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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 504 - (show annotations) (download)
Mon Mar 8 08:57:04 2010 UTC (4 years, 7 months ago) by ph10
File MIME type: text/plain
File size: 223108 byte(s)
Previous patch for fixing problem with recursion loop checking was incorrect.

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) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95
96 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
97 are simple data values; negative values are for special things like \d and so
98 on. Zero means further processing is needed (for things like \x), or the escape
99 is invalid. */
100
101 #ifndef EBCDIC
102
103 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
104 in UTF-8 mode. */
105
106 static const short int escapes[] = {
107 0, 0,
108 0, 0,
109 0, 0,
110 0, 0,
111 0, 0,
112 CHAR_COLON, CHAR_SEMICOLON,
113 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
114 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
115 CHAR_COMMERCIAL_AT, -ESC_A,
116 -ESC_B, -ESC_C,
117 -ESC_D, -ESC_E,
118 0, -ESC_G,
119 -ESC_H, 0,
120 0, -ESC_K,
121 0, 0,
122 0, 0,
123 -ESC_P, -ESC_Q,
124 -ESC_R, -ESC_S,
125 0, 0,
126 -ESC_V, -ESC_W,
127 -ESC_X, 0,
128 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
129 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
130 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
131 CHAR_GRAVE_ACCENT, 7,
132 -ESC_b, 0,
133 -ESC_d, ESC_e,
134 ESC_f, 0,
135 -ESC_h, 0,
136 0, -ESC_k,
137 0, 0,
138 ESC_n, 0,
139 -ESC_p, 0,
140 ESC_r, -ESC_s,
141 ESC_tee, 0,
142 -ESC_v, -ESC_w,
143 0, 0,
144 -ESC_z
145 };
146
147 #else
148
149 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
150
151 static const short int escapes[] = {
152 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
153 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
154 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
155 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
156 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
157 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
158 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
159 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
160 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
161 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
162 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
163 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
164 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
165 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
166 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
167 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
168 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
169 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
170 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
171 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
172 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
173 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
174 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
175 };
176 #endif
177
178
179 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
180 searched linearly. Put all the names into a single string, in order to reduce
181 the number of relocations when a shared library is dynamically linked. The
182 string is built from string macros so that it works in UTF-8 mode on EBCDIC
183 platforms. */
184
185 typedef struct verbitem {
186 int len;
187 int op;
188 } verbitem;
189
190 static const char verbnames[] =
191 STRING_ACCEPT0
192 STRING_COMMIT0
193 STRING_F0
194 STRING_FAIL0
195 STRING_PRUNE0
196 STRING_SKIP0
197 STRING_THEN;
198
199 static const verbitem verbs[] = {
200 { 6, OP_ACCEPT },
201 { 6, OP_COMMIT },
202 { 1, OP_FAIL },
203 { 4, OP_FAIL },
204 { 5, OP_PRUNE },
205 { 4, OP_SKIP },
206 { 4, OP_THEN }
207 };
208
209 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
210
211
212 /* Tables of names of POSIX character classes and their lengths. The names are
213 now all in a single string, to reduce the number of relocations when a shared
214 library is dynamically loaded. The list of lengths is terminated by a zero
215 length entry. The first three must be alpha, lower, upper, as this is assumed
216 for handling case independence. */
217
218 static const char posix_names[] =
219 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
220 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
221 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
222 STRING_word0 STRING_xdigit;
223
224 static const uschar posix_name_lengths[] = {
225 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
226
227 /* Table of class bit maps for each POSIX class. Each class is formed from a
228 base map, with an optional addition or removal of another map. Then, for some
229 classes, there is some additional tweaking: for [:blank:] the vertical space
230 characters are removed, and for [:alpha:] and [:alnum:] the underscore
231 character is removed. The triples in the table consist of the base map offset,
232 second map offset or -1 if no second map, and a non-negative value for map
233 addition or a negative value for map subtraction (if there are two maps). The
234 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
235 remove vertical space characters, 2 => remove underscore. */
236
237 static const int posix_class_maps[] = {
238 cbit_word, cbit_digit, -2, /* alpha */
239 cbit_lower, -1, 0, /* lower */
240 cbit_upper, -1, 0, /* upper */
241 cbit_word, -1, 2, /* alnum - word without underscore */
242 cbit_print, cbit_cntrl, 0, /* ascii */
243 cbit_space, -1, 1, /* blank - a GNU extension */
244 cbit_cntrl, -1, 0, /* cntrl */
245 cbit_digit, -1, 0, /* digit */
246 cbit_graph, -1, 0, /* graph */
247 cbit_print, -1, 0, /* print */
248 cbit_punct, -1, 0, /* punct */
249 cbit_space, -1, 0, /* space */
250 cbit_word, -1, 0, /* word - a Perl extension */
251 cbit_xdigit,-1, 0 /* xdigit */
252 };
253
254
255 #define STRING(a) # a
256 #define XSTRING(s) STRING(s)
257
258 /* The texts of compile-time error messages. These are "char *" because they
259 are passed to the outside world. Do not ever re-use any error number, because
260 they are documented. Always add a new error instead. Messages marked DEAD below
261 are no longer used. This used to be a table of strings, but in order to reduce
262 the number of relocations needed when a shared library is loaded dynamically,
263 it is now one long string. We cannot use a table of offsets, because the
264 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
265 simply count through to the one we want - this isn't a performance issue
266 because these strings are used only when there is a compilation error.
267
268 Each substring ends with \0 to insert a null character. This includes the final
269 substring, so that the whole string ends with \0\0, which can be detected when
270 counting through. */
271
272 static const char error_texts[] =
273 "no error\0"
274 "\\ at end of pattern\0"
275 "\\c at end of pattern\0"
276 "unrecognized character follows \\\0"
277 "numbers out of order in {} quantifier\0"
278 /* 5 */
279 "number too big in {} quantifier\0"
280 "missing terminating ] for character class\0"
281 "invalid escape sequence in character class\0"
282 "range out of order in character class\0"
283 "nothing to repeat\0"
284 /* 10 */
285 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
286 "internal error: unexpected repeat\0"
287 "unrecognized character after (? or (?-\0"
288 "POSIX named classes are supported only within a class\0"
289 "missing )\0"
290 /* 15 */
291 "reference to non-existent subpattern\0"
292 "erroffset passed as NULL\0"
293 "unknown option bit(s) set\0"
294 "missing ) after comment\0"
295 "parentheses nested too deeply\0" /** DEAD **/
296 /* 20 */
297 "regular expression is too large\0"
298 "failed to get memory\0"
299 "unmatched parentheses\0"
300 "internal error: code overflow\0"
301 "unrecognized character after (?<\0"
302 /* 25 */
303 "lookbehind assertion is not fixed length\0"
304 "malformed number or name after (?(\0"
305 "conditional group contains more than two branches\0"
306 "assertion expected after (?(\0"
307 "(?R or (?[+-]digits must be followed by )\0"
308 /* 30 */
309 "unknown POSIX class name\0"
310 "POSIX collating elements are not supported\0"
311 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
312 "spare error\0" /** DEAD **/
313 "character value in \\x{...} sequence is too large\0"
314 /* 35 */
315 "invalid condition (?(0)\0"
316 "\\C not allowed in lookbehind assertion\0"
317 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
318 "number after (?C is > 255\0"
319 "closing ) for (?C expected\0"
320 /* 40 */
321 "recursive call could loop indefinitely\0"
322 "unrecognized character after (?P\0"
323 "syntax error in subpattern name (missing terminator)\0"
324 "two named subpatterns have the same name\0"
325 "invalid UTF-8 string\0"
326 /* 45 */
327 "support for \\P, \\p, and \\X has not been compiled\0"
328 "malformed \\P or \\p sequence\0"
329 "unknown property name after \\P or \\p\0"
330 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
331 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
332 /* 50 */
333 "repeated subpattern is too long\0" /** DEAD **/
334 "octal value is greater than \\377 (not in UTF-8 mode)\0"
335 "internal error: overran compiling workspace\0"
336 "internal error: previously-checked referenced subpattern not found\0"
337 "DEFINE group contains more than one branch\0"
338 /* 55 */
339 "repeating a DEFINE group is not allowed\0"
340 "inconsistent NEWLINE options\0"
341 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
342 "a numbered reference must not be zero\0"
343 "(*VERB) with an argument is not supported\0"
344 /* 60 */
345 "(*VERB) not recognized\0"
346 "number is too big\0"
347 "subpattern name expected\0"
348 "digit expected after (?+\0"
349 "] is an invalid data character in JavaScript compatibility mode\0"
350 /* 65 */
351 "different names for subpatterns of the same number are not allowed\0";
352
353 /* Table to identify digits and hex digits. This is used when compiling
354 patterns. Note that the tables in chartables are dependent on the locale, and
355 may mark arbitrary characters as digits - but the PCRE compiling code expects
356 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
357 a private table here. It costs 256 bytes, but it is a lot faster than doing
358 character value tests (at least in some simple cases I timed), and in some
359 applications one wants PCRE to compile efficiently as well as match
360 efficiently.
361
362 For convenience, we use the same bit definitions as in chartables:
363
364 0x04 decimal digit
365 0x08 hexadecimal digit
366
367 Then we can use ctype_digit and ctype_xdigit in the code. */
368
369 #ifndef EBCDIC
370
371 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
372 UTF-8 mode. */
373
374 static const unsigned char digitab[] =
375 {
376 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
377 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
378 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
379 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
382 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
383 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
384 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
387 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
388 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
394 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
395 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
396 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
402 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
403 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
404 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
405 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
406 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
407 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
408
409 #else
410
411 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
412
413 static const unsigned char digitab[] =
414 {
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
420 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
421 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
422 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
423 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
424 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
425 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
426 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
431 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
433 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
439 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
444 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
445 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
446 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
447
448 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
449 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
450 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
451 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
453 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
457 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
458 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
460 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
462 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
465 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
466 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
467 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
468 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
469 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
470 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
471 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
472 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
473 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
474 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
475 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
476 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
477 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
478 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
479 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
480 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
481 #endif
482
483
484 /* Definition to allow mutual recursion */
485
486 static BOOL
487 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
488 int *, int *, branch_chain *, compile_data *, int *);
489
490
491
492 /*************************************************
493 * Find an error text *
494 *************************************************/
495
496 /* The error texts are now all in one long string, to save on relocations. As
497 some of the text is of unknown length, we can't use a table of offsets.
498 Instead, just count through the strings. This is not a performance issue
499 because it happens only when there has been a compilation error.
500
501 Argument: the error number
502 Returns: pointer to the error string
503 */
504
505 static const char *
506 find_error_text(int n)
507 {
508 const char *s = error_texts;
509 for (; n > 0; n--)
510 {
511 while (*s++ != 0) {};
512 if (*s == 0) return "Error text not found (please report)";
513 }
514 return s;
515 }
516
517
518 /*************************************************
519 * Handle escapes *
520 *************************************************/
521
522 /* This function is called when a \ has been encountered. It either returns a
523 positive value for a simple escape such as \n, or a negative value which
524 encodes one of the more complicated things such as \d. A backreference to group
525 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
526 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
527 ptr is pointing at the \. On exit, it is on the final character of the escape
528 sequence.
529
530 Arguments:
531 ptrptr points to the pattern position pointer
532 errorcodeptr points to the errorcode variable
533 bracount number of previous extracting brackets
534 options the options bits
535 isclass TRUE if inside a character class
536
537 Returns: zero or positive => a data character
538 negative => a special escape sequence
539 on error, errorcodeptr is set
540 */
541
542 static int
543 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
544 int options, BOOL isclass)
545 {
546 BOOL utf8 = (options & PCRE_UTF8) != 0;
547 const uschar *ptr = *ptrptr + 1;
548 int c, i;
549
550 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
551 ptr--; /* Set pointer back to the last byte */
552
553 /* If backslash is at the end of the pattern, it's an error. */
554
555 if (c == 0) *errorcodeptr = ERR1;
556
557 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
558 in a table. A non-zero result is something that can be returned immediately.
559 Otherwise further processing may be required. */
560
561 #ifndef EBCDIC /* ASCII/UTF-8 coding */
562 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
563 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
564
565 #else /* EBCDIC coding */
566 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
567 else if ((i = escapes[c - 0x48]) != 0) c = i;
568 #endif
569
570 /* Escapes that need further processing, or are illegal. */
571
572 else
573 {
574 const uschar *oldptr;
575 BOOL braced, negated;
576
577 switch (c)
578 {
579 /* A number of Perl escapes are not handled by PCRE. We give an explicit
580 error. */
581
582 case CHAR_l:
583 case CHAR_L:
584 case CHAR_N:
585 case CHAR_u:
586 case CHAR_U:
587 *errorcodeptr = ERR37;
588 break;
589
590 /* \g must be followed by one of a number of specific things:
591
592 (1) A number, either plain or braced. If positive, it is an absolute
593 backreference. If negative, it is a relative backreference. This is a Perl
594 5.10 feature.
595
596 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
597 is part of Perl's movement towards a unified syntax for back references. As
598 this is synonymous with \k{name}, we fudge it up by pretending it really
599 was \k.
600
601 (3) For Oniguruma compatibility we also support \g followed by a name or a
602 number either in angle brackets or in single quotes. However, these are
603 (possibly recursive) subroutine calls, _not_ backreferences. Just return
604 the -ESC_g code (cf \k). */
605
606 case CHAR_g:
607 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
608 {
609 c = -ESC_g;
610 break;
611 }
612
613 /* Handle the Perl-compatible cases */
614
615 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
616 {
617 const uschar *p;
618 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
619 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
620 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
621 {
622 c = -ESC_k;
623 break;
624 }
625 braced = TRUE;
626 ptr++;
627 }
628 else braced = FALSE;
629
630 if (ptr[1] == CHAR_MINUS)
631 {
632 negated = TRUE;
633 ptr++;
634 }
635 else negated = FALSE;
636
637 c = 0;
638 while ((digitab[ptr[1]] & ctype_digit) != 0)
639 c = c * 10 + *(++ptr) - CHAR_0;
640
641 if (c < 0) /* Integer overflow */
642 {
643 *errorcodeptr = ERR61;
644 break;
645 }
646
647 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
648 {
649 *errorcodeptr = ERR57;
650 break;
651 }
652
653 if (c == 0)
654 {
655 *errorcodeptr = ERR58;
656 break;
657 }
658
659 if (negated)
660 {
661 if (c > bracount)
662 {
663 *errorcodeptr = ERR15;
664 break;
665 }
666 c = bracount - (c - 1);
667 }
668
669 c = -(ESC_REF + c);
670 break;
671
672 /* The handling of escape sequences consisting of a string of digits
673 starting with one that is not zero is not straightforward. By experiment,
674 the way Perl works seems to be as follows:
675
676 Outside a character class, the digits are read as a decimal number. If the
677 number is less than 10, or if there are that many previous extracting
678 left brackets, then it is a back reference. Otherwise, up to three octal
679 digits are read to form an escaped byte. Thus \123 is likely to be octal
680 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
681 value is greater than 377, the least significant 8 bits are taken. Inside a
682 character class, \ followed by a digit is always an octal number. */
683
684 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
685 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
686
687 if (!isclass)
688 {
689 oldptr = ptr;
690 c -= CHAR_0;
691 while ((digitab[ptr[1]] & ctype_digit) != 0)
692 c = c * 10 + *(++ptr) - CHAR_0;
693 if (c < 0) /* Integer overflow */
694 {
695 *errorcodeptr = ERR61;
696 break;
697 }
698 if (c < 10 || c <= bracount)
699 {
700 c = -(ESC_REF + c);
701 break;
702 }
703 ptr = oldptr; /* Put the pointer back and fall through */
704 }
705
706 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
707 generates a binary zero byte and treats the digit as a following literal.
708 Thus we have to pull back the pointer by one. */
709
710 if ((c = *ptr) >= CHAR_8)
711 {
712 ptr--;
713 c = 0;
714 break;
715 }
716
717 /* \0 always starts an octal number, but we may drop through to here with a
718 larger first octal digit. The original code used just to take the least
719 significant 8 bits of octal numbers (I think this is what early Perls used
720 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
721 than 3 octal digits. */
722
723 case CHAR_0:
724 c -= CHAR_0;
725 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
726 c = c * 8 + *(++ptr) - CHAR_0;
727 if (!utf8 && c > 255) *errorcodeptr = ERR51;
728 break;
729
730 /* \x is complicated. \x{ddd} is a character number which can be greater
731 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
732 treated as a data character. */
733
734 case CHAR_x:
735 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
736 {
737 const uschar *pt = ptr + 2;
738 int count = 0;
739
740 c = 0;
741 while ((digitab[*pt] & ctype_xdigit) != 0)
742 {
743 register int cc = *pt++;
744 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
745 count++;
746
747 #ifndef EBCDIC /* ASCII/UTF-8 coding */
748 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
749 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
750 #else /* EBCDIC coding */
751 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
752 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
753 #endif
754 }
755
756 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
757 {
758 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
759 ptr = pt;
760 break;
761 }
762
763 /* If the sequence of hex digits does not end with '}', then we don't
764 recognize this construct; fall through to the normal \x handling. */
765 }
766
767 /* Read just a single-byte hex-defined char */
768
769 c = 0;
770 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
771 {
772 int cc; /* Some compilers don't like */
773 cc = *(++ptr); /* ++ in initializers */
774 #ifndef EBCDIC /* ASCII/UTF-8 coding */
775 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
776 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
777 #else /* EBCDIC coding */
778 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
779 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
780 #endif
781 }
782 break;
783
784 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
785 This coding is ASCII-specific, but then the whole concept of \cx is
786 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
787
788 case CHAR_c:
789 c = *(++ptr);
790 if (c == 0)
791 {
792 *errorcodeptr = ERR2;
793 break;
794 }
795
796 #ifndef EBCDIC /* ASCII/UTF-8 coding */
797 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
798 c ^= 0x40;
799 #else /* EBCDIC coding */
800 if (c >= CHAR_a && c <= CHAR_z) c += 64;
801 c ^= 0xC0;
802 #endif
803 break;
804
805 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
806 other alphanumeric following \ is an error if PCRE_EXTRA was set;
807 otherwise, for Perl compatibility, it is a literal. This code looks a bit
808 odd, but there used to be some cases other than the default, and there may
809 be again in future, so I haven't "optimized" it. */
810
811 default:
812 if ((options & PCRE_EXTRA) != 0) switch(c)
813 {
814 default:
815 *errorcodeptr = ERR3;
816 break;
817 }
818 break;
819 }
820 }
821
822 *ptrptr = ptr;
823 return c;
824 }
825
826
827
828 #ifdef SUPPORT_UCP
829 /*************************************************
830 * Handle \P and \p *
831 *************************************************/
832
833 /* This function is called after \P or \p has been encountered, provided that
834 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
835 pointing at the P or p. On exit, it is pointing at the final character of the
836 escape sequence.
837
838 Argument:
839 ptrptr points to the pattern position pointer
840 negptr points to a boolean that is set TRUE for negation else FALSE
841 dptr points to an int that is set to the detailed property value
842 errorcodeptr points to the error code variable
843
844 Returns: type value from ucp_type_table, or -1 for an invalid type
845 */
846
847 static int
848 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
849 {
850 int c, i, bot, top;
851 const uschar *ptr = *ptrptr;
852 char name[32];
853
854 c = *(++ptr);
855 if (c == 0) goto ERROR_RETURN;
856
857 *negptr = FALSE;
858
859 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
860 negation. */
861
862 if (c == CHAR_LEFT_CURLY_BRACKET)
863 {
864 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
865 {
866 *negptr = TRUE;
867 ptr++;
868 }
869 for (i = 0; i < (int)sizeof(name) - 1; i++)
870 {
871 c = *(++ptr);
872 if (c == 0) goto ERROR_RETURN;
873 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
874 name[i] = c;
875 }
876 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
877 name[i] = 0;
878 }
879
880 /* Otherwise there is just one following character */
881
882 else
883 {
884 name[0] = c;
885 name[1] = 0;
886 }
887
888 *ptrptr = ptr;
889
890 /* Search for a recognized property name using binary chop */
891
892 bot = 0;
893 top = _pcre_utt_size;
894
895 while (bot < top)
896 {
897 i = (bot + top) >> 1;
898 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
899 if (c == 0)
900 {
901 *dptr = _pcre_utt[i].value;
902 return _pcre_utt[i].type;
903 }
904 if (c > 0) bot = i + 1; else top = i;
905 }
906
907 *errorcodeptr = ERR47;
908 *ptrptr = ptr;
909 return -1;
910
911 ERROR_RETURN:
912 *errorcodeptr = ERR46;
913 *ptrptr = ptr;
914 return -1;
915 }
916 #endif
917
918
919
920
921 /*************************************************
922 * Check for counted repeat *
923 *************************************************/
924
925 /* This function is called when a '{' is encountered in a place where it might
926 start a quantifier. It looks ahead to see if it really is a quantifier or not.
927 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
928 where the ddds are digits.
929
930 Arguments:
931 p pointer to the first char after '{'
932
933 Returns: TRUE or FALSE
934 */
935
936 static BOOL
937 is_counted_repeat(const uschar *p)
938 {
939 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
940 while ((digitab[*p] & ctype_digit) != 0) p++;
941 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
942
943 if (*p++ != CHAR_COMMA) return FALSE;
944 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
945
946 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
947 while ((digitab[*p] & ctype_digit) != 0) p++;
948
949 return (*p == CHAR_RIGHT_CURLY_BRACKET);
950 }
951
952
953
954 /*************************************************
955 * Read repeat counts *
956 *************************************************/
957
958 /* Read an item of the form {n,m} and return the values. This is called only
959 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
960 so the syntax is guaranteed to be correct, but we need to check the values.
961
962 Arguments:
963 p pointer to first char after '{'
964 minp pointer to int for min
965 maxp pointer to int for max
966 returned as -1 if no max
967 errorcodeptr points to error code variable
968
969 Returns: pointer to '}' on success;
970 current ptr on error, with errorcodeptr set non-zero
971 */
972
973 static const uschar *
974 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
975 {
976 int min = 0;
977 int max = -1;
978
979 /* Read the minimum value and do a paranoid check: a negative value indicates
980 an integer overflow. */
981
982 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
983 if (min < 0 || min > 65535)
984 {
985 *errorcodeptr = ERR5;
986 return p;
987 }
988
989 /* Read the maximum value if there is one, and again do a paranoid on its size.
990 Also, max must not be less than min. */
991
992 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
993 {
994 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
995 {
996 max = 0;
997 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
998 if (max < 0 || max > 65535)
999 {
1000 *errorcodeptr = ERR5;
1001 return p;
1002 }
1003 if (max < min)
1004 {
1005 *errorcodeptr = ERR4;
1006 return p;
1007 }
1008 }
1009 }
1010
1011 /* Fill in the required variables, and pass back the pointer to the terminating
1012 '}'. */
1013
1014 *minp = min;
1015 *maxp = max;
1016 return p;
1017 }
1018
1019
1020
1021 /*************************************************
1022 * Subroutine for finding forward reference *
1023 *************************************************/
1024
1025 /* This recursive function is called only from find_parens() below. The
1026 top-level call starts at the beginning of the pattern. All other calls must
1027 start at a parenthesis. It scans along a pattern's text looking for capturing
1028 subpatterns, and counting them. If it finds a named pattern that matches the
1029 name it is given, it returns its number. Alternatively, if the name is NULL, it
1030 returns when it reaches a given numbered subpattern. We know that if (?P< is
1031 encountered, the name will be terminated by '>' because that is checked in the
1032 first pass. Recursion is used to keep track of subpatterns that reset the
1033 capturing group numbers - the (?| feature.
1034
1035 Arguments:
1036 ptrptr address of the current character pointer (updated)
1037 cd compile background data
1038 name name to seek, or NULL if seeking a numbered subpattern
1039 lorn name length, or subpattern number if name is NULL
1040 xmode TRUE if we are in /x mode
1041 count pointer to the current capturing subpattern number (updated)
1042
1043 Returns: the number of the named subpattern, or -1 if not found
1044 */
1045
1046 static int
1047 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1048 BOOL xmode, int *count)
1049 {
1050 uschar *ptr = *ptrptr;
1051 int start_count = *count;
1052 int hwm_count = start_count;
1053 BOOL dup_parens = FALSE;
1054
1055 /* If the first character is a parenthesis, check on the type of group we are
1056 dealing with. The very first call may not start with a parenthesis. */
1057
1058 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1059 {
1060 if (ptr[1] == CHAR_QUESTION_MARK &&
1061 ptr[2] == CHAR_VERTICAL_LINE)
1062 {
1063 ptr += 3;
1064 dup_parens = TRUE;
1065 }
1066
1067 /* Handle a normal, unnamed capturing parenthesis */
1068
1069 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1070 {
1071 *count += 1;
1072 if (name == NULL && *count == lorn) return *count;
1073 ptr++;
1074 }
1075
1076 /* Handle a condition. If it is an assertion, just carry on so that it
1077 is processed as normal. If not, skip to the closing parenthesis of the
1078 condition (there can't be any nested parens. */
1079
1080 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1081 {
1082 ptr += 2;
1083 if (ptr[1] != CHAR_QUESTION_MARK)
1084 {
1085 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1086 if (*ptr != 0) ptr++;
1087 }
1088 }
1089
1090 /* We have either (? or (* and not a condition */
1091
1092 else
1093 {
1094 ptr += 2;
1095 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1096
1097 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1098
1099 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1100 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1101 {
1102 int term;
1103 const uschar *thisname;
1104 *count += 1;
1105 if (name == NULL && *count == lorn) return *count;
1106 term = *ptr++;
1107 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1108 thisname = ptr;
1109 while (*ptr != term) ptr++;
1110 if (name != NULL && lorn == ptr - thisname &&
1111 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1112 return *count;
1113 term++;
1114 }
1115 }
1116 }
1117
1118 /* Past any initial parenthesis handling, scan for parentheses or vertical
1119 bars. */
1120
1121 for (; *ptr != 0; ptr++)
1122 {
1123 /* Skip over backslashed characters and also entire \Q...\E */
1124
1125 if (*ptr == CHAR_BACKSLASH)
1126 {
1127 if (*(++ptr) == 0) goto FAIL_EXIT;
1128 if (*ptr == CHAR_Q) for (;;)
1129 {
1130 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1131 if (*ptr == 0) goto FAIL_EXIT;
1132 if (*(++ptr) == CHAR_E) break;
1133 }
1134 continue;
1135 }
1136
1137 /* Skip over character classes; this logic must be similar to the way they
1138 are handled for real. If the first character is '^', skip it. Also, if the
1139 first few characters (either before or after ^) are \Q\E or \E we skip them
1140 too. This makes for compatibility with Perl. Note the use of STR macros to
1141 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1142
1143 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1144 {
1145 BOOL negate_class = FALSE;
1146 for (;;)
1147 {
1148 if (ptr[1] == CHAR_BACKSLASH)
1149 {
1150 if (ptr[2] == CHAR_E)
1151 ptr+= 2;
1152 else if (strncmp((const char *)ptr+2,
1153 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1154 ptr += 4;
1155 else
1156 break;
1157 }
1158 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1159 {
1160 negate_class = TRUE;
1161 ptr++;
1162 }
1163 else break;
1164 }
1165
1166 /* If the next character is ']', it is a data character that must be
1167 skipped, except in JavaScript compatibility mode. */
1168
1169 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1170 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1171 ptr++;
1172
1173 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1174 {
1175 if (*ptr == 0) return -1;
1176 if (*ptr == CHAR_BACKSLASH)
1177 {
1178 if (*(++ptr) == 0) goto FAIL_EXIT;
1179 if (*ptr == CHAR_Q) for (;;)
1180 {
1181 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1182 if (*ptr == 0) goto FAIL_EXIT;
1183 if (*(++ptr) == CHAR_E) break;
1184 }
1185 continue;
1186 }
1187 }
1188 continue;
1189 }
1190
1191 /* Skip comments in /x mode */
1192
1193 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1194 {
1195 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1196 if (*ptr == 0) goto FAIL_EXIT;
1197 continue;
1198 }
1199
1200 /* Check for the special metacharacters */
1201
1202 if (*ptr == CHAR_LEFT_PARENTHESIS)
1203 {
1204 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1205 if (rc > 0) return rc;
1206 if (*ptr == 0) goto FAIL_EXIT;
1207 }
1208
1209 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1210 {
1211 if (dup_parens && *count < hwm_count) *count = hwm_count;
1212 *ptrptr = ptr;
1213 return -1;
1214 }
1215
1216 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1217 {
1218 if (*count > hwm_count) hwm_count = *count;
1219 *count = start_count;
1220 }
1221 }
1222
1223 FAIL_EXIT:
1224 *ptrptr = ptr;
1225 return -1;
1226 }
1227
1228
1229
1230
1231 /*************************************************
1232 * Find forward referenced subpattern *
1233 *************************************************/
1234
1235 /* This function scans along a pattern's text looking for capturing
1236 subpatterns, and counting them. If it finds a named pattern that matches the
1237 name it is given, it returns its number. Alternatively, if the name is NULL, it
1238 returns when it reaches a given numbered subpattern. This is used for forward
1239 references to subpatterns. We used to be able to start this scan from the
1240 current compiling point, using the current count value from cd->bracount, and
1241 do it all in a single loop, but the addition of the possibility of duplicate
1242 subpattern numbers means that we have to scan from the very start, in order to
1243 take account of such duplicates, and to use a recursive function to keep track
1244 of the different types of group.
1245
1246 Arguments:
1247 cd compile background data
1248 name name to seek, or NULL if seeking a numbered subpattern
1249 lorn name length, or subpattern number if name is NULL
1250 xmode TRUE if we are in /x mode
1251
1252 Returns: the number of the found subpattern, or -1 if not found
1253 */
1254
1255 static int
1256 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1257 {
1258 uschar *ptr = (uschar *)cd->start_pattern;
1259 int count = 0;
1260 int rc;
1261
1262 /* If the pattern does not start with an opening parenthesis, the first call
1263 to find_parens_sub() will scan right to the end (if necessary). However, if it
1264 does start with a parenthesis, find_parens_sub() will return when it hits the
1265 matching closing parens. That is why we have to have a loop. */
1266
1267 for (;;)
1268 {
1269 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1270 if (rc > 0 || *ptr++ == 0) break;
1271 }
1272
1273 return rc;
1274 }
1275
1276
1277
1278
1279 /*************************************************
1280 * Find first significant op code *
1281 *************************************************/
1282
1283 /* This is called by several functions that scan a compiled expression looking
1284 for a fixed first character, or an anchoring op code etc. It skips over things
1285 that do not influence this. For some calls, a change of option is important.
1286 For some calls, it makes sense to skip negative forward and all backward
1287 assertions, and also the \b assertion; for others it does not.
1288
1289 Arguments:
1290 code pointer to the start of the group
1291 options pointer to external options
1292 optbit the option bit whose changing is significant, or
1293 zero if none are
1294 skipassert TRUE if certain assertions are to be skipped
1295
1296 Returns: pointer to the first significant opcode
1297 */
1298
1299 static const uschar*
1300 first_significant_code(const uschar *code, int *options, int optbit,
1301 BOOL skipassert)
1302 {
1303 for (;;)
1304 {
1305 switch ((int)*code)
1306 {
1307 case OP_OPT:
1308 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1309 *options = (int)code[1];
1310 code += 2;
1311 break;
1312
1313 case OP_ASSERT_NOT:
1314 case OP_ASSERTBACK:
1315 case OP_ASSERTBACK_NOT:
1316 if (!skipassert) return code;
1317 do code += GET(code, 1); while (*code == OP_ALT);
1318 code += _pcre_OP_lengths[*code];
1319 break;
1320
1321 case OP_WORD_BOUNDARY:
1322 case OP_NOT_WORD_BOUNDARY:
1323 if (!skipassert) return code;
1324 /* Fall through */
1325
1326 case OP_CALLOUT:
1327 case OP_CREF:
1328 case OP_NCREF:
1329 case OP_RREF:
1330 case OP_NRREF:
1331 case OP_DEF:
1332 code += _pcre_OP_lengths[*code];
1333 break;
1334
1335 default:
1336 return code;
1337 }
1338 }
1339 /* Control never reaches here */
1340 }
1341
1342
1343
1344
1345 /*************************************************
1346 * Find the fixed length of a branch *
1347 *************************************************/
1348
1349 /* Scan a branch and compute the fixed length of subject that will match it,
1350 if the length is fixed. This is needed for dealing with backward assertions.
1351 In UTF8 mode, the result is in characters rather than bytes. The branch is
1352 temporarily terminated with OP_END when this function is called.
1353
1354 This function is called when a backward assertion is encountered, so that if it
1355 fails, the error message can point to the correct place in the pattern.
1356 However, we cannot do this when the assertion contains subroutine calls,
1357 because they can be forward references. We solve this by remembering this case
1358 and doing the check at the end; a flag specifies which mode we are running in.
1359
1360 Arguments:
1361 code points to the start of the pattern (the bracket)
1362 options the compiling options
1363 atend TRUE if called when the pattern is complete
1364 cd the "compile data" structure
1365
1366 Returns: the fixed length,
1367 or -1 if there is no fixed length,
1368 or -2 if \C was encountered
1369 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1370 */
1371
1372 static int
1373 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1374 {
1375 int length = -1;
1376
1377 register int branchlength = 0;
1378 register uschar *cc = code + 1 + LINK_SIZE;
1379
1380 /* Scan along the opcodes for this branch. If we get to the end of the
1381 branch, check the length against that of the other branches. */
1382
1383 for (;;)
1384 {
1385 int d;
1386 uschar *ce, *cs;
1387 register int op = *cc;
1388 switch (op)
1389 {
1390 case OP_CBRA:
1391 case OP_BRA:
1392 case OP_ONCE:
1393 case OP_COND:
1394 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1395 if (d < 0) return d;
1396 branchlength += d;
1397 do cc += GET(cc, 1); while (*cc == OP_ALT);
1398 cc += 1 + LINK_SIZE;
1399 break;
1400
1401 /* Reached end of a branch; if it's a ket it is the end of a nested
1402 call. If it's ALT it is an alternation in a nested call. If it is
1403 END it's the end of the outer call. All can be handled by the same code. */
1404
1405 case OP_ALT:
1406 case OP_KET:
1407 case OP_KETRMAX:
1408 case OP_KETRMIN:
1409 case OP_END:
1410 if (length < 0) length = branchlength;
1411 else if (length != branchlength) return -1;
1412 if (*cc != OP_ALT) return length;
1413 cc += 1 + LINK_SIZE;
1414 branchlength = 0;
1415 break;
1416
1417 /* A true recursion implies not fixed length, but a subroutine call may
1418 be OK. If the subroutine is a forward reference, we can't deal with
1419 it until the end of the pattern, so return -3. */
1420
1421 case OP_RECURSE:
1422 if (!atend) return -3;
1423 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1424 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1425 if (cc > cs && cc < ce) return -1; /* Recursion */
1426 d = find_fixedlength(cs + 2, options, atend, cd);
1427 if (d < 0) return d;
1428 branchlength += d;
1429 cc += 1 + LINK_SIZE;
1430 break;
1431
1432 /* Skip over assertive subpatterns */
1433
1434 case OP_ASSERT:
1435 case OP_ASSERT_NOT:
1436 case OP_ASSERTBACK:
1437 case OP_ASSERTBACK_NOT:
1438 do cc += GET(cc, 1); while (*cc == OP_ALT);
1439 /* Fall through */
1440
1441 /* Skip over things that don't match chars */
1442
1443 case OP_REVERSE:
1444 case OP_CREF:
1445 case OP_NCREF:
1446 case OP_RREF:
1447 case OP_NRREF:
1448 case OP_DEF:
1449 case OP_OPT:
1450 case OP_CALLOUT:
1451 case OP_SOD:
1452 case OP_SOM:
1453 case OP_SET_SOM:
1454 case OP_EOD:
1455 case OP_EODN:
1456 case OP_CIRC:
1457 case OP_DOLL:
1458 case OP_NOT_WORD_BOUNDARY:
1459 case OP_WORD_BOUNDARY:
1460 cc += _pcre_OP_lengths[*cc];
1461 break;
1462
1463 /* Handle literal characters */
1464
1465 case OP_CHAR:
1466 case OP_CHARNC:
1467 case OP_NOT:
1468 branchlength++;
1469 cc += 2;
1470 #ifdef SUPPORT_UTF8
1471 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1472 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1473 #endif
1474 break;
1475
1476 /* Handle exact repetitions. The count is already in characters, but we
1477 need to skip over a multibyte character in UTF8 mode. */
1478
1479 case OP_EXACT:
1480 branchlength += GET2(cc,1);
1481 cc += 4;
1482 #ifdef SUPPORT_UTF8
1483 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1484 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1485 #endif
1486 break;
1487
1488 case OP_TYPEEXACT:
1489 branchlength += GET2(cc,1);
1490 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1491 cc += 4;
1492 break;
1493
1494 /* Handle single-char matchers */
1495
1496 case OP_PROP:
1497 case OP_NOTPROP:
1498 cc += 2;
1499 /* Fall through */
1500
1501 case OP_NOT_DIGIT:
1502 case OP_DIGIT:
1503 case OP_NOT_WHITESPACE:
1504 case OP_WHITESPACE:
1505 case OP_NOT_WORDCHAR:
1506 case OP_WORDCHAR:
1507 case OP_ANY:
1508 case OP_ALLANY:
1509 branchlength++;
1510 cc++;
1511 break;
1512
1513 /* The single-byte matcher isn't allowed */
1514
1515 case OP_ANYBYTE:
1516 return -2;
1517
1518 /* Check a class for variable quantification */
1519
1520 #ifdef SUPPORT_UTF8
1521 case OP_XCLASS:
1522 cc += GET(cc, 1) - 33;
1523 /* Fall through */
1524 #endif
1525
1526 case OP_CLASS:
1527 case OP_NCLASS:
1528 cc += 33;
1529
1530 switch (*cc)
1531 {
1532 case OP_CRSTAR:
1533 case OP_CRMINSTAR:
1534 case OP_CRQUERY:
1535 case OP_CRMINQUERY:
1536 return -1;
1537
1538 case OP_CRRANGE:
1539 case OP_CRMINRANGE:
1540 if (GET2(cc,1) != GET2(cc,3)) return -1;
1541 branchlength += GET2(cc,1);
1542 cc += 5;
1543 break;
1544
1545 default:
1546 branchlength++;
1547 }
1548 break;
1549
1550 /* Anything else is variable length */
1551
1552 default:
1553 return -1;
1554 }
1555 }
1556 /* Control never gets here */
1557 }
1558
1559
1560
1561
1562 /*************************************************
1563 * Scan compiled regex for specific bracket *
1564 *************************************************/
1565
1566 /* This little function scans through a compiled pattern until it finds a
1567 capturing bracket with the given number, or, if the number is negative, an
1568 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1569 so that it can be called from pcre_study() when finding the minimum matching
1570 length.
1571
1572 Arguments:
1573 code points to start of expression
1574 utf8 TRUE in UTF-8 mode
1575 number the required bracket number or negative to find a lookbehind
1576
1577 Returns: pointer to the opcode for the bracket, or NULL if not found
1578 */
1579
1580 const uschar *
1581 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1582 {
1583 for (;;)
1584 {
1585 register int c = *code;
1586 if (c == OP_END) return NULL;
1587
1588 /* XCLASS is used for classes that cannot be represented just by a bit
1589 map. This includes negated single high-valued characters. The length in
1590 the table is zero; the actual length is stored in the compiled code. */
1591
1592 if (c == OP_XCLASS) code += GET(code, 1);
1593
1594 /* Handle recursion */
1595
1596 else if (c == OP_REVERSE)
1597 {
1598 if (number < 0) return (uschar *)code;
1599 code += _pcre_OP_lengths[c];
1600 }
1601
1602 /* Handle capturing bracket */
1603
1604 else if (c == OP_CBRA)
1605 {
1606 int n = GET2(code, 1+LINK_SIZE);
1607 if (n == number) return (uschar *)code;
1608 code += _pcre_OP_lengths[c];
1609 }
1610
1611 /* Otherwise, we can get the item's length from the table, except that for
1612 repeated character types, we have to test for \p and \P, which have an extra
1613 two bytes of parameters. */
1614
1615 else
1616 {
1617 switch(c)
1618 {
1619 case OP_TYPESTAR:
1620 case OP_TYPEMINSTAR:
1621 case OP_TYPEPLUS:
1622 case OP_TYPEMINPLUS:
1623 case OP_TYPEQUERY:
1624 case OP_TYPEMINQUERY:
1625 case OP_TYPEPOSSTAR:
1626 case OP_TYPEPOSPLUS:
1627 case OP_TYPEPOSQUERY:
1628 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1629 break;
1630
1631 case OP_TYPEUPTO:
1632 case OP_TYPEMINUPTO:
1633 case OP_TYPEEXACT:
1634 case OP_TYPEPOSUPTO:
1635 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1636 break;
1637 }
1638
1639 /* Add in the fixed length from the table */
1640
1641 code += _pcre_OP_lengths[c];
1642
1643 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1644 a multi-byte character. The length in the table is a minimum, so we have to
1645 arrange to skip the extra bytes. */
1646
1647 #ifdef SUPPORT_UTF8
1648 if (utf8) switch(c)
1649 {
1650 case OP_CHAR:
1651 case OP_CHARNC:
1652 case OP_EXACT:
1653 case OP_UPTO:
1654 case OP_MINUPTO:
1655 case OP_POSUPTO:
1656 case OP_STAR:
1657 case OP_MINSTAR:
1658 case OP_POSSTAR:
1659 case OP_PLUS:
1660 case OP_MINPLUS:
1661 case OP_POSPLUS:
1662 case OP_QUERY:
1663 case OP_MINQUERY:
1664 case OP_POSQUERY:
1665 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1666 break;
1667 }
1668 #else
1669 (void)(utf8); /* Keep compiler happy by referencing function argument */
1670 #endif
1671 }
1672 }
1673 }
1674
1675
1676
1677 /*************************************************
1678 * Scan compiled regex for recursion reference *
1679 *************************************************/
1680
1681 /* This little function scans through a compiled pattern until it finds an
1682 instance of OP_RECURSE.
1683
1684 Arguments:
1685 code points to start of expression
1686 utf8 TRUE in UTF-8 mode
1687
1688 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1689 */
1690
1691 static const uschar *
1692 find_recurse(const uschar *code, BOOL utf8)
1693 {
1694 for (;;)
1695 {
1696 register int c = *code;
1697 if (c == OP_END) return NULL;
1698 if (c == OP_RECURSE) return code;
1699
1700 /* XCLASS is used for classes that cannot be represented just by a bit
1701 map. This includes negated single high-valued characters. The length in
1702 the table is zero; the actual length is stored in the compiled code. */
1703
1704 if (c == OP_XCLASS) code += GET(code, 1);
1705
1706 /* Otherwise, we can get the item's length from the table, except that for
1707 repeated character types, we have to test for \p and \P, which have an extra
1708 two bytes of parameters. */
1709
1710 else
1711 {
1712 switch(c)
1713 {
1714 case OP_TYPESTAR:
1715 case OP_TYPEMINSTAR:
1716 case OP_TYPEPLUS:
1717 case OP_TYPEMINPLUS:
1718 case OP_TYPEQUERY:
1719 case OP_TYPEMINQUERY:
1720 case OP_TYPEPOSSTAR:
1721 case OP_TYPEPOSPLUS:
1722 case OP_TYPEPOSQUERY:
1723 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1724 break;
1725
1726 case OP_TYPEPOSUPTO:
1727 case OP_TYPEUPTO:
1728 case OP_TYPEMINUPTO:
1729 case OP_TYPEEXACT:
1730 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1731 break;
1732 }
1733
1734 /* Add in the fixed length from the table */
1735
1736 code += _pcre_OP_lengths[c];
1737
1738 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1739 by a multi-byte character. The length in the table is a minimum, so we have
1740 to arrange to skip the extra bytes. */
1741
1742 #ifdef SUPPORT_UTF8
1743 if (utf8) switch(c)
1744 {
1745 case OP_CHAR:
1746 case OP_CHARNC:
1747 case OP_EXACT:
1748 case OP_UPTO:
1749 case OP_MINUPTO:
1750 case OP_POSUPTO:
1751 case OP_STAR:
1752 case OP_MINSTAR:
1753 case OP_POSSTAR:
1754 case OP_PLUS:
1755 case OP_MINPLUS:
1756 case OP_POSPLUS:
1757 case OP_QUERY:
1758 case OP_MINQUERY:
1759 case OP_POSQUERY:
1760 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1761 break;
1762 }
1763 #else
1764 (void)(utf8); /* Keep compiler happy by referencing function argument */
1765 #endif
1766 }
1767 }
1768 }
1769
1770
1771
1772 /*************************************************
1773 * Scan compiled branch for non-emptiness *
1774 *************************************************/
1775
1776 /* This function scans through a branch of a compiled pattern to see whether it
1777 can match the empty string or not. It is called from could_be_empty()
1778 below and from compile_branch() when checking for an unlimited repeat of a
1779 group that can match nothing. Note that first_significant_code() skips over
1780 backward and negative forward assertions when its final argument is TRUE. If we
1781 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1782 bracket whose current branch will already have been scanned.
1783
1784 Arguments:
1785 code points to start of search
1786 endcode points to where to stop
1787 utf8 TRUE if in UTF8 mode
1788 cd contains pointers to tables etc.
1789
1790 Returns: TRUE if what is matched could be empty
1791 */
1792
1793 static BOOL
1794 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1795 compile_data *cd)
1796 {
1797 register int c;
1798 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1799 code < endcode;
1800 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1801 {
1802 const uschar *ccode;
1803
1804 c = *code;
1805
1806 /* Skip over forward assertions; the other assertions are skipped by
1807 first_significant_code() with a TRUE final argument. */
1808
1809 if (c == OP_ASSERT)
1810 {
1811 do code += GET(code, 1); while (*code == OP_ALT);
1812 c = *code;
1813 continue;
1814 }
1815
1816 /* Groups with zero repeats can of course be empty; skip them. */
1817
1818 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1819 {
1820 code += _pcre_OP_lengths[c];
1821 do code += GET(code, 1); while (*code == OP_ALT);
1822 c = *code;
1823 continue;
1824 }
1825
1826 /* For a recursion/subroutine call, if its end has been reached, which
1827 implies a subroutine call, we can scan it. */
1828
1829 if (c == OP_RECURSE)
1830 {
1831 BOOL empty_branch = FALSE;
1832 const uschar *scode = cd->start_code + GET(code, 1);
1833 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1834 do
1835 {
1836 if (could_be_empty_branch(scode, endcode, utf8, cd))
1837 {
1838 empty_branch = TRUE;
1839 break;
1840 }
1841 scode += GET(scode, 1);
1842 }
1843 while (*scode == OP_ALT);
1844 if (!empty_branch) return FALSE; /* All branches are non-empty */
1845 continue;
1846 }
1847
1848 /* For other groups, scan the branches. */
1849
1850 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1851 {
1852 BOOL empty_branch;
1853 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1854
1855 /* If a conditional group has only one branch, there is a second, implied,
1856 empty branch, so just skip over the conditional, because it could be empty.
1857 Otherwise, scan the individual branches of the group. */
1858
1859 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1860 code += GET(code, 1);
1861 else
1862 {
1863 empty_branch = FALSE;
1864 do
1865 {
1866 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
1867 empty_branch = TRUE;
1868 code += GET(code, 1);
1869 }
1870 while (*code == OP_ALT);
1871 if (!empty_branch) return FALSE; /* All branches are non-empty */
1872 }
1873
1874 c = *code;
1875 continue;
1876 }
1877
1878 /* Handle the other opcodes */
1879
1880 switch (c)
1881 {
1882 /* Check for quantifiers after a class. XCLASS is used for classes that
1883 cannot be represented just by a bit map. This includes negated single
1884 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1885 actual length is stored in the compiled code, so we must update "code"
1886 here. */
1887
1888 #ifdef SUPPORT_UTF8
1889 case OP_XCLASS:
1890 ccode = code += GET(code, 1);
1891 goto CHECK_CLASS_REPEAT;
1892 #endif
1893
1894 case OP_CLASS:
1895 case OP_NCLASS:
1896 ccode = code + 33;
1897
1898 #ifdef SUPPORT_UTF8
1899 CHECK_CLASS_REPEAT:
1900 #endif
1901
1902 switch (*ccode)
1903 {
1904 case OP_CRSTAR: /* These could be empty; continue */
1905 case OP_CRMINSTAR:
1906 case OP_CRQUERY:
1907 case OP_CRMINQUERY:
1908 break;
1909
1910 default: /* Non-repeat => class must match */
1911 case OP_CRPLUS: /* These repeats aren't empty */
1912 case OP_CRMINPLUS:
1913 return FALSE;
1914
1915 case OP_CRRANGE:
1916 case OP_CRMINRANGE:
1917 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1918 break;
1919 }
1920 break;
1921
1922 /* Opcodes that must match a character */
1923
1924 case OP_PROP:
1925 case OP_NOTPROP:
1926 case OP_EXTUNI:
1927 case OP_NOT_DIGIT:
1928 case OP_DIGIT:
1929 case OP_NOT_WHITESPACE:
1930 case OP_WHITESPACE:
1931 case OP_NOT_WORDCHAR:
1932 case OP_WORDCHAR:
1933 case OP_ANY:
1934 case OP_ALLANY:
1935 case OP_ANYBYTE:
1936 case OP_CHAR:
1937 case OP_CHARNC:
1938 case OP_NOT:
1939 case OP_PLUS:
1940 case OP_MINPLUS:
1941 case OP_POSPLUS:
1942 case OP_EXACT:
1943 case OP_NOTPLUS:
1944 case OP_NOTMINPLUS:
1945 case OP_NOTPOSPLUS:
1946 case OP_NOTEXACT:
1947 case OP_TYPEPLUS:
1948 case OP_TYPEMINPLUS:
1949 case OP_TYPEPOSPLUS:
1950 case OP_TYPEEXACT:
1951 return FALSE;
1952
1953 /* These are going to continue, as they may be empty, but we have to
1954 fudge the length for the \p and \P cases. */
1955
1956 case OP_TYPESTAR:
1957 case OP_TYPEMINSTAR:
1958 case OP_TYPEPOSSTAR:
1959 case OP_TYPEQUERY:
1960 case OP_TYPEMINQUERY:
1961 case OP_TYPEPOSQUERY:
1962 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1963 break;
1964
1965 /* Same for these */
1966
1967 case OP_TYPEUPTO:
1968 case OP_TYPEMINUPTO:
1969 case OP_TYPEPOSUPTO:
1970 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1971 break;
1972
1973 /* End of branch */
1974
1975 case OP_KET:
1976 case OP_KETRMAX:
1977 case OP_KETRMIN:
1978 case OP_ALT:
1979 return TRUE;
1980
1981 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1982 MINUPTO, and POSUPTO may be followed by a multibyte character */
1983
1984 #ifdef SUPPORT_UTF8
1985 case OP_STAR:
1986 case OP_MINSTAR:
1987 case OP_POSSTAR:
1988 case OP_QUERY:
1989 case OP_MINQUERY:
1990 case OP_POSQUERY:
1991 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1992 break;
1993
1994 case OP_UPTO:
1995 case OP_MINUPTO:
1996 case OP_POSUPTO:
1997 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1998 break;
1999 #endif
2000
2001 /* None of the remaining opcodes are required to match a character. */
2002
2003 default:
2004 break;
2005 }
2006 }
2007
2008 return TRUE;
2009 }
2010
2011
2012
2013 /*************************************************
2014 * Scan compiled regex for non-emptiness *
2015 *************************************************/
2016
2017 /* This function is called to check for left recursive calls. We want to check
2018 the current branch of the current pattern to see if it could match the empty
2019 string. If it could, we must look outwards for branches at other levels,
2020 stopping when we pass beyond the bracket which is the subject of the recursion.
2021
2022 Arguments:
2023 code points to start of the recursion
2024 endcode points to where to stop (current RECURSE item)
2025 bcptr points to the chain of current (unclosed) branch starts
2026 utf8 TRUE if in UTF-8 mode
2027 cd pointers to tables etc
2028
2029 Returns: TRUE if what is matched could be empty
2030 */
2031
2032 static BOOL
2033 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2034 BOOL utf8, compile_data *cd)
2035 {
2036 while (bcptr != NULL && bcptr->current_branch >= code)
2037 {
2038 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2039 return FALSE;
2040 bcptr = bcptr->outer;
2041 }
2042 return TRUE;
2043 }
2044
2045
2046
2047 /*************************************************
2048 * Check for POSIX class syntax *
2049 *************************************************/
2050
2051 /* This function is called when the sequence "[:" or "[." or "[=" is
2052 encountered in a character class. It checks whether this is followed by a
2053 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2054 reach an unescaped ']' without the special preceding character, return FALSE.
2055
2056 Originally, this function only recognized a sequence of letters between the
2057 terminators, but it seems that Perl recognizes any sequence of characters,
2058 though of course unknown POSIX names are subsequently rejected. Perl gives an
2059 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2060 didn't consider this to be a POSIX class. Likewise for [:1234:].
2061
2062 The problem in trying to be exactly like Perl is in the handling of escapes. We
2063 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2064 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2065 below handles the special case of \], but does not try to do any other escape
2066 processing. This makes it different from Perl for cases such as [:l\ower:]
2067 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2068 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2069 I think.
2070
2071 Arguments:
2072 ptr pointer to the initial [
2073 endptr where to return the end pointer
2074
2075 Returns: TRUE or FALSE
2076 */
2077
2078 static BOOL
2079 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2080 {
2081 int terminator; /* Don't combine these lines; the Solaris cc */
2082 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2083 for (++ptr; *ptr != 0; ptr++)
2084 {
2085 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2086 {
2087 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2088 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2089 {
2090 *endptr = ptr;
2091 return TRUE;
2092 }
2093 }
2094 }
2095 return FALSE;
2096 }
2097
2098
2099
2100
2101 /*************************************************
2102 * Check POSIX class name *
2103 *************************************************/
2104
2105 /* This function is called to check the name given in a POSIX-style class entry
2106 such as [:alnum:].
2107
2108 Arguments:
2109 ptr points to the first letter
2110 len the length of the name
2111
2112 Returns: a value representing the name, or -1 if unknown
2113 */
2114
2115 static int
2116 check_posix_name(const uschar *ptr, int len)
2117 {
2118 const char *pn = posix_names;
2119 register int yield = 0;
2120 while (posix_name_lengths[yield] != 0)
2121 {
2122 if (len == posix_name_lengths[yield] &&
2123 strncmp((const char *)ptr, pn, len) == 0) return yield;
2124 pn += posix_name_lengths[yield] + 1;
2125 yield++;
2126 }
2127 return -1;
2128 }
2129
2130
2131 /*************************************************
2132 * Adjust OP_RECURSE items in repeated group *
2133 *************************************************/
2134
2135 /* OP_RECURSE items contain an offset from the start of the regex to the group
2136 that is referenced. This means that groups can be replicated for fixed
2137 repetition simply by copying (because the recursion is allowed to refer to
2138 earlier groups that are outside the current group). However, when a group is
2139 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2140 inserted before it, after it has been compiled. This means that any OP_RECURSE
2141 items within it that refer to the group itself or any contained groups have to
2142 have their offsets adjusted. That one of the jobs of this function. Before it
2143 is called, the partially compiled regex must be temporarily terminated with
2144 OP_END.
2145
2146 This function has been extended with the possibility of forward references for
2147 recursions and subroutine calls. It must also check the list of such references
2148 for the group we are dealing with. If it finds that one of the recursions in
2149 the current group is on this list, it adjusts the offset in the list, not the
2150 value in the reference (which is a group number).
2151
2152 Arguments:
2153 group points to the start of the group
2154 adjust the amount by which the group is to be moved
2155 utf8 TRUE in UTF-8 mode
2156 cd contains pointers to tables etc.
2157 save_hwm the hwm forward reference pointer at the start of the group
2158
2159 Returns: nothing
2160 */
2161
2162 static void
2163 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2164 uschar *save_hwm)
2165 {
2166 uschar *ptr = group;
2167
2168 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2169 {
2170 int offset;
2171 uschar *hc;
2172
2173 /* See if this recursion is on the forward reference list. If so, adjust the
2174 reference. */
2175
2176 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2177 {
2178 offset = GET(hc, 0);
2179 if (cd->start_code + offset == ptr + 1)
2180 {
2181 PUT(hc, 0, offset + adjust);
2182 break;
2183 }
2184 }
2185
2186 /* Otherwise, adjust the recursion offset if it's after the start of this
2187 group. */
2188
2189 if (hc >= cd->hwm)
2190 {
2191 offset = GET(ptr, 1);
2192 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2193 }
2194
2195 ptr += 1 + LINK_SIZE;
2196 }
2197 }
2198
2199
2200
2201 /*************************************************
2202 * Insert an automatic callout point *
2203 *************************************************/
2204
2205 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2206 callout points before each pattern item.
2207
2208 Arguments:
2209 code current code pointer
2210 ptr current pattern pointer
2211 cd pointers to tables etc
2212
2213 Returns: new code pointer
2214 */
2215
2216 static uschar *
2217 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2218 {
2219 *code++ = OP_CALLOUT;
2220 *code++ = 255;
2221 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2222 PUT(code, LINK_SIZE, 0); /* Default length */
2223 return code + 2*LINK_SIZE;
2224 }
2225
2226
2227
2228 /*************************************************
2229 * Complete a callout item *
2230 *************************************************/
2231
2232 /* A callout item contains the length of the next item in the pattern, which
2233 we can't fill in till after we have reached the relevant point. This is used
2234 for both automatic and manual callouts.
2235
2236 Arguments:
2237 previous_callout points to previous callout item
2238 ptr current pattern pointer
2239 cd pointers to tables etc
2240
2241 Returns: nothing
2242 */
2243
2244 static void
2245 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2246 {
2247 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2248 PUT(previous_callout, 2 + LINK_SIZE, length);
2249 }
2250
2251
2252
2253 #ifdef SUPPORT_UCP
2254 /*************************************************
2255 * Get othercase range *
2256 *************************************************/
2257
2258 /* This function is passed the start and end of a class range, in UTF-8 mode
2259 with UCP support. It searches up the characters, looking for internal ranges of
2260 characters in the "other" case. Each call returns the next one, updating the
2261 start address.
2262
2263 Arguments:
2264 cptr points to starting character value; updated
2265 d end value
2266 ocptr where to put start of othercase range
2267 odptr where to put end of othercase range
2268
2269 Yield: TRUE when range returned; FALSE when no more
2270 */
2271
2272 static BOOL
2273 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2274 unsigned int *odptr)
2275 {
2276 unsigned int c, othercase, next;
2277
2278 for (c = *cptr; c <= d; c++)
2279 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2280
2281 if (c > d) return FALSE;
2282
2283 *ocptr = othercase;
2284 next = othercase + 1;
2285
2286 for (++c; c <= d; c++)
2287 {
2288 if (UCD_OTHERCASE(c) != next) break;
2289 next++;
2290 }
2291
2292 *odptr = next - 1;
2293 *cptr = c;
2294
2295 return TRUE;
2296 }
2297 #endif /* SUPPORT_UCP */
2298
2299
2300
2301 /*************************************************
2302 * Check if auto-possessifying is possible *
2303 *************************************************/
2304
2305 /* This function is called for unlimited repeats of certain items, to see
2306 whether the next thing could possibly match the repeated item. If not, it makes
2307 sense to automatically possessify the repeated item.
2308
2309 Arguments:
2310 op_code the repeated op code
2311 this data for this item, depends on the opcode
2312 utf8 TRUE in UTF-8 mode
2313 utf8_char used for utf8 character bytes, NULL if not relevant
2314 ptr next character in pattern
2315 options options bits
2316 cd contains pointers to tables etc.
2317
2318 Returns: TRUE if possessifying is wanted
2319 */
2320
2321 static BOOL
2322 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2323 const uschar *ptr, int options, compile_data *cd)
2324 {
2325 int next;
2326
2327 /* Skip whitespace and comments in extended mode */
2328
2329 if ((options & PCRE_EXTENDED) != 0)
2330 {
2331 for (;;)
2332 {
2333 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2334 if (*ptr == CHAR_NUMBER_SIGN)
2335 {
2336 while (*(++ptr) != 0)
2337 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2338 }
2339 else break;
2340 }
2341 }
2342
2343 /* If the next item is one that we can handle, get its value. A non-negative
2344 value is a character, a negative value is an escape value. */
2345
2346 if (*ptr == CHAR_BACKSLASH)
2347 {
2348 int temperrorcode = 0;
2349 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2350 if (temperrorcode != 0) return FALSE;
2351 ptr++; /* Point after the escape sequence */
2352 }
2353
2354 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2355 {
2356 #ifdef SUPPORT_UTF8
2357 if (utf8) { GETCHARINC(next, ptr); } else
2358 #endif
2359 next = *ptr++;
2360 }
2361
2362 else return FALSE;
2363
2364 /* Skip whitespace and comments in extended mode */
2365
2366 if ((options & PCRE_EXTENDED) != 0)
2367 {
2368 for (;;)
2369 {
2370 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2371 if (*ptr == CHAR_NUMBER_SIGN)
2372 {
2373 while (*(++ptr) != 0)
2374 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2375 }
2376 else break;
2377 }
2378 }
2379
2380 /* If the next thing is itself optional, we have to give up. */
2381
2382 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2383 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2384 return FALSE;
2385
2386 /* Now compare the next item with the previous opcode. If the previous is a
2387 positive single character match, "item" either contains the character or, if
2388 "item" is greater than 127 in utf8 mode, the character's bytes are in
2389 utf8_char. */
2390
2391
2392 /* Handle cases when the next item is a character. */
2393
2394 if (next >= 0) switch(op_code)
2395 {
2396 case OP_CHAR:
2397 #ifdef SUPPORT_UTF8
2398 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2399 #else
2400 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2401 #endif
2402 return item != next;
2403
2404 /* For CHARNC (caseless character) we must check the other case. If we have
2405 Unicode property support, we can use it to test the other case of
2406 high-valued characters. */
2407
2408 case OP_CHARNC:
2409 #ifdef SUPPORT_UTF8
2410 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2411 #endif
2412 if (item == next) return FALSE;
2413 #ifdef SUPPORT_UTF8
2414 if (utf8)
2415 {
2416 unsigned int othercase;
2417 if (next < 128) othercase = cd->fcc[next]; else
2418 #ifdef SUPPORT_UCP
2419 othercase = UCD_OTHERCASE((unsigned int)next);
2420 #else
2421 othercase = NOTACHAR;
2422 #endif
2423 return (unsigned int)item != othercase;
2424 }
2425 else
2426 #endif /* SUPPORT_UTF8 */
2427 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2428
2429 /* For OP_NOT, "item" must be a single-byte character. */
2430
2431 case OP_NOT:
2432 if (item == next) return TRUE;
2433 if ((options & PCRE_CASELESS) == 0) return FALSE;
2434 #ifdef SUPPORT_UTF8
2435 if (utf8)
2436 {
2437 unsigned int othercase;
2438 if (next < 128) othercase = cd->fcc[next]; else
2439 #ifdef SUPPORT_UCP
2440 othercase = UCD_OTHERCASE(next);
2441 #else
2442 othercase = NOTACHAR;
2443 #endif
2444 return (unsigned int)item == othercase;
2445 }
2446 else
2447 #endif /* SUPPORT_UTF8 */
2448 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2449
2450 case OP_DIGIT:
2451 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2452
2453 case OP_NOT_DIGIT:
2454 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2455
2456 case OP_WHITESPACE:
2457 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2458
2459 case OP_NOT_WHITESPACE:
2460 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2461
2462 case OP_WORDCHAR:
2463 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2464
2465 case OP_NOT_WORDCHAR:
2466 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2467
2468 case OP_HSPACE:
2469 case OP_NOT_HSPACE:
2470 switch(next)
2471 {
2472 case 0x09:
2473 case 0x20:
2474 case 0xa0:
2475 case 0x1680:
2476 case 0x180e:
2477 case 0x2000:
2478 case 0x2001:
2479 case 0x2002:
2480 case 0x2003:
2481 case 0x2004:
2482 case 0x2005:
2483 case 0x2006:
2484 case 0x2007:
2485 case 0x2008:
2486 case 0x2009:
2487 case 0x200A:
2488 case 0x202f:
2489 case 0x205f:
2490 case 0x3000:
2491 return op_code != OP_HSPACE;
2492 default:
2493 return op_code == OP_HSPACE;
2494 }
2495
2496 case OP_VSPACE:
2497 case OP_NOT_VSPACE:
2498 switch(next)
2499 {
2500 case 0x0a:
2501 case 0x0b:
2502 case 0x0c:
2503 case 0x0d:
2504 case 0x85:
2505 case 0x2028:
2506 case 0x2029:
2507 return op_code != OP_VSPACE;
2508 default:
2509 return op_code == OP_VSPACE;
2510 }
2511
2512 default:
2513 return FALSE;
2514 }
2515
2516
2517 /* Handle the case when the next item is \d, \s, etc. */
2518
2519 switch(op_code)
2520 {
2521 case OP_CHAR:
2522 case OP_CHARNC:
2523 #ifdef SUPPORT_UTF8
2524 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2525 #endif
2526 switch(-next)
2527 {
2528 case ESC_d:
2529 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2530
2531 case ESC_D:
2532 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2533
2534 case ESC_s:
2535 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2536
2537 case ESC_S:
2538 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2539
2540 case ESC_w:
2541 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2542
2543 case ESC_W:
2544 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2545
2546 case ESC_h:
2547 case ESC_H:
2548 switch(item)
2549 {
2550 case 0x09:
2551 case 0x20:
2552 case 0xa0:
2553 case 0x1680:
2554 case 0x180e:
2555 case 0x2000:
2556 case 0x2001:
2557 case 0x2002:
2558 case 0x2003:
2559 case 0x2004:
2560 case 0x2005:
2561 case 0x2006:
2562 case 0x2007:
2563 case 0x2008:
2564 case 0x2009:
2565 case 0x200A:
2566 case 0x202f:
2567 case 0x205f:
2568 case 0x3000:
2569 return -next != ESC_h;
2570 default:
2571 return -next == ESC_h;
2572 }
2573
2574 case ESC_v:
2575 case ESC_V:
2576 switch(item)
2577 {
2578 case 0x0a:
2579 case 0x0b:
2580 case 0x0c:
2581 case 0x0d:
2582 case 0x85:
2583 case 0x2028:
2584 case 0x2029:
2585 return -next != ESC_v;
2586 default:
2587 return -next == ESC_v;
2588 }
2589
2590 default:
2591 return FALSE;
2592 }
2593
2594 case OP_DIGIT:
2595 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2596 next == -ESC_h || next == -ESC_v;
2597
2598 case OP_NOT_DIGIT:
2599 return next == -ESC_d;
2600
2601 case OP_WHITESPACE:
2602 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2603
2604 case OP_NOT_WHITESPACE:
2605 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2606
2607 case OP_HSPACE:
2608 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2609
2610 case OP_NOT_HSPACE:
2611 return next == -ESC_h;
2612
2613 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2614 case OP_VSPACE:
2615 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2616
2617 case OP_NOT_VSPACE:
2618 return next == -ESC_v;
2619
2620 case OP_WORDCHAR:
2621 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2622
2623 case OP_NOT_WORDCHAR:
2624 return next == -ESC_w || next == -ESC_d;
2625
2626 default:
2627 return FALSE;
2628 }
2629
2630 /* Control does not reach here */
2631 }
2632
2633
2634
2635 /*************************************************
2636 * Compile one branch *
2637 *************************************************/
2638
2639 /* Scan the pattern, compiling it into the a vector. If the options are
2640 changed during the branch, the pointer is used to change the external options
2641 bits. This function is used during the pre-compile phase when we are trying
2642 to find out the amount of memory needed, as well as during the real compile
2643 phase. The value of lengthptr distinguishes the two phases.
2644
2645 Arguments:
2646 optionsptr pointer to the option bits
2647 codeptr points to the pointer to the current code point
2648 ptrptr points to the current pattern pointer
2649 errorcodeptr points to error code variable
2650 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2651 reqbyteptr set to the last literal character required, else < 0
2652 bcptr points to current branch chain
2653 cd contains pointers to tables etc.
2654 lengthptr NULL during the real compile phase
2655 points to length accumulator during pre-compile phase
2656
2657 Returns: TRUE on success
2658 FALSE, with *errorcodeptr set non-zero on error
2659 */
2660
2661 static BOOL
2662 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2663 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2664 compile_data *cd, int *lengthptr)
2665 {
2666 int repeat_type, op_type;
2667 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2668 int bravalue = 0;
2669 int greedy_default, greedy_non_default;
2670 int firstbyte, reqbyte;
2671 int zeroreqbyte, zerofirstbyte;
2672 int req_caseopt, reqvary, tempreqvary;
2673 int options = *optionsptr;
2674 int after_manual_callout = 0;
2675 int length_prevgroup = 0;
2676 register int c;
2677 register uschar *code = *codeptr;
2678 uschar *last_code = code;
2679 uschar *orig_code = code;
2680 uschar *tempcode;
2681 BOOL inescq = FALSE;
2682 BOOL groupsetfirstbyte = FALSE;
2683 const uschar *ptr = *ptrptr;
2684 const uschar *tempptr;
2685 uschar *previous = NULL;
2686 uschar *previous_callout = NULL;
2687 uschar *save_hwm = NULL;
2688 uschar classbits[32];
2689
2690 #ifdef SUPPORT_UTF8
2691 BOOL class_utf8;
2692 BOOL utf8 = (options & PCRE_UTF8) != 0;
2693 uschar *class_utf8data;
2694 uschar *class_utf8data_base;
2695 uschar utf8_char[6];
2696 #else
2697 BOOL utf8 = FALSE;
2698 uschar *utf8_char = NULL;
2699 #endif
2700
2701 #ifdef PCRE_DEBUG
2702 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2703 #endif
2704
2705 /* Set up the default and non-default settings for greediness */
2706
2707 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2708 greedy_non_default = greedy_default ^ 1;
2709
2710 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2711 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2712 matches a non-fixed char first char; reqbyte just remains unset if we never
2713 find one.
2714
2715 When we hit a repeat whose minimum is zero, we may have to adjust these values
2716 to take the zero repeat into account. This is implemented by setting them to
2717 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2718 item types that can be repeated set these backoff variables appropriately. */
2719
2720 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2721
2722 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2723 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2724 value > 255. It is added into the firstbyte or reqbyte variables to record the
2725 case status of the value. This is used only for ASCII characters. */
2726
2727 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2728
2729 /* Switch on next character until the end of the branch */
2730
2731 for (;; ptr++)
2732 {
2733 BOOL negate_class;
2734 BOOL should_flip_negation;
2735 BOOL possessive_quantifier;
2736 BOOL is_quantifier;
2737 BOOL is_recurse;
2738 BOOL reset_bracount;
2739 int class_charcount;
2740 int class_lastchar;
2741 int newoptions;
2742 int recno;
2743 int refsign;
2744 int skipbytes;
2745 int subreqbyte;
2746 int subfirstbyte;
2747 int terminator;
2748 int mclength;
2749 uschar mcbuffer[8];
2750
2751 /* Get next byte in the pattern */
2752
2753 c = *ptr;
2754
2755 /* If we are in the pre-compile phase, accumulate the length used for the
2756 previous cycle of this loop. */
2757
2758 if (lengthptr != NULL)
2759 {
2760 #ifdef PCRE_DEBUG
2761 if (code > cd->hwm) cd->hwm = code; /* High water info */
2762 #endif
2763 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2764 {
2765 *errorcodeptr = ERR52;
2766 goto FAILED;
2767 }
2768
2769 /* There is at least one situation where code goes backwards: this is the
2770 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2771 the class is simply eliminated. However, it is created first, so we have to
2772 allow memory for it. Therefore, don't ever reduce the length at this point.
2773 */
2774
2775 if (code < last_code) code = last_code;
2776
2777 /* Paranoid check for integer overflow */
2778
2779 if (OFLOW_MAX - *lengthptr < code - last_code)
2780 {
2781 *errorcodeptr = ERR20;
2782 goto FAILED;
2783 }
2784
2785 *lengthptr += code - last_code;
2786 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2787
2788 /* If "previous" is set and it is not at the start of the work space, move
2789 it back to there, in order to avoid filling up the work space. Otherwise,
2790 if "previous" is NULL, reset the current code pointer to the start. */
2791
2792 if (previous != NULL)
2793 {
2794 if (previous > orig_code)
2795 {
2796 memmove(orig_code, previous, code - previous);
2797 code -= previous - orig_code;
2798 previous = orig_code;
2799 }
2800 }
2801 else code = orig_code;
2802
2803 /* Remember where this code item starts so we can pick up the length
2804 next time round. */
2805
2806 last_code = code;
2807 }
2808
2809 /* In the real compile phase, just check the workspace used by the forward
2810 reference list. */
2811
2812 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2813 {
2814 *errorcodeptr = ERR52;
2815 goto FAILED;
2816 }
2817
2818 /* If in \Q...\E, check for the end; if not, we have a literal */
2819
2820 if (inescq && c != 0)
2821 {
2822 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2823 {
2824 inescq = FALSE;
2825 ptr++;
2826 continue;
2827 }
2828 else
2829 {
2830 if (previous_callout != NULL)
2831 {
2832 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2833 complete_callout(previous_callout, ptr, cd);
2834 previous_callout = NULL;
2835 }
2836 if ((options & PCRE_AUTO_CALLOUT) != 0)
2837 {
2838 previous_callout = code;
2839 code = auto_callout(code, ptr, cd);
2840 }
2841 goto NORMAL_CHAR;
2842 }
2843 }
2844
2845 /* Fill in length of a previous callout, except when the next thing is
2846 a quantifier. */
2847
2848 is_quantifier =
2849 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2850 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2851
2852 if (!is_quantifier && previous_callout != NULL &&
2853 after_manual_callout-- <= 0)
2854 {
2855 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2856 complete_callout(previous_callout, ptr, cd);
2857 previous_callout = NULL;
2858 }
2859
2860 /* In extended mode, skip white space and comments */
2861
2862 if ((options & PCRE_EXTENDED) != 0)
2863 {
2864 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2865 if (c == CHAR_NUMBER_SIGN)
2866 {
2867 while (*(++ptr) != 0)
2868 {
2869 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2870 }
2871 if (*ptr != 0) continue;
2872
2873 /* Else fall through to handle end of string */
2874 c = 0;
2875 }
2876 }
2877
2878 /* No auto callout for quantifiers. */
2879
2880 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2881 {
2882 previous_callout = code;
2883 code = auto_callout(code, ptr, cd);
2884 }
2885
2886 switch(c)
2887 {
2888 /* ===================================================================*/
2889 case 0: /* The branch terminates at string end */
2890 case CHAR_VERTICAL_LINE: /* or | or ) */
2891 case CHAR_RIGHT_PARENTHESIS:
2892 *firstbyteptr = firstbyte;
2893 *reqbyteptr = reqbyte;
2894 *codeptr = code;
2895 *ptrptr = ptr;
2896 if (lengthptr != NULL)
2897 {
2898 if (OFLOW_MAX - *lengthptr < code - last_code)
2899 {
2900 *errorcodeptr = ERR20;
2901 goto FAILED;
2902 }
2903 *lengthptr += code - last_code; /* To include callout length */
2904 DPRINTF((">> end branch\n"));
2905 }
2906 return TRUE;
2907
2908
2909 /* ===================================================================*/
2910 /* Handle single-character metacharacters. In multiline mode, ^ disables
2911 the setting of any following char as a first character. */
2912
2913 case CHAR_CIRCUMFLEX_ACCENT:
2914 if ((options & PCRE_MULTILINE) != 0)
2915 {
2916 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2917 }
2918 previous = NULL;
2919 *code++ = OP_CIRC;
2920 break;
2921
2922 case CHAR_DOLLAR_SIGN:
2923 previous = NULL;
2924 *code++ = OP_DOLL;
2925 break;
2926
2927 /* There can never be a first char if '.' is first, whatever happens about
2928 repeats. The value of reqbyte doesn't change either. */
2929
2930 case CHAR_DOT:
2931 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2932 zerofirstbyte = firstbyte;
2933 zeroreqbyte = reqbyte;
2934 previous = code;
2935 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2936 break;
2937
2938
2939 /* ===================================================================*/
2940 /* Character classes. If the included characters are all < 256, we build a
2941 32-byte bitmap of the permitted characters, except in the special case
2942 where there is only one such character. For negated classes, we build the
2943 map as usual, then invert it at the end. However, we use a different opcode
2944 so that data characters > 255 can be handled correctly.
2945
2946 If the class contains characters outside the 0-255 range, a different
2947 opcode is compiled. It may optionally have a bit map for characters < 256,
2948 but those above are are explicitly listed afterwards. A flag byte tells
2949 whether the bitmap is present, and whether this is a negated class or not.
2950
2951 In JavaScript compatibility mode, an isolated ']' causes an error. In
2952 default (Perl) mode, it is treated as a data character. */
2953
2954 case CHAR_RIGHT_SQUARE_BRACKET:
2955 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2956 {
2957 *errorcodeptr = ERR64;
2958 goto FAILED;
2959 }
2960 goto NORMAL_CHAR;
2961
2962 case CHAR_LEFT_SQUARE_BRACKET:
2963 previous = code;
2964
2965 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2966 they are encountered at the top level, so we'll do that too. */
2967
2968 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2969 ptr[1] == CHAR_EQUALS_SIGN) &&
2970 check_posix_syntax(ptr, &tempptr))
2971 {
2972 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2973 goto FAILED;
2974 }
2975
2976 /* If the first character is '^', set the negation flag and skip it. Also,
2977 if the first few characters (either before or after ^) are \Q\E or \E we
2978 skip them too. This makes for compatibility with Perl. */
2979
2980 negate_class = FALSE;
2981 for (;;)
2982 {
2983 c = *(++ptr);
2984 if (c == CHAR_BACKSLASH)
2985 {
2986 if (ptr[1] == CHAR_E)
2987 ptr++;
2988 else if (strncmp((const char *)ptr+1,
2989 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2990 ptr += 3;
2991 else
2992 break;
2993 }
2994 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2995 negate_class = TRUE;
2996 else break;
2997 }
2998
2999 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3000 an initial ']' is taken as a data character -- the code below handles
3001 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3002 [^] must match any character, so generate OP_ALLANY. */
3003
3004 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3005 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3006 {
3007 *code++ = negate_class? OP_ALLANY : OP_FAIL;
3008 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3009 zerofirstbyte = firstbyte;
3010 break;
3011 }
3012
3013 /* If a class contains a negative special such as \S, we need to flip the
3014 negation flag at the end, so that support for characters > 255 works
3015 correctly (they are all included in the class). */
3016
3017 should_flip_negation = FALSE;
3018
3019 /* Keep a count of chars with values < 256 so that we can optimize the case
3020 of just a single character (as long as it's < 256). However, For higher
3021 valued UTF-8 characters, we don't yet do any optimization. */
3022
3023 class_charcount = 0;
3024 class_lastchar = -1;
3025
3026 /* Initialize the 32-char bit map to all zeros. We build the map in a
3027 temporary bit of memory, in case the class contains only 1 character (less
3028 than 256), because in that case the compiled code doesn't use the bit map.
3029 */
3030
3031 memset(classbits, 0, 32 * sizeof(uschar));
3032
3033 #ifdef SUPPORT_UTF8
3034 class_utf8 = FALSE; /* No chars >= 256 */
3035 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3036 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3037 #endif
3038
3039 /* Process characters until ] is reached. By writing this as a "do" it
3040 means that an initial ] is taken as a data character. At the start of the
3041 loop, c contains the first byte of the character. */
3042
3043 if (c != 0) do
3044 {
3045 const uschar *oldptr;
3046
3047 #ifdef SUPPORT_UTF8
3048 if (utf8 && c > 127)
3049 { /* Braces are required because the */
3050 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3051 }
3052
3053 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3054 data and reset the pointer. This is so that very large classes that
3055 contain a zillion UTF-8 characters no longer overwrite the work space
3056 (which is on the stack). */
3057
3058 if (lengthptr != NULL)
3059 {
3060 *lengthptr += class_utf8data - class_utf8data_base;
3061 class_utf8data = class_utf8data_base;
3062 }
3063
3064 #endif
3065
3066 /* Inside \Q...\E everything is literal except \E */
3067
3068 if (inescq)
3069 {
3070 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3071 {
3072 inescq = FALSE; /* Reset literal state */
3073 ptr++; /* Skip the 'E' */
3074 continue; /* Carry on with next */
3075 }
3076 goto CHECK_RANGE; /* Could be range if \E follows */
3077 }
3078
3079 /* Handle POSIX class names. Perl allows a negation extension of the
3080 form [:^name:]. A square bracket that doesn't match the syntax is
3081 treated as a literal. We also recognize the POSIX constructions
3082 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3083 5.6 and 5.8 do. */
3084
3085 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3086 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3087 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3088 {
3089 BOOL local_negate = FALSE;
3090 int posix_class, taboffset, tabopt;
3091 register const uschar *cbits = cd->cbits;
3092 uschar pbits[32];
3093
3094 if (ptr[1] != CHAR_COLON)
3095 {
3096 *errorcodeptr = ERR31;
3097 goto FAILED;
3098 }
3099
3100 ptr += 2;
3101 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3102 {
3103 local_negate = TRUE;
3104 should_flip_negation = TRUE; /* Note negative special */
3105 ptr++;
3106 }
3107
3108 posix_class = check_posix_name(ptr, tempptr - ptr);
3109 if (posix_class < 0)
3110 {
3111 *errorcodeptr = ERR30;
3112 goto FAILED;
3113 }
3114
3115 /* If matching is caseless, upper and lower are converted to
3116 alpha. This relies on the fact that the class table starts with
3117 alpha, lower, upper as the first 3 entries. */
3118
3119 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3120 posix_class = 0;
3121
3122 /* We build the bit map for the POSIX class in a chunk of local store
3123 because we may be adding and subtracting from it, and we don't want to
3124 subtract bits that may be in the main map already. At the end we or the
3125 result into the bit map that is being built. */
3126
3127 posix_class *= 3;
3128
3129 /* Copy in the first table (always present) */
3130
3131 memcpy(pbits, cbits + posix_class_maps[posix_class],
3132 32 * sizeof(uschar));
3133
3134 /* If there is a second table, add or remove it as required. */
3135
3136 taboffset = posix_class_maps[posix_class + 1];
3137 tabopt = posix_class_maps[posix_class + 2];
3138
3139 if (taboffset >= 0)
3140 {
3141 if (tabopt >= 0)
3142 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3143 else
3144 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3145 }
3146
3147 /* Not see if we need to remove any special characters. An option
3148 value of 1 removes vertical space and 2 removes underscore. */
3149
3150 if (tabopt < 0) tabopt = -tabopt;
3151 if (tabopt == 1) pbits[1] &= ~0x3c;
3152 else if (tabopt == 2) pbits[11] &= 0x7f;
3153
3154 /* Add the POSIX table or its complement into the main table that is
3155 being built and we are done. */
3156
3157 if (local_negate)
3158 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3159 else
3160 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3161
3162 ptr = tempptr + 1;
3163 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3164 continue; /* End of POSIX syntax handling */
3165 }
3166
3167 /* Backslash may introduce a single character, or it may introduce one
3168 of the specials, which just set a flag. The sequence \b is a special
3169 case. Inside a class (and only there) it is treated as backspace.
3170 Elsewhere it marks a word boundary. Other escapes have preset maps ready
3171 to 'or' into the one we are building. We assume they have more than one
3172 character in them, so set class_charcount bigger than one. */
3173
3174 if (c == CHAR_BACKSLASH)
3175 {
3176 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3177 if (*errorcodeptr != 0) goto FAILED;
3178
3179 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3180 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3181 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3182 else if (-c == ESC_Q) /* Handle start of quoted string */
3183 {
3184 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3185 {
3186 ptr += 2; /* avoid empty string */
3187 }
3188 else inescq = TRUE;
3189 continue;
3190 }
3191 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3192
3193 if (c < 0)
3194 {
3195 register const uschar *cbits = cd->cbits;
3196 class_charcount += 2; /* Greater than 1 is what matters */
3197
3198 /* Save time by not doing this in the pre-compile phase. */
3199
3200 if (lengthptr == NULL) switch (-c)
3201 {
3202 case ESC_d:
3203 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3204 continue;
3205
3206 case ESC_D:
3207 should_flip_negation = TRUE;
3208 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3209 continue;
3210
3211 case ESC_w:
3212 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3213 continue;
3214
3215 case ESC_W:
3216 should_flip_negation = TRUE;
3217 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3218 continue;
3219
3220 case ESC_s:
3221 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3222 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3223 continue;
3224
3225 case ESC_S:
3226 should_flip_negation = TRUE;
3227 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3228 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3229 continue;
3230
3231 default: /* Not recognized; fall through */
3232 break; /* Need "default" setting to stop compiler warning. */
3233 }
3234
3235 /* In the pre-compile phase, just do the recognition. */
3236
3237 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3238 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3239
3240 /* We need to deal with \H, \h, \V, and \v in both phases because
3241 they use extra memory. */
3242
3243 if (-c == ESC_h)
3244 {
3245 SETBIT(classbits, 0x09); /* VT */
3246 SETBIT(classbits, 0x20); /* SPACE */
3247 SETBIT(classbits, 0xa0); /* NSBP */
3248 #ifdef SUPPORT_UTF8
3249 if (utf8)
3250 {
3251 class_utf8 = TRUE;
3252 *class_utf8data++ = XCL_SINGLE;
3253 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3254 *class_utf8data++ = XCL_SINGLE;
3255 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3256 *class_utf8data++ = XCL_RANGE;
3257 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3258 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3259 *class_utf8data++ = XCL_SINGLE;
3260 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3261 *class_utf8data++ = XCL_SINGLE;
3262 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3263 *class_utf8data++ = XCL_SINGLE;
3264 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3265 }
3266 #endif
3267 continue;
3268 }
3269
3270 if (-c == ESC_H)
3271 {
3272 for (c = 0; c < 32; c++)
3273 {
3274 int x = 0xff;
3275 switch (c)
3276 {
3277 case 0x09/8: x ^= 1 << (0x09%8); break;
3278 case 0x20/8: x ^= 1 << (0x20%8); break;
3279 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3280 default: break;
3281 }
3282 classbits[c] |= x;
3283 }
3284
3285 #ifdef SUPPORT_UTF8
3286 if (utf8)
3287 {
3288 class_utf8 = TRUE;
3289 *class_utf8data++ = XCL_RANGE;
3290 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3291 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3292 *class_utf8data++ = XCL_RANGE;
3293 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3294 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3295 *class_utf8data++ = XCL_RANGE;
3296 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3297 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3298 *class_utf8data++ = XCL_RANGE;
3299 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3300 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3301 *class_utf8data++ = XCL_RANGE;
3302 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3303 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3304 *class_utf8data++ = XCL_RANGE;
3305 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3306 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3307 *class_utf8data++ = XCL_RANGE;
3308 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3309 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3310 }
3311 #endif
3312 continue;
3313 }
3314
3315 if (-c == ESC_v)
3316 {
3317 SETBIT(classbits, 0x0a); /* LF */
3318 SETBIT(classbits, 0x0b); /* VT */
3319 SETBIT(classbits, 0x0c); /* FF */
3320 SETBIT(classbits, 0x0d); /* CR */
3321 SETBIT(classbits, 0x85); /* NEL */
3322 #ifdef SUPPORT_UTF8
3323 if (utf8)
3324 {
3325 class_utf8 = TRUE;
3326 *class_utf8data++ = XCL_RANGE;
3327 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3328 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3329 }
3330 #endif
3331 continue;
3332 }
3333
3334 if (-c == ESC_V)
3335 {
3336 for (c = 0; c < 32; c++)
3337 {
3338 int x = 0xff;
3339 switch (c)
3340 {
3341 case 0x0a/8: x ^= 1 << (0x0a%8);
3342 x ^= 1 << (0x0b%8);
3343 x ^= 1 << (0x0c%8);
3344 x ^= 1 << (0x0d%8);
3345 break;
3346 case 0x85/8: x ^= 1 << (0x85%8); break;
3347 default: break;
3348 }
3349 classbits[c] |= x;
3350 }
3351
3352 #ifdef SUPPORT_UTF8
3353 if (utf8)
3354 {
3355 class_utf8 = TRUE;
3356 *class_utf8data++ = XCL_RANGE;
3357 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3358 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3359 *class_utf8data++ = XCL_RANGE;
3360 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3361 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3362 }
3363 #endif
3364 continue;
3365 }
3366
3367 /* We need to deal with \P and \p in both phases. */
3368
3369 #ifdef SUPPORT_UCP
3370 if (-c == ESC_p || -c == ESC_P)
3371 {
3372 BOOL negated;
3373 int pdata;
3374 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3375 if (ptype < 0) goto FAILED;
3376 class_utf8 = TRUE;
3377 *class_utf8data++ = ((-c == ESC_p) != negated)?
3378 XCL_PROP : XCL_NOTPROP;
3379 *class_utf8data++ = ptype;
3380 *class_utf8data++ = pdata;
3381 class_charcount -= 2; /* Not a < 256 character */
3382 continue;
3383 }
3384 #endif
3385 /* Unrecognized escapes are faulted if PCRE is running in its
3386 strict mode. By default, for compatibility with Perl, they are
3387 treated as literals. */
3388
3389 if ((options & PCRE_EXTRA) != 0)
3390 {
3391 *errorcodeptr = ERR7;
3392 goto FAILED;
3393 }
3394
3395 class_charcount -= 2; /* Undo the default count from above */
3396 c = *ptr; /* Get the final character and fall through */
3397 }
3398
3399 /* Fall through if we have a single character (c >= 0). This may be
3400 greater than 256 in UTF-8 mode. */
3401
3402 } /* End of backslash handling */
3403
3404 /* A single character may be followed by '-' to form a range. However,
3405 Perl does not permit ']' to be the end of the range. A '-' character
3406 at the end is treated as a literal. Perl ignores orphaned \E sequences
3407 entirely. The code for handling \Q and \E is messy. */
3408
3409 CHECK_RANGE:
3410 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3411 {
3412 inescq = FALSE;
3413 ptr += 2;
3414 }
3415
3416 oldptr = ptr;
3417
3418 /* Remember \r or \n */
3419
3420 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3421
3422 /* Check for range */
3423
3424 if (!inescq && ptr[1] == CHAR_MINUS)
3425 {
3426 int d;
3427 ptr += 2;
3428 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3429
3430 /* If we hit \Q (not followed by \E) at this point, go into escaped
3431 mode. */
3432
3433 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3434 {
3435 ptr += 2;
3436 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3437 { ptr += 2; continue; }
3438 inescq = TRUE;
3439 break;
3440 }
3441
3442 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3443 {
3444 ptr = oldptr;
3445 goto LONE_SINGLE_CHARACTER;
3446 }
3447
3448 #ifdef SUPPORT_UTF8
3449 if (utf8)
3450 { /* Braces are required because the */
3451 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3452 }
3453 else
3454 #endif
3455 d = *ptr; /* Not UTF-8 mode */
3456
3457 /* The second part of a range can be a single-character escape, but
3458 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3459 in such circumstances. */
3460
3461 if (!inescq && d == CHAR_BACKSLASH)
3462 {
3463 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3464 if (*errorcodeptr != 0) goto FAILED;
3465
3466 /* \b is backspace; \X is literal X; \R is literal R; any other
3467 special means the '-' was literal */
3468
3469 if (d < 0)
3470 {
3471 if (d == -ESC_b) d = CHAR_BS;
3472 else if (d == -ESC_X) d = CHAR_X;
3473 else if (d == -ESC_R) d = CHAR_R; else
3474 {
3475 ptr = oldptr;
3476 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3477 }
3478 }
3479 }
3480
3481 /* Check that the two values are in the correct order. Optimize
3482 one-character ranges */
3483
3484 if (d < c)
3485 {
3486 *errorcodeptr = ERR8;
3487 goto FAILED;
3488 }
3489
3490 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3491
3492 /* Remember \r or \n */
3493
3494 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3495
3496 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3497 matching, we have to use an XCLASS with extra data items. Caseless
3498 matching for characters > 127 is available only if UCP support is
3499 available. */
3500
3501 #ifdef SUPPORT_UTF8
3502 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3503 {
3504 class_utf8 = TRUE;
3505
3506 /* With UCP support, we can find the other case equivalents of
3507 the relevant characters. There may be several ranges. Optimize how
3508 they fit with the basic range. */
3509
3510 #ifdef SUPPORT_UCP
3511 if ((options & PCRE_CASELESS) != 0)
3512 {
3513 unsigned int occ, ocd;
3514 unsigned int cc = c;
3515 unsigned int origd = d;
3516 while (get_othercase_range(&cc, origd, &occ, &ocd))
3517 {
3518 if (occ >= (unsigned int)c &&
3519 ocd <= (unsigned int)d)
3520 continue; /* Skip embedded ranges */
3521
3522 if (occ < (unsigned int)c &&
3523 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3524 { /* if there is overlap, */
3525 c = occ; /* noting that if occ < c */
3526 continue; /* we can't have ocd > d */
3527 } /* because a subrange is */
3528 if (ocd > (unsigned int)d &&
3529 occ <= (unsigned int)d + 1) /* always shorter than */
3530 { /* the basic range. */
3531 d = ocd;
3532 continue;
3533 }
3534
3535 if (occ == ocd)
3536 {
3537 *class_utf8data++ = XCL_SINGLE;
3538 }
3539 else
3540 {
3541 *class_utf8data++ = XCL_RANGE;
3542 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3543 }
3544 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3545 }
3546 }
3547 #endif /* SUPPORT_UCP */
3548
3549 /* Now record the original range, possibly modified for UCP caseless
3550 overlapping ranges. */
3551
3552 *class_utf8data++ = XCL_RANGE;
3553 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3554 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3555
3556 /* With UCP support, we are done. Without UCP support, there is no
3557 caseless matching for UTF-8 characters > 127; we can use the bit map
3558 for the smaller ones. */
3559
3560 #ifdef SUPPORT_UCP
3561 continue; /* With next character in the class */
3562 #else
3563 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3564
3565 /* Adjust upper limit and fall through to set up the map */
3566
3567 d = 127;
3568
3569 #endif /* SUPPORT_UCP */
3570 }
3571 #endif /* SUPPORT_UTF8 */
3572
3573 /* We use the bit map for all cases when not in UTF-8 mode; else
3574 ranges that lie entirely within 0-127 when there is UCP support; else
3575 for partial ranges without UCP support. */
3576
3577 class_charcount += d - c + 1;
3578 class_lastchar = d;
3579
3580 /* We can save a bit of time by skipping this in the pre-compile. */
3581
3582 if (lengthptr == NULL) for (; c <= d; c++)
3583 {
3584 classbits[c/8] |= (1 << (c&7));
3585 if ((options & PCRE_CASELESS) != 0)
3586 {
3587 int uc = cd->fcc[c]; /* flip case */
3588 classbits[uc/8] |= (1 << (uc&7));
3589 }
3590 }
3591
3592 continue; /* Go get the next char in the class */
3593 }
3594
3595 /* Handle a lone single character - we can get here for a normal
3596 non-escape char, or after \ that introduces a single character or for an
3597 apparent range that isn't. */
3598
3599 LONE_SINGLE_CHARACTER:
3600
3601 /* Handle a character that cannot go in the bit map */
3602
3603 #ifdef SUPPORT_UTF8
3604 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3605 {
3606 class_utf8 = TRUE;
3607 *class_utf8data++ = XCL_SINGLE;
3608 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3609
3610 #ifdef SUPPORT_UCP
3611 if ((options & PCRE_CASELESS) != 0)
3612 {
3613 unsigned int othercase;
3614 if ((othercase = UCD_OTHERCASE(c)) != c)
3615 {
3616 *class_utf8data++ = XCL_SINGLE;
3617 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3618 }
3619 }
3620 #endif /* SUPPORT_UCP */
3621
3622 }
3623 else
3624 #endif /* SUPPORT_UTF8 */
3625
3626 /* Handle a single-byte character */
3627 {
3628 classbits[c/8] |= (1 << (c&7));
3629 if ((options & PCRE_CASELESS) != 0)
3630 {
3631 c = cd->fcc[c]; /* flip case */
3632 classbits[c/8] |= (1 << (c&7));
3633 }
3634 class_charcount++;
3635 class_lastchar = c;
3636 }
3637 }
3638
3639 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3640
3641 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3642
3643 if (c == 0) /* Missing terminating ']' */
3644 {
3645 *errorcodeptr = ERR6;
3646 goto FAILED;
3647 }
3648
3649
3650 /* This code has been disabled because it would mean that \s counts as
3651 an explicit \r or \n reference, and that's not really what is wanted. Now
3652 we set the flag only if there is a literal "\r" or "\n" in the class. */
3653
3654 #if 0
3655 /* Remember whether \r or \n are in this class */
3656
3657 if (negate_class)
3658 {
3659 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3660 }
3661 else
3662 {
3663 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3664 }
3665 #endif
3666
3667
3668 /* If class_charcount is 1, we saw precisely one character whose value is
3669 less than 256. As long as there were no characters >= 128 and there was no
3670 use of \p or \P, in other words, no use of any XCLASS features, we can
3671 optimize.
3672
3673 In UTF-8 mode, we can optimize the negative case only if there were no
3674 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3675 operate on single-bytes only. This is an historical hangover. Maybe one day
3676 we can tidy these opcodes to handle multi-byte characters.
3677
3678 The optimization throws away the bit map. We turn the item into a
3679 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3680 that OP_NOT does not support multibyte characters. In the positive case, it
3681 can cause firstbyte to be set. Otherwise, there can be no first char if
3682 this item is first, whatever repeat count may follow. In the case of
3683 reqbyte, save the previous value for reinstating. */
3684
3685 #ifdef SUPPORT_UTF8
3686 if (class_charcount == 1 && !class_utf8 &&
3687 (!utf8 || !negate_class || class_lastchar < 128))
3688 #else
3689 if (class_charcount == 1)
3690 #endif
3691 {
3692 zeroreqbyte = reqbyte;
3693
3694 /* The OP_NOT opcode works on one-byte characters only. */
3695
3696 if (negate_class)
3697 {
3698 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3699 zerofirstbyte = firstbyte;
3700 *code++ = OP_NOT;
3701 *code++ = class_lastchar;
3702 break;
3703 }
3704
3705 /* For a single, positive character, get the value into mcbuffer, and
3706 then we can handle this with the normal one-character code. */
3707
3708 #ifdef SUPPORT_UTF8
3709 if (utf8 && class_lastchar > 127)
3710 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3711 else
3712 #endif
3713 {
3714 mcbuffer[0] = class_lastchar;
3715 mclength = 1;
3716 }
3717 goto ONE_CHAR;
3718 } /* End of 1-char optimization */
3719
3720 /* The general case - not the one-char optimization. If this is the first
3721 thing in the branch, there can be no first char setting, whatever the
3722 repeat count. Any reqbyte setting must remain unchanged after any kind of
3723 repeat. */
3724
3725 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3726 zerofirstbyte = firstbyte;
3727 zeroreqbyte = reqbyte;
3728
3729 /* If there are characters with values > 255, we have to compile an
3730 extended class, with its own opcode, unless there was a negated special
3731 such as \S in the class, because in that case all characters > 255 are in
3732 the class, so any that were explicitly given as well can be ignored. If
3733 (when there are explicit characters > 255 that must be listed) there are no
3734 characters < 256, we can omit the bitmap in the actual compiled code. */
3735
3736 #ifdef SUPPORT_UTF8
3737 if (class_utf8 && !should_flip_negation)
3738 {
3739 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3740 *code++ = OP_XCLASS;
3741 code += LINK_SIZE;
3742 *code = negate_class? XCL_NOT : 0;
3743
3744 /* If the map is required, move up the extra data to make room for it;
3745 otherwise just move the code pointer to the end of the extra data. */
3746
3747 if (class_charcount > 0)
3748 {
3749 *code++ |= XCL_MAP;
3750 memmove(code + 32, code, class_utf8data - code);
3751 memcpy(code, classbits, 32);
3752 code = class_utf8data + 32;
3753 }
3754 else code = class_utf8data;
3755
3756 /* Now fill in the complete length of the item */
3757
3758 PUT(previous, 1, code - previous);
3759 break; /* End of class handling */
3760 }
3761 #endif
3762
3763 /* If there are no characters > 255, set the opcode to OP_CLASS or
3764 OP_NCLASS, depending on whether the whole class was negated and whether
3765 there were negative specials such as \S in the class. Then copy the 32-byte
3766 map into the code vector, negating it if necessary. */
3767
3768 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3769 if (negate_class)
3770 {
3771 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3772 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3773 }
3774 else
3775 {
3776 memcpy(code, classbits, 32);
3777 }
3778 code += 32;
3779 break;
3780
3781
3782 /* ===================================================================*/
3783 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3784 has been tested above. */
3785
3786 case CHAR_LEFT_CURLY_BRACKET:
3787 if (!is_quantifier) goto NORMAL_CHAR;
3788 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3789 if (*errorcodeptr != 0) goto FAILED;
3790 goto REPEAT;
3791
3792 case CHAR_ASTERISK:
3793 repeat_min = 0;
3794 repeat_max = -1;
3795 goto REPEAT;
3796
3797 case CHAR_PLUS:
3798 repeat_min = 1;
3799 repeat_max = -1;
3800 goto REPEAT;
3801
3802 case CHAR_QUESTION_MARK:
3803 repeat_min = 0;
3804 repeat_max = 1;
3805
3806 REPEAT:
3807 if (previous == NULL)
3808 {
3809 *errorcodeptr = ERR9;
3810 goto FAILED;
3811 }
3812
3813 if (repeat_min == 0)
3814 {
3815 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3816 reqbyte = zeroreqbyte; /* Ditto */
3817 }
3818
3819 /* Remember whether this is a variable length repeat */
3820
3821 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3822
3823 op_type = 0; /* Default single-char op codes */
3824 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3825
3826 /* Save start of previous item, in case we have to move it up to make space
3827 for an inserted OP_ONCE for the additional '+' extension. */
3828
3829 tempcode = previous;
3830
3831 /* If the next character is '+', we have a possessive quantifier. This
3832 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3833 If the next character is '?' this is a minimizing repeat, by default,
3834 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3835 repeat type to the non-default. */
3836
3837 if (ptr[1] == CHAR_PLUS)
3838 {
3839 repeat_type = 0; /* Force greedy */
3840 possessive_quantifier = TRUE;
3841 ptr++;
3842 }
3843 else if (ptr[1] == CHAR_QUESTION_MARK)
3844 {
3845 repeat_type = greedy_non_default;
3846 ptr++;
3847 }
3848 else repeat_type = greedy_default;
3849
3850 /* If previous was a character match, abolish the item and generate a
3851 repeat item instead. If a char item has a minumum of more than one, ensure
3852 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3853 the first thing in a branch because the x will have gone into firstbyte
3854 instead. */
3855
3856 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3857 {
3858 /* Deal with UTF-8 characters that take up more than one byte. It's
3859 easier to write this out separately than try to macrify it. Use c to
3860 hold the length of the character in bytes, plus 0x80 to flag that it's a
3861 length rather than a small character. */
3862
3863 #ifdef SUPPORT_UTF8
3864 if (utf8 && (code[-1] & 0x80) != 0)
3865 {
3866 uschar *lastchar = code - 1;
3867 while((*lastchar & 0xc0) == 0x80) lastchar--;
3868 c = code - lastchar; /* Length of UTF-8 character */
3869 memcpy(utf8_char, lastchar, c); /* Save the char */
3870 c |= 0x80; /* Flag c as a length */
3871 }
3872 else
3873 #endif
3874
3875 /* Handle the case of a single byte - either with no UTF8 support, or
3876 with UTF-8 disabled, or for a UTF-8 character < 128. */
3877
3878 {
3879 c = code[-1];
3880 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3881 }
3882
3883 /* If the repetition is unlimited, it pays to see if the next thing on
3884 the line is something that cannot possibly match this character. If so,
3885 automatically possessifying this item gains some performance in the case
3886 where the match fails. */
3887
3888 if (!possessive_quantifier &&
3889 repeat_max < 0 &&
3890 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3891 options, cd))
3892 {
3893 repeat_type = 0; /* Force greedy */
3894 possessive_quantifier = TRUE;
3895 }
3896
3897 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3898 }
3899
3900 /* If previous was a single negated character ([^a] or similar), we use
3901 one of the special opcodes, replacing it. The code is shared with single-
3902 character repeats by setting opt_type to add a suitable offset into
3903 repeat_type. We can also test for auto-possessification. OP_NOT is
3904 currently used only for single-byte chars. */
3905
3906 else if (*previous == OP_NOT)
3907 {
3908 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3909 c = previous[1];
3910 if (!possessive_quantifier &&
3911 repeat_max < 0 &&
3912 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3913 {
3914 repeat_type = 0; /* Force greedy */
3915 possessive_quantifier = TRUE;
3916 }
3917 goto OUTPUT_SINGLE_REPEAT;
3918 }
3919
3920 /* If previous was a character type match (\d or similar), abolish it and
3921 create a suitable repeat item. The code is shared with single-character
3922 repeats by setting op_type to add a suitable offset into repeat_type. Note
3923 the the Unicode property types will be present only when SUPPORT_UCP is
3924 defined, but we don't wrap the little bits of code here because it just
3925 makes it horribly messy. */
3926
3927 else if (*previous < OP_EODN)
3928 {
3929 uschar *oldcode;
3930 int prop_type, prop_value;
3931 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3932 c = *previous;
3933
3934 if (!possessive_quantifier &&
3935 repeat_max < 0 &&
3936 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3937 {
3938 repeat_type = 0; /* Force greedy */
3939 possessive_quantifier = TRUE;
3940 }
3941
3942 OUTPUT_SINGLE_REPEAT:
3943 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3944 {
3945 prop_type = previous[1];
3946 prop_value = previous[2];
3947 }
3948 else prop_type = prop_value = -1;
3949
3950 oldcode = code;
3951 code = previous; /* Usually overwrite previous item */
3952
3953 /* If the maximum is zero then the minimum must also be zero; Perl allows
3954 this case, so we do too - by simply omitting the item altogether. */
3955
3956 if (repeat_max == 0) goto END_REPEAT;
3957
3958 /*--------------------------------------------------------------------*/
3959 /* This code is obsolete from release 8.00; the restriction was finally
3960 removed: */
3961
3962 /* All real repeats make it impossible to handle partial matching (maybe
3963 one day we will be able to remove this restriction). */
3964
3965 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3966 /*--------------------------------------------------------------------*/
3967
3968 /* Combine the op_type with the repeat_type */
3969
3970 repeat_type += op_type;
3971
3972 /* A minimum of zero is handled either as the special case * or ?, or as
3973 an UPTO, with the maximum given. */
3974
3975 if (repeat_min == 0)
3976 {
3977 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3978 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3979 else
3980 {
3981 *code++ = OP_UPTO + repeat_type;
3982 PUT2INC(code, 0, repeat_max);
3983 }
3984 }
3985
3986 /* A repeat minimum of 1 is optimized into some special cases. If the
3987 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3988 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3989 one less than the maximum. */
3990
3991 else if (repeat_min == 1)
3992 {
3993 if (repeat_max == -1)
3994 *code++ = OP_PLUS + repeat_type;
3995 else
3996 {
3997 code = oldcode; /* leave previous item in place */
3998 if (repeat_max == 1) goto END_REPEAT;
3999 *code++ = OP_UPTO + repeat_type;
4000 PUT2INC(code, 0, repeat_max - 1);
4001 }
4002 }
4003
4004 /* The case {n,n} is just an EXACT, while the general case {n,m} is
4005 handled as an EXACT followed by an UPTO. */
4006
4007 else
4008 {
4009 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4010 PUT2INC(code, 0, repeat_min);
4011
4012 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4013 we have to insert the character for the previous code. For a repeated
4014 Unicode property match, there are two extra bytes that define the
4015 required property. In UTF-8 mode, long characters have their length in
4016 c, with the 0x80 bit as a flag. */
4017
4018 if (repeat_max < 0)
4019 {
4020 #ifdef SUPPORT_UTF8
4021 if (utf8 && c >= 128)
4022 {
4023 memcpy(code, utf8_char, c & 7);
4024 code += c & 7;
4025 }
4026 else
4027 #endif
4028 {
4029 *code++ = c;
4030 if (prop_type >= 0)
4031 {
4032 *code++ = prop_type;
4033 *code++ = prop_value;
4034 }
4035 }
4036 *code++ = OP_STAR + repeat_type;
4037 }
4038
4039 /* Else insert an UPTO if the max is greater than the min, again
4040 preceded by the character, for the previously inserted code. If the
4041 UPTO is just for 1 instance, we can use QUERY instead. */
4042
4043 else if (repeat_max != repeat_min)
4044 {
4045 #ifdef SUPPORT_UTF8
4046 if (utf8 && c >= 128)
4047 {
4048 memcpy(code, utf8_char, c & 7);
4049 code += c & 7;
4050 }
4051 else
4052 #endif
4053 *code++ = c;
4054 if (prop_type >= 0)
4055 {
4056 *code++ = prop_type;
4057 *code++ = prop_value;
4058 }
4059 repeat_max -= repeat_min;
4060
4061 if (repeat_max == 1)
4062 {
4063 *code++ = OP_QUERY + repeat_type;
4064 }
4065 else
4066 {
4067 *code++ = OP_UPTO + repeat_type;
4068 PUT2INC(code, 0, repeat_max);
4069 }
4070 }
4071 }
4072
4073 /* The character or character type itself comes last in all cases. */
4074
4075 #ifdef SUPPORT_UTF8
4076 if (utf8 && c >= 128)
4077 {
4078 memcpy(code, utf8_char, c & 7);
4079 code += c & 7;
4080 }
4081 else
4082 #endif
4083 *code++ = c;
4084
4085 /* For a repeated Unicode property match, there are two extra bytes that
4086 define the required property. */
4087
4088 #ifdef SUPPORT_UCP
4089 if (prop_type >= 0)
4090 {
4091 *code++ = prop_type;
4092 *code++ = prop_value;
4093 }
4094 #endif
4095 }
4096
4097 /* If previous was a character class or a back reference, we put the repeat
4098 stuff after it, but just skip the item if the repeat was {0,0}. */
4099
4100 else if (*previous == OP_CLASS ||
4101 *previous == OP_NCLASS ||
4102 #ifdef SUPPORT_UTF8
4103 *previous == OP_XCLASS ||
4104 #endif
4105 *previous == OP_REF)
4106 {
4107 if (repeat_max == 0)
4108 {
4109 code = previous;
4110 goto END_REPEAT;
4111 }
4112
4113 /*--------------------------------------------------------------------*/
4114 /* This code is obsolete from release 8.00; the restriction was finally
4115 removed: */
4116
4117 /* All real repeats make it impossible to handle partial matching (maybe
4118 one day we will be able to remove this restriction). */
4119
4120 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4121 /*--------------------------------------------------------------------*/
4122
4123 if (repeat_min == 0 && repeat_max == -1)
4124 *code++ = OP_CRSTAR + repeat_type;
4125 else if (repeat_min == 1 && repeat_max == -1)
4126 *code++ = OP_CRPLUS + repeat_type;
4127 else if (repeat_min == 0 && repeat_max == 1)
4128 *code++ = OP_CRQUERY + repeat_type;
4129 else
4130 {
4131 *code++ = OP_CRRANGE + repeat_type;
4132 PUT2INC(code, 0, repeat_min);
4133 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4134 PUT2INC(code, 0, repeat_max);
4135 }
4136 }
4137
4138 /* If previous was a bracket group, we may have to replicate it in certain
4139 cases. */
4140
4141 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4142 *previous == OP_ONCE || *previous == OP_COND)
4143 {
4144 register int i;
4145 int ketoffset = 0;
4146 int len = code - previous;
4147 uschar *bralink = NULL;
4148
4149 /* Repeating a DEFINE group is pointless */
4150
4151 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4152 {
4153 *errorcodeptr = ERR55;
4154 goto FAILED;
4155 }
4156
4157 /* If the maximum repeat count is unlimited, find the end of the bracket
4158 by scanning through from the start, and compute the offset back to it
4159 from the current code pointer. There may be an OP_OPT setting following
4160 the final KET, so we can't find the end just by going back from the code
4161 pointer. */
4162
4163 if (repeat_max == -1)
4164 {
4165 register uschar *ket = previous;
4166 do ket += GET(ket, 1); while (*ket != OP_KET);
4167 ketoffset = code - ket;
4168 }
4169
4170 /* The case of a zero minimum is special because of the need to stick
4171 OP_BRAZERO in front of it, and because the group appears once in the
4172 data, whereas in other cases it appears the minimum number of times. For
4173 this reason, it is simplest to treat this case separately, as otherwise
4174 the code gets far too messy. There are several special subcases when the
4175 minimum is zero. */
4176
4177 if (repeat_min == 0)
4178 {
4179 /* If the maximum is also zero, we used to just omit the group from the
4180 output altogether, like this:
4181
4182 ** if (repeat_max == 0)
4183 ** {
4184 ** code = previous;
4185 ** goto END_REPEAT;
4186 ** }
4187
4188 However, that fails when a group is referenced as a subroutine from
4189 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4190 so that it is skipped on execution. As we don't have a list of which
4191 groups are referenced, we cannot do this selectively.
4192
4193 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4194 and do no more at this point. However, we do need to adjust any
4195 OP_RECURSE calls inside the group that refer to the group itself or any
4196 internal or forward referenced group, because the offset is from the
4197 start of the whole regex. Temporarily terminate the pattern while doing
4198 this. */
4199
4200 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4201 {
4202 *code = OP_END;
4203 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4204 memmove(previous+1, previous, len);
4205 code++;
4206 if (repeat_max == 0)
4207 {
4208 *previous++ = OP_SKIPZERO;
4209 goto END_REPEAT;
4210 }
4211 *previous++ = OP_BRAZERO + repeat_type;
4212 }
4213
4214 /* If the maximum is greater than 1 and limited, we have to replicate
4215 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4216 The first one has to be handled carefully because it's the original
4217 copy, which has to be moved up. The remainder can be handled by code
4218 that is common with the non-zero minimum case below. We have to
4219 adjust the value or repeat_max, since one less copy is required. Once
4220 again, we may have to adjust any OP_RECURSE calls inside the group. */
4221
4222 else
4223 {
4224 int offset;
4225 *code = OP_END;
4226 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4227 memmove(previous + 2 + LINK_SIZE, previous, len);
4228 code += 2 + LINK_SIZE;
4229 *previous++ = OP_BRAZERO + repeat_type;
4230 *previous++ = OP_BRA;
4231
4232 /* We chain together the bracket offset fields that have to be
4233 filled in later when the ends of the brackets are reached. */
4234
4235 offset = (bralink == NULL)? 0 : previous - bralink;
4236 bralink = previous;
4237 PUTINC(previous, 0, offset);
4238 }
4239
4240 repeat_max--;
4241 }
4242
4243 /* If the minimum is greater than zero, replicate the group as many
4244 times as necessary, and adjust the maximum to the number of subsequent
4245 copies that we need. If we set a first char from the group, and didn't
4246 set a required char, copy the latter from the former. If there are any
4247 forward reference subroutine calls in the group, there will be entries on
4248 the workspace list; replicate these with an appropriate increment. */
4249
4250 else
4251 {
4252 if (repeat_min > 1)
4253 {
4254 /* In the pre-compile phase, we don't actually do the replication. We
4255 just adjust the length as if we had. Do some paranoid checks for
4256 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4257 integer type when available, otherwise double. */
4258
4259 if (lengthptr != NULL)
4260 {
4261 int delta = (repeat_min - 1)*length_prevgroup;
4262 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4263 (INT64_OR_DOUBLE)length_prevgroup >
4264 (INT64_OR_DOUBLE)INT_MAX ||
4265 OFLOW_MAX - *lengthptr < delta)
4266 {
4267 *errorcodeptr = ERR20;
4268 goto FAILED;
4269 }
4270 *lengthptr += delta;
4271 }
4272
4273 /* This is compiling for real */
4274
4275 else
4276 {
4277 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4278 for (i = 1; i < repeat_min; i++)
4279 {
4280 uschar *hc;
4281 uschar *this_hwm = cd->hwm;
4282 memcpy(code, previous, len);
4283 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4284 {
4285 PUT(cd->hwm, 0, GET(hc, 0) + len);
4286 cd->hwm += LINK_SIZE;
4287 }
4288 save_hwm = this_hwm;
4289 code += len;
4290 }
4291 }
4292 }
4293
4294 if (repeat_max > 0) repeat_max -= repeat_min;
4295 }
4296
4297 /* This code is common to both the zero and non-zero minimum cases. If
4298 the maximum is limited, it replicates the group in a nested fashion,
4299 remembering the bracket starts on a stack. In the case of a zero minimum,
4300 the first one was set up above. In all cases the repeat_max now specifies
4301 the number of additional copies needed. Again, we must remember to
4302 replicate entries on the forward reference list. */
4303
4304 if (repeat_max >= 0)
4305 {
4306 /* In the pre-compile phase, we don't actually do the replication. We
4307 just adjust the length as if we had. For each repetition we must add 1
4308 to the length for BRAZERO and for all but the last repetition we must
4309 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4310 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4311 a 64-bit integer type when available, otherwise double. */
4312
4313 if (lengthptr != NULL && repeat_max > 0)
4314 {
4315 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4316 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4317 if ((INT64_OR_DOUBLE)repeat_max *
4318 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4319 > (INT64_OR_DOUBLE)INT_MAX ||
4320 OFLOW_MAX - *lengthptr < delta)
4321 {
4322 *errorcodeptr = ERR20;
4323 goto FAILED;
4324 }
4325 *lengthptr += delta;
4326 }
4327
4328 /* This is compiling for real */
4329
4330 else for (i = repeat_max - 1; i >= 0; i--)
4331 {
4332 uschar *hc;
4333 uschar *this_hwm = cd->hwm;
4334
4335 *code++ = OP_BRAZERO + repeat_type;
4336
4337 /* All but the final copy start a new nesting, maintaining the
4338 chain of brackets outstanding. */
4339
4340 if (i != 0)
4341 {
4342 int offset;
4343 *code++ = OP_BRA;
4344 offset = (bralink == NULL)? 0 : code - bralink;
4345 bralink = code;
4346 PUTINC(code, 0, offset);
4347 }
4348
4349 memcpy(code, previous, len);
4350 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4351 {
4352 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4353 cd->hwm += LINK_SIZE;
4354 }
4355 save_hwm = this_hwm;
4356 code += len;
4357 }
4358
4359 /* Now chain through the pending brackets, and fill in their length
4360 fields (which are holding the chain links pro tem). */
4361
4362 while (bralink != NULL)
4363 {
4364 int oldlinkoffset;
4365 int offset = code - bralink + 1;
4366 uschar *bra = code - offset;
4367 oldlinkoffset = GET(bra, 1);
4368 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4369 *code++ = OP_KET;
4370 PUTINC(code, 0, offset);
4371 PUT(bra, 1, offset);
4372 }
4373 }
4374
4375 /* If the maximum is unlimited, set a repeater in the final copy. We
4376 can't just offset backwards from the current code point, because we
4377 don't know if there's been an options resetting after the ket. The
4378 correct offset was computed above.
4379
4380 Then, when we are doing the actual compile phase, check to see whether
4381 this group is a non-atomic one that could match an empty string. If so,
4382 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4383 that runtime checking can be done. [This check is also applied to
4384 atomic groups at runtime, but in a different way.] */
4385
4386 else
4387 {
4388 uschar *ketcode = code - ketoffset;
4389 uschar *bracode = ketcode - GET(ketcode, 1);
4390 *ketcode = OP_KETRMAX + repeat_type;
4391 if (lengthptr == NULL && *bracode != OP_ONCE)
4392 {
4393 uschar *scode = bracode;
4394 do
4395 {
4396 if (could_be_empty_branch(scode, ketcode, utf8, cd))
4397 {
4398 *bracode += OP_SBRA - OP_BRA;
4399 break;
4400 }
4401 scode += GET(scode, 1);
4402 }
4403 while (*scode == OP_ALT);
4404 }
4405 }
4406 }
4407
4408 /* If previous is OP_FAIL, it was generated by an empty class [] in
4409 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4410 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4411 error above. We can just ignore the repeat in JS case. */
4412
4413 else if (*previous == OP_FAIL) goto END_REPEAT;
4414
4415 /* Else there's some kind of shambles */
4416
4417 else
4418 {
4419 *errorcodeptr = ERR11;
4420 goto FAILED;
4421 }
4422
4423 /* If the character following a repeat is '+', or if certain optimization
4424 tests above succeeded, possessive_quantifier is TRUE. For some of the
4425 simpler opcodes, there is an special alternative opcode for this. For
4426 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4427 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4428 but the special opcodes can optimize it a bit. The repeated item starts at
4429 tempcode, not at previous, which might be the first part of a string whose
4430 (former) last char we repeated.
4431
4432 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4433 an 'upto' may follow. We skip over an 'exact' item, and then test the
4434 length of what remains before proceeding. */
4435
4436 if (possessive_quantifier)
4437 {
4438 int len;
4439
4440 if (*tempcode == OP_TYPEEXACT)
4441 tempcode += _pcre_OP_lengths[*tempcode] +
4442 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4443
4444 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4445 {
4446 tempcode += _pcre_OP_lengths[*tempcode];
4447 #ifdef SUPPORT_UTF8
4448 if (utf8 && tempcode[-1] >= 0xc0)
4449 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4450 #endif
4451 }
4452
4453 len = code - tempcode;
4454 if (len > 0) switch (*tempcode)
4455 {
4456 case OP_STAR: *tempcode = OP_POSSTAR; break;
4457 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4458 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4459 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4460
4461 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4462 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4463 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4464 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4465
4466 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4467 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4468 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4469 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4470
4471 /* Because we are moving code along, we must ensure that any
4472 pending recursive references are updated. */
4473
4474 default:
4475 *code = OP_END;
4476 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4477 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4478 code += 1 + LINK_SIZE;
4479 len += 1 + LINK_SIZE;
4480 tempcode[0] = OP_ONCE;
4481 *code++ = OP_KET;
4482 PUTINC(code, 0, len);
4483 PUT(tempcode, 1, len);
4484 break;
4485 }
4486 }
4487
4488 /* In all case we no longer have a previous item. We also set the
4489 "follows varying string" flag for subsequently encountered reqbytes if
4490 it isn't already set and we have just passed a varying length item. */
4491
4492 END_REPEAT:
4493 previous = NULL;
4494 cd->req_varyopt |= reqvary;
4495 break;
4496
4497
4498 /* ===================================================================*/
4499 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4500 lookbehind or option setting or condition or all the other extended
4501 parenthesis forms. */
4502
4503 case CHAR_LEFT_PARENTHESIS:
4504 newoptions = options;
4505 skipbytes = 0;
4506 bravalue = OP_CBRA;
4507 save_hwm = cd->hwm;
4508 reset_bracount = FALSE;
4509
4510 /* First deal with various "verbs" that can be introduced by '*'. */
4511
4512 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4513 {
4514 int i, namelen;
4515 const char *vn = verbnames;
4516 const uschar *name = ++ptr;
4517 previous = NULL;
4518 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4519 if (*ptr == CHAR_COLON)
4520 {
4521 *errorcodeptr = ERR59; /* Not supported */
4522 goto FAILED;
4523 }
4524 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4525 {
4526 *errorcodeptr = ERR60;
4527 goto FAILED;
4528 }
4529 namelen = ptr - name;
4530 for (i = 0; i < verbcount; i++)
4531 {
4532 if (namelen == verbs[i].len &&
4533 strncmp((char *)name, vn, namelen) == 0)
4534 {
4535 /* Check for open captures before ACCEPT */
4536
4537 if (verbs[i].op == OP_ACCEPT)
4538 {
4539 open_capitem *oc;
4540 cd->had_accept = TRUE;
4541 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4542 {
4543 *code++ = OP_CLOSE;
4544 PUT2INC(code, 0, oc->number);
4545 }
4546 }
4547 *code++ = verbs[i].op;
4548 break;
4549 }
4550 vn += verbs[i].len + 1;
4551 }
4552 if (i < verbcount) continue;
4553 *errorcodeptr = ERR60;
4554 goto FAILED;
4555 }
4556
4557 /* Deal with the extended parentheses; all are introduced by '?', and the
4558 appearance of any of them means that this is not a capturing group. */
4559
4560 else if (*ptr == CHAR_QUESTION_MARK)
4561 {
4562 int i, set, unset, namelen;
4563 int *optset;
4564 const uschar *name;
4565 uschar *slot;
4566
4567 switch (*(++ptr))
4568 {
4569 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4570 ptr++;
4571 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4572 if (*ptr == 0)
4573 {
4574 *errorcodeptr = ERR18;
4575 goto FAILED;
4576 }
4577 continue;
4578
4579
4580 /* ------------------------------------------------------------ */
4581 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4582 reset_bracount = TRUE;
4583 /* Fall through */
4584
4585 /* ------------------------------------------------------------ */
4586 case CHAR_COLON: /* Non-capturing bracket */
4587 bravalue = OP_BRA;
4588 ptr++;
4589 break;
4590
4591
4592 /* ------------------------------------------------------------ */
4593 case CHAR_LEFT_PARENTHESIS:
4594 bravalue = OP_COND; /* Conditional group */
4595
4596 /* A condition can be an assertion, a number (referring to a numbered
4597 group), a name (referring to a named group), or 'R', referring to
4598 recursion. R<digits> and R&name are also permitted for recursion tests.
4599
4600 There are several syntaxes for testing a named group: (?(name)) is used
4601 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4602
4603 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4604 be the recursive thing or the name 'R' (and similarly for 'R' followed
4605 by digits), and (b) a number could be a name that consists of digits.
4606 In both cases, we look for a name first; if not found, we try the other
4607 cases. */
4608
4609 /* For conditions that are assertions, check the syntax, and then exit
4610 the switch. This will take control down to where bracketed groups,
4611 including assertions, are processed. */
4612
4613 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4614 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4615 break;
4616
4617 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4618 below), and all need to skip 3 bytes at the start of the group. */
4619
4620 code[1+LINK_SIZE] = OP_CREF;
4621 skipbytes = 3;
4622 refsign = -1;
4623
4624 /* Check for a test for recursion in a named group. */
4625
4626 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4627 {
4628 terminator = -1;
4629 ptr += 2;
4630 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4631 }
4632
4633 /* Check for a test for a named group's having been set, using the Perl
4634 syntax (?(<name>) or (?('name') */
4635
4636 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4637 {
4638 terminator = CHAR_GREATER_THAN_SIGN;
4639 ptr++;
4640 }
4641 else if (ptr[1] == CHAR_APOSTROPHE)
4642 {
4643 terminator = CHAR_APOSTROPHE;
4644 ptr++;
4645 }
4646 else
4647 {
4648 terminator = 0;
4649 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4650 }
4651
4652 /* We now expect to read a name; any thing else is an error */
4653
4654 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4655 {
4656 ptr += 1; /* To get the right offset */
4657 *errorcodeptr = ERR28;
4658 goto FAILED;
4659 }
4660
4661 /* Read the name, but also get it as a number if it's all digits */
4662
4663 recno = 0;
4664 name = ++ptr;
4665 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4666 {
4667 if (recno >= 0)
4668 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4669 recno * 10 + *ptr - CHAR_0 : -1;
4670 ptr++;
4671 }
4672 namelen = ptr - name;
4673
4674 if ((terminator > 0 && *ptr++ != terminator) ||
4675 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4676 {
4677 ptr--; /* Error offset */
4678 *errorcodeptr = ERR26;
4679 goto FAILED;
4680 }
4681
4682 /* Do no further checking in the pre-compile phase. */
4683
4684 if (lengthptr != NULL) break;
4685
4686 /* In the real compile we do the work of looking for the actual
4687 reference. If the string started with "+" or "-" we require the rest to
4688 be digits, in which case recno will be set. */
4689
4690 if (refsign > 0)
4691 {
4692 if (recno <= 0)
4693 {
4694 *errorcodeptr = ERR58;
4695 goto FAILED;
4696 }
4697 recno = (refsign == CHAR_MINUS)?
4698 cd->bracount - recno + 1 : recno +cd->bracount;
4699 if (recno <= 0 || recno > cd->final_bracount)
4700 {
4701 *errorcodeptr = ERR15;
4702 goto FAILED;
4703 }
4704 PUT2(code, 2+LINK_SIZE, recno);
4705 break;
4706 }
4707
4708 /* Otherwise (did not start with "+" or "-"), start by looking for the
4709 name. If we find a name, add one to the opcode to change OP_CREF or
4710 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4711 except they record that the reference was originally to a name. The
4712 information is used to check duplicate names. */
4713
4714 slot = cd->name_table;
4715 for (i = 0; i < cd->names_found; i++)
4716 {
4717 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4718 slot += cd->name_entry_size;
4719 }
4720
4721 /* Found a previous named subpattern */
4722
4723 if (i < cd->names_found)
4724 {
4725 recno = GET2(slot, 0);
4726 PUT2(code, 2+LINK_SIZE, recno);
4727 code[1+LINK_SIZE]++;
4728 }
4729
4730 /* Search the pattern for a forward reference */
4731
4732 else if ((i = find_parens(cd, name, namelen,
4733 (options & PCRE_EXTENDED) != 0)) > 0)
4734 {
4735 PUT2(code, 2+LINK_SIZE, i);
4736 code[1+LINK_SIZE]++;
4737 }
4738
4739 /* If terminator == 0 it means that the name followed directly after
4740 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4741 some further alternatives to try. For the cases where terminator != 0
4742 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4743 now checked all the possibilities, so give an error. */
4744
4745 else if (terminator != 0)
4746 {
4747 *errorcodeptr = ERR15;
4748 goto FAILED;
4749 }
4750
4751 /* Check for (?(R) for recursion. Allow digits after R to specify a
4752 specific group number. */
4753
4754 else if (*name == CHAR_R)
4755 {
4756 recno = 0;
4757 for (i = 1; i < namelen; i++)
4758 {
4759 if ((digitab[name[i]] & ctype_digit) == 0)
4760 {
4761 *errorcodeptr = ERR15;
4762 goto FAILED;
4763 }
4764 recno = recno * 10 + name[i] - CHAR_0;
4765 }
4766 if (recno == 0) recno = RREF_ANY;
4767 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4768 PUT2(code, 2+LINK_SIZE, recno);
4769 }
4770
4771 /* Similarly, check for the (?(DEFINE) "condition", which is always
4772 false. */
4773
4774 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4775 {
4776 code[1+LINK_SIZE] = OP_DEF;
4777 skipbytes = 1;
4778 }
4779
4780 /* Check for the "name" actually being a subpattern number. We are
4781 in the second pass here, so final_bracount is set. */
4782
4783 else if (recno > 0 && recno <= cd->final_bracount)
4784 {
4785 PUT2(code, 2+LINK_SIZE, recno);
4786 }
4787
4788 /* Either an unidentified subpattern, or a reference to (?(0) */
4789
4790 else
4791 {
4792 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4793 goto FAILED;
4794 }
4795 break;
4796
4797
4798 /* ------------------------------------------------------------ */
4799 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4800 bravalue = OP_ASSERT;
4801 ptr++;
4802 break;
4803
4804
4805 /* ------------------------------------------------------------ */
4806 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4807 ptr++;
4808 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4809 {
4810 *code++ = OP_FAIL;
4811 previous = NULL;
4812 continue;
4813 }
4814 bravalue = OP_ASSERT_NOT;
4815 break;
4816
4817
4818 /* ------------------------------------------------------------ */
4819 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4820 switch (ptr[1])
4821 {
4822 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4823 bravalue = OP_ASSERTBACK;
4824 ptr += 2;
4825 break;
4826
4827 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4828 bravalue = OP_ASSERTBACK_NOT;
4829 ptr += 2;
4830 break;
4831
4832 default: /* Could be name define, else bad */
4833 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4834 ptr++; /* Correct offset for error */
4835 *errorcodeptr = ERR24;
4836 goto FAILED;
4837 }
4838 break;
4839
4840
4841 /* ------------------------------------------------------------ */
4842 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4843 bravalue = OP_ONCE;
4844 ptr++;
4845 break;
4846
4847
4848 /* ------------------------------------------------------------ */
4849 case CHAR_C: /* Callout - may be followed by digits; */
4850 previous_callout = code; /* Save for later completion */
4851 after_manual_callout = 1; /* Skip one item before completing */
4852 *code++ = OP_CALLOUT;
4853 {
4854 int n = 0;
4855 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4856 n = n * 10 + *ptr - CHAR_0;
4857 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4858 {
4859 *errorcodeptr = ERR39;
4860 goto FAILED;
4861 }
4862 if (n > 255)
4863 {
4864 *errorcodeptr = ERR38;
4865 goto FAILED;
4866 }
4867 *code++ = n;
4868 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4869 PUT(code, LINK_SIZE, 0); /* Default length */
4870 code += 2 * LINK_SIZE;
4871 }
4872 previous = NULL;
4873 continue;
4874
4875
4876 /* ------------------------------------------------------------ */
4877 case CHAR_P: /* Python-style named subpattern handling */
4878 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4879 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4880 {
4881 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4882 terminator = CHAR_RIGHT_PARENTHESIS;
4883 goto NAMED_REF_OR_RECURSE;
4884 }
4885 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4886 {
4887 *errorcodeptr = ERR41;
4888 goto FAILED;
4889 }
4890 /* Fall through to handle (?P< as (?< is handled */
4891
4892
4893 /* ------------------------------------------------------------ */
4894 DEFINE_NAME: /* Come here from (?< handling */
4895 case CHAR_APOSTROPHE:
4896 {
4897 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4898 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4899 name = ++ptr;
4900
4901 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4902 namelen = ptr - name;
4903
4904 /* In the pre-compile phase, just do a syntax check. */
4905
4906 if (lengthptr != NULL)
4907 {
4908 if (*ptr != terminator)
4909 {
4910 *errorcodeptr = ERR42;
4911 goto FAILED;
4912 }
4913 if (cd->names_found >= MAX_NAME_COUNT)
4914 {
4915 *errorcodeptr = ERR49;
4916 goto FAILED;
4917 }
4918 if (namelen + 3 > cd->name_entry_size)
4919 {
4920 cd->name_entry_size = namelen + 3;
4921 if (namelen > MAX_NAME_SIZE)
4922 {
4923 *errorcodeptr = ERR48;
4924 goto FAILED;
4925 }
4926 }
4927 }
4928
4929 /* In the real compile, create the entry in the table, maintaining
4930 alphabetical order. Duplicate names for different numbers are
4931 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
4932 number are always OK. (An existing number can be re-used if (?|
4933 appears in the pattern.) In either event, a duplicate name results in
4934 a duplicate entry in the table, even if the number is the same. This
4935 is because the number of names, and hence the table size, is computed
4936 in the pre-compile, and it affects various numbers and pointers which
4937 would all have to be modified, and the compiled code moved down, if
4938 duplicates with the same number were omitted from the table. This
4939 doesn't seem worth the hassle. However, *different* names for the
4940 same number are not permitted. */
4941
4942 else
4943 {
4944 BOOL dupname = FALSE;
4945 slot = cd->name_table;
4946
4947 for (i = 0; i < cd->names_found; i++)
4948 {
4949 int crc = memcmp(name, slot+2, namelen);
4950 if (crc == 0)
4951 {
4952 if (slot[2+namelen] == 0)
4953 {
4954 if (GET2(slot, 0) != cd->bracount + 1 &&
4955 (options & PCRE_DUPNAMES) == 0)
4956 {
4957 *errorcodeptr = ERR43;
4958 goto FAILED;
4959 }
4960 else dupname = TRUE;
4961 }
4962 else crc = -1; /* Current name is a substring */
4963 }
4964
4965 /* Make space in the table and break the loop for an earlier
4966 name. For a duplicate or later name, carry on. We do this for
4967 duplicates so that in the simple case (when ?(| is not used) they
4968 are in order of their numbers. */
4969
4970 if (crc < 0)
4971 {
4972 memmove(slot + cd->name_entry_size, slot,
4973 (cd->names_found - i) * cd->name_entry_size);
4974 break;
4975 }
4976
4977 /* Continue the loop for a later or duplicate name */
4978
4979 slot += cd->name_entry_size;
4980 }
4981
4982 /* For non-duplicate names, check for a duplicate number before
4983 adding the new name. */
4984
4985 if (!dupname)
4986 {
4987 uschar *cslot = cd->name_table;
4988 for (i = 0; i < cd->names_found; i++)
4989 {
4990 if (cslot != slot)
4991 {
4992 if (GET2(cslot, 0) == cd->bracount + 1)
4993 {
4994 *errorcodeptr = ERR65;
4995 goto FAILED;
4996 }
4997 }
4998 else i--;
4999 cslot += cd->name_entry_size;
5000 }
5001 }
5002
5003 PUT2(slot, 0, cd->bracount + 1);
5004 memcpy(slot + 2, name, namelen);
5005 slot[2+namelen] = 0;
5006 }
5007 }
5008
5009 /* In both pre-compile and compile, count the number of names we've
5010 encountered. */
5011
5012 cd->names_found++;
5013 ptr++; /* Move past > or ' */
5014 goto NUMBERED_GROUP;
5015
5016
5017 /* ------------------------------------------------------------ */
5018 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5019 terminator = CHAR_RIGHT_PARENTHESIS;
5020 is_recurse = TRUE;
5021 /* Fall through */
5022
5023 /* We come here from the Python syntax above that handles both
5024 references (?P=name) and recursion (?P>name), as well as falling
5025 through from the Perl recursion syntax (?&name). We also come here from
5026 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5027 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5028
5029 NAMED_REF_OR_RECURSE:
5030 name = ++ptr;
5031 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5032 namelen = ptr - name;
5033
5034 /* In the pre-compile phase, do a syntax check and set a dummy
5035 reference number. */
5036
5037 if (lengthptr != NULL)
5038 {
5039 if (namelen == 0)
5040 {
5041 *errorcodeptr = ERR62;
5042 goto FAILED;
5043 }
5044 if (*ptr != terminator)
5045 {
5046 *errorcodeptr = ERR42;
5047 goto FAILED;
5048 }
5049 if (namelen > MAX_NAME_SIZE)
5050 {
5051 *errorcodeptr = ERR48;
5052 goto FAILED;
5053 }
5054 recno = 0;
5055 }
5056
5057 /* In the real compile, seek the name in the table. We check the name
5058 first, and then check that we have reached the end of the name in the
5059 table. That way, if the name that is longer than any in the table,
5060 the comparison will fail without reading beyond the table entry. */
5061
5062 else
5063 {
5064 slot = cd->name_table;
5065 for (i = 0; i < cd->names_found; i++)
5066 {
5067 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5068 slot[2+namelen] == 0)
5069 break;
5070 slot += cd->name_entry_size;
5071 }
5072
5073 if (i < cd->names_found) /* Back reference */
5074 {
5075 recno = GET2(slot, 0);
5076 }
5077 else if ((recno = /* Forward back reference */
5078 find_parens(cd, name, namelen,
5079 (options & PCRE_EXTENDED) != 0)) <= 0)
5080 {
5081 *errorcodeptr = ERR15;
5082 goto FAILED;
5083 }
5084 }
5085
5086 /* In both phases, we can now go to the code than handles numerical
5087 recursion or backreferences. */
5088
5089 if (is_recurse) goto HANDLE_RECURSION;
5090 else goto HANDLE_REFERENCE;
5091
5092
5093 /* ------------------------------------------------------------ */
5094 case CHAR_R: /* Recursion */
5095 ptr++; /* Same as (?0) */
5096 /* Fall through */
5097
5098
5099 /* ------------------------------------------------------------ */
5100 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5101 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5102 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5103 {
5104 const uschar *called;
5105 terminator = CHAR_RIGHT_PARENTHESIS;
5106
5107 /* Come here from the \g<...> and \g'...' code (Oniguruma
5108 compatibility). However, the syntax has been checked to ensure that
5109 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5110 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5111 ever be taken. */
5112
5113 HANDLE_NUMERICAL_RECURSION:
5114
5115 if ((refsign = *ptr) == CHAR_PLUS)
5116 {
5117 ptr++;
5118 if ((digitab[*ptr] & ctype_digit) == 0)
5119 {
5120 *errorcodeptr = ERR63;
5121 goto FAILED;
5122 }
5123 }
5124 else if (refsign == CHAR_MINUS)
5125 {
5126 if ((digitab[ptr[1]] & ctype_digit) == 0)
5127 goto OTHER_CHAR_AFTER_QUERY;
5128 ptr++;
5129 }
5130
5131 recno = 0;
5132 while((digitab[*ptr] & ctype_digit) != 0)
5133 recno = recno * 10 + *ptr++ - CHAR_0;
5134
5135 if (*ptr != terminator)
5136 {
5137 *errorcodeptr = ERR29;
5138 goto FAILED;
5139 }
5140
5141 if (refsign == CHAR_MINUS)
5142 {
5143 if (recno == 0)
5144 {
5145 *errorcodeptr = ERR58;
5146 goto FAILED;
5147 }
5148 recno = cd->bracount - recno + 1;
5149 if (recno <= 0)
5150 {
5151 *errorcodeptr = ERR15;
5152 goto FAILED;
5153 }
5154 }
5155 else if (refsign == CHAR_PLUS)
5156 {
5157 if (recno == 0)
5158 {
5159 *errorcodeptr = ERR58;
5160 goto FAILED;
5161 }
5162 recno += cd->bracount;
5163 }
5164
5165 /* Come here from code above that handles a named recursion */
5166
5167 HANDLE_RECURSION:
5168
5169 previous = code;
5170 called = cd->start_code;
5171
5172 /* When we are actually compiling, find the bracket that is being
5173 referenced. Temporarily end the regex in case it doesn't exist before
5174 this point. If we end up with a forward reference, first check that
5175 the bracket does occur later so we can give the error (and position)
5176 now. Then remember this forward reference in the workspace so it can
5177 be filled in at the end. */
5178
5179 if (lengthptr == NULL)
5180 {
5181 *code = OP_END;
5182 if (recno != 0)
5183 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5184
5185 /* Forward reference */
5186
5187 if (called == NULL)
5188 {
5189 if (find_parens(cd, NULL, recno,
5190 (options & PCRE_EXTENDED) != 0) < 0)
5191 {
5192 *errorcodeptr = ERR15;
5193 goto FAILED;
5194 }
5195
5196 /* Fudge the value of "called" so that when it is inserted as an
5197 offset below, what it actually inserted is the reference number
5198 of the group. */
5199
5200 called = cd->start_code + recno;
5201 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5202 }
5203
5204 /* If not a forward reference, and the subpattern is still open,
5205 this is a recursive call. We check to see if this is a left
5206 recursion that could loop for ever, and diagnose that case. */
5207
5208 else if (GET(called, 1) == 0 &&
5209 could_be_empty(called, code, bcptr, utf8, cd))
5210 {
5211 *errorcodeptr = ERR40;
5212 goto FAILED;
5213 }
5214 }
5215
5216 /* Insert the recursion/subroutine item, automatically wrapped inside
5217 "once" brackets. Set up a "previous group" length so that a
5218 subsequent quantifier will work. */
5219
5220 *code = OP_ONCE;
5221 PUT(code, 1, 2 + 2*LINK_SIZE);
5222 code += 1 + LINK_SIZE;
5223
5224 *code = OP_RECURSE;
5225 PUT(code, 1, called - cd->start_code);
5226 code += 1 + LINK_SIZE;
5227
5228 *code = OP_KET;
5229 PUT(code, 1, 2 + 2*LINK_SIZE);
5230 code += 1 + LINK_SIZE;
5231
5232 length_prevgroup = 3 + 3*LINK_SIZE;
5233 }
5234
5235 /* Can't determine a first byte now */
5236
5237 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5238 continue;
5239
5240
5241 /* ------------------------------------------------------------ */
5242 default: /* Other characters: check option setting */
5243 OTHER_CHAR_AFTER_QUERY:
5244 set = unset = 0;
5245 optset = &set;
5246
5247 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5248 {
5249 switch (*ptr++)
5250 {
5251 case CHAR_MINUS: optset = &unset; break;
5252
5253 case CHAR_J: /* Record that it changed in the external options */
5254 *optset |= PCRE_DUPNAMES;
5255 cd->external_flags |= PCRE_JCHANGED;
5256 break;
5257
5258 case CHAR_i: *optset |= PCRE_CASELESS; break;
5259 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5260 case CHAR_s: *optset |= PCRE_DOTALL; break;
5261 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5262 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5263 case CHAR_X: *optset |= PCRE_EXTRA; break;
5264
5265 default: *errorcodeptr = ERR12;
5266 ptr--; /* Correct the offset */
5267 goto FAILED;
5268 }
5269 }
5270
5271 /* Set up the changed option bits, but don't change anything yet. */
5272
5273 newoptions = (options | set) & (~unset);
5274
5275 /* If the options ended with ')' this is not the start of a nested
5276 group with option changes, so the options change at this level. If this
5277 item is right at the start of the pattern, the options can be
5278 abstracted and made external in the pre-compile phase, and ignored in
5279 the compile phase. This can be helpful when matching -- for instance in
5280 caseless checking of required bytes.
5281
5282 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5283 definitely *not* at the start of the pattern because something has been
5284 compiled. In the pre-compile phase, however, the code pointer can have
5285 that value after the start, because it gets reset as code is discarded
5286 during the pre-compile. However, this can happen only at top level - if
5287 we are within parentheses, the starting BRA will still be present. At
5288 any parenthesis level, the length value can be used to test if anything
5289 has been compiled at that level. Thus, a test for both these conditions
5290 is necessary to ensure we correctly detect the start of the pattern in
5291 both phases.
5292
5293 If we are not at the pattern start, compile code to change the ims
5294 options if this setting actually changes any of them, and reset the
5295 greedy defaults and the case value for firstbyte and reqbyte. */
5296
5297 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5298 {
5299 if (code == cd->start_code + 1 + LINK_SIZE &&
5300 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5301 {
5302 cd->external_options = newoptions;
5303 }
5304 else
5305 {
5306 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5307 {
5308 *code++ = OP_OPT;
5309 *code++ = newoptions & PCRE_IMS;
5310 }
5311 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5312 greedy_non_default = greedy_default ^ 1;
5313 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5314 }
5315
5316 /* Change options at this level, and pass them back for use
5317 in subsequent branches. When not at the start of the pattern, this
5318 information is also necessary so that a resetting item can be
5319 compiled at the end of a group (if we are in a group). */
5320
5321 *optionsptr = options = newoptions;
5322 previous = NULL; /* This item can't be repeated */
5323 continue; /* It is complete */
5324 }
5325
5326 /* If the options ended with ':' we are heading into a nested group
5327 with possible change of options. Such groups are non-capturing and are
5328 not assertions of any kind. All we need to do is skip over the ':';
5329 the newoptions value is handled below. */
5330
5331 bravalue = OP_BRA;
5332 ptr++;
5333 } /* End of switch for character following (? */
5334 } /* End of (? handling */
5335
5336 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5337 all unadorned brackets become non-capturing and behave like (?:...)
5338 brackets. */
5339
5340 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5341 {
5342 bravalue = OP_BRA;
5343 }
5344
5345 /* Else we have a capturing group. */
5346
5347 else
5348 {
5349 NUMBERED_GROUP:
5350 cd->bracount += 1;
5351 PUT2(code, 1+LINK_SIZE, cd->bracount);
5352 skipbytes = 2;
5353 }
5354
5355 /* Process nested bracketed regex. Assertions may not be repeated, but
5356 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5357 non-register variable in order to be able to pass its address because some
5358 compilers complain otherwise. Pass in a new setting for the ims options if
5359 they have changed. */
5360
5361 previous = (bravalue >= OP_ONCE)? code : NULL;
5362 *code = bravalue;
5363 tempcode = code;
5364 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5365 length_prevgroup = 0; /* Initialize for pre-compile phase */
5366
5367 if (!compile_regex(
5368 newoptions, /* The complete new option state */
5369 options & PCRE_IMS, /* The previous ims option state */
5370 &tempcode, /* Where to put code (updated) */
5371 &ptr, /* Input pointer (updated) */
5372 errorcodeptr, /* Where to put an error message */
5373 (bravalue == OP_ASSERTBACK ||
5374 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5375 reset_bracount, /* True if (?| group */
5376 skipbytes, /* Skip over bracket number */
5377 &subfirstbyte, /* For possible first char */
5378 &subreqbyte, /* For possible last char */
5379 bcptr, /* Current branch chain */
5380 cd, /* Tables block */
5381 (lengthptr == NULL)? NULL : /* Actual compile phase */
5382 &length_prevgroup /* Pre-compile phase */
5383 ))
5384 goto FAILED;
5385
5386 /* At the end of compiling, code is still pointing to the start of the
5387 group, while tempcode has been updated to point past the end of the group
5388 and any option resetting that may follow it. The pattern pointer (ptr)
5389 is on the bracket. */
5390
5391 /* If this is a conditional bracket, check that there are no more than
5392 two branches in the group, or just one if it's a DEFINE group. We do this
5393 in the real compile phase, not in the pre-pass, where the whole group may
5394 not be available. */
5395
5396 if (bravalue == OP_COND && lengthptr == NULL)
5397 {
5398 uschar *tc = code;
5399 int condcount = 0;
5400
5401 do {
5402 condcount++;
5403 tc += GET(tc,1);
5404 }
5405 while (*tc != OP_KET);
5406
5407 /* A DEFINE group is never obeyed inline (the "condition" is always
5408 false). It must have only one branch. */
5409
5410 if (code[LINK_SIZE+1] == OP_DEF)
5411 {
5412 if (condcount > 1)
5413 {
5414 *errorcodeptr = ERR54;
5415 goto FAILED;
5416 }
5417 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5418 }
5419
5420 /* A "normal" conditional group. If there is just one branch, we must not
5421 make use of its firstbyte or reqbyte, because this is equivalent to an
5422 empty second branch. */
5423
5424 else
5425 {
5426 if (condcount > 2)
5427 {
5428 *errorcodeptr = ERR27;
5429 goto FAILED;
5430 }
5431 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5432 }
5433 }
5434
5435 /* Error if hit end of pattern */
5436
5437 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5438 {
5439 *errorcodeptr = ERR14;
5440 goto FAILED;
5441 }
5442
5443 /* In the pre-compile phase, update the length by the length of the group,
5444 less the brackets at either end. Then reduce the compiled code to just a
5445 set of non-capturing brackets so that it doesn't use much memory if it is
5446 duplicated by a quantifier.*/
5447
5448 if (lengthptr != NULL)
5449 {
5450 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5451 {
5452 *errorcodeptr = ERR20;
5453 goto FAILED;
5454 }
5455 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5456 *code++ = OP_BRA;
5457 PUTINC(code, 0, 1 + LINK_SIZE);
5458 *code++ = OP_KET;
5459 PUTINC(code, 0, 1 + LINK_SIZE);
5460 break; /* No need to waste time with special character handling */
5461 }
5462
5463 /* Otherwise update the main code pointer to the end of the group. */
5464
5465 code = tempcode;
5466
5467 /* For a DEFINE group, required and first character settings are not
5468 relevant. */
5469
5470 if (bravalue == OP_DEF) break;
5471
5472 /* Handle updating of the required and first characters for other types of
5473 group. Update for normal brackets of all kinds, and conditions with two
5474 branches (see code above). If the bracket is followed by a quantifier with
5475 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5476 zerofirstbyte outside the main loop so that they can be accessed for the
5477 back off. */
5478
5479 zeroreqbyte = reqbyte;
5480 zerofirstbyte = firstbyte;
5481 groupsetfirstbyte = FALSE;
5482
5483 if (bravalue >= OP_ONCE)
5484 {
5485 /* If we have not yet set a firstbyte in this branch, take it from the
5486 subpattern, remembering that it was set here so that a repeat of more
5487 than one can replicate it as reqbyte if necessary. If the subpattern has
5488 no firstbyte, set "none" for the whole branch. In both cases, a zero
5489 repeat forces firstbyte to "none". */
5490
5491 if (firstbyte == REQ_UNSET)
5492 {
5493 if (subfirstbyte >= 0)
5494 {
5495 firstbyte = subfirstbyte;
5496 groupsetfirstbyte = TRUE;
5497 }
5498 else firstbyte = REQ_NONE;
5499 zerofirstbyte = REQ_NONE;
5500 }
5501
5502 /* If firstbyte was previously set, convert the subpattern's firstbyte
5503 into reqbyte if there wasn't one, using the vary flag that was in
5504 existence beforehand. */
5505
5506 else if (subfirstbyte >= 0 && subreqbyte < 0)
5507 subreqbyte = subfirstbyte | tempreqvary;
5508
5509 /* If the subpattern set a required byte (or set a first byte that isn't
5510 really the first byte - see above), set it. */
5511
5512 if (subreqbyte >= 0) reqbyte = subreqbyte;
5513 }
5514
5515 /* For a forward assertion, we take the reqbyte, if set. This can be
5516 helpful if the pattern that follows the assertion doesn't set a different
5517 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5518 for an assertion, however because it leads to incorrect effect for patterns
5519 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5520 of a firstbyte. This is overcome by a scan at the end if there's no
5521 firstbyte, looking for an asserted first char. */
5522
5523 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5524 break; /* End of processing '(' */
5525
5526
5527 /* ===================================================================*/
5528 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5529 are arranged to be the negation of the corresponding OP_values. For the
5530 back references, the values are ESC_REF plus the reference number. Only
5531 back references and those types that consume a character may be repeated.
5532 We can test for values between ESC_b and ESC_Z for the latter; this may
5533 have to change if any new ones are ever created. */
5534
5535 case CHAR_BACKSLASH:
5536 tempptr = ptr;
5537 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5538 if (*errorcodeptr != 0) goto FAILED;
5539
5540 if (c < 0)
5541 {
5542 if (-c == ESC_Q) /* Handle start of quoted string */
5543 {
5544 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5545 ptr += 2; /* avoid empty string */
5546 else inescq = TRUE;
5547 continue;
5548 }
5549
5550 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5551
5552 /* For metasequences that actually match a character, we disable the
5553 setting of a first character if it hasn't already been set. */
5554
5555 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5556 firstbyte = REQ_NONE;
5557
5558 /* Set values to reset to if this is followed by a zero repeat. */
5559
5560 zerofirstbyte = firstbyte;
5561 zeroreqbyte = reqbyte;
5562
5563 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5564 is a subroutine call by number (Oniguruma syntax). In fact, the value
5565 -ESC_g is returned only for these cases. So we don't need to check for <
5566 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5567 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5568 that is a synonym for a named back reference). */
5569
5570 if (-c == ESC_g)
5571 {
5572 const uschar *p;
5573 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5574 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5575 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5576
5577 /* These two statements stop the compiler for warning about possibly
5578 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5579 fact, because we actually check for a number below, the paths that
5580 would actually be in error are never taken. */
5581
5582 skipbytes = 0;
5583 reset_bracount = FALSE;
5584
5585 /* Test for a name */
5586
5587 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5588 {
5589 BOOL isnumber = TRUE;
5590 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5591 {
5592 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5593 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5594 }
5595 if (*p != terminator)
5596 {
5597 *errorcodeptr = ERR57;
5598 break;
5599 }
5600 if (isnumber)
5601 {
5602 ptr++;
5603 goto HANDLE_NUMERICAL_RECURSION;
5604 }
5605 is_recurse = TRUE;
5606 goto NAMED_REF_OR_RECURSE;
5607 }
5608
5609 /* Test a signed number in angle brackets or quotes. */
5610
5611 p = ptr + 2;
5612 while ((digitab[*p] & ctype_digit) != 0) p++;
5613 if (*p != terminator)
5614 {
5615 *errorcodeptr = ERR57;
5616 break;
5617 }
5618 ptr++;
5619 goto HANDLE_NUMERICAL_RECURSION;
5620 }
5621
5622 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5623 We also support \k{name} (.NET syntax) */
5624
5625 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5626 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5627 {
5628 is_recurse = FALSE;
5629 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5630 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5631 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5632 goto NAMED_REF_OR_RECURSE;
5633 }
5634
5635 /* Back references are handled specially; must disable firstbyte if
5636 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5637 ':' later. */
5638
5639 if (-c >= ESC_REF)
5640 {
5641 open_capitem *oc;
5642 recno = -c - ESC_REF;
5643
5644 HANDLE_REFERENCE: /* Come here from named backref handling */
5645 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5646 previous = code;
5647 *code++ = OP_REF;
5648 PUT2INC(code, 0, recno);
5649 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5650 if (recno > cd->top_backref) cd->top_backref = recno;
5651
5652 /* Check to see if this back reference is recursive, that it, it
5653 is inside the group that it references. A flag is set so that the
5654 group can be made atomic. */
5655
5656 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5657 {
5658 if (oc->number == recno)
5659 {
5660 oc->flag = TRUE;
5661 break;
5662 }
5663 }
5664 }
5665
5666 /* So are Unicode property matches, if supported. */
5667
5668 #ifdef SUPPORT_UCP
5669 else if (-c == ESC_P || -c == ESC_p)
5670 {
5671 BOOL negated;
5672 int pdata;
5673 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5674 if (ptype < 0) goto FAILED;
5675 previous = code;
5676 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5677 *code++ = ptype;
5678 *code++ = pdata;
5679 }
5680 #else
5681
5682 /* If Unicode properties are not supported, \X, \P, and \p are not
5683 allowed. */
5684
5685 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5686 {
5687 *errorcodeptr = ERR45;
5688 goto FAILED;
5689 }
5690 #endif
5691
5692 /* For the rest (including \X when Unicode properties are supported), we
5693 can obtain the OP value by negating the escape value. */
5694
5695 else
5696 {
5697 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5698 *code++ = -c;
5699 }
5700 continue;
5701 }
5702
5703 /* We have a data character whose value is in c. In UTF-8 mode it may have
5704 a value > 127. We set its representation in the length/buffer, and then
5705 handle it as a data character. */
5706
5707 #ifdef SUPPORT_UTF8
5708 if (utf8 && c > 127)
5709 mclength = _pcre_ord2utf8(c, mcbuffer);
5710 else
5711 #endif
5712
5713 {
5714 mcbuffer[0] = c;
5715 mclength = 1;
5716 }
5717 goto ONE_CHAR;
5718
5719
5720 /* ===================================================================*/
5721 /* Handle a literal character. It is guaranteed not to be whitespace or #
5722 when the extended flag is set. If we are in UTF-8 mode, it may be a
5723 multi-byte literal character. */
5724
5725 default:
5726 NORMAL_CHAR:
5727 mclength = 1;
5728 mcbuffer[0] = c;
5729
5730 #ifdef SUPPORT_UTF8
5731 if (utf8 && c >= 0xc0)
5732 {
5733 while ((ptr[1] & 0xc0) == 0x80)
5734 mcbuffer[mclength++] = *(++ptr);
5735 }
5736 #endif
5737
5738 /* At this point we have the character's bytes in mcbuffer, and the length
5739 in mclength. When not in UTF-8 mode, the length is always 1. */
5740
5741 ONE_CHAR:
5742 previous = code;
5743 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5744 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5745
5746 /* Remember if \r or \n were seen */
5747
5748 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5749 cd->external_flags |= PCRE_HASCRORLF;
5750
5751 /* Set the first and required bytes appropriately. If no previous first
5752 byte, set it from this character, but revert to none on a zero repeat.
5753 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5754 repeat. */
5755
5756 if (firstbyte == REQ_UNSET)
5757 {
5758 zerofirstbyte = REQ_NONE;
5759 zeroreqbyte = reqbyte;
5760
5761 /* If the character is more than one byte long, we can set firstbyte
5762 only if it is not to be matched caselessly. */
5763
5764 if (mclength == 1 || req_caseopt == 0)
5765 {
5766 firstbyte = mcbuffer[0] | req_caseopt;
5767 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5768 }
5769 else firstbyte = reqbyte = REQ_NONE;
5770 }
5771
5772 /* firstbyte was previously set; we can set reqbyte only the length is
5773 1 or the matching is caseful. */
5774
5775 else
5776 {
5777 zerofirstbyte = firstbyte;
5778 zeroreqbyte = reqbyte;
5779 if (mclength == 1 || req_caseopt == 0)
5780 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5781 }
5782
5783 break; /* End of literal character handling */
5784 }
5785 } /* end of big loop */
5786
5787
5788 /* Control never reaches here by falling through, only by a goto for all the
5789 error states. Pass back the position in the pattern so that it can be displayed
5790 to the user for diagnosing the error. */
5791
5792 FAILED:
5793 *ptrptr = ptr;
5794 return FALSE;
5795 }
5796
5797
5798
5799
5800 /*************************************************
5801 * Compile sequence of alternatives *
5802 *************************************************/
5803
5804 /* On entry, ptr is pointing past the bracket character, but on return it
5805 points to the closing bracket, or vertical bar, or end of string. The code
5806 variable is pointing at the byte into which the BRA operator has been stored.
5807 If the ims options are changed at the start (for a (?ims: group) or during any
5808 branch, we need to insert an OP_OPT item at the start of every following branch
5809 to ensure they get set correctly at run time, and also pass the new options
5810 into every subsequent branch compile.
5811
5812 This function is used during the pre-compile phase when we are trying to find
5813 out the amount of memory needed, as well as during the real compile phase. The
5814 value of lengthptr distinguishes the two phases.
5815
5816 Arguments:
5817 options option bits, including any changes for this subpattern
5818 oldims previous settings of ims option bits
5819 codeptr -> the address of the current code pointer
5820 ptrptr -> the address of the current pattern pointer
5821 errorcodeptr -> pointer to error code variable
5822 lookbehind TRUE if this is a lookbehind assertion
5823 reset_bracount TRUE to reset the count for each branch
5824 skipbytes skip this many bytes at start (for brackets and OP_COND)
5825 firstbyteptr place to put the first required character, or a negative number
5826 reqbyteptr place to put the last required character, or a negative number
5827 bcptr pointer to the chain of currently open branches
5828 cd points to the data block with tables pointers etc.
5829 lengthptr NULL during the real compile phase
5830 points to length accumulator during pre-compile phase
5831
5832 Returns: TRUE on success
5833 */
5834
5835 static BOOL
5836 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5837 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5838 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5839 int *lengthptr)
5840 {
5841 const uschar *ptr = *ptrptr;
5842 uschar *code = *codeptr;
5843 uschar *last_branch = code;
5844 uschar *start_bracket = code;
5845 uschar *reverse_count = NULL;
5846 open_capitem capitem;
5847 int capnumber = 0;
5848 int firstbyte, reqbyte;
5849 int branchfirstbyte, branchreqbyte;
5850 int length;
5851 int orig_bracount;
5852 int max_bracount;
5853 int old_external_options = cd->external_options;
5854 branch_chain bc;
5855
5856 bc.outer = bcptr;
5857 bc.current_branch = code;
5858
5859 firstbyte = reqbyte = REQ_UNSET;
5860
5861 /* Accumulate the length for use in the pre-compile phase. Start with the
5862 length of the BRA and KET and any extra bytes that are required at the
5863 beginning. We accumulate in a local variable to save frequent testing of
5864 lenthptr for NULL. We cannot do this by looking at the value of code at the
5865 start and end of each alternative, because compiled items are discarded during
5866 the pre-compile phase so that the work space is not exceeded. */
5867
5868 length = 2 + 2*LINK_SIZE + skipbytes;
5869
5870 /* WARNING: If the above line is changed for any reason, you must also change
5871 the code that abstracts option settings at the start of the pattern and makes
5872 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5873 pre-compile phase to find out whether anything has yet been compiled or not. */
5874
5875 /* If this is a capturing subpattern, add to the chain of open capturing items
5876 so that we can detect them if (*ACCEPT) is encountered. This is also used to
5877 detect groups that contain recursive back references to themselves. */
5878
5879 if (*code == OP_CBRA)
5880 {
5881 capnumber = GET2(code, 1 + LINK_SIZE);
5882 capitem.number = capnumber;
5883 capitem.next = cd->open_caps;
5884 capitem.flag = FALSE;
5885 cd->open_caps = &capitem;
5886 }
5887
5888 /* Offset is set zero to mark that this bracket is still open */
5889
5890 PUT(code, 1, 0);
5891 code += 1 + LINK_SIZE + skipbytes;
5892
5893 /* Loop for each alternative branch */
5894
5895 orig_bracount = max_bracount = cd->bracount;
5896 for (;;)
5897 {
5898 /* For a (?| group, reset the capturing bracket count so that each branch
5899 uses the same numbers. */
5900
5901 if (reset_bracount) cd->bracount = orig_bracount;
5902
5903 /* Handle a change of ims options at the start of the branch */
5904
5905 if ((options & PCRE_IMS) != oldims)
5906 {
5907 *code++ = OP_OPT;
5908 *code++ = options & PCRE_IMS;
5909 length += 2;
5910 }
5911
5912 /* Set up dummy OP_REVERSE if lookbehind assertion */
5913
5914 if (lookbehind)
5915 {
5916 *code++ = OP_REVERSE;
5917 reverse_count = code;
5918 PUTINC(code, 0, 0);
5919 length += 1 + LINK_SIZE;
5920 }
5921
5922 /* Now compile the branch; in the pre-compile phase its length gets added
5923 into the length. */
5924
5925 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5926 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5927 {
5928 *ptrptr = ptr;
5929 return FALSE;
5930 }
5931
5932 /* If the external options have changed during this branch, it means that we
5933 are at the top level, and a leading option setting has been encountered. We
5934 need to re-set the original option values to take account of this so that,
5935 during the pre-compile phase, we know to allow for a re-set at the start of
5936 subsequent branches. */
5937
5938 if (old_external_options != cd->external_options)
5939 oldims = cd->external_options & PCRE_IMS;
5940
5941 /* Keep the highest bracket count in case (?| was used and some branch
5942 has fewer than the rest. */
5943
5944 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5945
5946 /* In the real compile phase, there is some post-processing to be done. */
5947
5948 if (lengthptr == NULL)
5949 {
5950 /* If this is the first branch, the firstbyte and reqbyte values for the
5951 branch become the values for the regex. */
5952
5953 if (*last_branch != OP_ALT)
5954 {
5955 firstbyte = branchfirstbyte;
5956 reqbyte = branchreqbyte;
5957 }
5958
5959 /* If this is not the first branch, the first char and reqbyte have to
5960 match the values from all the previous branches, except that if the
5961 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5962 and we set REQ_VARY for the regex. */
5963
5964 else
5965 {
5966 /* If we previously had a firstbyte, but it doesn't match the new branch,
5967 we have to abandon the firstbyte for the regex, but if there was
5968 previously no reqbyte, it takes on the value of the old firstbyte. */
5969
5970 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5971 {
5972 if (reqbyte < 0) reqbyte = firstbyte;
5973 firstbyte = REQ_NONE;
5974 }
5975
5976 /* If we (now or from before) have no firstbyte, a firstbyte from the
5977 branch becomes a reqbyte if there isn't a branch reqbyte. */
5978
5979 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5980 branchreqbyte = branchfirstbyte;
5981
5982 /* Now ensure that the reqbytes match */
5983
5984 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5985 reqbyte = REQ_NONE;
5986 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5987 }
5988
5989 /* If lookbehind, check that this branch matches a fixed-length string, and
5990 put the length into the OP_REVERSE item. Temporarily mark the end of the
5991 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
5992 because there may be forward references that we can't check here. Set a
5993 flag to cause another lookbehind check at the end. Why not do it all at the
5994 end? Because common, erroneous checks are picked up here and the offset of
5995 the problem can be shown. */
5996
5997 if (lookbehind)
5998 {
5999 int fixed_length;
6000 *code = OP_END;
6001 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
6002 DPRINTF(("fixed length = %d\n", fixed_length));
6003 if (fixed_length == -3)
6004 {
6005 cd->check_lookbehind = TRUE;
6006 }
6007 else if (fixed_length < 0)
6008 {
6009 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6010 *ptrptr = ptr;
6011 return FALSE;
6012 }
6013 else { PUT(reverse_count, 0, fixed_length); }
6014 }
6015 }
6016
6017 /* Reached end of expression, either ')' or end of pattern. In the real
6018 compile phase, go back through the alternative branches and reverse the chain
6019 of offsets, with the field in the BRA item now becoming an offset to the
6020 first alternative. If there are no alternatives, it points to the end of the
6021 group. The length in the terminating ket is always the length of the whole
6022 bracketed item. If any of the ims options were changed inside the group,
6023 compile a resetting op-code following, except at the very end of the pattern.
6024 Return leaving the pointer at the terminating char. */
6025
6026 if (*ptr != CHAR_VERTICAL_LINE)
6027 {
6028 if (lengthptr == NULL)
6029 {
6030 int branch_length = code - last_branch;
6031 do
6032 {
6033 int prev_length = GET(last_branch, 1);
6034 PUT(last_branch, 1, branch_length);
6035 branch_length = prev_length;
6036 last_branch -= branch_length;
6037 }
6038 while (branch_length > 0);
6039 }
6040
6041 /* Fill in the ket */
6042
6043 *code = OP_KET;
6044 PUT(code, 1, code - start_bracket);
6045 code += 1 + LINK_SIZE;
6046
6047 /* If it was a capturing subpattern, check to see if it contained any
6048 recursive back references. If so, we must wrap it in atomic brackets.
6049 In any event, remove the block from the chain. */
6050
6051 if (capnumber > 0)
6052 {
6053 if (cd->open_caps->flag)
6054 {
6055 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6056 code - start_bracket);
6057 *start_bracket = OP_ONCE;
6058 code += 1 + LINK_SIZE;
6059 PUT(start_bracket, 1, code - start_bracket);
6060 *code = OP_KET;
6061 PUT(code, 1, code - start_bracket);
6062 code += 1 + LINK_SIZE;
6063 length += 2 + 2*LINK_SIZE;
6064 }
6065 cd->open_caps = cd->open_caps->next;
6066 }
6067
6068 /* Reset options if needed. */
6069
6070 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6071 {
6072 *code++ = OP_OPT;
6073 *code++ = oldims;
6074 length += 2;
6075 }
6076
6077 /* Retain the highest bracket number, in case resetting was used. */
6078
6079 cd->bracount = max_bracount;
6080
6081 /* Set values to pass back */
6082
6083 *codeptr = code;
6084 *ptrptr = ptr;
6085 *firstbyteptr = firstbyte;
6086 *reqbyteptr = reqbyte;
6087 if (lengthptr != NULL)
6088 {
6089 if (OFLOW_MAX - *lengthptr < length)
6090 {
6091 *errorcodeptr = ERR20;
6092 return FALSE;
6093 }
6094 *lengthptr += length;
6095 }
6096 return TRUE;
6097 }
6098
6099 /* Another branch follows. In the pre-compile phase, we can move the code
6100 pointer back to where it was for the start of the first branch. (That is,
6101 pretend that each branch is the only one.)
6102
6103 In the real compile phase, insert an ALT node. Its length field points back
6104 to the previous branch while the bracket remains open. At the end the chain
6105 is reversed. It's done like this so that the start of the bracket has a
6106 zero offset until it is closed, making it possible to detect recursion. */
6107
6108 if (lengthptr != NULL)
6109 {
6110 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6111 length += 1 + LINK_SIZE;
6112 }
6113 else
6114 {
6115 *code = OP_ALT;
6116 PUT(code, 1, code - last_branch);
6117 bc.current_branch = last_branch = code;
6118 code += 1 + LINK_SIZE;
6119 }
6120
6121 ptr++;
6122 }
6123 /* Control never reaches here */
6124 }
6125
6126
6127
6128
6129 /*************************************************
6130 * Check for anchored expression *
6131 *************************************************/
6132
6133 /* Try to find out if this is an anchored regular expression. Consider each
6134 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6135 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6136 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6137 counts, since OP_CIRC can match in the middle.
6138
6139 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6140 This is the code for \G, which means "match at start of match position, taking
6141 into account the match offset".
6142
6143 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6144 because that will try the rest of the pattern at all possible matching points,
6145 so there is no point trying again.... er ....
6146
6147 .... except when the .* appears inside capturing parentheses, and there is a
6148 subsequent back reference to those parentheses. We haven't enough information
6149 to catch that case precisely.
6150
6151 At first, the best we could do was to detect when .* was in capturing brackets
6152 and the highest back reference was greater than or equal to that level.
6153 However, by keeping a bitmap of the first 31 back references, we can catch some
6154 of the more common cases more precisely.
6155
6156 Arguments:
6157 code points to start of expression (the bracket)
6158 options points to the options setting
6159 bracket_map a bitmap of which brackets we are inside while testing; this
6160 handles up to substring 31; after that we just have to take
6161 the less precise approach
6162 backref_map the back reference bitmap
6163
6164 Returns: TRUE or FALSE
6165 */
6166
6167 static BOOL
6168 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6169 unsigned int backref_map)
6170 {
6171 do {
6172 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6173 options, PCRE_MULTILINE, FALSE);
6174 register int op = *scode;
6175
6176 /* Non-capturing brackets */
6177
6178 if (op == OP_BRA)
6179 {
6180 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6181 }
6182
6183 /* Capturing brackets */
6184
6185 else if (op == OP_CBRA)
6186 {
6187 int n = GET2(scode, 1+LINK_SIZE);
6188 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6189 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6190 }
6191
6192 /* Other brackets */
6193
6194 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6195 {
6196 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6197 }
6198
6199 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6200 it isn't in brackets that are or may be referenced. */
6201
6202 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6203 op == OP_TYPEPOSSTAR))
6204 {
6205 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6206 return FALSE;
6207 }
6208
6209 /* Check for explicit anchoring */
6210
6211 else if (op != OP_SOD && op != OP_SOM &&
6212 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6213 return FALSE;
6214 code += GET(code, 1);
6215 }
6216 while (*code == OP_ALT); /* Loop for each alternative */
6217 return TRUE;
6218 }
6219
6220
6221
6222 /*************************************************
6223 * Check for starting with ^ or .* *
6224 *************************************************/
6225
6226 /* This is called to find out if every branch starts with ^ or .* so that
6227 "first char" processing can be done to speed things up in multiline
6228 matching and for non-DOTALL patterns that start with .* (which must start at
6229 the beginning or after \n). As in the case of is_anchored() (see above), we
6230 have to take account of back references to capturing brackets that contain .*
6231 because in that case we can't make the assumption.
6232
6233 Arguments:
6234 code points to start of expression (the bracket)
6235 bracket_map a bitmap of which brackets we are inside while testing; this
6236 handles up to substring 31; after that we just have to take
6237 the less precise approach
6238 backref_map the back reference bitmap
6239
6240 Returns: TRUE or FALSE
6241 */
6242
6243 static BOOL
6244 is_startline(const uschar *code, unsigned int bracket_map,
6245 unsigned int backref_map)
6246 {
6247 do {
6248 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6249 NULL, 0, FALSE);
6250 register int op = *scode;
6251
6252 /* If we are at the start of a conditional assertion group, *both* the
6253 conditional assertion *and* what follows the condition must satisfy the test
6254 for start of line. Other kinds of condition fail. Note that there may be an
6255 auto-callout at the start of a condition. */
6256
6257 if (op == OP_COND)
6258 {
6259 scode += 1 + LINK_SIZE;
6260 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6261 switch (*scode)
6262 {
6263 case OP_CREF:
6264 case OP_NCREF:
6265 case OP_RREF:
6266 case OP_NRREF:
6267 case OP_DEF:
6268 return FALSE;
6269
6270 default: /* Assertion */
6271 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6272 do scode += GET(scode, 1); while (*scode == OP_ALT);
6273 scode += 1 + LINK_SIZE;
6274 break;
6275 }
6276 scode = first_significant_code(scode, NULL, 0, FALSE);
6277 op = *scode;
6278 }
6279
6280 /* Non-capturing brackets */
6281
6282 if (op == OP_BRA)
6283 {
6284 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6285 }
6286
6287 /* Capturing brackets */
6288
6289 else if (op == OP_CBRA)
6290 {
6291 int n = GET2(scode, 1+LINK_SIZE);
6292 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6293 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6294 }
6295
6296 /* Other brackets */
6297
6298 else if (op == OP_ASSERT || op == OP_ONCE)
6299 {
6300 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6301 }
6302
6303 /* .* means "start at start or after \n" if it isn't in brackets that
6304 may be referenced. */
6305
6306 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6307 {
6308 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6309 }
6310
6311 /* Check for explicit circumflex */
6312
6313 else if (op != OP_CIRC) return FALSE;
6314
6315 /* Move on to the next alternative */
6316
6317 code += GET(code, 1);
6318 }
6319 while (*code == OP_ALT); /* Loop for each alternative */
6320 return TRUE;
6321 }
6322
6323
6324
6325 /*************************************************
6326 * Check for asserted fixed first char *
6327 *************************************************/
6328
6329 /* During compilation, the "first char" settings from forward assertions are
6330 discarded, because they can cause conflicts with actual literals that follow.
6331 However, if we end up without a first char setting for an unanchored pattern,
6332 it is worth scanning the regex to see if there is an initial asserted first
6333 char. If all branches start with the same asserted char, or with a bracket all
6334 of whose alternatives start with the same asserted char (recurse ad lib), then
6335 we return that char, otherwise -1.
6336
6337 Arguments:
6338 code points to start of expression (the bracket)
6339 options pointer to the options (used to check casing changes)
6340 inassert TRUE if in an assertion
6341
6342 Returns: -1 or the fixed first char
6343 */
6344
6345 static int
6346 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6347 {
6348 register int c = -1;
6349 do {
6350 int d;
6351 const uschar *scode =
6352 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6353 register int op = *scode;
6354
6355 switch(op)
6356 {
6357 default:
6358 return -1;
6359
6360 case OP_BRA:
6361 case OP_CBRA:
6362 case OP_ASSERT:
6363 case OP_ONCE:
6364 case OP_COND:
6365 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6366 return -1;
6367 if (c < 0) c = d; else if (c != d) return -1;
6368 break;
6369
6370 case OP_EXACT: /* Fall through */
6371 scode += 2;
6372
6373 case OP_CHAR:
6374 case OP_CHARNC:
6375 case OP_PLUS:
6376 case OP_MINPLUS:
6377 case OP_POSPLUS:
6378 if (!inassert) return -1;
6379 if (c < 0)
6380 {
6381 c = scode[1];
6382 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6383 }
6384 else if (c != scode[1]) return -1;
6385 break;
6386 }
6387
6388 code += GET(code, 1);
6389 }
6390 while (*code == OP_ALT);
6391 return c;
6392 }
6393
6394
6395
6396 /*************************************************
6397 * Compile a Regular Expression *
6398 *************************************************/
6399
6400 /* This function takes a string and returns a pointer to a block of store
6401 holding a compiled version of the expression. The original API for this
6402 function had no error code return variable; it is retained for backwards
6403 compatibility. The new function is given a new name.
6404
6405 Arguments:
6406 pattern the regular expression
6407 options various option bits
6408 errorcodeptr pointer to error code variable (pcre_compile2() only)
6409 can be NULL if you don't want a code value
6410 errorptr pointer to pointer to error text
6411 erroroffset ptr offset in pattern where error was detected
6412 tables pointer to character tables or NULL
6413
6414 Returns: pointer to compiled data block, or NULL on error,
6415 with errorptr and erroroffset set
6416 */
6417
6418 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6419 pcre_compile(const char *pattern, int options, const char **errorptr,
6420 int *erroroffset, const unsigned char *tables)
6421 {
6422 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6423 }
6424
6425
6426 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6427 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6428 const char **errorptr, int *erroroffset, const unsigned char *tables)
6429 {
6430 real_pcre *re;
6431 int length = 1; /* For final END opcode */
6432 int firstbyte, reqbyte, newline;
6433 int errorcode = 0;
6434 int skipatstart = 0;
6435 BOOL utf8 = (options & PCRE_UTF8) != 0;
6436 size_t size;
6437 uschar *code;
6438 const uschar *codestart;
6439 const uschar *ptr;
6440 compile_data compile_block;
6441 compile_data *cd = &compile_block;
6442
6443 /* This space is used for "compiling" into during the first phase, when we are
6444 computing the amount of memory that is needed. Compiled items are thrown away
6445 as soon as possible, so that a fairly large buffer should be sufficient for
6446 this purpose. The same space is used in the second phase for remembering where
6447 to fill in forward references to subpatterns. */
6448
6449 uschar cworkspace[COMPILE_WORK_SIZE];
6450
6451 /* Set this early so that early errors get offset 0. */
6452
6453 ptr = (const uschar *)pattern;
6454
6455 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6456 can do is just return NULL, but we can set a code value if there is a code
6457 pointer. */
6458
6459 if (errorptr == NULL)
6460 {
6461 if (errorcodeptr != NULL) *errorcodeptr = 99;
6462 return NULL;
6463 }
6464
6465 *errorptr = NULL;
6466 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6467
6468 /* However, we can give a message for this error */
6469
6470 if (erroroffset == NULL)
6471 {
6472 errorcode = ERR16;
6473 goto PCRE_EARLY_ERROR_RETURN2;
6474 }
6475
6476 *erroroffset = 0;
6477
6478 /* Set up pointers to the individual character tables */
6479
6480 if (tables == NULL) tables = _pcre_default_tables;
6481 cd->lcc = tables + lcc_offset;
6482 cd->fcc = tables + fcc_offset;
6483 cd->cbits = tables + cbits_offset;
6484 cd->ctypes = tables + ctypes_offset;
6485
6486 /* Check that all undefined public option bits are zero */
6487
6488 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6489 {
6490 errorcode = ERR17;
6491 goto PCRE_EARLY_ERROR_RETURN;
6492 }
6493
6494 /* Check for global one-time settings at the start of the pattern, and remember
6495 the offset for later. */
6496
6497 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6498 ptr[skipatstart+1] == CHAR_ASTERISK)
6499 {
6500 int newnl = 0;
6501 int newbsr = 0;
6502
6503 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6504 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6505
6506 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6507 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6508 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6509 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6510 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6511 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6512 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6513 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6514 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6515 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6516
6517 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6518 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6519 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6520 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6521
6522 if (newnl != 0)
6523 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6524 else if (newbsr != 0)
6525 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6526 else break;
6527 }
6528
6529 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6530
6531 #ifdef SUPPORT_UTF8
6532 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6533 (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
6534 {
6535 errorcode = ERR44;
6536 goto PCRE_EARLY_ERROR_RETURN2;
6537 }
6538 #else
6539 if (utf8)
6540 {
6541 errorcode = ERR32;
6542 goto PCRE_EARLY_ERROR_RETURN;
6543 }
6544 #endif
6545
6546 /* Check validity of \R options. */
6547
6548 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6549 {
6550 case 0:
6551 case PCRE_BSR_ANYCRLF:
6552 case PCRE_BSR_UNICODE:
6553 break;
6554 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6555 }
6556
6557 /* Handle different types of newline. The three bits give seven cases. The
6558 current code allows for fixed one- or two-byte sequences, plus "any" and
6559 "anycrlf". */
6560
6561 switch (options & PCRE_NEWLINE_BITS)
6562 {
6563 case 0: newline = NEWLINE; break; /* Build-time default */
6564 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6565 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6566 case PCRE_NEWLINE_CR+
6567 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6568 case PCRE_NEWLINE_ANY: newline = -1; break;
6569 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6570 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6571 }
6572
6573 if (newline == -2)
6574 {
6575 cd->nltype = NLTYPE_ANYCRLF;
6576 }
6577 else if (newline < 0)
6578 {
6579 cd->nltype = NLTYPE_ANY;
6580 }
6581 else
6582 {
6583 cd->nltype = NLTYPE_FIXED;
6584 if (newline > 255)
6585 {
6586 cd->nllen = 2;
6587 cd->nl[0] = (newline >> 8) & 255;
6588 cd->nl[1] = newline & 255;
6589 }
6590 else
6591 {
6592 cd->nllen = 1;
6593 cd->nl[0] = newline;
6594 }
6595 }
6596
6597 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6598 references to help in deciding whether (.*) can be treated as anchored or not.
6599 */
6600
6601 cd->top_backref = 0;
6602 cd->backref_map = 0;
6603
6604 /* Reflect pattern for debugging output */
6605
6606 DPRINTF(("------------------------------------------------------------------\n"));
6607 DPRINTF(("%s\n", pattern));
6608
6609 /* Pretend to compile the pattern while actually just accumulating the length
6610 of memory required. This behaviour is triggered by passing a non-NULL final
6611 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6612 to compile parts of the pattern into; the compiled code is discarded when it is
6613 no longer needed, so hopefully this workspace will never overflow, though there
6614 is a test for its doing so. */
6615
6616 cd->bracount = cd->final_bracount = 0;
6617 cd->names_found = 0;
6618 cd->name_entry_size = 0;
6619 cd->name_table = NULL;
6620 cd->start_workspace = cworkspace;
6621 cd->start_code = cworkspace;
6622 cd->hwm = cworkspace;
6623 cd->start_pattern = (const uschar *)pattern;
6624 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6625 cd->req_varyopt = 0;
6626 cd->external_options = options;
6627 cd->external_flags = 0;
6628 cd->open_caps = NULL;
6629
6630 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6631 don't need to look at the result of the function here. The initial options have
6632 been put into the cd block so that they can be changed if an option setting is
6633 found within the regex right at the beginning. Bringing initial option settings
6634 outside can help speed up starting point checks. */
6635
6636 ptr += skipatstart;
6637 code = cworkspace;
6638 *code = OP_BRA;
6639 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6640 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6641 &length);
6642 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6643
6644 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6645 cd->hwm - cworkspace));
6646
6647 if (length > MAX_PATTERN_SIZE)
6648 {
6649 errorcode = ERR20;
6650 goto PCRE_EARLY_ERROR_RETURN;
6651 }
6652
6653 /* Compute the size of data block needed and get it, either from malloc or
6654 externally provided function. Integer overflow should no longer be possible
6655 because nowadays we limit the maximum value of cd->names_found and
6656 cd->name_entry_size. */
6657
6658 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6659 re = (real_pcre *)(pcre_malloc)(size);
6660
6661 if (re == NULL)
6662 {
6663 errorcode = ERR21;
6664 goto PCRE_EARLY_ERROR_RETURN;
6665 }
6666
6667 /* Put in the magic number, and save the sizes, initial options, internal
6668 flags, and character table pointer. NULL is used for the default character
6669 tables. The nullpad field is at the end; it's there to help in the case when a
6670 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6671 pointers. */
6672