1500 case OP_POSQUERY:
1501 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1502 break;
1503 }
1504 #else
1505 (void)(utf8); /* Keep compiler happy by referencing function argument */
1506 #endif
1507 }
1508 }
1509 }
1510
1511
1512
1513 /*************************************************
1514 * Scan compiled regex for recursion reference *
1515 *************************************************/
1516
1517 /* This little function scans through a compiled pattern until it finds an
1518 instance of OP_RECURSE.
1519
1520 Arguments:
1521 code points to start of expression
1522 utf8 TRUE in UTF-8 mode
1523
1524 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1525 */
1526
1527 static const uschar *
1528 find_recurse(const uschar *code, BOOL utf8)
1529 {
1530 for (;;)
1531 {
1532 register int c = *code;
1533 if (c == OP_END) return NULL;
1534 if (c == OP_RECURSE) return code;
1535
1536 /* XCLASS is used for classes that cannot be represented just by a bit
1537 map. This includes negated single high-valued characters. The length in
1538 the table is zero; the actual length is stored in the compiled code. */
1539
1540 if (c == OP_XCLASS) code += GET(code, 1);
1541
1542 /* Otherwise, we can get the item's length from the table, except that for
1543 repeated character types, we have to test for \p and \P, which have an extra
1544 two bytes of parameters. */
1545
1546 else
1547 {
1548 switch(c)
1549 {
1550 case OP_TYPESTAR:
1551 case OP_TYPEMINSTAR:
1552 case OP_TYPEPLUS:
1553 case OP_TYPEMINPLUS:
1554 case OP_TYPEQUERY:
1555 case OP_TYPEMINQUERY:
1556 case OP_TYPEPOSSTAR:
1557 case OP_TYPEPOSPLUS:
1558 case OP_TYPEPOSQUERY:
1559 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1560 break;
1561
1562 case OP_TYPEPOSUPTO:
1563 case OP_TYPEUPTO:
1564 case OP_TYPEMINUPTO:
1565 case OP_TYPEEXACT:
1566 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1567 break;
1568 }
1569
1570 /* Add in the fixed length from the table */
1571
1572 code += _pcre_OP_lengths[c];
1573
1574 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1575 by a multi-byte character. The length in the table is a minimum, so we have
1576 to arrange to skip the extra bytes. */
1577
1578 #ifdef SUPPORT_UTF8
1579 if (utf8) switch(c)
1580 {
1581 case OP_CHAR:
1582 case OP_CHARNC:
1583 case OP_EXACT:
1584 case OP_UPTO:
1585 case OP_MINUPTO:
1586 case OP_POSUPTO:
1587 case OP_STAR:
1588 case OP_MINSTAR:
1589 case OP_POSSTAR:
1590 case OP_PLUS:
1591 case OP_MINPLUS:
1592 case OP_POSPLUS:
1593 case OP_QUERY:
1594 case OP_MINQUERY:
1595 case OP_POSQUERY:
1596 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1597 break;
1598 }
1599 #else
1600 (void)(utf8); /* Keep compiler happy by referencing function argument */
1601 #endif
1602 }
1603 }
1604 }
1605
1606
1607
1608 /*************************************************
1609 * Scan compiled branch for non-emptiness *
1610 *************************************************/
1611
1612 /* This function scans through a branch of a compiled pattern to see whether it
1613 can match the empty string or not. It is called from could_be_empty()
1614 below and from compile_branch() when checking for an unlimited repeat of a
1615 group that can match nothing. Note that first_significant_code() skips over
1616 backward and negative forward assertions when its final argument is TRUE. If we
1617 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1618 bracket whose current branch will already have been scanned.
1619
1620 Arguments:
1621 code points to start of search
1622 endcode points to where to stop
1623 utf8 TRUE if in UTF8 mode
1624
1625 Returns: TRUE if what is matched could be empty
1626 */
1627
1628 static BOOL
1629 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1630 {
1631 register int c;
1632 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1633 code < endcode;
1634 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1635 {
1636 const uschar *ccode;
1637
1638 c = *code;
1639
1640 /* Skip over forward assertions; the other assertions are skipped by
1641 first_significant_code() with a TRUE final argument. */
1642
1643 if (c == OP_ASSERT)
1644 {
1645 do code += GET(code, 1); while (*code == OP_ALT);
1646 c = *code;
1647 continue;
1648 }
1649
1650 /* Groups with zero repeats can of course be empty; skip them. */
1651
1652 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1653 {
1654 code += _pcre_OP_lengths[c];
1655 do code += GET(code, 1); while (*code == OP_ALT);
1656 c = *code;
1657 continue;
1658 }
1659
1660 /* For other groups, scan the branches. */
1661
1662 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1663 {
1664 BOOL empty_branch;
1665 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1666
1667 /* If a conditional group has only one branch, there is a second, implied,
1668 empty branch, so just skip over the conditional, because it could be empty.
1669 Otherwise, scan the individual branches of the group. */
1670
1671 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1672 code += GET(code, 1);
1673 else
1674 {
1675 empty_branch = FALSE;
1676 do
1677 {
1678 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1679 empty_branch = TRUE;
1680 code += GET(code, 1);
1681 }
1682 while (*code == OP_ALT);
1683 if (!empty_branch) return FALSE; /* All branches are non-empty */
1684 }
1685
1686 c = *code;
1687 continue;
1688 }
1689
1690 /* Handle the other opcodes */
1691
1692 switch (c)
1693 {
1694 /* Check for quantifiers after a class. XCLASS is used for classes that
1695 cannot be represented just by a bit map. This includes negated single
1696 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1697 actual length is stored in the compiled code, so we must update "code"
1698 here. */
1699
1700 #ifdef SUPPORT_UTF8
1701 case OP_XCLASS:
1702 ccode = code += GET(code, 1);
1703 goto CHECK_CLASS_REPEAT;
1704 #endif
1705
1706 case OP_CLASS:
1707 case OP_NCLASS:
1708 ccode = code + 33;
1709
1710 #ifdef SUPPORT_UTF8
1711 CHECK_CLASS_REPEAT:
1712 #endif
1713
1714 switch (*ccode)
1715 {
1716 case OP_CRSTAR: /* These could be empty; continue */
1717 case OP_CRMINSTAR:
1718 case OP_CRQUERY:
1719 case OP_CRMINQUERY:
1720 break;
1721
1722 default: /* Non-repeat => class must match */
1723 case OP_CRPLUS: /* These repeats aren't empty */
1724 case OP_CRMINPLUS:
1725 return FALSE;
1726
1727 case OP_CRRANGE:
1728 case OP_CRMINRANGE:
1729 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1730 break;
1731 }
1732 break;
1733
1734 /* Opcodes that must match a character */
1735
1736 case OP_PROP:
1737 case OP_NOTPROP:
1738 case OP_EXTUNI:
1739 case OP_NOT_DIGIT:
1740 case OP_DIGIT:
1741 case OP_NOT_WHITESPACE:
1742 case OP_WHITESPACE:
1743 case OP_NOT_WORDCHAR:
1744 case OP_WORDCHAR:
1745 case OP_ANY:
1746 case OP_ALLANY:
1747 case OP_ANYBYTE:
1748 case OP_CHAR:
1749 case OP_CHARNC:
1750 case OP_NOT:
1751 case OP_PLUS:
1752 case OP_MINPLUS:
1753 case OP_POSPLUS:
1754 case OP_EXACT:
1755 case OP_NOTPLUS:
1756 case OP_NOTMINPLUS:
1757 case OP_NOTPOSPLUS:
1758 case OP_NOTEXACT:
1759 case OP_TYPEPLUS:
1760 case OP_TYPEMINPLUS:
1761 case OP_TYPEPOSPLUS:
1762 case OP_TYPEEXACT:
1763 return FALSE;
1764
1765 /* These are going to continue, as they may be empty, but we have to
1766 fudge the length for the \p and \P cases. */
1767
1768 case OP_TYPESTAR:
1769 case OP_TYPEMINSTAR:
1770 case OP_TYPEPOSSTAR:
1771 case OP_TYPEQUERY:
1772 case OP_TYPEMINQUERY:
1773 case OP_TYPEPOSQUERY:
1774 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1775 break;
1776
1777 /* Same for these */
1778
1779 case OP_TYPEUPTO:
1780 case OP_TYPEMINUPTO:
1781 case OP_TYPEPOSUPTO:
1782 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1783 break;
1784
1785 /* End of branch */
1786
1787 case OP_KET:
1788 case OP_KETRMAX:
1789 case OP_KETRMIN:
1790 case OP_ALT:
1791 return TRUE;
1792
1793 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1794 MINUPTO, and POSUPTO may be followed by a multibyte character */
1795
1796 #ifdef SUPPORT_UTF8
1797 case OP_STAR:
1798 case OP_MINSTAR:
1799 case OP_POSSTAR:
1800 case OP_QUERY:
1801 case OP_MINQUERY:
1802 case OP_POSQUERY:
1803 case OP_UPTO:
1804 case OP_MINUPTO:
1805 case OP_POSUPTO:
1806 if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
1807 break;
1808 #endif
1809 }
1810 }
1811
1812 return TRUE;
1813 }
1814
1815
1816
1817 /*************************************************
1818 * Scan compiled regex for non-emptiness *
1819 *************************************************/
1820
1821 /* This function is called to check for left recursive calls. We want to check
1822 the current branch of the current pattern to see if it could match the empty
1823 string. If it could, we must look outwards for branches at other levels,
1824 stopping when we pass beyond the bracket which is the subject of the recursion.
1825
1826 Arguments:
1827 code points to start of the recursion
1828 endcode points to where to stop (current RECURSE item)
1829 bcptr points to the chain of current (unclosed) branch starts
1830 utf8 TRUE if in UTF-8 mode
1831
1832 Returns: TRUE if what is matched could be empty
1833 */
1834
1835 static BOOL
1836 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1837 BOOL utf8)
1838 {
1839 while (bcptr != NULL && bcptr->current >= code)
1840 {
1841 if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1842 bcptr = bcptr->outer;
1843 }
1844 return TRUE;
1845 }
1846
1847
1848
1849 /*************************************************
1850 * Check for POSIX class syntax *
1851 *************************************************/
1852
1853 /* This function is called when the sequence "[:" or "[." or "[=" is
1854 encountered in a character class. It checks whether this is followed by a
1855 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1856 reach an unescaped ']' without the special preceding character, return FALSE.
1857
1858 Originally, this function only recognized a sequence of letters between the
1859 terminators, but it seems that Perl recognizes any sequence of characters,
1860 though of course unknown POSIX names are subsequently rejected. Perl gives an
1861 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1862 didn't consider this to be a POSIX class. Likewise for [:1234:].
1863
1864 The problem in trying to be exactly like Perl is in the handling of escapes. We
1865 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1866 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1867 below handles the special case of \], but does not try to do any other escape
1868 processing. This makes it different from Perl for cases such as [:l\ower:]
1869 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1870 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1871 I think.
1872
1873 Arguments:
1874 ptr pointer to the initial [
1875 endptr where to return the end pointer
1876
1877 Returns: TRUE or FALSE
1878 */
1879
1880 static BOOL
1881 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1882 {
1883 int terminator; /* Don't combine these lines; the Solaris cc */
1884 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1885 for (++ptr; *ptr != 0; ptr++)
1886 {
1887 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
1888 {
1889 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
1890 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
1891 {
1892 *endptr = ptr;
1893 return TRUE;
1894 }
1895 }
1896 }
1897 return FALSE;
1898 }
1899
1900
1901
1902
1903 /*************************************************
1904 * Check POSIX class name *
1905 *************************************************/
1906
1907 /* This function is called to check the name given in a POSIX-style class entry
1908 such as [:alnum:].
1909
1910 Arguments:
1911 ptr points to the first letter
1912 len the length of the name
1913
1914 Returns: a value representing the name, or -1 if unknown
1915 */
1916
1917 static int
1918 check_posix_name(const uschar *ptr, int len)
1919 {
1920 const char *pn = posix_names;
1921 register int yield = 0;
1922 while (posix_name_lengths[yield] != 0)
1923 {
1924 if (len == posix_name_lengths[yield] &&
1925 strncmp((const char *)ptr, pn, len) == 0) return yield;
1926 pn += posix_name_lengths[yield] + 1;
1927 yield++;
1928 }
1929 return -1;
1930 }
1931
1932
1933 /*************************************************
1934 * Adjust OP_RECURSE items in repeated group *
1935 *************************************************/
1936
1937 /* OP_RECURSE items contain an offset from the start of the regex to the group
1938 that is referenced. This means that groups can be replicated for fixed
1939 repetition simply by copying (because the recursion is allowed to refer to
1940 earlier groups that are outside the current group). However, when a group is
1941 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
1942 inserted before it, after it has been compiled. This means that any OP_RECURSE
1943 items within it that refer to the group itself or any contained groups have to
1944 have their offsets adjusted. That one of the jobs of this function. Before it
1945 is called, the partially compiled regex must be temporarily terminated with
1946 OP_END.
1947
1948 This function has been extended with the possibility of forward references for
1949 recursions and subroutine calls. It must also check the list of such references
1950 for the group we are dealing with. If it finds that one of the recursions in
1951 the current group is on this list, it adjusts the offset in the list, not the
1952 value in the reference (which is a group number).
1953
1954 Arguments:
1955 group points to the start of the group
1956 adjust the amount by which the group is to be moved
1957 utf8 TRUE in UTF-8 mode
1958 cd contains pointers to tables etc.
1959 save_hwm the hwm forward reference pointer at the start of the group
1960
1961 Returns: nothing
1962 */
1963
1964 static void
1965 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
1966 uschar *save_hwm)
1967 {
1968 uschar *ptr = group;
1969
1970 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
1971 {
1972 int offset;
1973 uschar *hc;
1974
1975 /* See if this recursion is on the forward reference list. If so, adjust the
1976 reference. */
1977
1978 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
1979 {
1980 offset = GET(hc, 0);
1981 if (cd->start_code + offset == ptr + 1)
1982 {
1983 PUT(hc, 0, offset + adjust);
1984 break;
1985 }
1986 }
1987
1988 /* Otherwise, adjust the recursion offset if it's after the start of this
1989 group. */
1990
1991 if (hc >= cd->hwm)
1992 {
1993 offset = GET(ptr, 1);
1994 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
1995 }
1996
1997 ptr += 1 + LINK_SIZE;
1998 }
1999 }
2000
2001
2002
2003 /*************************************************
2004 * Insert an automatic callout point *
2005 *************************************************/
2006
2007 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2008 callout points before each pattern item.
2009
2010 Arguments:
2011 code current code pointer
2012 ptr current pattern pointer
2013 cd pointers to tables etc
2014
2015 Returns: new code pointer
2016 */
2017
2018 static uschar *
2019 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2020 {
2021 *code++ = OP_CALLOUT;
2022 *code++ = 255;
2023 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2024 PUT(code, LINK_SIZE, 0); /* Default length */
2025 return code + 2*LINK_SIZE;
2026 }
2027
2028
2029
2030 /*************************************************
2031 * Complete a callout item *
2032 *************************************************/
2033
2034 /* A callout item contains the length of the next item in the pattern, which
2035 we can't fill in till after we have reached the relevant point. This is used
2036 for both automatic and manual callouts.
2037
2038 Arguments:
2039 previous_callout points to previous callout item
2040 ptr current pattern pointer
2041 cd pointers to tables etc
2042
2043 Returns: nothing
2044 */
2045
2046 static void
2047 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2048 {
2049 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2050 PUT(previous_callout, 2 + LINK_SIZE, length);
2051 }
2052
2053
2054
2055 #ifdef SUPPORT_UCP
2056 /*************************************************
2057 * Get othercase range *
2058 *************************************************/
2059
2060 /* This function is passed the start and end of a class range, in UTF-8 mode
2061 with UCP support. It searches up the characters, looking for internal ranges of
2062 characters in the "other" case. Each call returns the next one, updating the
2063 start address.
2064
2065 Arguments:
2066 cptr points to starting character value; updated
2067 d end value
2068 ocptr where to put start of othercase range
2069 odptr where to put end of othercase range
2070
2071 Yield: TRUE when range returned; FALSE when no more
2072 */
2073
2074 static BOOL
2075 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2076 unsigned int *odptr)
2077 {
2078 unsigned int c, othercase, next;
2079
2080 for (c = *cptr; c <= d; c++)
2081 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2082
2083 if (c > d) return FALSE;
2084
2085 *ocptr = othercase;
2086 next = othercase + 1;
2087
2088 for (++c; c <= d; c++)
2089 {
2090 if (UCD_OTHERCASE(c) != next) break;
2091 next++;
2092 }
2093
2094 *odptr = next - 1;
2095 *cptr = c;
2096
2097 return TRUE;
2098 }
2099 #endif /* SUPPORT_UCP */
2100
2101
2102
2103 /*************************************************
2104 * Check if auto-possessifying is possible *
2105 *************************************************/
2106
2107 /* This function is called for unlimited repeats of certain items, to see
2108 whether the next thing could possibly match the repeated item. If not, it makes
2109 sense to automatically possessify the repeated item.
2110
2111 Arguments:
2112 op_code the repeated op code
2113 this data for this item, depends on the opcode
2114 utf8 TRUE in UTF-8 mode
2115 utf8_char used for utf8 character bytes, NULL if not relevant
2116 ptr next character in pattern
2117 options options bits
2118 cd contains pointers to tables etc.
2119
2120 Returns: TRUE if possessifying is wanted
2121 */
2122
2123 static BOOL
2124 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2125 const uschar *ptr, int options, compile_data *cd)
2126 {
2127 int next;
2128
2129 /* Skip whitespace and comments in extended mode */
2130
2131 if ((options & PCRE_EXTENDED) != 0)
2132 {
2133 for (;;)
2134 {
2135 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2136 if (*ptr == CHAR_NUMBER_SIGN)
2137 {
2138 while (*(++ptr) != 0)
2139 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2140 }
2141 else break;
2142 }
2143 }
2144
2145 /* If the next item is one that we can handle, get its value. A non-negative
2146 value is a character, a negative value is an escape value. */
2147
2148 if (*ptr == CHAR_BACKSLASH)
2149 {
2150 int temperrorcode = 0;
2151 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2152 if (temperrorcode != 0) return FALSE;
2153 ptr++; /* Point after the escape sequence */
2154 }
2155
2156 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2157 {
2158 #ifdef SUPPORT_UTF8
2159 if (utf8) { GETCHARINC(next, ptr); } else
2160 #endif
2161 next = *ptr++;
2162 }
2163
2164 else return FALSE;
2165
2166 /* Skip whitespace and comments in extended mode */
2167
2168 if ((options & PCRE_EXTENDED) != 0)
2169 {
2170 for (;;)
2171 {
2172 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2173 if (*ptr == CHAR_NUMBER_SIGN)
2174 {
2175 while (*(++ptr) != 0)
2176 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2177 }
2178 else break;
2179 }
2180 }
2181
2182 /* If the next thing is itself optional, we have to give up. */
2183
2184 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2185 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2186 return FALSE;
2187
2188 /* Now compare the next item with the previous opcode. If the previous is a
2189 positive single character match, "item" either contains the character or, if
2190 "item" is greater than 127 in utf8 mode, the character's bytes are in
2191 utf8_char. */
2192
2193
2194 /* Handle cases when the next item is a character. */
2195
2196 if (next >= 0) switch(op_code)
2197 {
2198 case OP_CHAR:
2199 #ifdef SUPPORT_UTF8
2200 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2201 #else
2202 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2203 #endif
2204 return item != next;
2205
2206 /* For CHARNC (caseless character) we must check the other case. If we have
2207 Unicode property support, we can use it to test the other case of
2208 high-valued characters. */
2209
2210 case OP_CHARNC:
2211 #ifdef SUPPORT_UTF8
2212 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2213 #endif
2214 if (item == next) return FALSE;
2215 #ifdef SUPPORT_UTF8
2216 if (utf8)
2217 {
2218 unsigned int othercase;
2219 if (next < 128) othercase = cd->fcc[next]; else
2220 #ifdef SUPPORT_UCP
2221 othercase = UCD_OTHERCASE((unsigned int)next);
2222 #else
2223 othercase = NOTACHAR;
2224 #endif
2225 return (unsigned int)item != othercase;
2226 }
2227 else
2228 #endif /* SUPPORT_UTF8 */
2229 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2230
2231 /* For OP_NOT, "item" must be a single-byte character. */
2232
2233 case OP_NOT:
2234 if (item == next) return TRUE;
2235 if ((options & PCRE_CASELESS) == 0) return FALSE;
2236 #ifdef SUPPORT_UTF8
2237 if (utf8)
2238 {
2239 unsigned int othercase;
2240 if (next < 128) othercase = cd->fcc[next]; else
2241 #ifdef SUPPORT_UCP
2242 othercase = UCD_OTHERCASE(next);
2243 #else
2244 othercase = NOTACHAR;
2245 #endif
2246 return (unsigned int)item == othercase;
2247 }
2248 else
2249 #endif /* SUPPORT_UTF8 */
2250 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2251
2252 case OP_DIGIT:
2253 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2254
2255 case OP_NOT_DIGIT:
2256 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2257
2258 case OP_WHITESPACE:
2259 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2260
2261 case OP_NOT_WHITESPACE:
2262 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2263
2264 case OP_WORDCHAR:
2265 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2266
2267 case OP_NOT_WORDCHAR:
2268 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2269
2270 case OP_HSPACE:
2271 case OP_NOT_HSPACE:
2272 switch(next)
2273 {
2274 case 0x09:
2275 case 0x20:
2276 case 0xa0:
2277 case 0x1680:
2278 case 0x180e:
2279 case 0x2000:
2280 case 0x2001:
2281 case 0x2002:
2282 case 0x2003:
2283 case 0x2004:
2284 case 0x2005:
2285 case 0x2006:
2286 case 0x2007:
2287 case 0x2008:
2288 case 0x2009:
2289 case 0x200A:
2290 case 0x202f:
2291 case 0x205f:
2292 case 0x3000:
2293 return op_code != OP_HSPACE;
2294 default:
2295 return op_code == OP_HSPACE;
2296 }
2297
2298 case OP_VSPACE:
2299 case OP_NOT_VSPACE:
2300 switch(next)
2301 {
2302 case 0x0a:
2303 case 0x0b:
2304 case 0x0c:
2305 case 0x0d:
2306 case 0x85:
2307 case 0x2028:
2308 case 0x2029:
2309 return op_code != OP_VSPACE;
2310 default:
2311 return op_code == OP_VSPACE;
2312 }
2313
2314 default:
2315 return FALSE;
2316 }
2317
2318
2319 /* Handle the case when the next item is \d, \s, etc. */
2320
2321 switch(op_code)
2322 {
2323 case OP_CHAR:
2324 case OP_CHARNC:
2325 #ifdef SUPPORT_UTF8
2326 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2327 #endif
2328 switch(-next)
2329 {
2330 case ESC_d:
2331 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2332
2333 case ESC_D:
2334 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2335
2336 case ESC_s:
2337 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2338
2339 case ESC_S:
2340 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2341
2342 case ESC_w:
2343 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2344
2345 case ESC_W:
2346 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2347
2348 case ESC_h:
2349 case ESC_H:
2350 switch(item)
2351 {
2352 case 0x09:
2353 case 0x20:
2354 case 0xa0:
2355 case 0x1680:
2356 case 0x180e:
2357 case 0x2000:
2358 case 0x2001:
2359 case 0x2002:
2360 case 0x2003:
2361 case 0x2004:
2362 case 0x2005:
2363 case 0x2006:
2364 case 0x2007:
2365 case 0x2008:
2366 case 0x2009:
2367 case 0x200A:
2368 case 0x202f:
2369 case 0x205f:
2370 case 0x3000:
2371 return -next != ESC_h;
2372 default:
2373 return -next == ESC_h;
2374 }
2375
2376 case ESC_v:
2377 case ESC_V:
2378 switch(item)
2379 {
2380 case 0x0a:
2381 case 0x0b:
2382 case 0x0c:
2383 case 0x0d:
2384 case 0x85:
2385 case 0x2028:
2386 case 0x2029:
2387 return -next != ESC_v;
2388 default:
2389 return -next == ESC_v;
2390 }
2391
2392 default:
2393 return FALSE;
2394 }
2395
2396 case OP_DIGIT:
2397 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2398 next == -ESC_h || next == -ESC_v;
2399
2400 case OP_NOT_DIGIT:
2401 return next == -ESC_d;
2402
2403 case OP_WHITESPACE:
2404 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2405
2406 case OP_NOT_WHITESPACE:
2407 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2408
2409 case OP_HSPACE:
2410 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2411
2412 case OP_NOT_HSPACE:
2413 return next == -ESC_h;
2414
2415 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2416 case OP_VSPACE:
2417 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2418
2419 case OP_NOT_VSPACE:
2420 return next == -ESC_v;
2421
2422 case OP_WORDCHAR:
2423 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2424
2425 case OP_NOT_WORDCHAR:
2426 return next == -ESC_w || next == -ESC_d;
2427
2428 default:
2429 return FALSE;
2430 }
2431
2432 /* Control does not reach here */
2433 }
2434
2435
2436
2437 /*************************************************
2438 * Compile one branch *
2439 *************************************************/
2440
2441 /* Scan the pattern, compiling it into the a vector. If the options are
2442 changed during the branch, the pointer is used to change the external options
2443 bits. This function is used during the pre-compile phase when we are trying
2444 to find out the amount of memory needed, as well as during the real compile
2445 phase. The value of lengthptr distinguishes the two phases.
2446
2447 Arguments:
2448 optionsptr pointer to the option bits
2449 codeptr points to the pointer to the current code point
2450 ptrptr points to the current pattern pointer
2451 errorcodeptr points to error code variable
2452 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2453 reqbyteptr set to the last literal character required, else < 0
2454 bcptr points to current branch chain
2455 cd contains pointers to tables etc.
2456 lengthptr NULL during the real compile phase
2457 points to length accumulator during pre-compile phase
2458
2459 Returns: TRUE on success
2460 FALSE, with *errorcodeptr set non-zero on error
2461 */
2462
2463 static BOOL
2464 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2465 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2466 compile_data *cd, int *lengthptr)
2467 {
2468 int repeat_type, op_type;
2469 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2470 int bravalue = 0;
2471 int greedy_default, greedy_non_default;
2472 int firstbyte, reqbyte;
2473 int zeroreqbyte, zerofirstbyte;
2474 int req_caseopt, reqvary, tempreqvary;
2475 int options = *optionsptr;
2476 int after_manual_callout = 0;
2477 int length_prevgroup = 0;
2478 register int c;
2479 register uschar *code = *codeptr;
2480 uschar *last_code = code;
2481 uschar *orig_code = code;
2482 uschar *tempcode;
2483 BOOL inescq = FALSE;
2484 BOOL groupsetfirstbyte = FALSE;
2485 const uschar *ptr = *ptrptr;
2486 const uschar *tempptr;
2487 uschar *previous = NULL;
2488 uschar *previous_callout = NULL;
2489 uschar *save_hwm = NULL;
2490 uschar classbits[32];
2491
2492 #ifdef SUPPORT_UTF8
2493 BOOL class_utf8;
2494 BOOL utf8 = (options & PCRE_UTF8) != 0;
2495 uschar *class_utf8data;
2496 uschar *class_utf8data_base;
2497 uschar utf8_char[6];
2498 #else
2499 BOOL utf8 = FALSE;
2500 uschar *utf8_char = NULL;
2501 #endif
2502
2503 #ifdef DEBUG
2504 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2505 #endif
2506
2507 /* Set up the default and non-default settings for greediness */
2508
2509 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2510 greedy_non_default = greedy_default ^ 1;
2511
2512 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2513 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2514 matches a non-fixed char first char; reqbyte just remains unset if we never
2515 find one.
2516
2517 When we hit a repeat whose minimum is zero, we may have to adjust these values
2518 to take the zero repeat into account. This is implemented by setting them to
2519 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2520 item types that can be repeated set these backoff variables appropriately. */
2521
2522 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2523
2524 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2525 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2526 value > 255. It is added into the firstbyte or reqbyte variables to record the
2527 case status of the value. This is used only for ASCII characters. */
2528
2529 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2530
2531 /* Switch on next character until the end of the branch */
2532
2533 for (;; ptr++)
2534 {
2535 BOOL negate_class;
2536 BOOL should_flip_negation;
2537 BOOL possessive_quantifier;
2538 BOOL is_quantifier;
2539 BOOL is_recurse;
2540 BOOL reset_bracount;
2541 int class_charcount;
2542 int class_lastchar;
2543 int newoptions;
2544 int recno;
2545 int refsign;
2546 int skipbytes;
2547 int subreqbyte;
2548 int subfirstbyte;
2549 int terminator;
2550 int mclength;
2551 uschar mcbuffer[8];
2552
2553 /* Get next byte in the pattern */
2554
2555 c = *ptr;
2556
2557 /* If we are in the pre-compile phase, accumulate the length used for the
2558 previous cycle of this loop. */
2559
2560 if (lengthptr != NULL)
2561 {
2562 #ifdef DEBUG
2563 if (code > cd->hwm) cd->hwm = code; /* High water info */
2564 #endif
2565 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2566 {
2567 *errorcodeptr = ERR52;
2568 goto FAILED;
2569 }
2570
2571 /* There is at least one situation where code goes backwards: this is the
2572 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2573 the class is simply eliminated. However, it is created first, so we have to
2574 allow memory for it. Therefore, don't ever reduce the length at this point.
2575 */
2576
2577 if (code < last_code) code = last_code;
2578
2579 /* Paranoid check for integer overflow */
2580
2581 if (OFLOW_MAX - *lengthptr < code - last_code)
2582 {
2583 *errorcodeptr = ERR20;
2584 goto FAILED;
2585 }
2586
2587 *lengthptr += code - last_code;
2588 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2589
2590 /* If "previous" is set and it is not at the start of the work space, move
2591 it back to there, in order to avoid filling up the work space. Otherwise,
2592 if "previous" is NULL, reset the current code pointer to the start. */
2593
2594 if (previous != NULL)
2595 {
2596 if (previous > orig_code)
2597 {
2598 memmove(orig_code, previous, code - previous);
2599 code -= previous - orig_code;
2600 previous = orig_code;
2601 }
2602 }
2603 else code = orig_code;
2604
2605 /* Remember where this code item starts so we can pick up the length
2606 next time round. */
2607
2608 last_code = code;
2609 }
2610
2611 /* In the real compile phase, just check the workspace used by the forward
2612 reference list. */
2613
2614 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2615 {
2616 *errorcodeptr = ERR52;
2617 goto FAILED;
2618 }
2619
2620 /* If in \Q...\E, check for the end; if not, we have a literal */
2621
2622 if (inescq && c != 0)
2623 {
2624 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2625 {
2626 inescq = FALSE;
2627 ptr++;
2628 continue;
2629 }
2630 else
2631 {
2632 if (previous_callout != NULL)
2633 {
2634 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2635 complete_callout(previous_callout, ptr, cd);
2636 previous_callout = NULL;
2637 }
2638 if ((options & PCRE_AUTO_CALLOUT) != 0)
2639 {
2640 previous_callout = code;
2641 code = auto_callout(code, ptr, cd);
2642 }
2643 goto NORMAL_CHAR;
2644 }
2645 }
2646
2647 /* Fill in length of a previous callout, except when the next thing is
2648 a quantifier. */
2649
2650 is_quantifier =
2651 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2652 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2653
2654 if (!is_quantifier && previous_callout != NULL &&
2655 after_manual_callout-- <= 0)
2656 {
2657 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2658 complete_callout(previous_callout, ptr, cd);
2659 previous_callout = NULL;
2660 }
2661
2662 /* In extended mode, skip white space and comments */
2663
2664 if ((options & PCRE_EXTENDED) != 0)
2665 {
2666 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2667 if (c == CHAR_NUMBER_SIGN)
2668 {
2669 while (*(++ptr) != 0)
2670 {
2671 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2672 }
2673 if (*ptr != 0) continue;
2674
2675 /* Else fall through to handle end of string */
2676 c = 0;
2677 }
2678 }
2679
2680 /* No auto callout for quantifiers. */
2681
2682 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2683 {
2684 previous_callout = code;
2685 code = auto_callout(code, ptr, cd);
2686 }
2687
2688 switch(c)
2689 {
2690 /* ===================================================================*/
2691 case 0: /* The branch terminates at string end */
2692 case CHAR_VERTICAL_LINE: /* or | or ) */
2693 case CHAR_RIGHT_PARENTHESIS:
2694 *firstbyteptr = firstbyte;
2695 *reqbyteptr = reqbyte;
2696 *codeptr = code;
2697 *ptrptr = ptr;
2698 if (lengthptr != NULL)
2699 {
2700 if (OFLOW_MAX - *lengthptr < code - last_code)
2701 {
2702 *errorcodeptr = ERR20;
2703 goto FAILED;
2704 }
2705 *lengthptr += code - last_code; /* To include callout length */
2706 DPRINTF((">> end branch\n"));
2707 }
2708 return TRUE;
2709
2710
2711 /* ===================================================================*/
2712 /* Handle single-character metacharacters. In multiline mode, ^ disables
2713 the setting of any following char as a first character. */
2714
2715 case CHAR_CIRCUMFLEX_ACCENT:
2716 if ((options & PCRE_MULTILINE) != 0)
2717 {
2718 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2719 }
2720 previous = NULL;
2721 *code++ = OP_CIRC;
2722 break;
2723
2724 case CHAR_DOLLAR_SIGN:
2725 previous = NULL;
2726 *code++ = OP_DOLL;
2727 break;
2728
2729 /* There can never be a first char if '.' is first, whatever happens about
2730 repeats. The value of reqbyte doesn't change either. */
2731
2732 case CHAR_DOT:
2733 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2734 zerofirstbyte = firstbyte;
2735 zeroreqbyte = reqbyte;
2736 previous = code;
2737 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2738 break;
2739
2740
2741 /* ===================================================================*/
2742 /* Character classes. If the included characters are all < 256, we build a
2743 32-byte bitmap of the permitted characters, except in the special case
2744 where there is only one such character. For negated classes, we build the
2745 map as usual, then invert it at the end. However, we use a different opcode
2746 so that data characters > 255 can be handled correctly.
2747
2748 If the class contains characters outside the 0-255 range, a different
2749 opcode is compiled. It may optionally have a bit map for characters < 256,
2750 but those above are are explicitly listed afterwards. A flag byte tells
2751 whether the bitmap is present, and whether this is a negated class or not.
2752
2753 In JavaScript compatibility mode, an isolated ']' causes an error. In
2754 default (Perl) mode, it is treated as a data character. */
2755
2756 case CHAR_RIGHT_SQUARE_BRACKET:
2757 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2758 {
2759 *errorcodeptr = ERR64;
2760 goto FAILED;
2761 }
2762 goto NORMAL_CHAR;
2763
2764 case CHAR_LEFT_SQUARE_BRACKET:
2765 previous = code;
2766
2767 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2768 they are encountered at the top level, so we'll do that too. */
2769
2770 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2771 ptr[1] == CHAR_EQUALS_SIGN) &&
2772 check_posix_syntax(ptr, &tempptr))
2773 {
2774 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2775 goto FAILED;
2776 }
2777
2778 /* If the first character is '^', set the negation flag and skip it. Also,
2779 if the first few characters (either before or after ^) are \Q\E or \E we
2780 skip them too. This makes for compatibility with Perl. */
2781
2782 negate_class = FALSE;
2783 for (;;)
2784 {
2785 c = *(++ptr);
2786 if (c == CHAR_BACKSLASH)
2787 {
2788 if (ptr[1] == CHAR_E)
2789 ptr++;
2790 else if (strncmp((const char *)ptr+1,
2791 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2792 ptr += 3;
2793 else
2794 break;
2795 }
2796 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2797 negate_class = TRUE;
2798 else break;
2799 }
2800
2801 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2802 an initial ']' is taken as a data character -- the code below handles
2803 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2804 [^] must match any character, so generate OP_ALLANY. */
2805
2806 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2807 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2808 {
2809 *code++ = negate_class? OP_ALLANY : OP_FAIL;
2810 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2811 zerofirstbyte = firstbyte;
2812 break;
2813 }
2814
2815 /* If a class contains a negative special such as \S, we need to flip the
2816 negation flag at the end, so that support for characters > 255 works
2817 correctly (they are all included in the class). */
2818
2819 should_flip_negation = FALSE;
2820
2821 /* Keep a count of chars with values < 256 so that we can optimize the case
2822 of just a single character (as long as it's < 256). However, For higher
2823 valued UTF-8 characters, we don't yet do any optimization. */
2824
2825 class_charcount = 0;
2826 class_lastchar = -1;
2827
2828 /* Initialize the 32-char bit map to all zeros. We build the map in a
2829 temporary bit of memory, in case the class contains only 1 character (less
2830 than 256), because in that case the compiled code doesn't use the bit map.
2831 */
2832
2833 memset(classbits, 0, 32 * sizeof(uschar));
2834
2835 #ifdef SUPPORT_UTF8
2836 class_utf8 = FALSE; /* No chars >= 256 */
2837 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2838 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2839 #endif
2840
2841 /* Process characters until ] is reached. By writing this as a "do" it
2842 means that an initial ] is taken as a data character. At the start of the
2843 loop, c contains the first byte of the character. */
2844
2845 if (c != 0) do
2846 {
2847 const uschar *oldptr;
2848
2849 #ifdef SUPPORT_UTF8
2850 if (utf8 && c > 127)
2851 { /* Braces are required because the */
2852 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2853 }
2854
2855 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2856 data and reset the pointer. This is so that very large classes that
2857 contain a zillion UTF-8 characters no longer overwrite the work space
2858 (which is on the stack). */
2859
2860 if (lengthptr != NULL)
2861 {
2862 *lengthptr += class_utf8data - class_utf8data_base;
2863 class_utf8data = class_utf8data_base;
2864 }
2865
2866 #endif
2867
2868 /* Inside \Q...\E everything is literal except \E */
2869
2870 if (inescq)
2871 {
2872 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
2873 {
2874 inescq = FALSE; /* Reset literal state */
2875 ptr++; /* Skip the 'E' */
2876 continue; /* Carry on with next */
2877 }
2878 goto CHECK_RANGE; /* Could be range if \E follows */
2879 }
2880
2881 /* Handle POSIX class names. Perl allows a negation extension of the
2882 form [:^name:]. A square bracket that doesn't match the syntax is
2883 treated as a literal. We also recognize the POSIX constructions
2884 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2885 5.6 and 5.8 do. */
2886
2887 if (c == CHAR_LEFT_SQUARE_BRACKET &&
2888 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2889 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
2890 {
2891 BOOL local_negate = FALSE;
2892 int posix_class, taboffset, tabopt;
2893 register const uschar *cbits = cd->cbits;
2894 uschar pbits[32];
2895
2896 if (ptr[1] != CHAR_COLON)
2897 {
2898 *errorcodeptr = ERR31;
2899 goto FAILED;
2900 }
2901
2902 ptr += 2;
2903 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
2904 {
2905 local_negate = TRUE;
2906 should_flip_negation = TRUE; /* Note negative special */
2907 ptr++;
2908 }
2909
2910 posix_class = check_posix_name(ptr, tempptr - ptr);
2911 if (posix_class < 0)
2912 {
2913 *errorcodeptr = ERR30;
2914 goto FAILED;
2915 }
2916
2917 /* If matching is caseless, upper and lower are converted to
2918 alpha. This relies on the fact that the class table starts with
2919 alpha, lower, upper as the first 3 entries. */
2920
2921 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
2922 posix_class = 0;
2923
2924 /* We build the bit map for the POSIX class in a chunk of local store
2925 because we may be adding and subtracting from it, and we don't want to
2926 subtract bits that may be in the main map already. At the end we or the
2927 result into the bit map that is being built. */
2928
2929 posix_class *= 3;
2930
2931 /* Copy in the first table (always present) */
2932
2933 memcpy(pbits, cbits + posix_class_maps[posix_class],
2934 32 * sizeof(uschar));
2935
2936 /* If there is a second table, add or remove it as required. */
2937
2938 taboffset = posix_class_maps[posix_class + 1];
2939 tabopt = posix_class_maps[posix_class + 2];
2940
2941 if (taboffset >= 0)
2942 {
2943 if (tabopt >= 0)
2944 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
2945 else
2946 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
2947 }
2948
2949 /* Not see if we need to remove any special characters. An option
2950 value of 1 removes vertical space and 2 removes underscore. */
2951
2952 if (tabopt < 0) tabopt = -tabopt;
2953 if (tabopt == 1) pbits[1] &= ~0x3c;
2954 else if (tabopt == 2) pbits[11] &= 0x7f;
2955
2956 /* Add the POSIX table or its complement into the main table that is
2957 being built and we are done. */
2958
2959 if (local_negate)
2960 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
2961 else
2962 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
2963
2964 ptr = tempptr + 1;
2965 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
2966 continue; /* End of POSIX syntax handling */
2967 }
2968
2969 /* Backslash may introduce a single character, or it may introduce one
2970 of the specials, which just set a flag. The sequence \b is a special
2971 case. Inside a class (and only there) it is treated as backspace.
2972 Elsewhere it marks a word boundary. Other escapes have preset maps ready
2973 to 'or' into the one we are building. We assume they have more than one
2974 character in them, so set class_charcount bigger than one. */
2975
2976 if (c == CHAR_BACKSLASH)
2977 {
2978 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
2979 if (*errorcodeptr != 0) goto FAILED;
2980
2981 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
2982 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
2983 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
2984 else if (-c == ESC_Q) /* Handle start of quoted string */
2985 {
2986 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
2987 {
2988 ptr += 2; /* avoid empty string */
2989 }
2990 else inescq = TRUE;
2991 continue;
2992 }
2993 else if (-c == ESC_E) continue; /* Ignore orphan \E */
2994
2995 if (c < 0)
2996 {
2997 register const uschar *cbits = cd->cbits;
2998 class_charcount += 2; /* Greater than 1 is what matters */
2999
3000 /* Save time by not doing this in the pre-compile phase. */
3001
3002 if (lengthptr == NULL) switch (-c)
3003 {
3004 case ESC_d:
3005 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3006 continue;
3007
3008 case ESC_D:
3009 should_flip_negation = TRUE;
3010 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3011 continue;
3012
3013 case ESC_w:
3014 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3015 continue;
3016
3017 case ESC_W:
3018 should_flip_negation = TRUE;
3019 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3020 continue;
3021
3022 case ESC_s:
3023 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3024 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3025 continue;
3026
3027 case ESC_S:
3028 should_flip_negation = TRUE;
3029 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3030 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3031 continue;
3032
3033 default: /* Not recognized; fall through */
3034 break; /* Need "default" setting to stop compiler warning. */
3035 }
3036
3037 /* In the pre-compile phase, just do the recognition. */
3038
3039 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3040 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3041
3042 /* We need to deal with \H, \h, \V, and \v in both phases because
3043 they use extra memory. */
3044
3045 if (-c == ESC_h)
3046 {
3047 SETBIT(classbits, 0x09); /* VT */
3048 SETBIT(classbits, 0x20); /* SPACE */
3049 SETBIT(classbits, 0xa0); /* NSBP */
3050 #ifdef SUPPORT_UTF8
3051 if (utf8)
3052 {
3053 class_utf8 = TRUE;
3054 *class_utf8data++ = XCL_SINGLE;
3055 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3056 *class_utf8data++ = XCL_SINGLE;
3057 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3058 *class_utf8data++ = XCL_RANGE;
3059 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3060 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3061 *class_utf8data++ = XCL_SINGLE;
3062 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3063 *class_utf8data++ = XCL_SINGLE;
3064 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3065 *class_utf8data++ = XCL_SINGLE;
3066 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3067 }
3068 #endif
3069 continue;
3070 }
3071
3072 if (-c == ESC_H)
3073 {
3074 for (c = 0; c < 32; c++)
3075 {
3076 int x = 0xff;
3077 switch (c)
3078 {
3079 case 0x09/8: x ^= 1 << (0x09%8); break;
3080 case 0x20/8: x ^= 1 << (0x20%8); break;
3081 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3082 default: break;
3083 }
3084 classbits[c] |= x;
3085 }
3086
3087 #ifdef SUPPORT_UTF8
3088 if (utf8)
3089 {
3090 class_utf8 = TRUE;
3091 *class_utf8data++ = XCL_RANGE;
3092 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3093 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3094 *class_utf8data++ = XCL_RANGE;
3095 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3096 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3097 *class_utf8data++ = XCL_RANGE;
3098 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3099 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3100 *class_utf8data++ = XCL_RANGE;
3101 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3102 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3103 *class_utf8data++ = XCL_RANGE;
3104 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3105 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3106 *class_utf8data++ = XCL_RANGE;
3107 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3108 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3109 *class_utf8data++ = XCL_RANGE;
3110 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3111 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3112 }
3113 #endif
3114 continue;
3115 }
3116
3117 if (-c == ESC_v)
3118 {
3119 SETBIT(classbits, 0x0a); /* LF */
3120 SETBIT(classbits, 0x0b); /* VT */
3121 SETBIT(classbits, 0x0c); /* FF */
3122 SETBIT(classbits, 0x0d); /* CR */
3123 SETBIT(classbits, 0x85); /* NEL */
3124 #ifdef SUPPORT_UTF8
3125 if (utf8)
3126 {
3127 class_utf8 = TRUE;
3128 *class_utf8data++ = XCL_RANGE;
3129 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3130 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3131 }
3132 #endif
3133 continue;
3134 }
3135
3136 if (-c == ESC_V)
3137 {
3138 for (c = 0; c < 32; c++)
3139 {
3140 int x = 0xff;
3141 switch (c)
3142 {
3143 case 0x0a/8: x ^= 1 << (0x0a%8);
3144 x ^= 1 << (0x0b%8);
3145 x ^= 1 << (0x0c%8);
3146 x ^= 1 << (0x0d%8);
3147 break;
3148 case 0x85/8: x ^= 1 << (0x85%8); break;
3149 default: break;
3150 }
3151 classbits[c] |= x;
3152 }
3153
3154 #ifdef SUPPORT_UTF8
3155 if (utf8)
3156 {
3157 class_utf8 = TRUE;
3158 *class_utf8data++ = XCL_RANGE;
3159 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3160 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3161 *class_utf8data++ = XCL_RANGE;
3162 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3163 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3164 }
3165 #endif
3166 continue;
3167 }
3168
3169 /* We need to deal with \P and \p in both phases. */
3170
3171 #ifdef SUPPORT_UCP
3172 if (-c == ESC_p || -c == ESC_P)
3173 {
3174 BOOL negated;
3175 int pdata;
3176 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3177 if (ptype < 0) goto FAILED;
3178 class_utf8 = TRUE;
3179 *class_utf8data++ = ((-c == ESC_p) != negated)?
3180 XCL_PROP : XCL_NOTPROP;
3181 *class_utf8data++ = ptype;
3182 *class_utf8data++ = pdata;
3183 class_charcount -= 2; /* Not a < 256 character */
3184 continue;
3185 }
3186 #endif
3187 /* Unrecognized escapes are faulted if PCRE is running in its
3188 strict mode. By default, for compatibility with Perl, they are
3189 treated as literals. */
3190
3191 if ((options & PCRE_EXTRA) != 0)
3192 {
3193 *errorcodeptr = ERR7;
3194 goto FAILED;
3195 }
3196
3197 class_charcount -= 2; /* Undo the default count from above */
3198 c = *ptr; /* Get the final character and fall through */
3199 }
3200
3201 /* Fall through if we have a single character (c >= 0). This may be
3202 greater than 256 in UTF-8 mode. */
3203
3204 } /* End of backslash handling */
3205
3206 /* A single character may be followed by '-' to form a range. However,
3207 Perl does not permit ']' to be the end of the range. A '-' character
3208 at the end is treated as a literal. Perl ignores orphaned \E sequences
3209 entirely. The code for handling \Q and \E is messy. */
3210
3211 CHECK_RANGE:
3212 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3213 {
3214 inescq = FALSE;
3215 ptr += 2;
3216 }
3217
3218 oldptr = ptr;
3219
3220 /* Remember \r or \n */
3221
3222 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3223
3224 /* Check for range */
3225
3226 if (!inescq && ptr[1] == CHAR_MINUS)
3227 {
3228 int d;
3229 ptr += 2;
3230 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3231
3232 /* If we hit \Q (not followed by \E) at this point, go into escaped
3233 mode. */
3234
3235 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3236 {
3237 ptr += 2;
3238 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3239 { ptr += 2; continue; }
3240 inescq = TRUE;
3241 break;
3242 }
3243
3244 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3245 {
3246 ptr = oldptr;
3247 goto LONE_SINGLE_CHARACTER;
3248 }
3249
3250 #ifdef SUPPORT_UTF8
3251 if (utf8)
3252 { /* Braces are required because the */
3253 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3254 }
3255 else
3256 #endif
3257 d = *ptr; /* Not UTF-8 mode */
3258
3259 /* The second part of a range can be a single-character escape, but
3260 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3261 in such circumstances. */
3262
3263 if (!inescq && d == CHAR_BACKSLASH)
3264 {
3265 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3266 if (*errorcodeptr != 0) goto FAILED;
3267
3268 /* \b is backspace; \X is literal X; \R is literal R; any other
3269 special means the '-' was literal */
3270
3271 if (d < 0)
3272 {
3273 if (d == -ESC_b) d = CHAR_BS;
3274 else if (d == -ESC_X) d = CHAR_X;
3275 else if (d == -ESC_R) d = CHAR_R; else
3276 {
3277 ptr = oldptr;
3278 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3279 }
3280 }
3281 }
3282
3283 /* Check that the two values are in the correct order. Optimize
3284 one-character ranges */
3285
3286 if (d < c)
3287 {
3288 *errorcodeptr = ERR8;
3289 goto FAILED;
3290 }
3291
3292 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3293
3294 /* Remember \r or \n */
3295
3296 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3297
3298 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3299 matching, we have to use an XCLASS with extra data items. Caseless
3300 matching for characters > 127 is available only if UCP support is
3301 available. */
3302
3303 #ifdef SUPPORT_UTF8
3304 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3305 {
3306 class_utf8 = TRUE;
3307
3308 /* With UCP support, we can find the other case equivalents of
3309 the relevant characters. There may be several ranges. Optimize how
3310 they fit with the basic range. */
3311
3312 #ifdef SUPPORT_UCP
3313 if ((options & PCRE_CASELESS) != 0)
3314 {
3315 unsigned int occ, ocd;
3316 unsigned int cc = c;
3317 unsigned int origd = d;
3318 while (get_othercase_range(&cc, origd, &occ, &ocd))
3319 {
3320 if (occ >= (unsigned int)c &&
3321 ocd <= (unsigned int)d)
3322 continue; /* Skip embedded ranges */
3323
3324 if (occ < (unsigned int)c &&
3325 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3326 { /* if there is overlap, */
3327 c = occ; /* noting that if occ < c */
3328 continue; /* we can't have ocd > d */
3329 } /* because a subrange is */
3330 if (ocd > (unsigned int)d &&
3331 occ <= (unsigned int)d + 1) /* always shorter than */
3332 { /* the basic range. */
3333 d = ocd;
3334 continue;
3335 }
3336
3337 if (occ == ocd)
3338 {
3339 *class_utf8data++ = XCL_SINGLE;
3340 }
3341 else
3342 {
3343 *class_utf8data++ = XCL_RANGE;
3344 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3345 }
3346 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3347 }
3348 }
3349 #endif /* SUPPORT_UCP */
3350
3351 /* Now record the original range, possibly modified for UCP caseless
3352 overlapping ranges. */
3353
3354 *class_utf8data++ = XCL_RANGE;
3355 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3356 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3357
3358 /* With UCP support, we are done. Without UCP support, there is no
3359 caseless matching for UTF-8 characters > 127; we can use the bit map
3360 for the smaller ones. */
3361
3362 #ifdef SUPPORT_UCP
3363 continue; /* With next character in the class */
3364 #else
3365 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3366
3367 /* Adjust upper limit and fall through to set up the map */
3368
3369 d = 127;
3370
3371 #endif /* SUPPORT_UCP */
3372 }
3373 #endif /* SUPPORT_UTF8 */
3374
3375 /* We use the bit map for all cases when not in UTF-8 mode; else
3376 ranges that lie entirely within 0-127 when there is UCP support; else
3377 for partial ranges without UCP support. */
3378
3379 class_charcount += d - c + 1;
3380 class_lastchar = d;
3381
3382 /* We can save a bit of time by skipping this in the pre-compile. */
3383
3384 if (lengthptr == NULL) for (; c <= d; c++)
3385 {
3386 classbits[c/8] |= (1 << (c&7));
3387 if ((options & PCRE_CASELESS) != 0)
3388 {
3389 int uc = cd->fcc[c]; /* flip case */
3390 classbits[uc/8] |= (1 << (uc&7));
3391 }
3392 }
3393
3394 continue; /* Go get the next char in the class */
3395 }
3396
3397 /* Handle a lone single character - we can get here for a normal
3398 non-escape char, or after \ that introduces a single character or for an
3399 apparent range that isn't. */
3400
3401 LONE_SINGLE_CHARACTER:
3402
3403 /* Handle a character that cannot go in the bit map */
3404
3405 #ifdef SUPPORT_UTF8
3406 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3407 {
3408 class_utf8 = TRUE;
3409 *class_utf8data++ = XCL_SINGLE;
3410 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3411
3412 #ifdef SUPPORT_UCP
3413 if ((options & PCRE_CASELESS) != 0)
3414 {
3415 unsigned int othercase;
3416 if ((othercase = UCD_OTHERCASE(c)) != c)
3417 {
3418 *class_utf8data++ = XCL_SINGLE;
3419 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3420 }
3421 }
3422 #endif /* SUPPORT_UCP */
3423
3424 }
3425 else
3426 #endif /* SUPPORT_UTF8 */
3427
3428 /* Handle a single-byte character */
3429 {
3430 classbits[c/8] |= (1 << (c&7));
3431 if ((options & PCRE_CASELESS) != 0)
3432 {
3433 c = cd->fcc[c]; /* flip case */
3434 classbits[c/8] |= (1 << (c&7));
3435 }
3436 class_charcount++;
3437 class_lastchar = c;
3438 }
3439 }
3440
3441 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3442
3443 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3444
3445 if (c == 0) /* Missing terminating ']' */
3446 {
3447 *errorcodeptr = ERR6;
3448 goto FAILED;
3449 }
3450
3451
3452 /* This code has been disabled because it would mean that \s counts as
3453 an explicit \r or \n reference, and that's not really what is wanted. Now
3454 we set the flag only if there is a literal "\r" or "\n" in the class. */
3455
3456 #if 0
3457 /* Remember whether \r or \n are in this class */
3458
3459 if (negate_class)
3460 {
3461 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3462 }
3463 else
3464 {
3465 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3466 }
3467 #endif
3468
3469
3470 /* If class_charcount is 1, we saw precisely one character whose value is
3471 less than 256. As long as there were no characters >= 128 and there was no
3472 use of \p or \P, in other words, no use of any XCLASS features, we can
3473 optimize.
3474
3475 In UTF-8 mode, we can optimize the negative case only if there were no
3476 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3477 operate on single-bytes only. This is an historical hangover. Maybe one day
3478 we can tidy these opcodes to handle multi-byte characters.
3479
3480 The optimization throws away the bit map. We turn the item into a
3481 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3482 that OP_NOT does not support multibyte characters. In the positive case, it
3483 can cause firstbyte to be set. Otherwise, there can be no first char if
3484 this item is first, whatever repeat count may follow. In the case of
3485 reqbyte, save the previous value for reinstating. */
3486
3487 #ifdef SUPPORT_UTF8
3488 if (class_charcount == 1 && !class_utf8 &&
3489 (!utf8 || !negate_class || class_lastchar < 128))
3490 #else
3491 if (class_charcount == 1)
3492 #endif
3493 {
3494 zeroreqbyte = reqbyte;
3495
3496 /* The OP_NOT opcode works on one-byte characters only. */
3497
3498 if (negate_class)
3499 {
3500 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3501 zerofirstbyte = firstbyte;
3502 *code++ = OP_NOT;
3503 *code++ = class_lastchar;
3504 break;
3505 }
3506
3507 /* For a single, positive character, get the value into mcbuffer, and
3508 then we can handle this with the normal one-character code. */
3509
3510 #ifdef SUPPORT_UTF8
3511 if (utf8 && class_lastchar > 127)
3512 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3513 else
3514 #endif
3515 {
3516 mcbuffer[0] = class_lastchar;
3517 mclength = 1;
3518 }
3519 goto ONE_CHAR;
3520 } /* End of 1-char optimization */
3521
3522 /* The general case - not the one-char optimization. If this is the first
3523 thing in the branch, there can be no first char setting, whatever the
3524 repeat count. Any reqbyte setting must remain unchanged after any kind of
3525 repeat. */
3526
3527 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3528 zerofirstbyte = firstbyte;
3529 zeroreqbyte = reqbyte;
3530
3531 /* If there are characters with values > 255, we have to compile an
3532 extended class, with its own opcode, unless there was a negated special
3533 such as \S in the class, because in that case all characters > 255 are in
3534 the class, so any that were explicitly given as well can be ignored. If
3535 (when there are explicit characters > 255 that must be listed) there are no
3536 characters < 256, we can omit the bitmap in the actual compiled code. */
3537
3538 #ifdef SUPPORT_UTF8
3539 if (class_utf8 && !should_flip_negation)
3540 {
3541 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3542 *code++ = OP_XCLASS;
3543 code += LINK_SIZE;
3544 *code = negate_class? XCL_NOT : 0;
3545
3546 /* If the map is required, move up the extra data to make room for it;
3547 otherwise just move the code pointer to the end of the extra data. */
3548
3549 if (class_charcount > 0)
3550 {
3551 *code++ |= XCL_MAP;
3552 memmove(code + 32, code, class_utf8data - code);
3553 memcpy(code, classbits, 32);
3554 code = class_utf8data + 32;
3555 }
3556 else code = class_utf8data;
3557
3558 /* Now fill in the complete length of the item */
3559
3560 PUT(previous, 1, code - previous);
3561 break; /* End of class handling */
3562 }
3563 #endif
3564
3565 /* If there are no characters > 255, set the opcode to OP_CLASS or
3566 OP_NCLASS, depending on whether the whole class was negated and whether
3567 there were negative specials such as \S in the class. Then copy the 32-byte
3568 map into the code vector, negating it if necessary. */
3569
3570 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3571 if (negate_class)
3572 {
3573 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3574 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3575 }
3576 else
3577 {
3578 memcpy(code, classbits, 32);
3579 }
3580 code += 32;
3581 break;
3582
3583
3584 /* ===================================================================*/
3585 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3586 has been tested above. */
3587
3588 case CHAR_LEFT_CURLY_BRACKET:
3589 if (!is_quantifier) goto NORMAL_CHAR;
3590 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3591 if (*errorcodeptr != 0) goto FAILED;
3592 goto REPEAT;
3593
3594 case CHAR_ASTERISK:
3595 repeat_min = 0;
3596 repeat_max = -1;
3597 goto REPEAT;
3598
3599 case CHAR_PLUS:
3600 repeat_min = 1;
3601 repeat_max = -1;
3602 goto REPEAT;
3603
3604 case CHAR_QUESTION_MARK:
3605 repeat_min = 0;
3606 repeat_max = 1;
3607
3608 REPEAT:
3609 if (previous == NULL)
3610 {
3611 *errorcodeptr = ERR9;
3612 goto FAILED;
3613 }
3614
3615 if (repeat_min == 0)
3616 {
3617 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3618 reqbyte = zeroreqbyte; /* Ditto */
3619 }
3620
3621 /* Remember whether this is a variable length repeat */
3622
3623 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3624
3625 op_type = 0; /* Default single-char op codes */
3626 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3627
3628 /* Save start of previous item, in case we have to move it up to make space
3629 for an inserted OP_ONCE for the additional '+' extension. */
3630
3631 tempcode = previous;
3632
3633 /* If the next character is '+', we have a possessive quantifier. This
3634 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3635 If the next character is '?' this is a minimizing repeat, by default,
3636 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3637 repeat type to the non-default. */
3638
3639 if (ptr[1] == CHAR_PLUS)
3640 {
3641 repeat_type = 0; /* Force greedy */
3642 possessive_quantifier = TRUE;
3643 ptr++;
3644 }
3645 else if (ptr[1] == CHAR_QUESTION_MARK)
3646 {
3647 repeat_type = greedy_non_default;
3648 ptr++;
3649 }
3650 else repeat_type = greedy_default;
3651
3652 /* If previous was a character match, abolish the item and generate a
3653 repeat item instead. If a char item has a minumum of more than one, ensure
3654 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3655 the first thing in a branch because the x will have gone into firstbyte
3656 instead. */
3657
3658 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3659 {
3660 /* Deal with UTF-8 characters that take up more than one byte. It's
3661 easier to write this out separately than try to macrify it. Use c to
3662 hold the length of the character in bytes, plus 0x80 to flag that it's a
3663 length rather than a small character. */
3664
3665 #ifdef SUPPORT_UTF8
3666 if (utf8 && (code[-1] & 0x80) != 0)
3667 {
3668 uschar *lastchar = code - 1;
3669 while((*lastchar & 0xc0) == 0x80) lastchar--;
3670 c = code - lastchar; /* Length of UTF-8 character */
3671 memcpy(utf8_char, lastchar, c); /* Save the char */
3672 c |= 0x80; /* Flag c as a length */
3673 }
3674 else
3675 #endif
3676
3677 /* Handle the case of a single byte - either with no UTF8 support, or
3678 with UTF-8 disabled, or for a UTF-8 character < 128. */
3679
3680 {
3681 c = code[-1];
3682 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3683 }
3684
3685 /* If the repetition is unlimited, it pays to see if the next thing on
3686 the line is something that cannot possibly match this character. If so,
3687 automatically possessifying this item gains some performance in the case
3688 where the match fails. */
3689
3690 if (!possessive_quantifier &&
3691 repeat_max < 0 &&
3692 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3693 options, cd))
3694 {
3695 repeat_type = 0; /* Force greedy */
3696 possessive_quantifier = TRUE;
3697 }
3698
3699 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3700 }
3701
3702 /* If previous was a single negated character ([^a] or similar), we use
3703 one of the special opcodes, replacing it. The code is shared with single-
3704 character repeats by setting opt_type to add a suitable offset into
3705 repeat_type. We can also test for auto-possessification. OP_NOT is
3706 currently used only for single-byte chars. */
3707
3708 else if (*previous == OP_NOT)
3709 {
3710 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3711 c = previous[1];
3712 if (!possessive_quantifier &&
3713 repeat_max < 0 &&
3714 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3715 {
3716 repeat_type = 0; /* Force greedy */
3717 possessive_quantifier = TRUE;
3718 }
3719 goto OUTPUT_SINGLE_REPEAT;
3720 }
3721
3722 /* If previous was a character type match (\d or similar), abolish it and
3723 create a suitable repeat item. The code is shared with single-character
3724 repeats by setting op_type to add a suitable offset into repeat_type. Note
3725 the the Unicode property types will be present only when SUPPORT_UCP is
3726 defined, but we don't wrap the little bits of code here because it just
3727 makes it horribly messy. */
3728
3729 else if (*previous < OP_EODN)
3730 {
3731 uschar *oldcode;
3732 int prop_type, prop_value;
3733 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3734 c = *previous;
3735
3736 if (!possessive_quantifier &&
3737 repeat_max < 0 &&
3738 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3739 {
3740 repeat_type = 0; /* Force greedy */
3741 possessive_quantifier = TRUE;
3742 }
3743
3744 OUTPUT_SINGLE_REPEAT:
3745 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3746 {
3747 prop_type = previous[1];
3748 prop_value = previous[2];
3749 }
3750 else prop_type = prop_value = -1;
3751
3752 oldcode = code;
3753 code = previous; /* Usually overwrite previous item */
3754
3755 /* If the maximum is zero then the minimum must also be zero; Perl allows
3756 this case, so we do too - by simply omitting the item altogether. */
3757
3758 if (repeat_max == 0) goto END_REPEAT;
3759
3760 /* All real repeats make it impossible to handle partial matching (maybe
3761 one day we will be able to remove this restriction). */
3762
3763 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3764
3765 /* Combine the op_type with the repeat_type */
3766
3767 repeat_type += op_type;
3768
3769 /* A minimum of zero is handled either as the special case * or ?, or as
3770 an UPTO, with the maximum given. */
3771
3772 if (repeat_min == 0)
3773 {
3774 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3775 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3776 else
3777 {
3778 *code++ = OP_UPTO + repeat_type;
3779 PUT2INC(code, 0, repeat_max);
3780 }
3781 }
3782
3783 /* A repeat minimum of 1 is optimized into some special cases. If the
3784 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3785 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3786 one less than the maximum. */
3787
3788 else if (repeat_min == 1)
3789 {
3790 if (repeat_max == -1)
3791 *code++ = OP_PLUS + repeat_type;
3792 else
3793 {
3794 code = oldcode; /* leave previous item in place */
3795 if (repeat_max == 1) goto END_REPEAT;
3796 *code++ = OP_UPTO + repeat_type;
3797 PUT2INC(code, 0, repeat_max - 1);
3798 }
3799 }
3800
3801 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3802 handled as an EXACT followed by an UPTO. */
3803
3804 else
3805 {
3806 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3807 PUT2INC(code, 0, repeat_min);
3808
3809 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3810 we have to insert the character for the previous code. For a repeated
3811 Unicode property match, there are two extra bytes that define the
3812 required property. In UTF-8 mode, long characters have their length in
3813 c, with the 0x80 bit as a flag. */
3814
3815 if (repeat_max < 0)
3816 {
3817 #ifdef SUPPORT_UTF8
3818 if (utf8 && c >= 128)
3819 {
3820 memcpy(code, utf8_char, c & 7);
3821 code += c & 7;
3822 }
3823 else
3824 #endif
3825 {
3826 *code++ = c;
3827 if (prop_type >= 0)
3828 {
3829 *code++ = prop_type;
3830 *code++ = prop_value;
3831 }
3832 }
3833 *code++ = OP_STAR + repeat_type;
3834 }
3835
3836 /* Else insert an UPTO if the max is greater than the min, again
3837 preceded by the character, for the previously inserted code. If the
3838 UPTO is just for 1 instance, we can use QUERY instead. */
3839
3840 else if (repeat_max != repeat_min)
3841 {
3842 #ifdef SUPPORT_UTF8
3843 if (utf8 && c >= 128)
3844 {
3845 memcpy(code, utf8_char, c & 7);
3846 code += c & 7;
3847 }
3848 else
3849 #endif
3850 *code++ = c;
3851 if (prop_type >= 0)
3852 {
3853 *code++ = prop_type;
3854 *code++ = prop_value;
3855 }
3856 repeat_max -= repeat_min;
3857
3858 if (repeat_max == 1)
3859 {
3860 *code++ = OP_QUERY + repeat_type;
3861 }
3862 else
3863 {
3864 *code++ = OP_UPTO + repeat_type;
3865 PUT2INC(code, 0, repeat_max);
3866 }
3867 }
3868 }
3869
3870 /* The character or character type itself comes last in all cases. */
3871
3872 #ifdef SUPPORT_UTF8
3873 if (utf8 && c >= 128)
3874 {
3875 memcpy(code, utf8_char, c & 7);
3876 code += c & 7;
3877 }
3878 else
3879 #endif
3880 *code++ = c;
3881
3882 /* For a repeated Unicode property match, there are two extra bytes that
3883 define the required property. */
3884
3885 #ifdef SUPPORT_UCP
3886 if (prop_type >= 0)
3887 {
3888 *code++ = prop_type;
3889 *code++ = prop_value;
3890 }
3891 #endif
3892 }
3893
3894 /* If previous was a character class or a back reference, we put the repeat
3895 stuff after it, but just skip the item if the repeat was {0,0}. */
3896
3897 else if (*previous == OP_CLASS ||
3898 *previous == OP_NCLASS ||
3899 #ifdef SUPPORT_UTF8
3900 *previous == OP_XCLASS ||
3901 #endif
3902 *previous == OP_REF)
3903 {
3904 if (repeat_max == 0)
3905 {
3906 code = previous;
3907 goto END_REPEAT;
3908 }
3909
3910 /* All real repeats make it impossible to handle partial matching (maybe
3911 one day we will be able to remove this restriction). */
3912
3913 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3914
3915 if (repeat_min == 0 && repeat_max == -1)
3916 *code++ = OP_CRSTAR + repeat_type;
3917 else if (repeat_min == 1 && repeat_max == -1)
3918 *code++ = OP_CRPLUS + repeat_type;
3919 else if (repeat_min == 0 && repeat_max == 1)
3920 *code++ = OP_CRQUERY + repeat_type;
3921 else
3922 {
3923 *code++ = OP_CRRANGE + repeat_type;
3924 PUT2INC(code, 0, repeat_min);
3925 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
3926 PUT2INC(code, 0, repeat_max);
3927 }
3928 }
3929
3930 /* If previous was a bracket group, we may have to replicate it in certain
3931 cases. */
3932
3933 else if (*previous == OP_BRA || *previous == OP_CBRA ||
3934 *previous == OP_ONCE || *previous == OP_COND)
3935 {
3936 register int i;
3937 int ketoffset = 0;
3938 int len = code - previous;
3939 uschar *bralink = NULL;
3940
3941 /* Repeating a DEFINE group is pointless */
3942
3943 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
3944 {
3945 *errorcodeptr = ERR55;
3946 goto FAILED;
3947 }
3948
3949 /* If the maximum repeat count is unlimited, find the end of the bracket
3950 by scanning through from the start, and compute the offset back to it
3951 from the current code pointer. There may be an OP_OPT setting following
3952 the final KET, so we can't find the end just by going back from the code
3953 pointer. */
3954
3955 if (repeat_max == -1)
3956 {
3957 register uschar *ket = previous;
3958 do ket += GET(ket, 1); while (*ket != OP_KET);
3959 ketoffset = code - ket;
3960 }
3961
3962 /* The case of a zero minimum is special because of the need to stick
3963 OP_BRAZERO in front of it, and because the group appears once in the
3964 data, whereas in other cases it appears the minimum number of times. For
3965 this reason, it is simplest to treat this case separately, as otherwise
3966 the code gets far too messy. There are several special subcases when the
3967 minimum is zero. */
3968
3969 if (repeat_min == 0)
3970 {
3971 /* If the maximum is also zero, we used to just omit the group from the
3972 output altogether, like this:
3973
3974 ** if (repeat_max == 0)
3975 ** {
3976 ** code = previous;
3977 ** goto END_REPEAT;
3978 ** }
3979
3980 However, that fails when a group is referenced as a subroutine from
3981 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
3982 so that it is skipped on execution. As we don't have a list of which
3983 groups are referenced, we cannot do this selectively.
3984
3985 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
3986 and do no more at this point. However, we do need to adjust any
3987 OP_RECURSE calls inside the group that refer to the group itself or any
3988 internal or forward referenced group, because the offset is from the
3989 start of the whole regex. Temporarily terminate the pattern while doing
3990 this. */
3991
3992 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
3993 {
3994 *code = OP_END;
3995 adjust_recurse(previous, 1, utf8, cd, save_hwm);
3996 memmove(previous+1, previous, len);
3997 code++;
3998 if (repeat_max == 0)
3999 {
4000 *previous++ = OP_SKIPZERO;
4001 goto END_REPEAT;
4002 }
4003 *previous++ = OP_BRAZERO + repeat_type;
4004 }
4005
4006 /* If the maximum is greater than 1 and limited, we have to replicate
4007 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4008 The first one has to be handled carefully because it's the original
4009 copy, which has to be moved up. The remainder can be handled by code
4010 that is common with the non-zero minimum case below. We have to
4011 adjust the value or repeat_max, since one less copy is required. Once
4012 again, we may have to adjust any OP_RECURSE calls inside the group. */
4013
4014 else
4015 {
4016 int offset;
4017 *code = OP_END;
4018 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4019 memmove(previous + 2 + LINK_SIZE, previous, len);
4020 code += 2 + LINK_SIZE;
4021 *previous++ = OP_BRAZERO + repeat_type;
4022 *previous++ = OP_BRA;
4023
4024 /* We chain together the bracket offset fields that have to be
4025 filled in later when the ends of the brackets are reached. */
4026
4027 offset = (bralink == NULL)? 0 : previous - bralink;
4028 bralink = previous;
4029 PUTINC(previous, 0, offset);
4030 }
4031
4032 repeat_max--;
4033 }
4034
4035 /* If the minimum is greater than zero, replicate the group as many
4036 times as necessary, and adjust the maximum to the number of subsequent
4037 copies that we need. If we set a first char from the group, and didn't
4038 set a required char, copy the latter from the former. If there are any
4039 forward reference subroutine calls in the group, there will be entries on
4040 the workspace list; replicate these with an appropriate increment. */
4041
4042 else
4043 {
4044 if (repeat_min > 1)
4045 {
4046 /* In the pre-compile phase, we don't actually do the replication. We
4047 just adjust the length as if we had. Do some paranoid checks for
4048 potential integer overflow. */
4049
4050 if (lengthptr != NULL)
4051 {
4052 int delta = (repeat_min - 1)*length_prevgroup;
4053 if ((double)(repeat_min - 1)*(double)length_prevgroup >
4054 (double)INT_MAX ||
4055 OFLOW_MAX - *lengthptr < delta)
4056 {
4057 *errorcodeptr = ERR20;
4058 goto FAILED;
4059 }
4060 *lengthptr += delta;
4061 }
4062
4063 /* This is compiling for real */
4064
4065 else
4066 {
4067 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4068 for (i = 1; i < repeat_min; i++)
4069 {
4070 uschar *hc;
4071 uschar *this_hwm = cd->hwm;
4072 memcpy(code, previous, len);
4073 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4074 {
4075 PUT(cd->hwm, 0, GET(hc, 0) + len);
4076 cd->hwm += LINK_SIZE;
4077 }
4078 save_hwm = this_hwm;
4079 code += len;
4080 }
4081 }
4082 }
4083
4084 if (repeat_max > 0) repeat_max -= repeat_min;
4085 }
4086
4087 /* This code is common to both the zero and non-zero minimum cases. If
4088 the maximum is limited, it replicates the group in a nested fashion,
4089 remembering the bracket starts on a stack. In the case of a zero minimum,
4090 the first one was set up above. In all cases the repeat_max now specifies
4091 the number of additional copies needed. Again, we must remember to
4092 replicate entries on the forward reference list. */
4093
4094 if (repeat_max >= 0)
4095 {
4096 /* In the pre-compile phase, we don't actually do the replication. We
4097 just adjust the length as if we had. For each repetition we must add 1
4098 to the length for BRAZERO and for all but the last repetition we must
4099 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4100 paranoid checks to avoid integer overflow. */
4101
4102 if (lengthptr != NULL && repeat_max > 0)
4103 {
4104 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4105 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4106 if ((double)repeat_max *
4107 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4108 > (double)INT_MAX ||
4109 OFLOW_MAX - *lengthptr < delta)
4110 {
4111 *errorcodeptr = ERR20;
4112 goto FAILED;
4113 }
4114 *lengthptr += delta;
4115 }
4116
4117 /* This is compiling for real */
4118
4119 else for (i = repeat_max - 1; i >= 0; i--)
4120 {
4121 uschar *hc;
4122 uschar *this_hwm = cd->hwm;
4123
4124 *code++ = OP_BRAZERO + repeat_type;
4125
4126 /* All but the final copy start a new nesting, maintaining the
4127 chain of brackets outstanding. */
4128
4129 if (i != 0)
4130 {
4131 int offset;
4132 *code++ = OP_BRA;
4133 offset = (bralink == NULL)? 0 : code - bralink;
4134 bralink = code;
4135 PUTINC(code, 0, offset);
4136 }
4137
4138 memcpy(code, previous, len);
4139 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4140 {
4141 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4142 cd->hwm += LINK_SIZE;
4143 }
4144 save_hwm = this_hwm;
4145 code += len;
4146 }
4147
4148 /* Now chain through the pending brackets, and fill in their length
4149 fields (which are holding the chain links pro tem). */
4150
4151 while (bralink != NULL)
4152 {
4153 int oldlinkoffset;
4154 int offset = code - bralink + 1;
4155 uschar *bra = code - offset;
4156 oldlinkoffset = GET(bra, 1);
4157 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4158 *code++ = OP_KET;
4159 PUTINC(code, 0, offset);
4160 PUT(bra, 1, offset);
4161 }
4162 }
4163
4164 /* If the maximum is unlimited, set a repeater in the final copy. We
4165 can't just offset backwards from the current code point, because we
4166 don't know if there's been an options resetting after the ket. The
4167 correct offset was computed above.
4168
4169 Then, when we are doing the actual compile phase, check to see whether
4170 this group is a non-atomic one that could match an empty string. If so,
4171 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4172 that runtime checking can be done. [This check is also applied to
4173 atomic groups at runtime, but in a different way.] */
4174
4175 else
4176 {
4177 uschar *ketcode = code - ketoffset;
4178 uschar *bracode = ketcode - GET(ketcode, 1);
4179 *ketcode = OP_KETRMAX + repeat_type;
4180 if (lengthptr == NULL && *bracode != OP_ONCE)
4181 {
4182 uschar *scode = bracode;
4183 do
4184 {
4185 if (could_be_empty_branch(scode, ketcode, utf8))
4186 {
4187 *bracode += OP_SBRA - OP_BRA;
4188 break;
4189 }
4190 scode += GET(scode, 1);
4191 }
4192 while (*scode == OP_ALT);
4193 }
4194 }
4195 }
4196
4197 /* If previous is OP_FAIL, it was generated by an empty class [] in
4198 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4199 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4200 error above. We can just ignore the repeat in JS case. */
4201
4202 else if (*previous == OP_FAIL) goto END_REPEAT;
4203
4204 /* Else there's some kind of shambles */
4205
4206 else
4207 {
4208 *errorcodeptr = ERR11;
4209 goto FAILED;
4210 }
4211
4212 /* If the character following a repeat is '+', or if certain optimization
4213 tests above succeeded, possessive_quantifier is TRUE. For some of the
4214 simpler opcodes, there is an special alternative opcode for this. For
4215 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4216 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4217 but the special opcodes can optimize it a bit. The repeated item starts at
4218 tempcode, not at previous, which might be the first part of a string whose
4219 (former) last char we repeated.
4220
4221 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4222 an 'upto' may follow. We skip over an 'exact' item, and then test the
4223 length of what remains before proceeding. */
4224
4225 if (possessive_quantifier)
4226 {
4227 int len;
4228 if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
4229 *tempcode == OP_NOTEXACT)
4230 tempcode += _pcre_OP_lengths[*tempcode] +
4231 ((*tempcode == OP_TYPEEXACT &&
4232 (tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP))? 2:0);
4233 len = code - tempcode;
4234 if (len > 0) switch (*tempcode)
4235 {
4236 case OP_STAR: *tempcode = OP_POSSTAR; break;
4237 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4238 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4239 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4240
4241 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4242 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4243 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4244 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4245
4246 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4247 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4248 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4249 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4250
4251 default:
4252 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4253 code += 1 + LINK_SIZE;
4254 len += 1 + LINK_SIZE;
4255 tempcode[0] = OP_ONCE;
4256 *code++ = OP_KET;
4257 PUTINC(code, 0, len);
4258 PUT(tempcode, 1, len);
4259 break;
4260 }
4261 }
4262
4263 /* In all case we no longer have a previous item. We also set the
4264 "follows varying string" flag for subsequently encountered reqbytes if
4265 it isn't already set and we have just passed a varying length item. */
4266
4267 END_REPEAT:
4268 previous = NULL;
4269 cd->req_varyopt |= reqvary;
4270 break;
4271
4272
4273 /* ===================================================================*/
4274 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4275 lookbehind or option setting or condition or all the other extended
4276 parenthesis forms. */
4277
4278 case CHAR_LEFT_PARENTHESIS:
4279 newoptions = options;
4280 skipbytes = 0;
4281 bravalue = OP_CBRA;
4282 save_hwm = cd->hwm;
4283 reset_bracount = FALSE;
4284
4285 /* First deal with various "verbs" that can be introduced by '*'. */
4286
4287 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4288 {
4289 int i, namelen;
4290 const char *vn = verbnames;
4291 const uschar *name = ++ptr;
4292 previous = NULL;
4293 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4294 if (*ptr == CHAR_COLON)
4295 {
4296 *errorcodeptr = ERR59; /* Not supported */
4297 goto FAILED;
4298 }
4299 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4300 {
4301 *errorcodeptr = ERR60;
4302 goto FAILED;
4303 }
4304 namelen = ptr - name;
4305 for (i = 0; i < verbcount; i++)
4306 {
4307 if (namelen == verbs[i].len &&
4308 strncmp((char *)name, vn, namelen) == 0)
4309 {
4310 *code = verbs[i].op;
4311 if (*code++ == OP_ACCEPT) cd->had_accept = TRUE;
4312 break;
4313 }
4314 vn += verbs[i].len + 1;
4315 }
4316 if (i < verbcount) continue;
4317 *errorcodeptr = ERR60;
4318 goto FAILED;
4319 }
4320
4321 /* Deal with the extended parentheses; all are introduced by '?', and the
4322 appearance of any of them means that this is not a capturing group. */
4323
4324 else if (*ptr == CHAR_QUESTION_MARK)
4325 {
4326 int i, set, unset, namelen;
4327 int *optset;
4328 const uschar *name;
4329 uschar *slot;
4330
4331 switch (*(++ptr))
4332 {
4333 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4334 ptr++;
4335 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4336 if (*ptr == 0)
4337 {
4338 *errorcodeptr = ERR18;
4339 goto FAILED;
4340 }
4341 continue;
4342
4343
4344 /* ------------------------------------------------------------ */
4345 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4346 reset_bracount = TRUE;
4347 /* Fall through */
4348
4349 /* ------------------------------------------------------------ */
4350 case CHAR_COLON: /* Non-capturing bracket */
4351 bravalue = OP_BRA;
4352 ptr++;
4353 break;
4354
4355
4356 /* ------------------------------------------------------------ */
4357 case CHAR_LEFT_PARENTHESIS:
4358 bravalue = OP_COND; /* Conditional group */
4359
4360 /* A condition can be an assertion, a number (referring to a numbered
4361 group), a name (referring to a named group), or 'R', referring to
4362 recursion. R<digits> and R&name are also permitted for recursion tests.
4363
4364 There are several syntaxes for testing a named group: (?(name)) is used
4365 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4366
4367 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4368 be the recursive thing or the name 'R' (and similarly for 'R' followed
4369 by digits), and (b) a number could be a name that consists of digits.
4370 In both cases, we look for a name first; if not found, we try the other
4371 cases. */
4372
4373 /* For conditions that are assertions, check the syntax, and then exit
4374 the switch. This will take control down to where bracketed groups,
4375 including assertions, are processed. */
4376
4377 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4378 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4379 break;
4380
4381 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4382 below), and all need to skip 3 bytes at the start of the group. */
4383
4384 code[1+LINK_SIZE] = OP_CREF;
4385 skipbytes = 3;
4386 refsign = -1;
4387
4388 /* Check for a test for recursion in a named group. */
4389
4390 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4391 {
4392 terminator = -1;
4393 ptr += 2;
4394 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4395 }
4396
4397 /* Check for a test for a named group's having been set, using the Perl
4398 syntax (?(<name>) or (?('name') */
4399
4400 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4401 {
4402 terminator = CHAR_GREATER_THAN_SIGN;
4403 ptr++;
4404 }
4405 else if (ptr[1] == CHAR_APOSTROPHE)
4406 {
4407 terminator = CHAR_APOSTROPHE;
4408 ptr++;
4409 }
4410 else
4411 {
4412 terminator = 0;
4413 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4414 }
4415
4416 /* We now expect to read a name; any thing else is an error */
4417
4418 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4419 {
4420 ptr += 1; /* To get the right offset */
4421 *errorcodeptr = ERR28;
4422 goto FAILED;
4423 }
4424
4425 /* Read the name, but also get it as a number if it's all digits */
4426
4427 recno = 0;
4428 name = ++ptr;
4429 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4430 {
4431 if (recno >= 0)
4432 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4433 recno * 10 + *ptr - CHAR_0 : -1;
4434 ptr++;
4435 }
4436 namelen = ptr - name;
4437
4438 if ((terminator > 0 && *ptr++ != terminator) ||
4439 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4440 {
4441 ptr--; /* Error offset */
4442 *errorcodeptr = ERR26;
4443 goto FAILED;
4444 }
4445
4446 /* Do no further checking in the pre-compile phase. */
4447
4448 if (lengthptr != NULL) break;
4449
4450 /* In the real compile we do the work of looking for the actual
4451 reference. If the string started with "+" or "-" we require the rest to
4452 be digits, in which case recno will be set. */
4453
4454 if (refsign > 0)
4455 {
4456 if (recno <= 0)
4457 {
4458 *errorcodeptr = ERR58;
4459 goto FAILED;
4460 }
4461 recno = (refsign == CHAR_MINUS)?
4462 cd->bracount - recno + 1 : recno +cd->bracount;
4463 if (recno <= 0 || recno > cd->final_bracount)
4464 {
4465 *errorcodeptr = ERR15;
4466 goto FAILED;
4467 }
4468 PUT2(code, 2+LINK_SIZE, recno);
4469 break;
4470 }
4471
4472 /* Otherwise (did not start with "+" or "-"), start by looking for the
4473 name. */
4474
4475 slot = cd->name_table;
4476 for (i = 0; i < cd->names_found; i++)
4477 {
4478 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4479 slot += cd->name_entry_size;
4480 }
4481
4482 /* Found a previous named subpattern */
4483
4484 if (i < cd->names_found)
4485 {
4486 recno = GET2(slot, 0);
4487 PUT2(code, 2+LINK_SIZE, recno);
4488 }
4489
4490 /* Search the pattern for a forward reference */
4491
4492 else if ((i = find_parens(ptr, cd, name, namelen,
4493 (options & PCRE_EXTENDED) != 0)) > 0)
4494 {
4495 PUT2(code, 2+LINK_SIZE, i);
4496 }
4497
4498 /* If terminator == 0 it means that the name followed directly after
4499 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4500 some further alternatives to try. For the cases where terminator != 0
4501 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4502 now checked all the possibilities, so give an error. */
4503
4504 else if (terminator != 0)
4505 {
4506 *errorcodeptr = ERR15;
4507 goto FAILED;
4508 }
4509
4510 /* Check for (?(R) for recursion. Allow digits after R to specify a
4511 specific group number. */
4512
4513 else if (*name == CHAR_R)
4514 {
4515 recno = 0;
4516 for (i = 1; i < namelen; i++)
4517 {
4518 if ((digitab[name[i]] & ctype_digit) == 0)
4519 {
4520 *errorcodeptr = ERR15;
4521 goto FAILED;
4522 }
4523 recno = recno * 10 + name[i] - CHAR_0;
4524 }
4525 if (recno == 0) recno = RREF_ANY;
4526 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4527 PUT2(code, 2+LINK_SIZE, recno);
4528 }
4529
4530 /* Similarly, check for the (?(DEFINE) "condition", which is always
4531 false. */
4532
4533 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4534 {
4535 code[1+LINK_SIZE] = OP_DEF;
4536 skipbytes = 1;
4537 }
4538
4539 /* Check for the "name" actually being a subpattern number. We are
4540 in the second pass here, so final_bracount is set. */
4541
4542 else if (recno > 0 && recno <= cd->final_bracount)
4543 {
4544 PUT2(code, 2+LINK_SIZE, recno);
4545 }
4546
4547 /* Either an unidentified subpattern, or a reference to (?(0) */
4548
4549 else
4550 {
4551 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4552 goto FAILED;
4553 }
4554 break;
4555
4556
4557 /* ------------------------------------------------------------ */
4558 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4559 bravalue = OP_ASSERT;
4560 ptr++;
4561 break;
4562
4563
4564 /* ------------------------------------------------------------ */
4565 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4566 ptr++;
4567 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4568 {
4569 *code++ = OP_FAIL;
4570 previous = NULL;
4571 continue;
4572 }
4573 bravalue = OP_ASSERT_NOT;
4574 break;
4575
4576
4577 /* ------------------------------------------------------------ */
4578 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4579 switch (ptr[1])
4580 {
4581 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4582 bravalue = OP_ASSERTBACK;
4583 ptr += 2;
4584 break;
4585
4586 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4587 bravalue = OP_ASSERTBACK_NOT;
4588 ptr += 2;
4589 break;
4590
4591 default: /* Could be name define, else bad */
4592 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4593 ptr++; /* Correct offset for error */
4594 *errorcodeptr = ERR24;
4595 goto FAILED;
4596 }
4597 break;
4598
4599
4600 /* ------------------------------------------------------------ */
4601 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4602 bravalue = OP_ONCE;
4603 ptr++;
4604 break;
4605
4606
4607 /* ------------------------------------------------------------ */
4608 case CHAR_C: /* Callout - may be followed by digits; */
4609 previous_callout = code; /* Save for later completion */
4610 after_manual_callout = 1; /* Skip one item before completing */
4611 *code++ = OP_CALLOUT;
4612 {
4613 int n = 0;
4614 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4615 n = n * 10 + *ptr - CHAR_0;
4616 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4617 {
4618 *errorcodeptr = ERR39;
4619 goto FAILED;
4620 }
4621 if (n > 255)
4622 {
4623 *errorcodeptr = ERR38;
4624 goto FAILED;
4625 }
4626 *code++ = n;
4627 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4628 PUT(code, LINK_SIZE, 0); /* Default length */
4629 code += 2 * LINK_SIZE;
4630 }
4631 previous = NULL;
4632 continue;
4633
4634
4635 /* ------------------------------------------------------------ */
4636 case CHAR_P: /* Python-style named subpattern handling */
4637 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4638 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4639 {
4640 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4641 terminator = CHAR_RIGHT_PARENTHESIS;
4642 goto NAMED_REF_OR_RECURSE;
4643 }
4644 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4645 {
4646 *errorcodeptr = ERR41;
4647 goto FAILED;
4648 }
4649 /* Fall through to handle (?P< as (?< is handled */
4650
4651
4652 /* ------------------------------------------------------------ */
4653 DEFINE_NAME: /* Come here from (?< handling */
4654 case CHAR_APOSTROPHE:
4655 {
4656 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4657 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4658 name = ++ptr;
4659
4660 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4661 namelen = ptr - name;
4662
4663 /* In the pre-compile phase, just do a syntax check. */
4664
4665 if (lengthptr != NULL)
4666 {
4667 if (*ptr != terminator)
4668 {
4669 *errorcodeptr = ERR42;
4670 goto FAILED;
4671 }
4672 if (cd->names_found >= MAX_NAME_COUNT)
4673 {
4674 *errorcodeptr = ERR49;
4675 goto FAILED;
4676 }
4677 if (namelen + 3 > cd->name_entry_size)
4678 {
4679 cd->name_entry_size = namelen + 3;
4680 if (namelen > MAX_NAME_SIZE)
4681 {
4682 *errorcodeptr = ERR48;
4683 goto FAILED;
4684 }
4685 }
4686 }
4687
4688 /* In the real compile, create the entry in the table */
4689
4690 else
4691 {
4692 slot = cd->name_table;
4693 for (i = 0; i < cd->names_found; i++)
4694 {
4695 int crc = memcmp(name, slot+2, namelen);
4696 if (crc == 0)
4697 {
4698 if (slot[2+namelen] == 0)
4699 {
4700 if ((options & PCRE_DUPNAMES) == 0)
4701 {
4702 *errorcodeptr = ERR43;
4703 goto FAILED;
4704 }
4705 }
4706 else crc = -1; /* Current name is substring */
4707 }
4708 if (crc < 0)
4709 {
4710 memmove(slot + cd->name_entry_size, slot,
4711 (cd->names_found - i) * cd->name_entry_size);
4712 break;
4713 }
4714 slot += cd->name_entry_size;
4715 }
4716
4717 PUT2(slot, 0, cd->bracount + 1);
4718 memcpy(slot + 2, name, namelen);
4719 slot[2+namelen] = 0;
4720 }
4721 }
4722
4723 /* In both cases, count the number of names we've encountered. */
4724
4725 ptr++; /* Move past > or ' */
4726 cd->names_found++;
4727 goto NUMBERED_GROUP;
4728
4729
4730 /* ------------------------------------------------------------ */
4731 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
4732 terminator = CHAR_RIGHT_PARENTHESIS;
4733 is_recurse = TRUE;
4734 /* Fall through */
4735
4736 /* We come here from the Python syntax above that handles both
4737 references (?P=name) and recursion (?P>name), as well as falling
4738 through from the Perl recursion syntax (?&name). We also come here from
4739 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4740 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
4741
4742 NAMED_REF_OR_RECURSE:
4743 name = ++ptr;
4744 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4745 namelen = ptr - name;
4746
4747 /* In the pre-compile phase, do a syntax check and set a dummy
4748 reference number. */
4749
4750 if (lengthptr != NULL)
4751 {
4752 if (namelen == 0)
4753 {
4754 *errorcodeptr = ERR62;
4755 goto FAILED;
4756 }
4757 if (*ptr != terminator)
4758 {
4759 *errorcodeptr = ERR42;
4760 goto FAILED;
4761 }
4762 if (namelen > MAX_NAME_SIZE)
4763 {
4764 *errorcodeptr = ERR48;
4765 goto FAILED;
4766 }
4767 recno = 0;
4768 }
4769
4770 /* In the real compile, seek the name in the table. We check the name
4771 first, and then check that we have reached the end of the name in the
4772 table. That way, if the name that is longer than any in the table,
4773 the comparison will fail without reading beyond the table entry. */
4774
4775 else
4776 {
4777 slot = cd->name_table;
4778 for (i = 0; i < cd->names_found; i++)
4779 {
4780 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
4781 slot[2+namelen] == 0)
4782 break;
4783 slot += cd->name_entry_size;
4784 }
4785
4786 if (i < cd->names_found) /* Back reference */
4787 {
4788 recno = GET2(slot, 0);
4789 }
4790 else if ((recno = /* Forward back reference */
4791 find_parens(ptr, cd, name, namelen,
4792 (options & PCRE_EXTENDED) != 0)) <= 0)
4793 {
4794 *errorcodeptr = ERR15;
4795 goto FAILED;
4796 }
4797 }
4798
4799 /* In both phases, we can now go to the code than handles numerical
4800 recursion or backreferences. */
4801
4802 if (is_recurse) goto HANDLE_RECURSION;
4803 else goto HANDLE_REFERENCE;
4804
4805
4806 /* ------------------------------------------------------------ */
4807 case CHAR_R: /* Recursion */
4808 ptr++; /* Same as (?0) */
4809 /* Fall through */
4810
4811
4812 /* ------------------------------------------------------------ */
4813 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
4814 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
4815 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
4816 {
4817 const uschar *called;
4818 terminator = CHAR_RIGHT_PARENTHESIS;
4819
4820 /* Come here from the \g<...> and \g'...' code (Oniguruma
4821 compatibility). However, the syntax has been checked to ensure that
4822 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
4823 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
4824 ever be taken. */
4825
4826 HANDLE_NUMERICAL_RECURSION:
4827
4828 if ((refsign = *ptr) == CHAR_PLUS)
4829 {
4830 ptr++;
4831 if ((digitab[*ptr] & ctype_digit) == 0)
4832 {
4833 *errorcodeptr = ERR63;
4834 goto FAILED;
4835 }
4836 }
4837 else if (refsign == CHAR_MINUS)
4838 {
4839 if ((digitab[ptr[1]] & ctype_digit) == 0)
4840 goto OTHER_CHAR_AFTER_QUERY;
4841 ptr++;
4842 }
4843
4844 recno = 0;
4845 while((digitab[*ptr] & ctype_digit) != 0)
4846 recno = recno * 10 + *ptr++ - CHAR_0;
4847
4848 if (*ptr != terminator)
4849 {
4850 *errorcodeptr = ERR29;
4851 goto FAILED;
4852 }
4853
4854 if (refsign == CHAR_MINUS)
4855 {
4856 if (recno == 0)
4857 {
4858 *errorcodeptr = ERR58;
4859 goto FAILED;
4860 }
4861 recno = cd->bracount - recno + 1;
4862 if (recno <= 0)
4863 {
4864 *errorcodeptr = ERR15;
4865 goto FAILED;
4866 }
4867 }
4868 else if (refsign == CHAR_PLUS)
4869 {
4870 if (recno == 0)
4871 {
4872 *errorcodeptr = ERR58;
4873 goto FAILED;
4874 }
4875 recno += cd->bracount;
4876 }
4877
4878 /* Come here from code above that handles a named recursion */
4879
4880 HANDLE_RECURSION:
4881
4882 previous = code;
4883 called = cd->start_code;
4884
4885 /* When we are actually compiling, find the bracket that is being
4886 referenced. Temporarily end the regex in case it doesn't exist before
4887 this point. If we end up with a forward reference, first check that
4888 the bracket does occur later so we can give the error (and position)
4889 now. Then remember this forward reference in the workspace so it can
4890 be filled in at the end. */
4891
4892 if (lengthptr == NULL)
4893 {
4894 *code = OP_END;
4895 if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
4896
4897 /* Forward reference */
4898
4899 if (called == NULL)
4900 {
4901 if (find_parens(ptr, cd, NULL, recno,
4902 (options & PCRE_EXTENDED) != 0) < 0)
4903 {
4904 *errorcodeptr = ERR15;
4905 goto FAILED;
4906 }
4907 called = cd->start_code + recno;
4908 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
4909 }
4910
4911 /* If not a forward reference, and the subpattern is still open,
4912 this is a recursive call. We check to see if this is a left
4913 recursion that could loop for ever, and diagnose that case. */
4914
4915 else if (GET(called, 1) == 0 &&
4916 could_be_empty(called, code, bcptr, utf8))
4917 {
4918 *errorcodeptr = ERR40;
4919 goto FAILED;
4920 }
4921 }
4922
4923 /* Insert the recursion/subroutine item, automatically wrapped inside
4924 "once" brackets. Set up a "previous group" length so that a
4925 subsequent quantifier will work. */
4926
4927 *code = OP_ONCE;
4928 PUT(code, 1, 2 + 2*LINK_SIZE);
4929 code += 1 + LINK_SIZE;
4930
4931 *code = OP_RECURSE;
4932 PUT(code, 1, called - cd->start_code);
4933 code += 1 + LINK_SIZE;
4934
4935 *code = OP_KET;
4936 PUT(code, 1, 2 + 2*LINK_SIZE);
4937 code += 1 + LINK_SIZE;
4938
4939 length_prevgroup = 3 + 3*LINK_SIZE;
4940 }
4941
4942 /* Can't determine a first byte now */
4943
4944 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4945 continue;
4946
4947
4948 /* ------------------------------------------------------------ */
4949 default: /* Other characters: check option setting */
4950 OTHER_CHAR_AFTER_QUERY:
4951 set = unset = 0;
4952 optset = &set;
4953
4954 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
4955 {
4956 switch (*ptr++)
4957 {
4958 case CHAR_MINUS: optset = &unset; break;
4959
4960 case CHAR_J: /* Record that it changed in the external options */
4961 *optset |= PCRE_DUPNAMES;
4962 cd->external_flags |= PCRE_JCHANGED;
4963 break;
4964
4965 case CHAR_i: *optset |= PCRE_CASELESS; break;
4966 case CHAR_m: *optset |= PCRE_MULTILINE; break;
4967 case CHAR_s: *optset |= PCRE_DOTALL; break;
4968 case CHAR_x: *optset |= PCRE_EXTENDED; break;
4969 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
4970 case CHAR_X: *optset |= PCRE_EXTRA; break;
4971
4972 default: *errorcodeptr = ERR12;
4973 ptr--; /* Correct the offset */
4974 goto FAILED;
4975 }
4976 }
4977
4978 /* Set up the changed option bits, but don't change anything yet. */
4979
4980 newoptions = (options | set) & (~unset);
4981
4982 /* If the options ended with ')' this is not the start of a nested
4983 group with option changes, so the options change at this level. If this
4984 item is right at the start of the pattern, the options can be
4985 abstracted and made external in the pre-compile phase, and ignored in
4986 the compile phase. This can be helpful when matching -- for instance in
4987 caseless checking of required bytes.
4988
4989 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
4990 definitely *not* at the start of the pattern because something has been
4991 compiled. In the pre-compile phase, however, the code pointer can have
4992 that value after the start, because it gets reset as code is discarded
4993 during the pre-compile. However, this can happen only at top level - if
4994 we are within parentheses, the starting BRA will still be present. At
4995 any parenthesis level, the length value can be used to test if anything
4996 has been compiled at that level. Thus, a test for both these conditions
4997 is necessary to ensure we correctly detect the start of the pattern in
4998 both phases.
4999
5000 If we are not at the pattern start, compile code to change the ims
5001 options if this setting actually changes any of them, and reset the
5002 greedy defaults and the case value for firstbyte and reqbyte. */
5003
5004 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5005 {
5006 if (code == cd->start_code + 1 + LINK_SIZE &&
5007 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5008 {
5009 cd->external_options = newoptions;
5010 }
5011 else
5012 {
5013 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5014 {
5015 *code++ = OP_OPT;
5016 *code++ = newoptions & PCRE_IMS;
5017 }
5018 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5019 greedy_non_default = greedy_default ^ 1;
5020 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5021 }
5022
5023 /* Change options at this level, and pass them back for use
5024 in subsequent branches. When not at the start of the pattern, this
5025 information is also necessary so that a resetting item can be
5026 compiled at the end of a group (if we are in a group). */
5027
5028 *optionsptr = options = newoptions;
5029 previous = NULL; /* This item can't be repeated */
5030 continue; /* It is complete */
5031 }
5032
5033 /* If the options ended with ':' we are heading into a nested group
5034 with possible change of options. Such groups are non-capturing and are
5035 not assertions of any kind. All we need to do is skip over the ':';
5036 the newoptions value is handled below. */
5037
5038 bravalue = OP_BRA;
5039 ptr++;
5040 } /* End of switch for character following (? */
5041 } /* End of (? handling */
5042
5043 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5044 all unadorned brackets become non-capturing and behave like (?:...)
5045 brackets. */
5046
5047 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5048 {
5049 bravalue = OP_BRA;
5050 }
5051
5052 /* Else we have a capturing group. */
5053
5054 else
5055 {
5056 NUMBERED_GROUP:
5057 cd->bracount += 1;
5058 PUT2(code, 1+LINK_SIZE, cd->bracount);
5059 skipbytes = 2;
5060 }
5061
5062 /* Process nested bracketed regex. Assertions may not be repeated, but
5063 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5064 non-register variable in order to be able to pass its address because some
5065 compilers complain otherwise. Pass in a new setting for the ims options if
5066 they have changed. */
5067
5068 previous = (bravalue >= OP_ONCE)? code : NULL;
5069 *code = bravalue;
5070 tempcode = code;
5071 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5072 length_prevgroup = 0; /* Initialize for pre-compile phase */
5073
5074 if (!compile_regex(
5075 newoptions, /* The complete new option state */
5076 options & PCRE_IMS, /* The previous ims option state */
5077 &tempcode, /* Where to put code (updated) */
5078 &ptr, /* Input pointer (updated) */
5079 errorcodeptr, /* Where to put an error message */
5080 (bravalue == OP_ASSERTBACK ||
5081 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5082 reset_bracount, /* True if (?| group */
5083 skipbytes, /* Skip over bracket number */
5084 &subfirstbyte, /* For possible first char */
5085 &subreqbyte, /* For possible last char */
5086 bcptr, /* Current branch chain */
5087 cd, /* Tables block */
5088 (lengthptr == NULL)? NULL : /* Actual compile phase */
5089 &length_prevgroup /* Pre-compile phase */
5090 ))
5091 goto FAILED;
5092
5093 /* At the end of compiling, code is still pointing to the start of the
5094 group, while tempcode has been updated to point past the end of the group
5095 and any option resetting that may follow it. The pattern pointer (ptr)
5096 is on the bracket. */
5097
5098 /* If this is a conditional bracket, check that there are no more than
5099 two branches in the group, or just one if it's a DEFINE group. We do this
5100 in the real compile phase, not in the pre-pass, where the whole group may
5101 not be available. */
5102
5103 if (bravalue == OP_COND && lengthptr == NULL)
5104 {
5105 uschar *tc = code;
5106 int condcount = 0;
5107
5108 do {
5109 condcount++;
5110 tc += GET(tc,1);
5111 }
5112 while (*tc != OP_KET);
5113
5114 /* A DEFINE group is never obeyed inline (the "condition" is always
5115 false). It must have only one branch. */
5116
5117 if (code[LINK_SIZE+1] == OP_DEF)
5118 {
5119 if (condcount > 1)
5120 {
5121 *errorcodeptr = ERR54;
5122 goto FAILED;
5123 }
5124 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5125 }
5126
5127 /* A "normal" conditional group. If there is just one branch, we must not
5128 make use of its firstbyte or reqbyte, because this is equivalent to an
5129 empty second branch. */
5130
5131 else
5132 {
5133 if (condcount > 2)
5134 {
5135 *errorcodeptr = ERR27;
5136 goto FAILED;
5137 }
5138 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5139 }
5140 }
5141
5142 /* Error if hit end of pattern */
5143
5144 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5145 {
5146 *errorcodeptr = ERR14;
5147 goto FAILED;
5148 }
5149
5150 /* In the pre-compile phase, update the length by the length of the group,
5151 less the brackets at either end. Then reduce the compiled code to just a
5152 set of non-capturing brackets so that it doesn't use much memory if it is
5153 duplicated by a quantifier.*/
5154
5155 if (lengthptr != NULL)
5156 {
5157 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5158 {
5159 *errorcodeptr = ERR20;
5160 goto FAILED;
5161 }
5162 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5163 *code++ = OP_BRA;
5164 PUTINC(code, 0, 1 + LINK_SIZE);
5165 *code++ = OP_KET;
5166 PUTINC(code, 0, 1 + LINK_SIZE);
5167 break; /* No need to waste time with special character handling */
5168 }
5169
5170 /* Otherwise update the main code pointer to the end of the group. */
5171
5172 code = tempcode;
5173
5174 /* For a DEFINE group, required and first character settings are not
5175 relevant. */
5176
5177 if (bravalue == OP_DEF) break;
5178
5179 /* Handle updating of the required and first characters for other types of
5180 group. Update for normal brackets of all kinds, and conditions with two
5181 branches (see code above). If the bracket is followed by a quantifier with
5182 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5183 zerofirstbyte outside the main loop so that they can be accessed for the
5184 back off. */
5185
5186 zeroreqbyte = reqbyte;
5187 zerofirstbyte = firstbyte;
5188 groupsetfirstbyte = FALSE;
5189
5190 if (bravalue >= OP_ONCE)
5191 {
5192 /* If we have not yet set a firstbyte in this branch, take it from the
5193 subpattern, remembering that it was set here so that a repeat of more
5194 than one can replicate it as reqbyte if necessary. If the subpattern has
5195 no firstbyte, set "none" for the whole branch. In both cases, a zero
5196 repeat forces firstbyte to "none". */
5197
5198 if (firstbyte == REQ_UNSET)
5199 {
5200 if (subfirstbyte >= 0)
5201 {
5202 firstbyte = subfirstbyte;
5203 groupsetfirstbyte = TRUE;
5204 }
5205 else firstbyte = REQ_NONE;
5206 zerofirstbyte = REQ_NONE;
5207 }
5208
5209 /* If firstbyte was previously set, convert the subpattern's firstbyte
5210 into reqbyte if there wasn't one, using the vary flag that was in
5211 existence beforehand. */
5212
5213 else if (subfirstbyte >= 0 && subreqbyte < 0)
5214 subreqbyte = subfirstbyte | tempreqvary;
5215
5216 /* If the subpattern set a required byte (or set a first byte that isn't
5217 really the first byte - see above), set it. */
5218
5219 if (subreqbyte >= 0) reqbyte = subreqbyte;
5220 }
5221
5222 /* For a forward assertion, we take the reqbyte, if set. This can be
5223 helpful if the pattern that follows the assertion doesn't set a different
5224 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5225 for an assertion, however because it leads to incorrect effect for patterns
5226 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5227 of a firstbyte. This is overcome by a scan at the end if there's no
5228 firstbyte, looking for an asserted first char. */
5229
5230 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5231 break; /* End of processing '(' */
5232
5233
5234 /* ===================================================================*/
5235 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5236 are arranged to be the negation of the corresponding OP_values. For the
5237 back references, the values are ESC_REF plus the reference number. Only
5238 back references and those types that consume a character may be repeated.
5239 We can test for values between ESC_b and ESC_Z for the latter; this may
5240 have to change if any new ones are ever created. */
5241
5242 case CHAR_BACKSLASH:
5243 tempptr = ptr;
5244 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5245 if (*errorcodeptr != 0) goto FAILED;
5246
5247 if (c < 0)
5248 {
5249 if (-c == ESC_Q) /* Handle start of quoted string */
5250 {
5251 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5252 ptr += 2; /* avoid empty string */
5253 else inescq = TRUE;
5254 continue;
5255 }
5256
5257 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5258
5259 /* For metasequences that actually match a character, we disable the
5260 setting of a first character if it hasn't already been set. */
5261
5262 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5263 firstbyte = REQ_NONE;
5264
5265 /* Set values to reset to if this is followed by a zero repeat. */
5266
5267 zerofirstbyte = firstbyte;
5268 zeroreqbyte = reqbyte;
5269
5270 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5271 is a subroutine call by number (Oniguruma syntax). In fact, the value
5272 -ESC_g is returned only for these cases. So we don't need to check for <
5273 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5274 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5275 that is a synonym for a named back reference). */
5276
5277 if (-c == ESC_g)
5278 {
5279 const uschar *p;
5280 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5281 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5282 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5283
5284 /* These two statements stop the compiler for warning about possibly
5285 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5286 fact, because we actually check for a number below, the paths that
5287 would actually be in error are never taken. */
5288
5289 skipbytes = 0;
5290 reset_bracount = FALSE;
5291
5292 /* Test for a name */
5293
5294 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5295 {
5296 BOOL isnumber = TRUE;
5297 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5298 {
5299 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5300 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5301 }
5302 if (*p != terminator)
5303 {
5304 *errorcodeptr = ERR57;
5305 break;
5306 }
5307 if (isnumber)
5308 {
5309 ptr++;
5310 goto HANDLE_NUMERICAL_RECURSION;
5311 }
5312 is_recurse = TRUE;
5313 goto NAMED_REF_OR_RECURSE;
5314 }
5315
5316 /* Test a signed number in angle brackets or quotes. */
5317
5318 p = ptr + 2;
5319 while ((digitab[*p] & ctype_digit) != 0) p++;
5320 if (*p != terminator)
5321 {
5322 *errorcodeptr = ERR57;
5323 break;
5324 }
5325 ptr++;
5326 goto HANDLE_NUMERICAL_RECURSION;
5327 }
5328
5329 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5330 We also support \k{name} (.NET syntax) */
5331
5332 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5333 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5334 {
5335 is_recurse = FALSE;
5336 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5337 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5338 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5339 goto NAMED_REF_OR_RECURSE;
5340 }
5341
5342 /* Back references are handled specially; must disable firstbyte if
5343 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5344 ':' later. */
5345
5346 if (-c >= ESC_REF)
5347 {
5348 recno = -c - ESC_REF;
5349
5350 HANDLE_REFERENCE: /* Come here from named backref handling */
5351 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5352 previous = code;
5353 *code++ = OP_REF;
5354 PUT2INC(code, 0, recno);
5355 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5356 if (recno > cd->top_backref) cd->top_backref = recno;
5357 }
5358
5359 /* So are Unicode property matches, if supported. */
5360
5361 #ifdef SUPPORT_UCP
5362 else if (-c == ESC_P || -c == ESC_p)
5363 {
5364 BOOL negated;
5365 int pdata;
5366 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5367 if (ptype < 0) goto FAILED;
5368 previous = code;
5369 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5370 *code++ = ptype;
5371 *code++ = pdata;
5372 }
5373 #else
5374
5375 /* If Unicode properties are not supported, \X, \P, and \p are not
5376 allowed. */
5377
5378 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5379 {
5380 *errorcodeptr = ERR45;
5381 goto FAILED;
5382 }
5383 #endif
5384
5385 /* For the rest (including \X when Unicode properties are supported), we
5386 can obtain the OP value by negating the escape value. */
5387
5388 else
5389 {
5390 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5391 *code++ = -c;
5392 }
5393 continue;
5394 }
5395
5396 /* We have a data character whose value is in c. In UTF-8 mode it may have
5397 a value > 127. We set its representation in the length/buffer, and then
5398 handle it as a data character. */
5399
5400 #ifdef SUPPORT_UTF8
5401 if (utf8 && c > 127)
5402 mclength = _pcre_ord2utf8(c, mcbuffer);
5403 else
5404 #endif
5405
5406 {
5407 mcbuffer[0] = c;
5408 mclength = 1;
5409 }
5410 goto ONE_CHAR;
5411
5412
5413 /* ===================================================================*/
5414 /* Handle a literal character. It is guaranteed not to be whitespace or #
5415 when the extended flag is set. If we are in UTF-8 mode, it may be a
5416 multi-byte literal character. */
5417
5418 default:
5419 NORMAL_CHAR:
5420 mclength = 1;
5421 mcbuffer[0] = c;
5422
5423 #ifdef SUPPORT_UTF8
5424 if (utf8 && c >= 0xc0)
5425 {
5426 while ((ptr[1] & 0xc0) == 0x80)
5427 mcbuffer[mclength++] = *(++ptr);
5428 }
5429 #endif
5430
5431 /* At this point we have the character's bytes in mcbuffer, and the length
5432 in mclength. When not in UTF-8 mode, the length is always 1. */
5433
5434 ONE_CHAR:
5435 previous = code;
5436 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5437 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5438
5439 /* Remember if \r or \n were seen */
5440
5441 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5442 cd->external_flags |= PCRE_HASCRORLF;
5443
5444 /* Set the first and required bytes appropriately. If no previous first
5445 byte, set it from this character, but revert to none on a zero repeat.
5446 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5447 repeat. */
5448
5449 if (firstbyte == REQ_UNSET)
5450 {
5451 zerofirstbyte = REQ_NONE;
5452 zeroreqbyte = reqbyte;
5453
5454 /* If the character is more than one byte long, we can set firstbyte
5455 only if it is not to be matched caselessly. */
5456
5457 if (mclength == 1 || req_caseopt == 0)
5458 {
5459 firstbyte = mcbuffer[0] | req_caseopt;
5460 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5461 }
5462 else firstbyte = reqbyte = REQ_NONE;
5463 }
5464
5465 /* firstbyte was previously set; we can set reqbyte only the length is
5466 1 or the matching is caseful. */
5467
5468 else
5469 {
5470 zerofirstbyte = firstbyte;
5471 zeroreqbyte = reqbyte;
5472 if (mclength == 1 || req_caseopt == 0)
5473 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5474 }
5475
5476 break; /* End of literal character handling */
5477 }
5478 } /* end of big loop */
5479
5480
5481 /* Control never reaches here by falling through, only by a goto for all the
5482 error states. Pass back the position in the pattern so that it can be displayed
5483 to the user for diagnosing the error. */
5484
5485 FAILED:
5486 *ptrptr = ptr;
5487 return FALSE;
5488 }
5489
5490
5491
5492
5493 /*************************************************
5494 * Compile sequence of alternatives *
5495 *************************************************/
5496
5497 /* On entry, ptr is pointing past the bracket character, but on return it
5498 points to the closing bracket, or vertical bar, or end of string. The code
5499 variable is pointing at the byte into which the BRA operator has been stored.
5500 If the ims options are changed at the start (for a (?ims: group) or during any
5501 branch, we need to insert an OP_OPT item at the start of every following branch
5502 to ensure they get set correctly at run time, and also pass the new options
5503 into every subsequent branch compile.
5504
5505 This function is used during the pre-compile phase when we are trying to find
5506 out the amount of memory needed, as well as during the real compile phase. The
5507 value of lengthptr distinguishes the two phases.
5508
5509 Arguments:
5510 options option bits, including any changes for this subpattern
5511 oldims previous settings of ims option bits
5512 codeptr -> the address of the current code pointer
5513 ptrptr -> the address of the current pattern pointer
5514 errorcodeptr -> pointer to error code variable
5515 lookbehind TRUE if this is a lookbehind assertion
5516 reset_bracount TRUE to reset the count for each branch
5517 skipbytes skip this many bytes at start (for brackets and OP_COND)
5518 firstbyteptr place to put the first required character, or a negative number
5519 reqbyteptr place to put the last required character, or a negative number
5520 bcptr pointer to the chain of currently open branches
5521 cd points to the data block with tables pointers etc.
5522 lengthptr NULL during the real compile phase
5523 points to length accumulator during pre-compile phase
5524
5525 Returns: TRUE on success
5526 */
5527
5528 static BOOL
5529 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5530 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5531 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5532 int *lengthptr)
5533 {
5534 const uschar *ptr = *ptrptr;
5535 uschar *code = *codeptr;
5536 uschar *last_branch = code;
5537 uschar *start_bracket = code;
5538 uschar *reverse_count = NULL;
5539 int firstbyte, reqbyte;
5540 int branchfirstbyte, branchreqbyte;
5541 int length;
5542 int orig_bracount;
5543 int max_bracount;
5544 branch_chain bc;
5545
5546 bc.outer = bcptr;
5547 bc.current = code;
5548
5549 firstbyte = reqbyte = REQ_UNSET;
5550
5551 /* Accumulate the length for use in the pre-compile phase. Start with the
5552 length of the BRA and KET and any extra bytes that are required at the
5553 beginning. We accumulate in a local variable to save frequent testing of
5554 lenthptr for NULL. We cannot do this by looking at the value of code at the
5555 start and end of each alternative, because compiled items are discarded during
5556 the pre-compile phase so that the work space is not exceeded. */
5557
5558 length = 2 + 2*LINK_SIZE + skipbytes;
5559
5560 /* WARNING: If the above line is changed for any reason, you must also change
5561 the code that abstracts option settings at the start of the pattern and makes
5562 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5563 pre-compile phase to find out whether anything has yet been compiled or not. */
5564
5565 /* Offset is set zero to mark that this bracket is still open */
5566
5567 PUT(code, 1, 0);
5568 code += 1 + LINK_SIZE + skipbytes;
5569
5570 /* Loop for each alternative branch */
5571
5572 orig_bracount = max_bracount = cd->bracount;
5573 for (;;)
5574 {
5575 /* For a (?| group, reset the capturing bracket count so that each branch
5576 uses the same numbers. */
5577
5578 if (reset_bracount) cd->bracount = orig_bracount;
5579
5580 /* Handle a change of ims options at the start of the branch */
5581
5582 if ((options & PCRE_IMS) != oldims)
5583 {
5584 *code++ = OP_OPT;
5585 *code++ = options & PCRE_IMS;
5586 length += 2;
5587 }
5588
5589 /* Set up dummy OP_REVERSE if lookbehind assertion */
5590
5591 if (lookbehind)
5592 {
5593 *code++ = OP_REVERSE;
5594 reverse_count = code;
5595 PUTINC(code, 0, 0);
5596 length += 1 + LINK_SIZE;
5597 }
5598
5599 /* Now compile the branch; in the pre-compile phase its length gets added
5600 into the length. */
5601
5602 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5603 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5604 {
5605 *ptrptr = ptr;
5606 return FALSE;
5607 }
5608
5609 /* Keep the highest bracket count in case (?| was used and some branch
5610 has fewer than the rest. */
5611
5612 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5613
5614 /* In the real compile phase, there is some post-processing to be done. */
5615
5616 if (lengthptr == NULL)
5617 {
5618 /* If this is the first branch, the firstbyte and reqbyte values for the
5619 branch become the values for the regex. */
5620
5621 if (*last_branch != OP_ALT)
5622 {
5623 firstbyte = branchfirstbyte;
5624 reqbyte = branchreqbyte;
5625 }
5626
5627 /* If this is not the first branch, the first char and reqbyte have to
5628 match the values from all the previous branches, except that if the
5629 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5630 and we set REQ_VARY for the regex. */
5631
5632 else
5633 {
5634 /* If we previously had a firstbyte, but it doesn't match the new branch,
5635 we have to abandon the firstbyte for the regex, but if there was
5636 previously no reqbyte, it takes on the value of the old firstbyte. */
5637
5638 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5639 {
5640 if (reqbyte < 0) reqbyte = firstbyte;
5641 firstbyte = REQ_NONE;
5642 }
5643
5644 /* If we (now or from before) have no firstbyte, a firstbyte from the
5645 branch becomes a reqbyte if there isn't a branch reqbyte. */
5646
5647 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5648 branchreqbyte = branchfirstbyte;
5649
5650 /* Now ensure that the reqbytes match */
5651
5652 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5653 reqbyte = REQ_NONE;
5654 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5655 }
5656
5657 /* If lookbehind, check that this branch matches a fixed-length string, and
5658 put the length into the OP_REVERSE item. Temporarily mark the end of the
5659 branch with OP_END. */
5660
5661 if (lookbehind)
5662 {
5663 int fixed_length;
5664 *code = OP_END;
5665 fixed_length = find_fixedlength(last_branch, options);
5666 DPRINTF(("fixed length = %d\n", fixed_length));
5667 if (fixed_length < 0)
5668 {
5669 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5670 *ptrptr = ptr;
5671 return FALSE;
5672 }
5673 PUT(reverse_count, 0, fixed_length);
5674 }
5675 }
5676
5677 /* Reached end of expression, either ')' or end of pattern. In the real
5678 compile phase, go back through the alternative branches and reverse the chain
5679 of offsets, with the field in the BRA item now becoming an offset to the
5680 first alternative. If there are no alternatives, it points to the end of the
5681 group. The length in the terminating ket is always the length of the whole
5682 bracketed item. If any of the ims options were changed inside the group,
5683 compile a resetting op-code following, except at the very end of the pattern.
5684 Return leaving the pointer at the terminating char. */
5685
5686 if (*ptr != CHAR_VERTICAL_LINE)
5687 {
5688 if (lengthptr == NULL)
5689 {
5690 int branch_length = code - last_branch;
5691 do
5692 {
5693 int prev_length = GET(last_branch, 1);
5694 PUT(last_branch, 1, branch_length);
5695 branch_length = prev_length;
5696 last_branch -= branch_length;
5697 }
5698 while (branch_length > 0);
5699 }
5700
5701 /* Fill in the ket */
5702
5703 *code = OP_KET;
5704 PUT(code, 1, code - start_bracket);
5705 code += 1 + LINK_SIZE;
5706
5707 /* Resetting option if needed */
5708
5709 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
5710 {
5711 *code++ = OP_OPT;
5712 *code++ = oldims;
5713 length += 2;
5714 }
5715
5716 /* Retain the highest bracket number, in case resetting was used. */
5717
5718 cd->bracount = max_bracount;
5719
5720 /* Set values to pass back */
5721
5722 *codeptr = code;
5723 *ptrptr = ptr;
5724 *firstbyteptr = firstbyte;
5725 *reqbyteptr = reqbyte;
5726 if (lengthptr != NULL)
5727 {
5728 if (OFLOW_MAX - *lengthptr < length)
5729 {
5730 *errorcodeptr = ERR20;
5731 return FALSE;
5732 }
5733 *lengthptr += length;
5734 }
5735 return TRUE;
5736 }
5737
5738 /* Another branch follows. In the pre-compile phase, we can move the code
5739 pointer back to where it was for the start of the first branch. (That is,
5740 pretend that each branch is the only one.)
5741
5742 In the real compile phase, insert an ALT node. Its length field points back
5743 to the previous branch while the bracket remains open. At the end the chain
5744 is reversed. It's done like this so that the start of the bracket has a
5745 zero offset until it is closed, making it possible to detect recursion. */
5746
5747 if (lengthptr != NULL)
5748 {
5749 code = *codeptr + 1 + LINK_SIZE + skipbytes;
5750 length += 1 + LINK_SIZE;
5751 }
5752 else
5753 {
5754 *code = OP_ALT;
5755 PUT(code, 1, code - last_branch);
5756 bc.current = last_branch = code;
5757 code += 1 + LINK_SIZE;
5758 }
5759
5760 ptr++;
5761 }
5762 /* Control never reaches here */
5763 }
5764
5765
5766
5767
5768 /*************************************************
5769 * Check for anchored expression *
5770 *************************************************/
5771
5772 /* Try to find out if this is an anchored regular expression. Consider each
5773 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
5774 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
5775 it's anchored. However, if this is a multiline pattern, then only OP_SOD
5776 counts, since OP_CIRC can match in the middle.
5777
5778 We can also consider a regex to be anchored if OP_SOM starts all its branches.
5779 This is the code for \G, which means "match at start of match position, taking
5780 into account the match offset".
5781
5782 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
5783 because that will try the rest of the pattern at all possible matching points,
5784 so there is no point trying again.... er ....
5785
5786 .... except when the .* appears inside capturing parentheses, and there is a
5787 subsequent back reference to those parentheses. We haven't enough information
5788 to catch that case precisely.
5789
5790 At first, the best we could do was to detect when .* was in capturing brackets
5791 and the highest back reference was greater than or equal to that level.
5792 However, by keeping a bitmap of the first 31 back references, we can catch some
5793 of the more common cases more precisely.
5794
5795 Arguments:
5796 code points to start of expression (the bracket)
5797 options points to the options setting
5798 bracket_map a bitmap of which brackets we are inside while testing; this
5799 handles up to substring 31; after that we just have to take
5800 the less precise approach
5801 backref_map the back reference bitmap
5802
5803 Returns: TRUE or FALSE
5804 */
5805
5806 static BOOL
5807 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
5808 unsigned int backref_map)
5809 {
5810 do {
5811 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5812 options, PCRE_MULTILINE, FALSE);
5813 register int op = *scode;
5814
5815 /* Non-capturing brackets */
5816
5817 if (op == OP_BRA)
5818 {
5819 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5820 }
5821
5822 /* Capturing brackets */
5823
5824 else if (op == OP_CBRA)
5825 {
5826 int n = GET2(scode, 1+LINK_SIZE);
5827 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5828 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
5829 }
5830
5831 /* Other brackets */
5832
5833 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
5834 {
5835 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
5836 }
5837
5838 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
5839 it isn't in brackets that are or may be referenced. */
5840
5841 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
5842 op == OP_TYPEPOSSTAR))
5843 {
5844 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
5845 return FALSE;
5846 }
5847
5848 /* Check for explicit anchoring */
5849
5850 else if (op != OP_SOD && op != OP_SOM &&
5851 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
5852 return FALSE;
5853 code += GET(code, 1);
5854 }
5855 while (*code == OP_ALT); /* Loop for each alternative */
5856 return TRUE;
5857 }
5858
5859
5860
5861 /*************************************************
5862 * Check for starting with ^ or .* *
5863 *************************************************/
5864
5865 /* This is called to find out if every branch starts with ^ or .* so that
5866 "first char" processing can be done to speed things up in multiline
5867 matching and for non-DOTALL patterns that start with .* (which must start at
5868 the beginning or after \n). As in the case of is_anchored() (see above), we
5869 have to take account of back references to capturing brackets that contain .*
5870 because in that case we can't make the assumption.
5871
5872 Arguments:
5873 code points to start of expression (the bracket)
5874 bracket_map a bitmap of which brackets we are inside while testing; this
5875 handles up to substring 31; after that we just have to take
5876 the less precise approach
5877 backref_map the back reference bitmap
5878
5879 Returns: TRUE or FALSE
5880 */
5881
5882 static BOOL
5883 is_startline(const uschar *code, unsigned int bracket_map,
5884 unsigned int backref_map)
5885 {
5886 do {
5887 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
5888 NULL, 0, FALSE);
5889 register int op = *scode;
5890
5891 /* If we are at the start of a conditional assertion group, *both* the
5892 conditional assertion *and* what follows the condition must satisfy the test
5893 for start of line. Other kinds of condition fail. Note that there may be an
5894 auto-callout at the start of a condition. */
5895
5896 if (op == OP_COND)
5897 {
5898 scode += 1 + LINK_SIZE;
5899 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
5900 switch (*scode)
5901 {
5902 case OP_CREF:
5903 case OP_RREF:
5904 case OP_DEF:
5905 return FALSE;
5906
5907 default: /* Assertion */
5908 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5909 do scode += GET(scode, 1); while (*scode == OP_ALT);
5910 scode += 1 + LINK_SIZE;
5911 break;
5912 }
5913 scode = first_significant_code(scode, NULL, 0, FALSE);
5914 op = *scode;
5915 }
5916
5917 /* Non-capturing brackets */
5918
5919 if (op == OP_BRA)
5920 {
5921 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5922 }
5923
5924 /* Capturing brackets */
5925
5926 else if (op == OP_CBRA)
5927 {
5928 int n = GET2(scode, 1+LINK_SIZE);
5929 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
5930 if (!is_startline(scode, new_map, backref_map)) return FALSE;
5931 }
5932
5933 /* Other brackets */
5934
5935 else if (op == OP_ASSERT || op == OP_ONCE)
5936 {
5937 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
5938 }
5939
5940 /* .* means "start at start or after \n" if it isn't in brackets that
5941 may be referenced. */
5942
5943 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
5944 {
5945 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
5946 }
5947
5948 /* Check for explicit circumflex */
5949
5950 else if (op != OP_CIRC) return FALSE;
5951
5952 /* Move on to the next alternative */
5953
5954 code += GET(code, 1);
5955 }
5956 while (*code == OP_ALT); /* Loop for each alternative */
5957 return TRUE;
5958 }
5959
5960
5961
5962 /*************************************************
5963 * Check for asserted fixed first char *
5964 *************************************************/
5965
5966 /* During compilation, the "first char" settings from forward assertions are
5967 discarded, because they can cause conflicts with actual literals that follow.
5968 However, if we end up without a first char setting for an unanchored pattern,
5969 it is worth scanning the regex to see if there is an initial asserted first
5970 char. If all branches start with the same asserted char, or with a bracket all
5971 of whose alternatives start with the same asserted char (recurse ad lib), then
5972 we return that char, otherwise -1.
5973
5974 Arguments:
5975 code points to start of expression (the bracket)
5976 options pointer to the options (used to check casing changes)
5977 inassert TRUE if in an assertion
5978
5979 Returns: -1 or the fixed first char
5980 */
5981
5982 static int
5983 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
5984 {
5985 register int c = -1;
5986 do {
5987 int d;
5988 const uschar *scode =
5989 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
5990 register int op = *scode;
5991
5992 switch(op)
5993 {
5994 default:
5995 return -1;
5996
5997 case OP_BRA:
5998 case OP_CBRA:
5999 case OP_ASSERT:
6000 case OP_ONCE:
6001 case OP_COND:
6002 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6003 return -1;
6004 if (c < 0) c = d; else if (c != d) return -1;
6005 break;
6006
6007 case OP_EXACT: /* Fall through */
6008 scode += 2;
6009
6010 case OP_CHAR:
6011 case OP_CHARNC:
6012 case OP_PLUS:
6013 case OP_MINPLUS:
6014 case OP_POSPLUS:
6015 if (!inassert) return -1;
6016 if (c < 0)
6017 {
6018 c = scode[1];
6019 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6020 }
6021 else if (c != scode[1]) return -1;
6022 break;
6023 }
6024
6025 code += GET(code, 1);
6026 }
6027 while (*code == OP_ALT);
6028 return c;
6029 }
6030
6031
6032
6033 /*************************************************
6034 * Compile a Regular Expression *
6035 *************************************************/
6036
6037 /* This function takes a string and returns a pointer to a block of store
6038 holding a compiled version of the expression. The original API for this
6039 function had no error code return variable; it is retained for backwards
6040 compatibility. The new function is given a new name.
6041
6042 Arguments:
6043 pattern the regular expression
6044 options various option bits
6045 errorcodeptr pointer to error code variable (pcre_compile2() only)
6046 can be NULL if you don't want a code value
6047 errorptr pointer to pointer to error text
6048 erroroffset ptr offset in pattern where error was detected
6049 tables pointer to character tables or NULL
6050
6051 Returns: pointer to compiled data block, or NULL on error,
6052 with errorptr and erroroffset set
6053 */
6054
6055 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6056 pcre_compile(const char *pattern, int options, const char **errorptr,
6057 int *erroroffset, const unsigned char *tables)
6058 {
6059 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6060 }
6061
6062
6063 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6064 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6065 const char **errorptr, int *erroroffset, const unsigned char *tables)
6066 {
6067 real_pcre *re;
6068 int length = 1; /* For final END opcode */
6069 int firstbyte, reqbyte, newline;
6070 int errorcode = 0;
6071 int skipatstart = 0;
6072 #ifdef SUPPORT_UTF8
6073 BOOL utf8;
6074 #endif
6075 size_t size;
6076 uschar *code;
6077 const uschar *codestart;
6078 const uschar *ptr;
6079 compile_data compile_block;
6080 compile_data *cd = &compile_block;
6081
6082 /* This space is used for "compiling" into during the first phase, when we are
6083 computing the amount of memory that is needed. Compiled items are thrown away
6084 as soon as possible, so that a fairly large buffer should be sufficient for
6085 this purpose. The same space is used in the second phase for remembering where
6086 to fill in forward references to subpatterns. */
6087
6088 uschar cworkspace[COMPILE_WORK_SIZE];
6089
6090 /* Set this early so that early errors get offset 0. */
6091
6092 ptr = (const uschar *)pattern;
6093
6094 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6095 can do is just return NULL, but we can set a code value if there is a code
6096 pointer. */
6097
6098 if (errorptr == NULL)
6099 {
6100 if (errorcodeptr != NULL) *errorcodeptr = 99;
6101 return NULL;
6102 }
6103
6104 *errorptr = NULL;
6105 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6106
6107 /* However, we can give a message for this error */
6108
6109 if (erroroffset == NULL)
6110 {
6111 errorcode = ERR16;
6112 goto PCRE_EARLY_ERROR_RETURN2;
6113 }
6114
6115 *erroroffset = 0;
6116
6117 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6118
6119 #ifdef SUPPORT_UTF8
6120 utf8 = (options & PCRE_UTF8) != 0;
6121 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6122 (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
6123 {
6124 errorcode = ERR44;
6125 goto PCRE_EARLY_ERROR_RETURN2;
6126 }
6127 #else
6128 if ((options & PCRE_UTF8) != 0)
6129 {
6130 errorcode = ERR32;
6131 goto PCRE_EARLY_ERROR_RETURN;
6132 }
6133 #endif
6134
6135 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6136 {
6137 errorcode = ERR17;
6138 goto PCRE_EARLY_ERROR_RETURN;
6139 }
6140
6141 /* Set up pointers to the individual character tables */
6142
6143 if (tables == NULL) tables = _pcre_default_tables;
6144 cd->lcc = tables + lcc_offset;
6145 cd->fcc = tables + fcc_offset;
6146 cd->cbits = tables + cbits_offset;
6147 cd->ctypes = tables + ctypes_offset;
6148
6149 /* Check for global one-time settings at the start of the pattern, and remember
6150 the offset for later. */
6151
6152 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6153 ptr[skipatstart+1] == CHAR_ASTERISK)
6154 {
6155 int newnl = 0;
6156 int newbsr = 0;
6157
6158 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6159 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6160 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6161 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6162 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6163 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6164 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6165 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6166 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6167 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6168
6169 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6170 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6171 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6172 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6173
6174 if (newnl != 0)
6175 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6176 else if (newbsr != 0)
6177 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6178 else break;
6179 }
6180
6181 /* Check validity of \R options. */
6182
6183 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6184 {
6185 case 0:
6186 case PCRE_BSR_ANYCRLF:
6187 case PCRE_BSR_UNICODE:
6188 break;
6189 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6190 }
6191
6192 /* Handle different types of newline. The three bits give seven cases. The
6193 current code allows for fixed one- or two-byte sequences, plus "any" and
6194 "anycrlf". */
6195
6196 switch (options & PCRE_NEWLINE_BITS)
6197 {
6198 case 0: newline = NEWLINE; break; /* Build-time default */
6199 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6200 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6201 case PCRE_NEWLINE_CR+
6202 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6203 case PCRE_NEWLINE_ANY: newline = -1; break;
6204 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6205 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6206 }
6207
6208 if (newline == -2)
6209 {
6210 cd->nltype = NLTYPE_ANYCRLF;
6211 }
6212 else if (newline < 0)
6213 {
6214 cd->nltype = NLTYPE_ANY;
6215 }
6216 else
6217 {
6218 cd->nltype = NLTYPE_FIXED;
6219 if (newline > 255)
6220 {
6221 cd->nllen = 2;
6222 cd->nl[0] = (newline >> 8) & 255;
6223 cd->nl[1] = newline & 255;
6224 }
6225 else
6226 {
6227 cd->nllen = 1;
6228 cd->nl[0] = newline;
6229 }
6230 }
6231
6232 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6233 references to help in deciding whether (.*) can be treated as anchored or not.
6234 */
6235
6236 cd->top_backref = 0;
6237 cd->backref_map = 0;
6238
6239 /* Reflect pattern for debugging output */
6240
6241 DPRINTF(("------------------------------------------------------------------\n"));
6242 DPRINTF(("%s\n", pattern));
6243
6244 /* Pretend to compile the pattern while actually just accumulating the length
6245 of memory required. This behaviour is triggered by passing a non-NULL final
6246 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6247 to compile parts of the pattern into; the compiled code is discarded when it is
6248 no longer needed, so hopefully this workspace will never overflow, though there
6249 is a test for its doing so. */
6250
6251 cd->bracount = cd->final_bracount = 0;
6252 cd->names_found = 0;
6253 cd->name_entry_size = 0;
6254 cd->name_table = NULL;
6255 cd->start_workspace = cworkspace;
6256 cd->start_code = cworkspace;
6257 cd->hwm = cworkspace;
6258 cd->start_pattern = (const uschar *)pattern;
6259 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6260 cd->req_varyopt = 0;
6261 cd->external_options = options;
6262 cd->external_flags = 0;
6263
6264 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6265 don't need to look at the result of the function here. The initial options have
6266 been put into the cd block so that they can be changed if an option setting is
6267 found within the regex right at the beginning. Bringing initial option settings
6268 outside can help speed up starting point checks. */
6269
6270 ptr += skipatstart;
6271 code = cworkspace;
6272 *code = OP_BRA;
6273 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6274 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6275 &length);
6276 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6277
6278 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6279 cd->hwm - cworkspace));
6280
6281 if (length > MAX_PATTERN_SIZE)
6282 {
6283 errorcode = ERR20;
6284 goto PCRE_EARLY_ERROR_RETURN;
6285 }
6286
6287 /* Compute the size of data block needed and get it, either from malloc or
6288 externally provided function. Integer overflow should no longer be possible
6289 because nowadays we limit the maximum value of cd->names_found and
6290 cd->name_entry_size. */
6291
6292 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6293 re = (real_pcre *)(pcre_malloc)(size);
6294
6295 if (re == NULL)
6296 {
6297 errorcode = ERR21;
6298 goto PCRE_EARLY_ERROR_RETURN;
6299 }
6300
6301 /* Put in the magic number, and save the sizes, initial options, internal
6302 flags, and character table pointer. NULL is used for the default character
6303 tables. The nullpad field is at the end; it's there to help in the case when a
6304 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6305 pointers. */
6306
6307 re->magic_number = MAGIC_NUMBER;
6308 re->size = size;
6309 re->options = cd->external_options;
6310 re->flags = cd->external_flags;
6311 re->dummy1 = 0;
6312 re->first_byte = 0;
6313 re->req_byte = 0;
6314 re->name_table_offset = sizeof(real_pcre);
6315 re->name_entry_size = cd->name_entry_size;
6316 re->name_count = cd->names_found;
6317 re->ref_count = 0;
6318 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6319 re->nullpad = NULL;
6320
6321 /* The starting points of the name/number translation table and of the code are
6322 passed around in the compile data block. The start/end pattern and initial
6323 options are already set from the pre-compile phase, as is the name_entry_size
6324 field. Reset the bracket count and the names_found field. Also reset the hwm
6325 field; this time it's used for remembering forward references to subpatterns.
6326 */
6327
6328 cd->final_bracount = cd->bracount; /* Save for checking forward references */
6329 cd->bracount = 0;
6330 cd->names_found = 0;
6331 cd->name_table = (uschar *)re + re->name_table_offset;
6332 codestart = cd->name_table + re->name_entry_size * re->name_count;
6333 cd->start_code = codestart;
6334 cd->hwm = cworkspace;
6335 cd->req_varyopt = 0;
6336 cd->had_accept = FALSE;
6337
6338 /* Set up a starting, non-extracting bracket, then compile the expression. On
6339 error, errorcode will be set non-zero, so we don't need to look at the result
6340 of the function here. */
6341
6342 ptr = (const uschar *)pattern + skipatstart;
6343 code = (uschar *)codestart;
6344 *code = OP_BRA;
6345 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
6346 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
6347 re->top_bracket = cd->bracount;
6348 re->top_backref = cd->top_backref;
6349 re->flags = cd->external_flags;
6350
6351 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
6352
6353 /* If not reached end of pattern on success, there's an excess bracket. */
6354
6355 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
6356
6357 /* Fill in the terminating state and check for disastrous overflow, but
6358 if debugging, leave the test till after things are printed out. */
6359
6360 *code++ = OP_END;
6361
6362 #ifndef DEBUG
6363 if (code - codestart > length) errorcode = ERR23;
6364 #endif
6365
6366 /* Fill in any forward references that are required. */
6367
6368 while (errorcode == 0 && cd->hwm > cworkspace)
6369 {
6370 int offset, recno;
6371 const uschar *groupptr;
6372 cd->hwm -= LINK_SIZE;
6373 offset = GET(cd->hwm, 0);
6374 recno = GET(codestart, offset);
6375 groupptr = find_bracket(codestart, (re->options & PCRE_UTF8) != 0, recno);
6376 if (groupptr == NULL) errorcode = ERR53;
6377 else PUT(((uschar *)codestart), offset, groupptr - codestart);
6378 }
6379
6380 /* Give an error if there's back reference to a non-existent capturing
6381 subpattern. */
6382
6383 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
6384
6385 /* Failed to compile, or error while post-processing */
6386
6387 if (errorcode != 0)
6388 {
6389 (pcre_free)(re);
6390 PCRE_EARLY_ERROR_RETURN:
6391 *erroroffset = ptr - (const uschar *)pattern;
6392 PCRE_EARLY_ERROR_RETURN2:
6393 *errorptr = find_error_text(errorcode);
6394 if (errorcodeptr != NULL) *errorcodeptr = errorcode;
6395 return NULL;
6396 }
6397
6398 /* If the anchored option was not passed, set the flag if we can determine that
6399 the pattern is anchored by virtue of ^ characters or \A or anything else (such
6400 as starting with .* when DOTALL is set).
6401
6402 Otherwise, if we know what the first byte has to be, save it, because that
6403 speeds up unanchored matches no end. If not, see if we can set the
6404 PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
6405 start with ^. and also when all branches start with .* for non-DOTALL matches.
6406 */
6407
6408 if ((re->options & PCRE_ANCHORED) == 0)
6409 {
6410 int temp_options = re->options; /* May get changed during these scans */
6411 if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
6412 re->options |= PCRE_ANCHORED;
6413 else
6414 {
6415 if (firstbyte < 0)
6416 firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
6417 if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */
6418 {
6419 int ch = firstbyte & 255;
6420 re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
6421 cd->fcc[ch] == ch)? ch : firstbyte;
6422 re->flags |= PCRE_FIRSTSET;
6423 }
6424 else if (is_startline(codestart, 0, cd->backref_map))
6425 re->flags |= PCRE_STARTLINE;
6426 }
6427 }
6428
6429 /* For an anchored pattern, we use the "required byte" only if it follows a
6430 variable length item in the regex. Remove the caseless flag for non-caseable
6431 bytes. */
6432
6433 if (reqbyte >= 0 &&
6434 ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0))
6435 {
6436 int ch = reqbyte & 255;
6437 re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
6438 cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
6439 re->flags |= PCRE_REQCHSET;
6440 }
6441
6442 /* Print out the compiled data if debugging is enabled. This is never the
6443 case when building a production library. */
6444
6445 #ifdef DEBUG
6446
6447 printf("Length = %d top_bracket = %d top_backref = %d\n",
6448 length, re->top_bracket, re->top_backref);
6449
6450 printf("Options=%08x\n", re->options);
6451
6452 if ((re->flags & PCRE_FIRSTSET) != 0)
6453 {
6454 int ch = re->first_byte & 255;
6455 const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
6456 "" : " (caseless)";
6457 if (isprint(ch)) printf("First char = %c%s\n", ch, caseless);
6458 else printf("First char = \\x%02x%s\n", ch, caseless);
6459 }
6460
6461 if ((re->flags & PCRE_REQCHSET) != 0)
6462 {
6463 int ch = re->req_byte & 255;
6464 const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
6465 "" : " (caseless)";
6466 if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless);
6467 else printf("Req char = \\x%02x%s\n", ch, caseless);
6468 }
6469
6470 pcre_printint(re, stdout, TRUE);
6471
6472 /* This check is done here in the debugging case so that the code that
6473 was compiled can be seen. */
6474
6475 if (code - codestart > length)
6476 {
6477 (pcre_free)(re);
6478 *errorptr = find_error_text(ERR23);
6479 *erroroffset = ptr - (uschar *)pattern;
6480 if (errorcodeptr != NULL) *errorcodeptr = ERR23;
6481 return NULL;
6482 }
6483 #endif /* DEBUG */
6484
6485 return (pcre *)re;
6486 }
6487
6488 /* 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