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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 455 - (show annotations) (download)
Sat Sep 26 19:12:32 2009 UTC (4 years, 10 months ago) by ph10
File MIME type: text/plain
File size: 217318 byte(s)
Added lower bound length-finding to pcre_study() and use it when matching; make 
the value available via pcre_fullinfo(); also fixed bugs connected with
pcre_study() in pcre_dfa_exec(). 

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2009 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When DEBUG is defined, we need the pcre_printint() function, which is also
57 used by pcretest. DEBUG is not defined when building a production library. */
58
59 #ifdef DEBUG
60 #include "pcre_printint.src"
61 #endif
62
63
64 /* Macro for setting individual bits in class bitmaps. */
65
66 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
67
68 /* Maximum length value to check against when making sure that the integer that
69 holds the compiled pattern length does not overflow. We make it a bit less than
70 INT_MAX to allow for adding in group terminating bytes, so that we don't have
71 to check them every time. */
72
73 #define OFLOW_MAX (INT_MAX - 20)
74
75
76 /*************************************************
77 * Code parameters and static tables *
78 *************************************************/
79
80 /* This value specifies the size of stack workspace that is used during the
81 first pre-compile phase that determines how much memory is required. The regex
82 is partly compiled into this space, but the compiled parts are discarded as
83 soon as they can be, so that hopefully there will never be an overrun. The code
84 does, however, check for an overrun. The largest amount I've seen used is 218,
85 so this number is very generous.
86
87 The same workspace is used during the second, actual compile phase for
88 remembering forward references to groups so that they can be filled in at the
89 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
90 is 4 there is plenty of room. */
91
92 #define COMPILE_WORK_SIZE (4096)
93
94
95 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
96 are simple data values; negative values are for special things like \d and so
97 on. Zero means further processing is needed (for things like \x), or the escape
98 is invalid. */
99
100 #ifndef EBCDIC
101
102 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
103 in UTF-8 mode. */
104
105 static const short int escapes[] = {
106 0, 0,
107 0, 0,
108 0, 0,
109 0, 0,
110 0, 0,
111 CHAR_COLON, CHAR_SEMICOLON,
112 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
113 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
114 CHAR_COMMERCIAL_AT, -ESC_A,
115 -ESC_B, -ESC_C,
116 -ESC_D, -ESC_E,
117 0, -ESC_G,
118 -ESC_H, 0,
119 0, -ESC_K,
120 0, 0,
121 0, 0,
122 -ESC_P, -ESC_Q,
123 -ESC_R, -ESC_S,
124 0, 0,
125 -ESC_V, -ESC_W,
126 -ESC_X, 0,
127 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
128 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
129 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
130 CHAR_GRAVE_ACCENT, 7,
131 -ESC_b, 0,
132 -ESC_d, ESC_e,
133 ESC_f, 0,
134 -ESC_h, 0,
135 0, -ESC_k,
136 0, 0,
137 ESC_n, 0,
138 -ESC_p, 0,
139 ESC_r, -ESC_s,
140 ESC_tee, 0,
141 -ESC_v, -ESC_w,
142 0, 0,
143 -ESC_z
144 };
145
146 #else
147
148 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
149
150 static const short int escapes[] = {
151 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
152 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
153 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
154 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
155 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
156 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
157 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
158 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
159 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
160 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
161 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
162 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
163 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
164 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
165 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
166 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
167 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
168 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
169 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
170 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
171 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
172 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
173 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
174 };
175 #endif
176
177
178 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
179 searched linearly. Put all the names into a single string, in order to reduce
180 the number of relocations when a shared library is dynamically linked. The
181 string is built from string macros so that it works in UTF-8 mode on EBCDIC
182 platforms. */
183
184 typedef struct verbitem {
185 int len;
186 int op;
187 } verbitem;
188
189 static const char verbnames[] =
190 STRING_ACCEPT0
191 STRING_COMMIT0
192 STRING_F0
193 STRING_FAIL0
194 STRING_PRUNE0
195 STRING_SKIP0
196 STRING_THEN;
197
198 static const verbitem verbs[] = {
199 { 6, OP_ACCEPT },
200 { 6, OP_COMMIT },
201 { 1, OP_FAIL },
202 { 4, OP_FAIL },
203 { 5, OP_PRUNE },
204 { 4, OP_SKIP },
205 { 4, OP_THEN }
206 };
207
208 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
209
210
211 /* Tables of names of POSIX character classes and their lengths. The names are
212 now all in a single string, to reduce the number of relocations when a shared
213 library is dynamically loaded. The list of lengths is terminated by a zero
214 length entry. The first three must be alpha, lower, upper, as this is assumed
215 for handling case independence. */
216
217 static const char posix_names[] =
218 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
219 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
220 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
221 STRING_word0 STRING_xdigit;
222
223 static const uschar posix_name_lengths[] = {
224 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
225
226 /* Table of class bit maps for each POSIX class. Each class is formed from a
227 base map, with an optional addition or removal of another map. Then, for some
228 classes, there is some additional tweaking: for [:blank:] the vertical space
229 characters are removed, and for [:alpha:] and [:alnum:] the underscore
230 character is removed. The triples in the table consist of the base map offset,
231 second map offset or -1 if no second map, and a non-negative value for map
232 addition or a negative value for map subtraction (if there are two maps). The
233 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
234 remove vertical space characters, 2 => remove underscore. */
235
236 static const int posix_class_maps[] = {
237 cbit_word, cbit_digit, -2, /* alpha */
238 cbit_lower, -1, 0, /* lower */
239 cbit_upper, -1, 0, /* upper */
240 cbit_word, -1, 2, /* alnum - word without underscore */
241 cbit_print, cbit_cntrl, 0, /* ascii */
242 cbit_space, -1, 1, /* blank - a GNU extension */
243 cbit_cntrl, -1, 0, /* cntrl */
244 cbit_digit, -1, 0, /* digit */
245 cbit_graph, -1, 0, /* graph */
246 cbit_print, -1, 0, /* print */
247 cbit_punct, -1, 0, /* punct */
248 cbit_space, -1, 0, /* space */
249 cbit_word, -1, 0, /* word - a Perl extension */
250 cbit_xdigit,-1, 0 /* xdigit */
251 };
252
253
254 #define STRING(a) # a
255 #define XSTRING(s) STRING(s)
256
257 /* The texts of compile-time error messages. These are "char *" because they
258 are passed to the outside world. Do not ever re-use any error number, because
259 they are documented. Always add a new error instead. Messages marked DEAD below
260 are no longer used. This used to be a table of strings, but in order to reduce
261 the number of relocations needed when a shared library is loaded dynamically,
262 it is now one long string. We cannot use a table of offsets, because the
263 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
264 simply count through to the one we want - this isn't a performance issue
265 because these strings are used only when there is a compilation error. */
266
267 static const char error_texts[] =
268 "no error\0"
269 "\\ at end of pattern\0"
270 "\\c at end of pattern\0"
271 "unrecognized character follows \\\0"
272 "numbers out of order in {} quantifier\0"
273 /* 5 */
274 "number too big in {} quantifier\0"
275 "missing terminating ] for character class\0"
276 "invalid escape sequence in character class\0"
277 "range out of order in character class\0"
278 "nothing to repeat\0"
279 /* 10 */
280 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
281 "internal error: unexpected repeat\0"
282 "unrecognized character after (? or (?-\0"
283 "POSIX named classes are supported only within a class\0"
284 "missing )\0"
285 /* 15 */
286 "reference to non-existent subpattern\0"
287 "erroffset passed as NULL\0"
288 "unknown option bit(s) set\0"
289 "missing ) after comment\0"
290 "parentheses nested too deeply\0" /** DEAD **/
291 /* 20 */
292 "regular expression is too large\0"
293 "failed to get memory\0"
294 "unmatched parentheses\0"
295 "internal error: code overflow\0"
296 "unrecognized character after (?<\0"
297 /* 25 */
298 "lookbehind assertion is not fixed length\0"
299 "malformed number or name after (?(\0"
300 "conditional group contains more than two branches\0"
301 "assertion expected after (?(\0"
302 "(?R or (?[+-]digits must be followed by )\0"
303 /* 30 */
304 "unknown POSIX class name\0"
305 "POSIX collating elements are not supported\0"
306 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
307 "spare error\0" /** DEAD **/
308 "character value in \\x{...} sequence is too large\0"
309 /* 35 */
310 "invalid condition (?(0)\0"
311 "\\C not allowed in lookbehind assertion\0"
312 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
313 "number after (?C is > 255\0"
314 "closing ) for (?C expected\0"
315 /* 40 */
316 "recursive call could loop indefinitely\0"
317 "unrecognized character after (?P\0"
318 "syntax error in subpattern name (missing terminator)\0"
319 "two named subpatterns have the same name\0"
320 "invalid UTF-8 string\0"
321 /* 45 */
322 "support for \\P, \\p, and \\X has not been compiled\0"
323 "malformed \\P or \\p sequence\0"
324 "unknown property name after \\P or \\p\0"
325 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
326 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
327 /* 50 */
328 "repeated subpattern is too long\0" /** DEAD **/
329 "octal value is greater than \\377 (not in UTF-8 mode)\0"
330 "internal error: overran compiling workspace\0"
331 "internal error: previously-checked referenced subpattern not found\0"
332 "DEFINE group contains more than one branch\0"
333 /* 55 */
334 "repeating a DEFINE group is not allowed\0"
335 "inconsistent NEWLINE options\0"
336 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
337 "a numbered reference must not be zero\0"
338 "(*VERB) with an argument is not supported\0"
339 /* 60 */
340 "(*VERB) not recognized\0"
341 "number is too big\0"
342 "subpattern name expected\0"
343 "digit expected after (?+\0"
344 "] is an invalid data character in JavaScript compatibility mode";
345
346
347 /* Table to identify digits and hex digits. This is used when compiling
348 patterns. Note that the tables in chartables are dependent on the locale, and
349 may mark arbitrary characters as digits - but the PCRE compiling code expects
350 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
351 a private table here. It costs 256 bytes, but it is a lot faster than doing
352 character value tests (at least in some simple cases I timed), and in some
353 applications one wants PCRE to compile efficiently as well as match
354 efficiently.
355
356 For convenience, we use the same bit definitions as in chartables:
357
358 0x04 decimal digit
359 0x08 hexadecimal digit
360
361 Then we can use ctype_digit and ctype_xdigit in the code. */
362
363 #ifndef EBCDIC
364
365 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
366 UTF-8 mode. */
367
368 static const unsigned char digitab[] =
369 {
370 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
371 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
372 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
373 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
374 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
375 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
376 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
377 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
378 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
379 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
382 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
383 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
384 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
387 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
388 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
394 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
395 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
396 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
402
403 #else
404
405 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
406
407 static const unsigned char digitab[] =
408 {
409 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
410 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
411 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
412 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
413 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
414 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
420 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
421 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
422 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
423 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
424 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
425 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
426 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
431 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
433 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
439 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
440 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
441
442 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
443 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
444 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
445 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
446 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
447 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
448 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
449 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
450 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
451 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
452 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
453 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
454 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
456 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
457 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
458 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
459 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
460 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
461 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
462 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
463 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
464 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
465 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
466 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
467 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
468 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
469 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
470 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
471 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
472 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
473 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
474 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
475 #endif
476
477
478 /* Definition to allow mutual recursion */
479
480 static BOOL
481 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
482 int *, int *, branch_chain *, compile_data *, int *);
483
484
485
486 /*************************************************
487 * Find an error text *
488 *************************************************/
489
490 /* The error texts are now all in one long string, to save on relocations. As
491 some of the text is of unknown length, we can't use a table of offsets.
492 Instead, just count through the strings. This is not a performance issue
493 because it happens only when there has been a compilation error.
494
495 Argument: the error number
496 Returns: pointer to the error string
497 */
498
499 static const char *
500 find_error_text(int n)
501 {
502 const char *s = error_texts;
503 for (; n > 0; n--) while (*s++ != 0) {};
504 return s;
505 }
506
507
508 /*************************************************
509 * Handle escapes *
510 *************************************************/
511
512 /* This function is called when a \ has been encountered. It either returns a
513 positive value for a simple escape such as \n, or a negative value which
514 encodes one of the more complicated things such as \d. A backreference to group
515 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
516 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
517 ptr is pointing at the \. On exit, it is on the final character of the escape
518 sequence.
519
520 Arguments:
521 ptrptr points to the pattern position pointer
522 errorcodeptr points to the errorcode variable
523 bracount number of previous extracting brackets
524 options the options bits
525 isclass TRUE if inside a character class
526
527 Returns: zero or positive => a data character
528 negative => a special escape sequence
529 on error, errorcodeptr is set
530 */
531
532 static int
533 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
534 int options, BOOL isclass)
535 {
536 BOOL utf8 = (options & PCRE_UTF8) != 0;
537 const uschar *ptr = *ptrptr + 1;
538 int c, i;
539
540 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
541 ptr--; /* Set pointer back to the last byte */
542
543 /* If backslash is at the end of the pattern, it's an error. */
544
545 if (c == 0) *errorcodeptr = ERR1;
546
547 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
548 in a table. A non-zero result is something that can be returned immediately.
549 Otherwise further processing may be required. */
550
551 #ifndef EBCDIC /* ASCII/UTF-8 coding */
552 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
553 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
554
555 #else /* EBCDIC coding */
556 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
557 else if ((i = escapes[c - 0x48]) != 0) c = i;
558 #endif
559
560 /* Escapes that need further processing, or are illegal. */
561
562 else
563 {
564 const uschar *oldptr;
565 BOOL braced, negated;
566
567 switch (c)
568 {
569 /* A number of Perl escapes are not handled by PCRE. We give an explicit
570 error. */
571
572 case CHAR_l:
573 case CHAR_L:
574 case CHAR_N:
575 case CHAR_u:
576 case CHAR_U:
577 *errorcodeptr = ERR37;
578 break;
579
580 /* \g must be followed by one of a number of specific things:
581
582 (1) A number, either plain or braced. If positive, it is an absolute
583 backreference. If negative, it is a relative backreference. This is a Perl
584 5.10 feature.
585
586 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
587 is part of Perl's movement towards a unified syntax for back references. As
588 this is synonymous with \k{name}, we fudge it up by pretending it really
589 was \k.
590
591 (3) For Oniguruma compatibility we also support \g followed by a name or a
592 number either in angle brackets or in single quotes. However, these are
593 (possibly recursive) subroutine calls, _not_ backreferences. Just return
594 the -ESC_g code (cf \k). */
595
596 case CHAR_g:
597 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
598 {
599 c = -ESC_g;
600 break;
601 }
602
603 /* Handle the Perl-compatible cases */
604
605 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
606 {
607 const uschar *p;
608 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
609 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
610 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
611 {
612 c = -ESC_k;
613 break;
614 }
615 braced = TRUE;
616 ptr++;
617 }
618 else braced = FALSE;
619
620 if (ptr[1] == CHAR_MINUS)
621 {
622 negated = TRUE;
623 ptr++;
624 }
625 else negated = FALSE;
626
627 c = 0;
628 while ((digitab[ptr[1]] & ctype_digit) != 0)
629 c = c * 10 + *(++ptr) - CHAR_0;
630
631 if (c < 0) /* Integer overflow */
632 {
633 *errorcodeptr = ERR61;
634 break;
635 }
636
637 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
638 {
639 *errorcodeptr = ERR57;
640 break;
641 }
642
643 if (c == 0)
644 {
645 *errorcodeptr = ERR58;
646 break;
647 }
648
649 if (negated)
650 {
651 if (c > bracount)
652 {
653 *errorcodeptr = ERR15;
654 break;
655 }
656 c = bracount - (c - 1);
657 }
658
659 c = -(ESC_REF + c);
660 break;
661
662 /* The handling of escape sequences consisting of a string of digits
663 starting with one that is not zero is not straightforward. By experiment,
664 the way Perl works seems to be as follows:
665
666 Outside a character class, the digits are read as a decimal number. If the
667 number is less than 10, or if there are that many previous extracting
668 left brackets, then it is a back reference. Otherwise, up to three octal
669 digits are read to form an escaped byte. Thus \123 is likely to be octal
670 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
671 value is greater than 377, the least significant 8 bits are taken. Inside a
672 character class, \ followed by a digit is always an octal number. */
673
674 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
675 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
676
677 if (!isclass)
678 {
679 oldptr = ptr;
680 c -= CHAR_0;
681 while ((digitab[ptr[1]] & ctype_digit) != 0)
682 c = c * 10 + *(++ptr) - CHAR_0;
683 if (c < 0) /* Integer overflow */
684 {
685 *errorcodeptr = ERR61;
686 break;
687 }
688 if (c < 10 || c <= bracount)
689 {
690 c = -(ESC_REF + c);
691 break;
692 }
693 ptr = oldptr; /* Put the pointer back and fall through */
694 }
695
696 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
697 generates a binary zero byte and treats the digit as a following literal.
698 Thus we have to pull back the pointer by one. */
699
700 if ((c = *ptr) >= CHAR_8)
701 {
702 ptr--;
703 c = 0;
704 break;
705 }
706
707 /* \0 always starts an octal number, but we may drop through to here with a
708 larger first octal digit. The original code used just to take the least
709 significant 8 bits of octal numbers (I think this is what early Perls used
710 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
711 than 3 octal digits. */
712
713 case CHAR_0:
714 c -= CHAR_0;
715 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
716 c = c * 8 + *(++ptr) - CHAR_0;
717 if (!utf8 && c > 255) *errorcodeptr = ERR51;
718 break;
719
720 /* \x is complicated. \x{ddd} is a character number which can be greater
721 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
722 treated as a data character. */
723
724 case CHAR_x:
725 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
726 {
727 const uschar *pt = ptr + 2;
728 int count = 0;
729
730 c = 0;
731 while ((digitab[*pt] & ctype_xdigit) != 0)
732 {
733 register int cc = *pt++;
734 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
735 count++;
736
737 #ifndef EBCDIC /* ASCII/UTF-8 coding */
738 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
739 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
740 #else /* EBCDIC coding */
741 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
742 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
743 #endif
744 }
745
746 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
747 {
748 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
749 ptr = pt;
750 break;
751 }
752
753 /* If the sequence of hex digits does not end with '}', then we don't
754 recognize this construct; fall through to the normal \x handling. */
755 }
756
757 /* Read just a single-byte hex-defined char */
758
759 c = 0;
760 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
761 {
762 int cc; /* Some compilers don't like */
763 cc = *(++ptr); /* ++ in initializers */
764 #ifndef EBCDIC /* ASCII/UTF-8 coding */
765 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
766 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
767 #else /* EBCDIC coding */
768 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
769 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
770 #endif
771 }
772 break;
773
774 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
775 This coding is ASCII-specific, but then the whole concept of \cx is
776 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
777
778 case CHAR_c:
779 c = *(++ptr);
780 if (c == 0)
781 {
782 *errorcodeptr = ERR2;
783 break;
784 }
785
786 #ifndef EBCDIC /* ASCII/UTF-8 coding */
787 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
788 c ^= 0x40;
789 #else /* EBCDIC coding */
790 if (c >= CHAR_a && c <= CHAR_z) c += 64;
791 c ^= 0xC0;
792 #endif
793 break;
794
795 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
796 other alphanumeric following \ is an error if PCRE_EXTRA was set;
797 otherwise, for Perl compatibility, it is a literal. This code looks a bit
798 odd, but there used to be some cases other than the default, and there may
799 be again in future, so I haven't "optimized" it. */
800
801 default:
802 if ((options & PCRE_EXTRA) != 0) switch(c)
803 {
804 default:
805 *errorcodeptr = ERR3;
806 break;
807 }
808 break;
809 }
810 }
811
812 *ptrptr = ptr;
813 return c;
814 }
815
816
817
818 #ifdef SUPPORT_UCP
819 /*************************************************
820 * Handle \P and \p *
821 *************************************************/
822
823 /* This function is called after \P or \p has been encountered, provided that
824 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
825 pointing at the P or p. On exit, it is pointing at the final character of the
826 escape sequence.
827
828 Argument:
829 ptrptr points to the pattern position pointer
830 negptr points to a boolean that is set TRUE for negation else FALSE
831 dptr points to an int that is set to the detailed property value
832 errorcodeptr points to the error code variable
833
834 Returns: type value from ucp_type_table, or -1 for an invalid type
835 */
836
837 static int
838 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
839 {
840 int c, i, bot, top;
841 const uschar *ptr = *ptrptr;
842 char name[32];
843
844 c = *(++ptr);
845 if (c == 0) goto ERROR_RETURN;
846
847 *negptr = FALSE;
848
849 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
850 negation. */
851
852 if (c == CHAR_LEFT_CURLY_BRACKET)
853 {
854 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
855 {
856 *negptr = TRUE;
857 ptr++;
858 }
859 for (i = 0; i < (int)sizeof(name) - 1; i++)
860 {
861 c = *(++ptr);
862 if (c == 0) goto ERROR_RETURN;
863 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
864 name[i] = c;
865 }
866 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
867 name[i] = 0;
868 }
869
870 /* Otherwise there is just one following character */
871
872 else
873 {
874 name[0] = c;
875 name[1] = 0;
876 }
877
878 *ptrptr = ptr;
879
880 /* Search for a recognized property name using binary chop */
881
882 bot = 0;
883 top = _pcre_utt_size;
884
885 while (bot < top)
886 {
887 i = (bot + top) >> 1;
888 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
889 if (c == 0)
890 {
891 *dptr = _pcre_utt[i].value;
892 return _pcre_utt[i].type;
893 }
894 if (c > 0) bot = i + 1; else top = i;
895 }
896
897 *errorcodeptr = ERR47;
898 *ptrptr = ptr;
899 return -1;
900
901 ERROR_RETURN:
902 *errorcodeptr = ERR46;
903 *ptrptr = ptr;
904 return -1;
905 }
906 #endif
907
908
909
910
911 /*************************************************
912 * Check for counted repeat *
913 *************************************************/
914
915 /* This function is called when a '{' is encountered in a place where it might
916 start a quantifier. It looks ahead to see if it really is a quantifier or not.
917 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
918 where the ddds are digits.
919
920 Arguments:
921 p pointer to the first char after '{'
922
923 Returns: TRUE or FALSE
924 */
925
926 static BOOL
927 is_counted_repeat(const uschar *p)
928 {
929 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
930 while ((digitab[*p] & ctype_digit) != 0) p++;
931 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
932
933 if (*p++ != CHAR_COMMA) return FALSE;
934 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
935
936 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
937 while ((digitab[*p] & ctype_digit) != 0) p++;
938
939 return (*p == CHAR_RIGHT_CURLY_BRACKET);
940 }
941
942
943
944 /*************************************************
945 * Read repeat counts *
946 *************************************************/
947
948 /* Read an item of the form {n,m} and return the values. This is called only
949 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
950 so the syntax is guaranteed to be correct, but we need to check the values.
951
952 Arguments:
953 p pointer to first char after '{'
954 minp pointer to int for min
955 maxp pointer to int for max
956 returned as -1 if no max
957 errorcodeptr points to error code variable
958
959 Returns: pointer to '}' on success;
960 current ptr on error, with errorcodeptr set non-zero
961 */
962
963 static const uschar *
964 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
965 {
966 int min = 0;
967 int max = -1;
968
969 /* Read the minimum value and do a paranoid check: a negative value indicates
970 an integer overflow. */
971
972 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
973 if (min < 0 || min > 65535)
974 {
975 *errorcodeptr = ERR5;
976 return p;
977 }
978
979 /* Read the maximum value if there is one, and again do a paranoid on its size.
980 Also, max must not be less than min. */
981
982 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
983 {
984 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
985 {
986 max = 0;
987 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
988 if (max < 0 || max > 65535)
989 {
990 *errorcodeptr = ERR5;
991 return p;
992 }
993 if (max < min)
994 {
995 *errorcodeptr = ERR4;
996 return p;
997 }
998 }
999 }
1000
1001 /* Fill in the required variables, and pass back the pointer to the terminating
1002 '}'. */
1003
1004 *minp = min;
1005 *maxp = max;
1006 return p;
1007 }
1008
1009
1010
1011 /*************************************************
1012 * Subroutine for finding forward reference *
1013 *************************************************/
1014
1015 /* This recursive function is called only from find_parens() below. The
1016 top-level call starts at the beginning of the pattern. All other calls must
1017 start at a parenthesis. It scans along a pattern's text looking for capturing
1018 subpatterns, and counting them. If it finds a named pattern that matches the
1019 name it is given, it returns its number. Alternatively, if the name is NULL, it
1020 returns when it reaches a given numbered subpattern. We know that if (?P< is
1021 encountered, the name will be terminated by '>' because that is checked in the
1022 first pass. Recursion is used to keep track of subpatterns that reset the
1023 capturing group numbers - the (?| feature.
1024
1025 Arguments:
1026 ptrptr address of the current character pointer (updated)
1027 cd compile background data
1028 name name to seek, or NULL if seeking a numbered subpattern
1029 lorn name length, or subpattern number if name is NULL
1030 xmode TRUE if we are in /x mode
1031 count pointer to the current capturing subpattern number (updated)
1032
1033 Returns: the number of the named subpattern, or -1 if not found
1034 */
1035
1036 static int
1037 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1038 BOOL xmode, int *count)
1039 {
1040 uschar *ptr = *ptrptr;
1041 int start_count = *count;
1042 int hwm_count = start_count;
1043 BOOL dup_parens = FALSE;
1044
1045 /* If the first character is a parenthesis, check on the type of group we are
1046 dealing with. The very first call may not start with a parenthesis. */
1047
1048 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1049 {
1050 if (ptr[1] == CHAR_QUESTION_MARK &&
1051 ptr[2] == CHAR_VERTICAL_LINE)
1052 {
1053 ptr += 3;
1054 dup_parens = TRUE;
1055 }
1056
1057 /* Handle a normal, unnamed capturing parenthesis */
1058
1059 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1060 {
1061 *count += 1;
1062 if (name == NULL && *count == lorn) return *count;
1063 ptr++;
1064 }
1065
1066 /* Handle a condition. If it is an assertion, just carry on so that it
1067 is processed as normal. If not, skip to the closing parenthesis of the
1068 condition (there can't be any nested parens. */
1069
1070 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1071 {
1072 ptr += 2;
1073 if (ptr[1] != CHAR_QUESTION_MARK)
1074 {
1075 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1076 if (*ptr != 0) ptr++;
1077 }
1078 }
1079
1080 /* We have either (? or (* and not a condition */
1081
1082 else
1083 {
1084 ptr += 2;
1085 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1086
1087 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1088
1089 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1090 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1091 {
1092 int term;
1093 const uschar *thisname;
1094 *count += 1;
1095 if (name == NULL && *count == lorn) return *count;
1096 term = *ptr++;
1097 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1098 thisname = ptr;
1099 while (*ptr != term) ptr++;
1100 if (name != NULL && lorn == ptr - thisname &&
1101 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1102 return *count;
1103 term++;
1104 }
1105 }
1106 }
1107
1108 /* Past any initial parenthesis handling, scan for parentheses or vertical
1109 bars. */
1110
1111 for (; *ptr != 0; ptr++)
1112 {
1113 /* Skip over backslashed characters and also entire \Q...\E */
1114
1115 if (*ptr == CHAR_BACKSLASH)
1116 {
1117 if (*(++ptr) == 0) goto FAIL_EXIT;
1118 if (*ptr == CHAR_Q) for (;;)
1119 {
1120 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1121 if (*ptr == 0) goto FAIL_EXIT;
1122 if (*(++ptr) == CHAR_E) break;
1123 }
1124 continue;
1125 }
1126
1127 /* Skip over character classes; this logic must be similar to the way they
1128 are handled for real. If the first character is '^', skip it. Also, if the
1129 first few characters (either before or after ^) are \Q\E or \E we skip them
1130 too. This makes for compatibility with Perl. Note the use of STR macros to
1131 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1132
1133 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1134 {
1135 BOOL negate_class = FALSE;
1136 for (;;)
1137 {
1138 if (ptr[1] == CHAR_BACKSLASH)
1139 {
1140 if (ptr[2] == CHAR_E)
1141 ptr+= 2;
1142 else if (strncmp((const char *)ptr+2,
1143 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1144 ptr += 4;
1145 else
1146 break;
1147 }
1148 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1149 {
1150 negate_class = TRUE;
1151 ptr++;
1152 }
1153 else break;
1154 }
1155
1156 /* If the next character is ']', it is a data character that must be
1157 skipped, except in JavaScript compatibility mode. */
1158
1159 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1160 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1161 ptr++;
1162
1163 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1164 {
1165 if (*ptr == 0) return -1;
1166 if (*ptr == CHAR_BACKSLASH)
1167 {
1168 if (*(++ptr) == 0) goto FAIL_EXIT;
1169 if (*ptr == CHAR_Q) for (;;)
1170 {
1171 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1172 if (*ptr == 0) goto FAIL_EXIT;
1173 if (*(++ptr) == CHAR_E) break;
1174 }
1175 continue;
1176 }
1177 }
1178 continue;
1179 }
1180
1181 /* Skip comments in /x mode */
1182
1183 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1184 {
1185 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1186 if (*ptr == 0) goto FAIL_EXIT;
1187 continue;
1188 }
1189
1190 /* Check for the special metacharacters */
1191
1192 if (*ptr == CHAR_LEFT_PARENTHESIS)
1193 {
1194 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1195 if (rc > 0) return rc;
1196 if (*ptr == 0) goto FAIL_EXIT;
1197 }
1198
1199 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1200 {
1201 if (dup_parens && *count < hwm_count) *count = hwm_count;
1202 *ptrptr = ptr;
1203 return -1;
1204 }
1205
1206 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1207 {
1208 if (*count > hwm_count) hwm_count = *count;
1209 *count = start_count;
1210 }
1211 }
1212
1213 FAIL_EXIT:
1214 *ptrptr = ptr;
1215 return -1;
1216 }
1217
1218
1219
1220
1221 /*************************************************
1222 * Find forward referenced subpattern *
1223 *************************************************/
1224
1225 /* This function scans along a pattern's text looking for capturing
1226 subpatterns, and counting them. If it finds a named pattern that matches the
1227 name it is given, it returns its number. Alternatively, if the name is NULL, it
1228 returns when it reaches a given numbered subpattern. This is used for forward
1229 references to subpatterns. We used to be able to start this scan from the
1230 current compiling point, using the current count value from cd->bracount, and
1231 do it all in a single loop, but the addition of the possibility of duplicate
1232 subpattern numbers means that we have to scan from the very start, in order to
1233 take account of such duplicates, and to use a recursive function to keep track
1234 of the different types of group.
1235
1236 Arguments:
1237 cd compile background data
1238 name name to seek, or NULL if seeking a numbered subpattern
1239 lorn name length, or subpattern number if name is NULL
1240 xmode TRUE if we are in /x mode
1241
1242 Returns: the number of the found subpattern, or -1 if not found
1243 */
1244
1245 static int
1246 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1247 {
1248 uschar *ptr = (uschar *)cd->start_pattern;
1249 int count = 0;
1250 int rc;
1251
1252 /* If the pattern does not start with an opening parenthesis, the first call
1253 to find_parens_sub() will scan right to the end (if necessary). However, if it
1254 does start with a parenthesis, find_parens_sub() will return when it hits the
1255 matching closing parens. That is why we have to have a loop. */
1256
1257 for (;;)
1258 {
1259 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1260 if (rc > 0 || *ptr++ == 0) break;
1261 }
1262
1263 return rc;
1264 }
1265
1266
1267
1268
1269 /*************************************************
1270 * Find first significant op code *
1271 *************************************************/
1272
1273 /* This is called by several functions that scan a compiled expression looking
1274 for a fixed first character, or an anchoring op code etc. It skips over things
1275 that do not influence this. For some calls, a change of option is important.
1276 For some calls, it makes sense to skip negative forward and all backward
1277 assertions, and also the \b assertion; for others it does not.
1278
1279 Arguments:
1280 code pointer to the start of the group
1281 options pointer to external options
1282 optbit the option bit whose changing is significant, or
1283 zero if none are
1284 skipassert TRUE if certain assertions are to be skipped
1285
1286 Returns: pointer to the first significant opcode
1287 */
1288
1289 static const uschar*
1290 first_significant_code(const uschar *code, int *options, int optbit,
1291 BOOL skipassert)
1292 {
1293 for (;;)
1294 {
1295 switch ((int)*code)
1296 {
1297 case OP_OPT:
1298 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1299 *options = (int)code[1];
1300 code += 2;
1301 break;
1302
1303 case OP_ASSERT_NOT:
1304 case OP_ASSERTBACK:
1305 case OP_ASSERTBACK_NOT:
1306 if (!skipassert) return code;
1307 do code += GET(code, 1); while (*code == OP_ALT);
1308 code += _pcre_OP_lengths[*code];
1309 break;
1310
1311 case OP_WORD_BOUNDARY:
1312 case OP_NOT_WORD_BOUNDARY:
1313 if (!skipassert) return code;
1314 /* Fall through */
1315
1316 case OP_CALLOUT:
1317 case OP_CREF:
1318 case OP_RREF:
1319 case OP_DEF:
1320 code += _pcre_OP_lengths[*code];
1321 break;
1322
1323 default:
1324 return code;
1325 }
1326 }
1327 /* Control never reaches here */
1328 }
1329
1330
1331
1332
1333 /*************************************************
1334 * Find the fixed length of a branch *
1335 *************************************************/
1336
1337 /* Scan a branch and compute the fixed length of subject that will match it,
1338 if the length is fixed. This is needed for dealing with backward assertions.
1339 In UTF8 mode, the result is in characters rather than bytes. The branch is
1340 temporarily terminated with OP_END when this function is called.
1341
1342 This function is called when a backward assertion is encountered, so that if it
1343 fails, the error message can point to the correct place in the pattern.
1344 However, we cannot do this when the assertion contains subroutine calls,
1345 because they can be forward references. We solve this by remembering this case
1346 and doing the check at the end; a flag specifies which mode we are running in.
1347
1348 Arguments:
1349 code points to the start of the pattern (the bracket)
1350 options the compiling options
1351 atend TRUE if called when the pattern is complete
1352 cd the "compile data" structure
1353
1354 Returns: the fixed length,
1355 or -1 if there is no fixed length,
1356 or -2 if \C was encountered
1357 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1358 */
1359
1360 static int
1361 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1362 {
1363 int length = -1;
1364
1365 register int branchlength = 0;
1366 register uschar *cc = code + 1 + LINK_SIZE;
1367
1368 /* Scan along the opcodes for this branch. If we get to the end of the
1369 branch, check the length against that of the other branches. */
1370
1371 for (;;)
1372 {
1373 int d;
1374 uschar *ce, *cs;
1375 register int op = *cc;
1376 switch (op)
1377 {
1378 case OP_CBRA:
1379 case OP_BRA:
1380 case OP_ONCE:
1381 case OP_COND:
1382 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1383 if (d < 0) return d;
1384 branchlength += d;
1385 do cc += GET(cc, 1); while (*cc == OP_ALT);
1386 cc += 1 + LINK_SIZE;
1387 break;
1388
1389 /* Reached end of a branch; if it's a ket it is the end of a nested
1390 call. If it's ALT it is an alternation in a nested call. If it is
1391 END it's the end of the outer call. All can be handled by the same code. */
1392
1393 case OP_ALT:
1394 case OP_KET:
1395 case OP_KETRMAX:
1396 case OP_KETRMIN:
1397 case OP_END:
1398 if (length < 0) length = branchlength;
1399 else if (length != branchlength) return -1;
1400 if (*cc != OP_ALT) return length;
1401 cc += 1 + LINK_SIZE;
1402 branchlength = 0;
1403 break;
1404
1405 /* A true recursion implies not fixed length, but a subroutine call may
1406 be OK. If the subroutine is a forward reference, we can't deal with
1407 it until the end of the pattern, so return -3. */
1408
1409 case OP_RECURSE:
1410 if (!atend) return -3;
1411 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1412 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1413 if (cc > cs && cc < ce) return -1; /* Recursion */
1414 d = find_fixedlength(cs + 2, options, atend, cd);
1415 if (d < 0) return d;
1416 branchlength += d;
1417 cc += 1 + LINK_SIZE;
1418 break;
1419
1420 /* Skip over assertive subpatterns */
1421
1422 case OP_ASSERT:
1423 case OP_ASSERT_NOT:
1424 case OP_ASSERTBACK:
1425 case OP_ASSERTBACK_NOT:
1426 do cc += GET(cc, 1); while (*cc == OP_ALT);
1427 /* Fall through */
1428
1429 /* Skip over things that don't match chars */
1430
1431 case OP_REVERSE:
1432 case OP_CREF:
1433 case OP_RREF:
1434 case OP_DEF:
1435 case OP_OPT:
1436 case OP_CALLOUT:
1437 case OP_SOD:
1438 case OP_SOM:
1439 case OP_EOD:
1440 case OP_EODN:
1441 case OP_CIRC:
1442 case OP_DOLL:
1443 case OP_NOT_WORD_BOUNDARY:
1444 case OP_WORD_BOUNDARY:
1445 cc += _pcre_OP_lengths[*cc];
1446 break;
1447
1448 /* Handle literal characters */
1449
1450 case OP_CHAR:
1451 case OP_CHARNC:
1452 case OP_NOT:
1453 branchlength++;
1454 cc += 2;
1455 #ifdef SUPPORT_UTF8
1456 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1457 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1458 #endif
1459 break;
1460
1461 /* Handle exact repetitions. The count is already in characters, but we
1462 need to skip over a multibyte character in UTF8 mode. */
1463
1464 case OP_EXACT:
1465 branchlength += GET2(cc,1);
1466 cc += 4;
1467 #ifdef SUPPORT_UTF8
1468 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1469 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1470 #endif
1471 break;
1472
1473 case OP_TYPEEXACT:
1474 branchlength += GET2(cc,1);
1475 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1476 cc += 4;
1477 break;
1478
1479 /* Handle single-char matchers */
1480
1481 case OP_PROP:
1482 case OP_NOTPROP:
1483 cc += 2;
1484 /* Fall through */
1485
1486 case OP_NOT_DIGIT:
1487 case OP_DIGIT:
1488 case OP_NOT_WHITESPACE:
1489 case OP_WHITESPACE:
1490 case OP_NOT_WORDCHAR:
1491 case OP_WORDCHAR:
1492 case OP_ANY:
1493 case OP_ALLANY:
1494 branchlength++;
1495 cc++;
1496 break;
1497
1498 /* The single-byte matcher isn't allowed */
1499
1500 case OP_ANYBYTE:
1501 return -2;
1502
1503 /* Check a class for variable quantification */
1504
1505 #ifdef SUPPORT_UTF8
1506 case OP_XCLASS:
1507 cc += GET(cc, 1) - 33;
1508 /* Fall through */
1509 #endif
1510
1511 case OP_CLASS:
1512 case OP_NCLASS:
1513 cc += 33;
1514
1515 switch (*cc)
1516 {
1517 case OP_CRSTAR:
1518 case OP_CRMINSTAR:
1519 case OP_CRQUERY:
1520 case OP_CRMINQUERY:
1521 return -1;
1522
1523 case OP_CRRANGE:
1524 case OP_CRMINRANGE:
1525 if (GET2(cc,1) != GET2(cc,3)) return -1;
1526 branchlength += GET2(cc,1);
1527 cc += 5;
1528 break;
1529
1530 default:
1531 branchlength++;
1532 }
1533 break;
1534
1535 /* Anything else is variable length */
1536
1537 default:
1538 return -1;
1539 }
1540 }
1541 /* Control never gets here */
1542 }
1543
1544
1545
1546
1547 /*************************************************
1548 * Scan compiled regex for specific bracket *
1549 *************************************************/
1550
1551 /* This little function scans through a compiled pattern until it finds a
1552 capturing bracket with the given number, or, if the number is negative, an
1553 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1554 so that it can be called from pcre_study() when finding the minimum matching
1555 length.
1556
1557 Arguments:
1558 code points to start of expression
1559 utf8 TRUE in UTF-8 mode
1560 number the required bracket number or negative to find a lookbehind
1561
1562 Returns: pointer to the opcode for the bracket, or NULL if not found
1563 */
1564
1565 const uschar *
1566 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1567 {
1568 for (;;)
1569 {
1570 register int c = *code;
1571 if (c == OP_END) return NULL;
1572
1573 /* XCLASS is used for classes that cannot be represented just by a bit
1574 map. This includes negated single high-valued characters. The length in
1575 the table is zero; the actual length is stored in the compiled code. */
1576
1577 if (c == OP_XCLASS) code += GET(code, 1);
1578
1579 /* Handle recursion */
1580
1581 else if (c == OP_REVERSE)
1582 {
1583 if (number < 0) return (uschar *)code;
1584 code += _pcre_OP_lengths[c];
1585 }
1586
1587 /* Handle capturing bracket */
1588
1589 else if (c == OP_CBRA)
1590 {
1591 int n = GET2(code, 1+LINK_SIZE);
1592 if (n == number) return (uschar *)code;
1593 code += _pcre_OP_lengths[c];
1594 }
1595
1596 /* Otherwise, we can get the item's length from the table, except that for
1597 repeated character types, we have to test for \p and \P, which have an extra
1598 two bytes of parameters. */
1599
1600 else
1601 {
1602 switch(c)
1603 {
1604 case OP_TYPESTAR:
1605 case OP_TYPEMINSTAR:
1606 case OP_TYPEPLUS:
1607 case OP_TYPEMINPLUS:
1608 case OP_TYPEQUERY:
1609 case OP_TYPEMINQUERY:
1610 case OP_TYPEPOSSTAR:
1611 case OP_TYPEPOSPLUS:
1612 case OP_TYPEPOSQUERY:
1613 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1614 break;
1615
1616 case OP_TYPEUPTO:
1617 case OP_TYPEMINUPTO:
1618 case OP_TYPEEXACT:
1619 case OP_TYPEPOSUPTO:
1620 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1621 break;
1622 }
1623
1624 /* Add in the fixed length from the table */
1625
1626 code += _pcre_OP_lengths[c];
1627
1628 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1629 a multi-byte character. The length in the table is a minimum, so we have to
1630 arrange to skip the extra bytes. */
1631
1632 #ifdef SUPPORT_UTF8
1633 if (utf8) switch(c)
1634 {
1635 case OP_CHAR:
1636 case OP_CHARNC:
1637 case OP_EXACT:
1638 case OP_UPTO:
1639 case OP_MINUPTO:
1640 case OP_POSUPTO:
1641 case OP_STAR:
1642 case OP_MINSTAR:
1643 case OP_POSSTAR:
1644 case OP_PLUS:
1645 case OP_MINPLUS:
1646 case OP_POSPLUS:
1647 case OP_QUERY:
1648 case OP_MINQUERY:
1649 case OP_POSQUERY:
1650 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1651 break;
1652 }
1653 #else
1654 (void)(utf8); /* Keep compiler happy by referencing function argument */
1655 #endif
1656 }
1657 }
1658 }
1659
1660
1661
1662 /*************************************************
1663 * Scan compiled regex for recursion reference *
1664 *************************************************/
1665
1666 /* This little function scans through a compiled pattern until it finds an
1667 instance of OP_RECURSE.
1668
1669 Arguments:
1670 code points to start of expression
1671 utf8 TRUE in UTF-8 mode
1672
1673 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1674 */
1675
1676 static const uschar *
1677 find_recurse(const uschar *code, BOOL utf8)
1678 {
1679 for (;;)
1680 {
1681 register int c = *code;
1682 if (c == OP_END) return NULL;
1683 if (c == OP_RECURSE) return code;
1684
1685 /* XCLASS is used for classes that cannot be represented just by a bit
1686 map. This includes negated single high-valued characters. The length in
1687 the table is zero; the actual length is stored in the compiled code. */
1688
1689 if (c == OP_XCLASS) code += GET(code, 1);
1690
1691 /* Otherwise, we can get the item's length from the table, except that for
1692 repeated character types, we have to test for \p and \P, which have an extra
1693 two bytes of parameters. */
1694
1695 else
1696 {
1697 switch(c)
1698 {
1699 case OP_TYPESTAR:
1700 case OP_TYPEMINSTAR:
1701 case OP_TYPEPLUS:
1702 case OP_TYPEMINPLUS:
1703 case OP_TYPEQUERY:
1704 case OP_TYPEMINQUERY:
1705 case OP_TYPEPOSSTAR:
1706 case OP_TYPEPOSPLUS:
1707 case OP_TYPEPOSQUERY:
1708 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1709 break;
1710
1711 case OP_TYPEPOSUPTO:
1712 case OP_TYPEUPTO:
1713 case OP_TYPEMINUPTO:
1714 case OP_TYPEEXACT:
1715 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1716 break;
1717 }
1718
1719 /* Add in the fixed length from the table */
1720
1721 code += _pcre_OP_lengths[c];
1722
1723 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1724 by a multi-byte character. The length in the table is a minimum, so we have
1725 to arrange to skip the extra bytes. */
1726
1727 #ifdef SUPPORT_UTF8
1728 if (utf8) switch(c)
1729 {
1730 case OP_CHAR:
1731 case OP_CHARNC:
1732 case OP_EXACT:
1733 case OP_UPTO:
1734 case OP_MINUPTO:
1735 case OP_POSUPTO:
1736 case OP_STAR:
1737 case OP_MINSTAR:
1738 case OP_POSSTAR:
1739 case OP_PLUS:
1740 case OP_MINPLUS:
1741 case OP_POSPLUS:
1742 case OP_QUERY:
1743 case OP_MINQUERY:
1744 case OP_POSQUERY:
1745 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1746 break;
1747 }
1748 #else
1749 (void)(utf8); /* Keep compiler happy by referencing function argument */
1750 #endif
1751 }
1752 }
1753 }
1754
1755
1756
1757 /*************************************************
1758 * Scan compiled branch for non-emptiness *
1759 *************************************************/
1760
1761 /* This function scans through a branch of a compiled pattern to see whether it
1762 can match the empty string or not. It is called from could_be_empty()
1763 below and from compile_branch() when checking for an unlimited repeat of a
1764 group that can match nothing. Note that first_significant_code() skips over
1765 backward and negative forward assertions when its final argument is TRUE. If we
1766 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1767 bracket whose current branch will already have been scanned.
1768
1769 Arguments:
1770 code points to start of search
1771 endcode points to where to stop
1772 utf8 TRUE if in UTF8 mode
1773
1774 Returns: TRUE if what is matched could be empty
1775 */
1776
1777 static BOOL
1778 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1779 {
1780 register int c;
1781 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1782 code < endcode;
1783 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1784 {
1785 const uschar *ccode;
1786
1787 c = *code;
1788
1789 /* Skip over forward assertions; the other assertions are skipped by
1790 first_significant_code() with a TRUE final argument. */
1791
1792 if (c == OP_ASSERT)
1793 {
1794 do code += GET(code, 1); while (*code == OP_ALT);
1795 c = *code;
1796 continue;
1797 }
1798
1799 /* Groups with zero repeats can of course be empty; skip them. */
1800
1801 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1802 {
1803 code += _pcre_OP_lengths[c];
1804 do code += GET(code, 1); while (*code == OP_ALT);
1805 c = *code;
1806 continue;
1807 }
1808
1809 /* For other groups, scan the branches. */
1810
1811 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1812 {
1813 BOOL empty_branch;
1814 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1815
1816 /* If a conditional group has only one branch, there is a second, implied,
1817 empty branch, so just skip over the conditional, because it could be empty.
1818 Otherwise, scan the individual branches of the group. */
1819
1820 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1821 code += GET(code, 1);
1822 else
1823 {
1824 empty_branch = FALSE;
1825 do
1826 {
1827 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1828 empty_branch = TRUE;
1829 code += GET(code, 1);
1830 }
1831 while (*code == OP_ALT);
1832 if (!empty_branch) return FALSE; /* All branches are non-empty */
1833 }
1834
1835 c = *code;
1836 continue;
1837 }
1838
1839 /* Handle the other opcodes */
1840
1841 switch (c)
1842 {
1843 /* Check for quantifiers after a class. XCLASS is used for classes that
1844 cannot be represented just by a bit map. This includes negated single
1845 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1846 actual length is stored in the compiled code, so we must update "code"
1847 here. */
1848
1849 #ifdef SUPPORT_UTF8
1850 case OP_XCLASS:
1851 ccode = code += GET(code, 1);
1852 goto CHECK_CLASS_REPEAT;
1853 #endif
1854
1855 case OP_CLASS:
1856 case OP_NCLASS:
1857 ccode = code + 33;
1858
1859 #ifdef SUPPORT_UTF8
1860 CHECK_CLASS_REPEAT:
1861 #endif
1862
1863 switch (*ccode)
1864 {
1865 case OP_CRSTAR: /* These could be empty; continue */
1866 case OP_CRMINSTAR:
1867 case OP_CRQUERY:
1868 case OP_CRMINQUERY:
1869 break;
1870
1871 default: /* Non-repeat => class must match */
1872 case OP_CRPLUS: /* These repeats aren't empty */
1873 case OP_CRMINPLUS:
1874 return FALSE;
1875
1876 case OP_CRRANGE:
1877 case OP_CRMINRANGE:
1878 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1879 break;
1880 }
1881 break;
1882
1883 /* Opcodes that must match a character */
1884
1885 case OP_PROP:
1886 case OP_NOTPROP:
1887 case OP_EXTUNI:
1888 case OP_NOT_DIGIT:
1889 case OP_DIGIT:
1890 case OP_NOT_WHITESPACE:
1891 case OP_WHITESPACE:
1892 case OP_NOT_WORDCHAR:
1893 case OP_WORDCHAR:
1894 case OP_ANY:
1895 case OP_ALLANY:
1896 case OP_ANYBYTE:
1897 case OP_CHAR:
1898 case OP_CHARNC:
1899 case OP_NOT:
1900 case OP_PLUS:
1901 case OP_MINPLUS:
1902 case OP_POSPLUS:
1903 case OP_EXACT:
1904 case OP_NOTPLUS:
1905 case OP_NOTMINPLUS:
1906 case OP_NOTPOSPLUS:
1907 case OP_NOTEXACT:
1908 case OP_TYPEPLUS:
1909 case OP_TYPEMINPLUS:
1910 case OP_TYPEPOSPLUS:
1911 case OP_TYPEEXACT:
1912 return FALSE;
1913
1914 /* These are going to continue, as they may be empty, but we have to
1915 fudge the length for the \p and \P cases. */
1916
1917 case OP_TYPESTAR:
1918 case OP_TYPEMINSTAR:
1919 case OP_TYPEPOSSTAR:
1920 case OP_TYPEQUERY:
1921 case OP_TYPEMINQUERY:
1922 case OP_TYPEPOSQUERY:
1923 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1924 break;
1925
1926 /* Same for these */
1927
1928 case OP_TYPEUPTO:
1929 case OP_TYPEMINUPTO:
1930 case OP_TYPEPOSUPTO:
1931 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1932 break;
1933
1934 /* End of branch */
1935
1936 case OP_KET:
1937 case OP_KETRMAX:
1938 case OP_KETRMIN:
1939 case OP_ALT:
1940 return TRUE;
1941
1942 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1943 MINUPTO, and POSUPTO may be followed by a multibyte character */
1944
1945 #ifdef SUPPORT_UTF8
1946 case OP_STAR:
1947 case OP_MINSTAR:
1948 case OP_POSSTAR:
1949 case OP_QUERY:
1950 case OP_MINQUERY:
1951 case OP_POSQUERY:
1952 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1953 break;
1954
1955 case OP_UPTO:
1956 case OP_MINUPTO:
1957 case OP_POSUPTO:
1958 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1959 break;
1960 #endif
1961 }
1962 }
1963
1964 return TRUE;
1965 }
1966
1967
1968
1969 /*************************************************
1970 * Scan compiled regex for non-emptiness *
1971 *************************************************/
1972
1973 /* This function is called to check for left recursive calls. We want to check
1974 the current branch of the current pattern to see if it could match the empty
1975 string. If it could, we must look outwards for branches at other levels,
1976 stopping when we pass beyond the bracket which is the subject of the recursion.
1977
1978 Arguments:
1979 code points to start of the recursion
1980 endcode points to where to stop (current RECURSE item)
1981 bcptr points to the chain of current (unclosed) branch starts
1982 utf8 TRUE if in UTF-8 mode
1983
1984 Returns: TRUE if what is matched could be empty
1985 */
1986
1987 static BOOL
1988 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1989 BOOL utf8)
1990 {
1991 while (bcptr != NULL && bcptr->current >= code)
1992 {
1993 if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1994 bcptr = bcptr->outer;
1995 }
1996 return TRUE;
1997 }
1998
1999
2000
2001 /*************************************************
2002 * Check for POSIX class syntax *
2003 *************************************************/
2004
2005 /* This function is called when the sequence "[:" or "[." or "[=" is
2006 encountered in a character class. It checks whether this is followed by a
2007 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2008 reach an unescaped ']' without the special preceding character, return FALSE.
2009
2010 Originally, this function only recognized a sequence of letters between the
2011 terminators, but it seems that Perl recognizes any sequence of characters,
2012 though of course unknown POSIX names are subsequently rejected. Perl gives an
2013 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2014 didn't consider this to be a POSIX class. Likewise for [:1234:].
2015
2016 The problem in trying to be exactly like Perl is in the handling of escapes. We
2017 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2018 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2019 below handles the special case of \], but does not try to do any other escape
2020 processing. This makes it different from Perl for cases such as [:l\ower:]
2021 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2022 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2023 I think.
2024
2025 Arguments:
2026 ptr pointer to the initial [
2027 endptr where to return the end pointer
2028
2029 Returns: TRUE or FALSE
2030 */
2031
2032 static BOOL
2033 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2034 {
2035 int terminator; /* Don't combine these lines; the Solaris cc */
2036 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2037 for (++ptr; *ptr != 0; ptr++)
2038 {
2039 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2040 {
2041 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2042 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2043 {
2044 *endptr = ptr;
2045 return TRUE;
2046 }
2047 }
2048 }
2049 return FALSE;
2050 }
2051
2052
2053
2054
2055 /*************************************************
2056 * Check POSIX class name *
2057 *************************************************/
2058
2059 /* This function is called to check the name given in a POSIX-style class entry
2060 such as [:alnum:].
2061
2062 Arguments:
2063 ptr points to the first letter
2064 len the length of the name
2065
2066 Returns: a value representing the name, or -1 if unknown
2067 */
2068
2069 static int
2070 check_posix_name(const uschar *ptr, int len)
2071 {
2072 const char *pn = posix_names;
2073 register int yield = 0;
2074 while (posix_name_lengths[yield] != 0)
2075 {
2076 if (len == posix_name_lengths[yield] &&
2077 strncmp((const char *)ptr, pn, len) == 0) return yield;
2078 pn += posix_name_lengths[yield] + 1;
2079 yield++;
2080 }
2081 return -1;
2082 }
2083
2084
2085 /*************************************************
2086 * Adjust OP_RECURSE items in repeated group *
2087 *************************************************/
2088
2089 /* OP_RECURSE items contain an offset from the start of the regex to the group
2090 that is referenced. This means that groups can be replicated for fixed
2091 repetition simply by copying (because the recursion is allowed to refer to
2092 earlier groups that are outside the current group). However, when a group is
2093 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2094 inserted before it, after it has been compiled. This means that any OP_RECURSE
2095 items within it that refer to the group itself or any contained groups have to
2096 have their offsets adjusted. That one of the jobs of this function. Before it
2097 is called, the partially compiled regex must be temporarily terminated with
2098 OP_END.
2099
2100 This function has been extended with the possibility of forward references for
2101 recursions and subroutine calls. It must also check the list of such references
2102 for the group we are dealing with. If it finds that one of the recursions in
2103 the current group is on this list, it adjusts the offset in the list, not the
2104 value in the reference (which is a group number).
2105
2106 Arguments:
2107 group points to the start of the group
2108 adjust the amount by which the group is to be moved
2109 utf8 TRUE in UTF-8 mode
2110 cd contains pointers to tables etc.
2111 save_hwm the hwm forward reference pointer at the start of the group
2112
2113 Returns: nothing
2114 */
2115
2116 static void
2117 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2118 uschar *save_hwm)
2119 {
2120 uschar *ptr = group;
2121
2122 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2123 {
2124 int offset;
2125 uschar *hc;
2126
2127 /* See if this recursion is on the forward reference list. If so, adjust the
2128 reference. */
2129
2130 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2131 {
2132 offset = GET(hc, 0);
2133 if (cd->start_code + offset == ptr + 1)
2134 {
2135 PUT(hc, 0, offset + adjust);
2136 break;
2137 }
2138 }
2139
2140 /* Otherwise, adjust the recursion offset if it's after the start of this
2141 group. */
2142
2143 if (hc >= cd->hwm)
2144 {
2145 offset = GET(ptr, 1);
2146 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2147 }
2148
2149 ptr += 1 + LINK_SIZE;
2150 }
2151 }
2152
2153
2154
2155 /*************************************************
2156 * Insert an automatic callout point *
2157 *************************************************/
2158
2159 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2160 callout points before each pattern item.
2161
2162 Arguments:
2163 code current code pointer
2164 ptr current pattern pointer
2165 cd pointers to tables etc
2166
2167 Returns: new code pointer
2168 */
2169
2170 static uschar *
2171 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2172 {
2173 *code++ = OP_CALLOUT;
2174 *code++ = 255;
2175 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2176 PUT(code, LINK_SIZE, 0); /* Default length */
2177 return code + 2*LINK_SIZE;
2178 }
2179
2180
2181
2182 /*************************************************
2183 * Complete a callout item *
2184 *************************************************/
2185
2186 /* A callout item contains the length of the next item in the pattern, which
2187 we can't fill in till after we have reached the relevant point. This is used
2188 for both automatic and manual callouts.
2189
2190 Arguments:
2191 previous_callout points to previous callout item
2192 ptr current pattern pointer
2193 cd pointers to tables etc
2194
2195 Returns: nothing
2196 */
2197
2198 static void
2199 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2200 {
2201 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2202 PUT(previous_callout, 2 + LINK_SIZE, length);
2203 }
2204
2205
2206
2207 #ifdef SUPPORT_UCP
2208 /*************************************************
2209 * Get othercase range *
2210 *************************************************/
2211
2212 /* This function is passed the start and end of a class range, in UTF-8 mode
2213 with UCP support. It searches up the characters, looking for internal ranges of
2214 characters in the "other" case. Each call returns the next one, updating the
2215 start address.
2216
2217 Arguments:
2218 cptr points to starting character value; updated
2219 d end value
2220 ocptr where to put start of othercase range
2221 odptr where to put end of othercase range
2222
2223 Yield: TRUE when range returned; FALSE when no more
2224 */
2225
2226 static BOOL
2227 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2228 unsigned int *odptr)
2229 {
2230 unsigned int c, othercase, next;
2231
2232 for (c = *cptr; c <= d; c++)
2233 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2234
2235 if (c > d) return FALSE;
2236
2237 *ocptr = othercase;
2238 next = othercase + 1;
2239
2240 for (++c; c <= d; c++)
2241 {
2242 if (UCD_OTHERCASE(c) != next) break;
2243 next++;
2244 }
2245
2246 *odptr = next - 1;
2247 *cptr = c;
2248
2249 return TRUE;
2250 }
2251 #endif /* SUPPORT_UCP */
2252
2253
2254
2255 /*************************************************
2256 * Check if auto-possessifying is possible *
2257 *************************************************/
2258
2259 /* This function is called for unlimited repeats of certain items, to see
2260 whether the next thing could possibly match the repeated item. If not, it makes
2261 sense to automatically possessify the repeated item.
2262
2263 Arguments:
2264 op_code the repeated op code
2265 this data for this item, depends on the opcode
2266 utf8 TRUE in UTF-8 mode
2267 utf8_char used for utf8 character bytes, NULL if not relevant
2268 ptr next character in pattern
2269 options options bits
2270 cd contains pointers to tables etc.
2271
2272 Returns: TRUE if possessifying is wanted
2273 */
2274
2275 static BOOL
2276 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2277 const uschar *ptr, int options, compile_data *cd)
2278 {
2279 int next;
2280
2281 /* Skip whitespace and comments in extended mode */
2282
2283 if ((options & PCRE_EXTENDED) != 0)
2284 {
2285 for (;;)
2286 {
2287 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2288 if (*ptr == CHAR_NUMBER_SIGN)
2289 {
2290 while (*(++ptr) != 0)
2291 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2292 }
2293 else break;
2294 }
2295 }
2296
2297 /* If the next item is one that we can handle, get its value. A non-negative
2298 value is a character, a negative value is an escape value. */
2299
2300 if (*ptr == CHAR_BACKSLASH)
2301 {
2302 int temperrorcode = 0;
2303 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2304 if (temperrorcode != 0) return FALSE;
2305 ptr++; /* Point after the escape sequence */
2306 }
2307
2308 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2309 {
2310 #ifdef SUPPORT_UTF8
2311 if (utf8) { GETCHARINC(next, ptr); } else
2312 #endif
2313 next = *ptr++;
2314 }
2315
2316 else return FALSE;
2317
2318 /* Skip whitespace and comments in extended mode */
2319
2320 if ((options & PCRE_EXTENDED) != 0)
2321 {
2322 for (;;)
2323 {
2324 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2325 if (*ptr == CHAR_NUMBER_SIGN)
2326 {
2327 while (*(++ptr) != 0)
2328 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2329 }
2330 else break;
2331 }
2332 }
2333
2334 /* If the next thing is itself optional, we have to give up. */
2335
2336 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2337 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2338 return FALSE;
2339
2340 /* Now compare the next item with the previous opcode. If the previous is a
2341 positive single character match, "item" either contains the character or, if
2342 "item" is greater than 127 in utf8 mode, the character's bytes are in
2343 utf8_char. */
2344
2345
2346 /* Handle cases when the next item is a character. */
2347
2348 if (next >= 0) switch(op_code)
2349 {
2350 case OP_CHAR:
2351 #ifdef SUPPORT_UTF8
2352 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2353 #else
2354 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2355 #endif
2356 return item != next;
2357
2358 /* For CHARNC (caseless character) we must check the other case. If we have
2359 Unicode property support, we can use it to test the other case of
2360 high-valued characters. */
2361
2362 case OP_CHARNC:
2363 #ifdef SUPPORT_UTF8
2364 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2365 #endif
2366 if (item == next) return FALSE;
2367 #ifdef SUPPORT_UTF8
2368 if (utf8)
2369 {
2370 unsigned int othercase;
2371 if (next < 128) othercase = cd->fcc[next]; else
2372 #ifdef SUPPORT_UCP
2373 othercase = UCD_OTHERCASE((unsigned int)next);
2374 #else
2375 othercase = NOTACHAR;
2376 #endif
2377 return (unsigned int)item != othercase;
2378 }
2379 else
2380 #endif /* SUPPORT_UTF8 */
2381 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2382
2383 /* For OP_NOT, "item" must be a single-byte character. */
2384
2385 case OP_NOT:
2386 if (item == next) return TRUE;
2387 if ((options & PCRE_CASELESS) == 0) return FALSE;
2388 #ifdef SUPPORT_UTF8
2389 if (utf8)
2390 {
2391 unsigned int othercase;
2392 if (next < 128) othercase = cd->fcc[next]; else
2393 #ifdef SUPPORT_UCP
2394 othercase = UCD_OTHERCASE(next);
2395 #else
2396 othercase = NOTACHAR;
2397 #endif
2398 return (unsigned int)item == othercase;
2399 }
2400 else
2401 #endif /* SUPPORT_UTF8 */
2402 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2403
2404 case OP_DIGIT:
2405 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2406
2407 case OP_NOT_DIGIT:
2408 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2409
2410 case OP_WHITESPACE:
2411 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2412
2413 case OP_NOT_WHITESPACE:
2414 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2415
2416 case OP_WORDCHAR:
2417 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2418
2419 case OP_NOT_WORDCHAR:
2420 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2421
2422 case OP_HSPACE:
2423 case OP_NOT_HSPACE:
2424 switch(next)
2425 {
2426 case 0x09:
2427 case 0x20:
2428 case 0xa0:
2429 case 0x1680:
2430 case 0x180e:
2431 case 0x2000:
2432 case 0x2001:
2433 case 0x2002:
2434 case 0x2003:
2435 case 0x2004:
2436 case 0x2005:
2437 case 0x2006:
2438 case 0x2007:
2439 case 0x2008:
2440 case 0x2009:
2441 case 0x200A:
2442 case 0x202f:
2443 case 0x205f:
2444 case 0x3000:
2445 return op_code != OP_HSPACE;
2446 default:
2447 return op_code == OP_HSPACE;
2448 }
2449
2450 case OP_VSPACE:
2451 case OP_NOT_VSPACE:
2452 switch(next)
2453 {
2454 case 0x0a:
2455 case 0x0b:
2456 case 0x0c:
2457 case 0x0d:
2458 case 0x85:
2459 case 0x2028:
2460 case 0x2029:
2461 return op_code != OP_VSPACE;
2462 default:
2463 return op_code == OP_VSPACE;
2464 }
2465
2466 default:
2467 return FALSE;
2468 }
2469
2470
2471 /* Handle the case when the next item is \d, \s, etc. */
2472
2473 switch(op_code)
2474 {
2475 case OP_CHAR:
2476 case OP_CHARNC:
2477 #ifdef SUPPORT_UTF8
2478 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2479 #endif
2480 switch(-next)
2481 {
2482 case ESC_d:
2483 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2484
2485 case ESC_D:
2486 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2487
2488 case ESC_s:
2489 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2490
2491 case ESC_S:
2492 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2493
2494 case ESC_w:
2495 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2496
2497 case ESC_W:
2498 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2499
2500 case ESC_h:
2501 case ESC_H:
2502 switch(item)
2503 {
2504 case 0x09:
2505 case 0x20:
2506 case 0xa0:
2507 case 0x1680:
2508 case 0x180e:
2509 case 0x2000:
2510 case 0x2001:
2511 case 0x2002:
2512 case 0x2003:
2513 case 0x2004:
2514 case 0x2005:
2515 case 0x2006:
2516 case 0x2007:
2517 case 0x2008:
2518 case 0x2009:
2519 case 0x200A:
2520 case 0x202f:
2521 case 0x205f:
2522 case 0x3000:
2523 return -next != ESC_h;
2524 default:
2525 return -next == ESC_h;
2526 }
2527
2528 case ESC_v:
2529 case ESC_V:
2530 switch(item)
2531 {
2532 case 0x0a:
2533 case 0x0b:
2534 case 0x0c:
2535 case 0x0d:
2536 case 0x85:
2537 case 0x2028:
2538 case 0x2029:
2539 return -next != ESC_v;
2540 default:
2541 return -next == ESC_v;
2542 }
2543
2544 default:
2545 return FALSE;
2546 }
2547
2548 case OP_DIGIT:
2549 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2550 next == -ESC_h || next == -ESC_v;
2551
2552 case OP_NOT_DIGIT:
2553 return next == -ESC_d;
2554
2555 case OP_WHITESPACE:
2556 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2557
2558 case OP_NOT_WHITESPACE:
2559 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2560
2561 case OP_HSPACE:
2562 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2563
2564 case OP_NOT_HSPACE:
2565 return next == -ESC_h;
2566
2567 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2568 case OP_VSPACE:
2569 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2570
2571 case OP_NOT_VSPACE:
2572 return next == -ESC_v;
2573
2574 case OP_WORDCHAR:
2575 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2576
2577 case OP_NOT_WORDCHAR:
2578 return next == -ESC_w || next == -ESC_d;
2579
2580 default:
2581 return FALSE;
2582 }
2583
2584 /* Control does not reach here */
2585 }
2586
2587
2588
2589 /*************************************************
2590 * Compile one branch *
2591 *************************************************/
2592
2593 /* Scan the pattern, compiling it into the a vector. If the options are
2594 changed during the branch, the pointer is used to change the external options
2595 bits. This function is used during the pre-compile phase when we are trying
2596 to find out the amount of memory needed, as well as during the real compile
2597 phase. The value of lengthptr distinguishes the two phases.
2598
2599 Arguments:
2600 optionsptr pointer to the option bits
2601 codeptr points to the pointer to the current code point
2602 ptrptr points to the current pattern pointer
2603 errorcodeptr points to error code variable
2604 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2605 reqbyteptr set to the last literal character required, else < 0
2606 bcptr points to current branch chain
2607 cd contains pointers to tables etc.
2608 lengthptr NULL during the real compile phase
2609 points to length accumulator during pre-compile phase
2610
2611 Returns: TRUE on success
2612 FALSE, with *errorcodeptr set non-zero on error
2613 */
2614
2615 static BOOL
2616 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2617 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2618 compile_data *cd, int *lengthptr)
2619 {
2620 int repeat_type, op_type;
2621 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2622 int bravalue = 0;
2623 int greedy_default, greedy_non_default;
2624 int firstbyte, reqbyte;
2625 int zeroreqbyte, zerofirstbyte;
2626 int req_caseopt, reqvary, tempreqvary;
2627 int options = *optionsptr;
2628 int after_manual_callout = 0;
2629 int length_prevgroup = 0;
2630 register int c;
2631 register uschar *code = *codeptr;
2632 uschar *last_code = code;
2633 uschar *orig_code = code;
2634 uschar *tempcode;
2635 BOOL inescq = FALSE;
2636 BOOL groupsetfirstbyte = FALSE;
2637 const uschar *ptr = *ptrptr;
2638 const uschar *tempptr;
2639 uschar *previous = NULL;
2640 uschar *previous_callout = NULL;
2641 uschar *save_hwm = NULL;
2642 uschar classbits[32];
2643
2644 #ifdef SUPPORT_UTF8
2645 BOOL class_utf8;
2646 BOOL utf8 = (options & PCRE_UTF8) != 0;
2647 uschar *class_utf8data;
2648 uschar *class_utf8data_base;
2649 uschar utf8_char[6];
2650 #else
2651 BOOL utf8 = FALSE;
2652 uschar *utf8_char = NULL;
2653 #endif
2654
2655 #ifdef DEBUG
2656 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2657 #endif
2658
2659 /* Set up the default and non-default settings for greediness */
2660
2661 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2662 greedy_non_default = greedy_default ^ 1;
2663
2664 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2665 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2666 matches a non-fixed char first char; reqbyte just remains unset if we never
2667 find one.
2668
2669 When we hit a repeat whose minimum is zero, we may have to adjust these values
2670 to take the zero repeat into account. This is implemented by setting them to
2671 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2672 item types that can be repeated set these backoff variables appropriately. */
2673
2674 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2675
2676 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2677 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2678 value > 255. It is added into the firstbyte or reqbyte variables to record the
2679 case status of the value. This is used only for ASCII characters. */
2680
2681 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2682
2683 /* Switch on next character until the end of the branch */
2684
2685 for (;; ptr++)
2686 {
2687 BOOL negate_class;
2688 BOOL should_flip_negation;
2689 BOOL possessive_quantifier;
2690 BOOL is_quantifier;
2691 BOOL is_recurse;
2692 BOOL reset_bracount;
2693 int class_charcount;
2694 int class_lastchar;
2695 int newoptions;
2696 int recno;
2697 int refsign;
2698 int skipbytes;
2699 int subreqbyte;
2700 int subfirstbyte;
2701 int terminator;
2702 int mclength;
2703 uschar mcbuffer[8];
2704
2705 /* Get next byte in the pattern */
2706
2707 c = *ptr;
2708
2709 /* If we are in the pre-compile phase, accumulate the length used for the
2710 previous cycle of this loop. */
2711
2712 if (lengthptr != NULL)
2713 {
2714 #ifdef DEBUG
2715 if (code > cd->hwm) cd->hwm = code; /* High water info */
2716 #endif
2717 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2718 {
2719 *errorcodeptr = ERR52;
2720 goto FAILED;
2721 }
2722
2723 /* There is at least one situation where code goes backwards: this is the
2724 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2725 the class is simply eliminated. However, it is created first, so we have to
2726 allow memory for it. Therefore, don't ever reduce the length at this point.
2727 */
2728
2729 if (code < last_code) code = last_code;
2730
2731 /* Paranoid check for integer overflow */
2732
2733 if (OFLOW_MAX - *lengthptr < code - last_code)
2734 {
2735 *errorcodeptr = ERR20;
2736 goto FAILED;
2737 }
2738
2739 *lengthptr += code - last_code;
2740 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2741
2742 /* If "previous" is set and it is not at the start of the work space, move
2743 it back to there, in order to avoid filling up the work space. Otherwise,
2744 if "previous" is NULL, reset the current code pointer to the start. */
2745
2746 if (previous != NULL)
2747 {
2748 if (previous > orig_code)
2749 {
2750 memmove(orig_code, previous, code - previous);
2751 code -= previous - orig_code;
2752 previous = orig_code;
2753 }
2754 }
2755 else code = orig_code;
2756
2757 /* Remember where this code item starts so we can pick up the length
2758 next time round. */
2759
2760 last_code = code;
2761 }
2762
2763 /* In the real compile phase, just check the workspace used by the forward
2764 reference list. */
2765
2766 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2767 {
2768 *errorcodeptr = ERR52;
2769 goto FAILED;
2770 }
2771
2772 /* If in \Q...\E, check for the end; if not, we have a literal */
2773
2774 if (inescq && c != 0)
2775 {
2776 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2777 {
2778 inescq = FALSE;
2779 ptr++;
2780 continue;
2781 }
2782 else
2783 {
2784 if (previous_callout != NULL)
2785 {
2786 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2787 complete_callout(previous_callout, ptr, cd);
2788 previous_callout = NULL;
2789 }
2790 if ((options & PCRE_AUTO_CALLOUT) != 0)
2791 {
2792 previous_callout = code;
2793 code = auto_callout(code, ptr, cd);
2794 }
2795 goto NORMAL_CHAR;
2796 }
2797 }
2798
2799 /* Fill in length of a previous callout, except when the next thing is
2800 a quantifier. */
2801
2802 is_quantifier =
2803 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2804 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2805
2806 if (!is_quantifier && previous_callout != NULL &&
2807 after_manual_callout-- <= 0)
2808 {
2809 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2810 complete_callout(previous_callout, ptr, cd);
2811 previous_callout = NULL;
2812 }
2813
2814 /* In extended mode, skip white space and comments */
2815
2816 if ((options & PCRE_EXTENDED) != 0)
2817 {
2818 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2819 if (c == CHAR_NUMBER_SIGN)
2820 {
2821 while (*(++ptr) != 0)
2822 {
2823 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2824 }
2825 if (*ptr != 0) continue;
2826
2827 /* Else fall through to handle end of string */
2828 c = 0;
2829 }
2830 }
2831
2832 /* No auto callout for quantifiers. */
2833
2834 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2835 {
2836 previous_callout = code;
2837 code = auto_callout(code, ptr, cd);
2838 }
2839
2840 switch(c)
2841 {
2842 /* ===================================================================*/
2843 case 0: /* The branch terminates at string end */
2844 case CHAR_VERTICAL_LINE: /* or | or ) */
2845 case CHAR_RIGHT_PARENTHESIS:
2846 *firstbyteptr = firstbyte;
2847 *reqbyteptr = reqbyte;
2848 *codeptr = code;
2849 *ptrptr = ptr;
2850 if (lengthptr != NULL)
2851 {
2852 if (OFLOW_MAX - *lengthptr < code - last_code)
2853 {
2854 *errorcodeptr = ERR20;
2855 goto FAILED;
2856 }
2857 *lengthptr += code - last_code; /* To include callout length */
2858 DPRINTF((">> end branch\n"));
2859 }
2860 return TRUE;
2861
2862
2863 /* ===================================================================*/
2864 /* Handle single-character metacharacters. In multiline mode, ^ disables
2865 the setting of any following char as a first character. */
2866
2867 case CHAR_CIRCUMFLEX_ACCENT:
2868 if ((options & PCRE_MULTILINE) != 0)
2869 {
2870 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2871 }
2872 previous = NULL;
2873 *code++ = OP_CIRC;
2874 break;
2875
2876 case CHAR_DOLLAR_SIGN:
2877 previous = NULL;
2878 *code++ = OP_DOLL;
2879 break;
2880
2881 /* There can never be a first char if '.' is first, whatever happens about
2882 repeats. The value of reqbyte doesn't change either. */
2883
2884 case CHAR_DOT:
2885 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2886 zerofirstbyte = firstbyte;
2887 zeroreqbyte = reqbyte;
2888 previous = code;
2889 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2890 break;
2891
2892
2893 /* ===================================================================*/
2894 /* Character classes. If the included characters are all < 256, we build a
2895 32-byte bitmap of the permitted characters, except in the special case
2896 where there is only one such character. For negated classes, we build the
2897 map as usual, then invert it at the end. However, we use a different opcode
2898 so that data characters > 255 can be handled correctly.
2899
2900 If the class contains characters outside the 0-255 range, a different
2901 opcode is compiled. It may optionally have a bit map for characters < 256,
2902 but those above are are explicitly listed afterwards. A flag byte tells
2903 whether the bitmap is present, and whether this is a negated class or not.
2904
2905 In JavaScript compatibility mode, an isolated ']' causes an error. In
2906 default (Perl) mode, it is treated as a data character. */
2907
2908 case CHAR_RIGHT_SQUARE_BRACKET:
2909 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2910 {
2911 *errorcodeptr = ERR64;
2912 goto FAILED;
2913 }
2914 goto NORMAL_CHAR;
2915
2916 case CHAR_LEFT_SQUARE_BRACKET:
2917 previous = code;
2918
2919 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2920 they are encountered at the top level, so we'll do that too. */
2921
2922 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2923 ptr[1] == CHAR_EQUALS_SIGN) &&
2924 check_posix_syntax(ptr, &tempptr))
2925 {
2926 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2927 goto FAILED;
2928 }
2929
2930 /* If the first character is '^', set the negation flag and skip it. Also,
2931 if the first few characters (either before or after ^) are \Q\E or \E we
2932 skip them too. This makes for compatibility with Perl. */
2933
2934 negate_class = FALSE;
2935 for (;;)
2936 {
2937 c = *(++ptr);
2938 if (c == CHAR_BACKSLASH)
2939 {
2940 if (ptr[1] == CHAR_E)
2941 ptr++;
2942 else if (strncmp((const char *)ptr+1,
2943 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2944 ptr += 3;
2945 else
2946 break;
2947 }
2948 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2949 negate_class = TRUE;
2950 else break;
2951 }
2952
2953 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2954 an initial ']' is taken as a data character -- the code below handles
2955 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2956 [^] must match any character, so generate OP_ALLANY. */
2957
2958 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2959 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2960 {
2961 *code++ = negate_class? OP_ALLANY : OP_FAIL;
2962 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2963 zerofirstbyte = firstbyte;
2964 break;
2965 }
2966
2967 /* If a class contains a negative special such as \S, we need to flip the
2968 negation flag at the end, so that support for characters > 255 works
2969 correctly (they are all included in the class). */
2970
2971 should_flip_negation = FALSE;
2972
2973 /* Keep a count of chars with values < 256 so that we can optimize the case
2974 of just a single character (as long as it's < 256). However, For higher
2975 valued UTF-8 characters, we don't yet do any optimization. */
2976
2977 class_charcount = 0;
2978 class_lastchar = -1;
2979
2980 /* Initialize the 32-char bit map to all zeros. We build the map in a
2981 temporary bit of memory, in case the class contains only 1 character (less
2982 than 256), because in that case the compiled code doesn't use the bit map.
2983 */
2984
2985 memset(classbits, 0, 32 * sizeof(uschar));
2986
2987 #ifdef SUPPORT_UTF8
2988 class_utf8 = FALSE; /* No chars >= 256 */
2989 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2990 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2991 #endif
2992
2993 /* Process characters until ] is reached. By writing this as a "do" it
2994 means that an initial ] is taken as a data character. At the start of the
2995 loop, c contains the first byte of the character. */
2996
2997 if (c != 0) do
2998 {
2999 const uschar *oldptr;
3000
3001 #ifdef SUPPORT_UTF8
3002 if (utf8 && c > 127)
3003 { /* Braces are required because the */
3004 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3005 }
3006
3007 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3008 data and reset the pointer. This is so that very large classes that
3009 contain a zillion UTF-8 characters no longer overwrite the work space
3010 (which is on the stack). */
3011
3012 if (lengthptr != NULL)
3013 {
3014 *lengthptr += class_utf8data - class_utf8data_base;
3015 class_utf8data = class_utf8data_base;
3016 }
3017
3018 #endif
3019
3020 /* Inside \Q...\E everything is literal except \E */
3021
3022 if (inescq)
3023 {
3024 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3025 {
3026 inescq = FALSE; /* Reset literal state */
3027 ptr++; /* Skip the 'E' */
3028 continue; /* Carry on with next */
3029 }
3030 goto CHECK_RANGE; /* Could be range if \E follows */
3031 }
3032
3033 /* Handle POSIX class names. Perl allows a negation extension of the
3034 form [:^name:]. A square bracket that doesn't match the syntax is
3035 treated as a literal. We also recognize the POSIX constructions
3036 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3037 5.6 and 5.8 do. */
3038
3039 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3040 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3041 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3042 {
3043 BOOL local_negate = FALSE;
3044 int posix_class, taboffset, tabopt;
3045 register const uschar *cbits = cd->cbits;
3046 uschar pbits[32];
3047
3048 if (ptr[1] != CHAR_COLON)
3049 {
3050 *errorcodeptr = ERR31;
3051 goto FAILED;
3052 }
3053
3054 ptr += 2;
3055 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3056 {
3057 local_negate = TRUE;
3058 should_flip_negation = TRUE; /* Note negative special */
3059 ptr++;
3060 }
3061
3062 posix_class = check_posix_name(ptr, tempptr - ptr);
3063 if (posix_class < 0)
3064 {
3065 *errorcodeptr = ERR30;
3066 goto FAILED;
3067 }
3068
3069 /* If matching is caseless, upper and lower are converted to
3070 alpha. This relies on the fact that the class table starts with
3071 alpha, lower, upper as the first 3 entries. */
3072
3073 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3074 posix_class = 0;
3075
3076 /* We build the bit map for the POSIX class in a chunk of local store
3077 because we may be adding and subtracting from it, and we don't want to
3078 subtract bits that may be in the main map already. At the end we or the
3079 result into the bit map that is being built. */
3080
3081 posix_class *= 3;
3082
3083 /* Copy in the first table (always present) */
3084
3085 memcpy(pbits, cbits + posix_class_maps[posix_class],
3086 32 * sizeof(uschar));
3087
3088 /* If there is a second table, add or remove it as required. */
3089
3090 taboffset = posix_class_maps[posix_class + 1];
3091 tabopt = posix_class_maps[posix_class + 2];
3092
3093 if (taboffset >= 0)
3094 {
3095 if (tabopt >= 0)
3096 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3097 else
3098 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3099 }
3100
3101 /* Not see if we need to remove any special characters. An option
3102 value of 1 removes vertical space and 2 removes underscore. */
3103
3104 if (tabopt < 0) tabopt = -tabopt;
3105 if (tabopt == 1) pbits[1] &= ~0x3c;
3106 else if (tabopt == 2) pbits[11] &= 0x7f;
3107
3108 /* Add the POSIX table or its complement into the main table that is
3109 being built and we are done. */
3110
3111 if (local_negate)
3112 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3113 else
3114 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3115
3116 ptr = tempptr + 1;
3117 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3118 continue; /* End of POSIX syntax handling */
3119 }
3120
3121 /* Backslash may introduce a single character, or it may introduce one
3122 of the specials, which just set a flag. The sequence \b is a special
3123 case. Inside a class (and only there) it is treated as backspace.
3124 Elsewhere it marks a word boundary. Other escapes have preset maps ready
3125 to 'or' into the one we are building. We assume they have more than one
3126 character in them, so set class_charcount bigger than one. */
3127
3128 if (c == CHAR_BACKSLASH)
3129 {
3130 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3131 if (*errorcodeptr != 0) goto FAILED;
3132
3133 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3134 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3135 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3136 else if (-c == ESC_Q) /* Handle start of quoted string */
3137 {
3138 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3139 {
3140 ptr += 2; /* avoid empty string */
3141 }
3142 else inescq = TRUE;
3143 continue;
3144 }
3145 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3146
3147 if (c < 0)
3148 {
3149 register const uschar *cbits = cd->cbits;
3150 class_charcount += 2; /* Greater than 1 is what matters */
3151
3152 /* Save time by not doing this in the pre-compile phase. */
3153
3154 if (lengthptr == NULL) switch (-c)
3155 {
3156 case ESC_d:
3157 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3158 continue;
3159
3160 case ESC_D:
3161 should_flip_negation = TRUE;
3162 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3163 continue;
3164
3165 case ESC_w:
3166 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3167 continue;
3168
3169 case ESC_W:
3170 should_flip_negation = TRUE;
3171 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3172 continue;
3173
3174 case ESC_s:
3175 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3176 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3177 continue;
3178
3179 case ESC_S:
3180 should_flip_negation = TRUE;
3181 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3182 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3183 continue;
3184
3185 default: /* Not recognized; fall through */
3186 break; /* Need "default" setting to stop compiler warning. */
3187 }
3188
3189 /* In the pre-compile phase, just do the recognition. */
3190
3191 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3192 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3193
3194 /* We need to deal with \H, \h, \V, and \v in both phases because
3195 they use extra memory. */
3196
3197 if (-c == ESC_h)
3198 {
3199 SETBIT(classbits, 0x09); /* VT */
3200 SETBIT(classbits, 0x20); /* SPACE */
3201 SETBIT(classbits, 0xa0); /* NSBP */
3202 #ifdef SUPPORT_UTF8
3203 if (utf8)
3204 {
3205 class_utf8 = TRUE;
3206 *class_utf8data++ = XCL_SINGLE;
3207 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3208 *class_utf8data++ = XCL_SINGLE;
3209 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3210 *class_utf8data++ = XCL_RANGE;
3211 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3212 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3213 *class_utf8data++ = XCL_SINGLE;
3214 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3215 *class_utf8data++ = XCL_SINGLE;
3216 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3217 *class_utf8data++ = XCL_SINGLE;
3218 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3219 }
3220 #endif
3221 continue;
3222 }
3223
3224 if (-c == ESC_H)
3225 {
3226 for (c = 0; c < 32; c++)
3227 {
3228 int x = 0xff;
3229 switch (c)
3230 {
3231 case 0x09/8: x ^= 1 << (0x09%8); break;
3232 case 0x20/8: x ^= 1 << (0x20%8); break;
3233 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3234 default: break;
3235 }
3236 classbits[c] |= x;
3237 }
3238
3239 #ifdef SUPPORT_UTF8
3240 if (utf8)
3241 {
3242 class_utf8 = TRUE;
3243 *class_utf8data++ = XCL_RANGE;
3244 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3245 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3246 *class_utf8data++ = XCL_RANGE;
3247 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3248 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3249 *class_utf8data++ = XCL_RANGE;
3250 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3251 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3252 *class_utf8data++ = XCL_RANGE;
3253 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3254 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3255 *class_utf8data++ = XCL_RANGE;
3256 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3257 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3258 *class_utf8data++ = XCL_RANGE;
3259 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3260 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3261 *class_utf8data++ = XCL_RANGE;
3262 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3263 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3264 }
3265 #endif
3266 continue;
3267 }
3268
3269 if (-c == ESC_v)
3270 {
3271 SETBIT(classbits, 0x0a); /* LF */
3272 SETBIT(classbits, 0x0b); /* VT */
3273 SETBIT(classbits, 0x0c); /* FF */
3274 SETBIT(classbits, 0x0d); /* CR */
3275 SETBIT(classbits, 0x85); /* NEL */
3276 #ifdef SUPPORT_UTF8
3277 if (utf8)
3278 {
3279 class_utf8 = TRUE;
3280 *class_utf8data++ = XCL_RANGE;
3281 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3282 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3283 }
3284 #endif
3285 continue;
3286 }
3287
3288 if (-c == ESC_V)
3289 {
3290 for (c = 0; c < 32; c++)
3291 {
3292 int x = 0xff;
3293 switch (c)
3294 {
3295 case 0x0a/8: x ^= 1 << (0x0a%8);
3296 x ^= 1 << (0x0b%8);
3297 x ^= 1 << (0x0c%8);
3298 x ^= 1 << (0x0d%8);
3299 break;
3300 case 0x85/8: x ^= 1 << (0x85%8); break;
3301 default: break;
3302 }
3303 classbits[c] |= x;
3304 }
3305
3306 #ifdef SUPPORT_UTF8
3307 if (utf8)
3308 {
3309 class_utf8 = TRUE;
3310 *class_utf8data++ = XCL_RANGE;
3311 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3312 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3313 *class_utf8data++ = XCL_RANGE;
3314 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3315 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3316 }
3317 #endif
3318 continue;
3319 }
3320
3321 /* We need to deal with \P and \p in both phases. */
3322
3323 #ifdef SUPPORT_UCP
3324 if (-c == ESC_p || -c == ESC_P)
3325 {
3326 BOOL negated;
3327 int pdata;
3328 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3329 if (ptype < 0) goto FAILED;
3330 class_utf8 = TRUE;
3331 *class_utf8data++ = ((-c == ESC_p) != negated)?
3332 XCL_PROP : XCL_NOTPROP;
3333 *class_utf8data++ = ptype;
3334 *class_utf8data++ = pdata;
3335 class_charcount -= 2; /* Not a < 256 character */
3336 continue;
3337 }
3338 #endif
3339 /* Unrecognized escapes are faulted if PCRE is running in its
3340 strict mode. By default, for compatibility with Perl, they are
3341 treated as literals. */
3342
3343 if ((options & PCRE_EXTRA) != 0)
3344 {
3345 *errorcodeptr = ERR7;
3346 goto FAILED;
3347 }
3348
3349 class_charcount -= 2; /* Undo the default count from above */
3350 c = *ptr; /* Get the final character and fall through */
3351 }
3352
3353 /* Fall through if we have a single character (c >= 0). This may be
3354 greater than 256 in UTF-8 mode. */
3355
3356 } /* End of backslash handling */
3357
3358 /* A single character may be followed by '-' to form a range. However,
3359 Perl does not permit ']' to be the end of the range. A '-' character
3360 at the end is treated as a literal. Perl ignores orphaned \E sequences
3361 entirely. The code for handling \Q and \E is messy. */
3362
3363 CHECK_RANGE:
3364 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3365 {
3366 inescq = FALSE;
3367 ptr += 2;
3368 }
3369
3370 oldptr = ptr;
3371
3372 /* Remember \r or \n */
3373
3374 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3375
3376 /* Check for range */
3377
3378 if (!inescq && ptr[1] == CHAR_MINUS)
3379 {
3380 int d;
3381 ptr += 2;
3382 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3383
3384 /* If we hit \Q (not followed by \E) at this point, go into escaped
3385 mode. */
3386
3387 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3388 {
3389 ptr += 2;
3390 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3391 { ptr += 2; continue; }
3392 inescq = TRUE;
3393 break;
3394 }
3395
3396 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3397 {
3398 ptr = oldptr;
3399 goto LONE_SINGLE_CHARACTER;
3400 }
3401
3402 #ifdef SUPPORT_UTF8
3403 if (utf8)
3404 { /* Braces are required because the */
3405 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3406 }
3407 else
3408 #endif
3409 d = *ptr; /* Not UTF-8 mode */
3410
3411 /* The second part of a range can be a single-character escape, but
3412 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3413 in such circumstances. */
3414
3415 if (!inescq && d == CHAR_BACKSLASH)
3416 {
3417 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3418 if (*errorcodeptr != 0) goto FAILED;
3419
3420 /* \b is backspace; \X is literal X; \R is literal R; any other
3421 special means the '-' was literal */
3422
3423 if (d < 0)
3424 {
3425 if (d == -ESC_b) d = CHAR_BS;
3426 else if (d == -ESC_X) d = CHAR_X;
3427 else if (d == -ESC_R) d = CHAR_R; else
3428 {
3429 ptr = oldptr;
3430 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3431 }
3432 }
3433 }
3434
3435 /* Check that the two values are in the correct order. Optimize
3436 one-character ranges */
3437
3438 if (d < c)
3439 {
3440 *errorcodeptr = ERR8;
3441 goto FAILED;
3442 }
3443
3444 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3445
3446 /* Remember \r or \n */
3447
3448 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3449
3450 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3451 matching, we have to use an XCLASS with extra data items. Caseless
3452 matching for characters > 127 is available only if UCP support is
3453 available. */
3454
3455 #ifdef SUPPORT_UTF8
3456 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3457 {
3458 class_utf8 = TRUE;
3459
3460 /* With UCP support, we can find the other case equivalents of
3461 the relevant characters. There may be several ranges. Optimize how
3462 they fit with the basic range. */
3463
3464 #ifdef SUPPORT_UCP
3465 if ((options & PCRE_CASELESS) != 0)
3466 {
3467 unsigned int occ, ocd;
3468 unsigned int cc = c;
3469 unsigned int origd = d;
3470 while (get_othercase_range(&cc, origd, &occ, &ocd))
3471 {
3472 if (occ >= (unsigned int)c &&
3473 ocd <= (unsigned int)d)
3474 continue; /* Skip embedded ranges */
3475
3476 if (occ < (unsigned int)c &&
3477 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3478 { /* if there is overlap, */
3479 c = occ; /* noting that if occ < c */
3480 continue; /* we can't have ocd > d */
3481 } /* because a subrange is */
3482 if (ocd > (unsigned int)d &&
3483 occ <= (unsigned int)d + 1) /* always shorter than */
3484 { /* the basic range. */
3485 d = ocd;
3486 continue;
3487 }
3488
3489 if (occ == ocd)
3490 {
3491 *class_utf8data++ = XCL_SINGLE;
3492 }
3493 else
3494 {
3495 *class_utf8data++ = XCL_RANGE;
3496 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3497 }
3498 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3499 }
3500 }
3501 #endif /* SUPPORT_UCP */
3502
3503 /* Now record the original range, possibly modified for UCP caseless
3504 overlapping ranges. */
3505
3506 *class_utf8data++ = XCL_RANGE;
3507 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3508 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3509
3510 /* With UCP support, we are done. Without UCP support, there is no
3511 caseless matching for UTF-8 characters > 127; we can use the bit map
3512 for the smaller ones. */
3513
3514 #ifdef SUPPORT_UCP
3515 continue; /* With next character in the class */
3516 #else
3517 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3518
3519 /* Adjust upper limit and fall through to set up the map */
3520
3521 d = 127;
3522
3523 #endif /* SUPPORT_UCP */
3524 }
3525 #endif /* SUPPORT_UTF8 */
3526
3527 /* We use the bit map for all cases when not in UTF-8 mode; else
3528 ranges that lie entirely within 0-127 when there is UCP support; else
3529 for partial ranges without UCP support. */
3530
3531 class_charcount += d - c + 1;
3532 class_lastchar = d;
3533
3534 /* We can save a bit of time by skipping this in the pre-compile. */
3535
3536 if (lengthptr == NULL) for (; c <= d; c++)
3537 {
3538 classbits[c/8] |= (1 << (c&7));
3539 if ((options & PCRE_CASELESS) != 0)
3540 {
3541 int uc = cd->fcc[c]; /* flip case */
3542 classbits[uc/8] |= (1 << (uc&7));
3543 }
3544 }
3545
3546 continue; /* Go get the next char in the class */
3547 }
3548
3549 /* Handle a lone single character - we can get here for a normal
3550 non-escape char, or after \ that introduces a single character or for an
3551 apparent range that isn't. */
3552
3553 LONE_SINGLE_CHARACTER:
3554
3555 /* Handle a character that cannot go in the bit map */
3556
3557 #ifdef SUPPORT_UTF8
3558 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3559 {
3560 class_utf8 = TRUE;
3561 *class_utf8data++ = XCL_SINGLE;
3562 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3563
3564 #ifdef SUPPORT_UCP
3565 if ((options & PCRE_CASELESS) != 0)
3566 {
3567 unsigned int othercase;
3568 if ((othercase = UCD_OTHERCASE(c)) != c)
3569 {
3570 *class_utf8data++ = XCL_SINGLE;
3571 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3572 }
3573 }
3574 #endif /* SUPPORT_UCP */
3575
3576 }
3577 else
3578 #endif /* SUPPORT_UTF8 */
3579
3580 /* Handle a single-byte character */
3581 {
3582 classbits[c/8] |= (1 << (c&7));
3583 if ((options & PCRE_CASELESS) != 0)
3584 {
3585 c = cd->fcc[c]; /* flip case */
3586 classbits[c/8] |= (1 << (c&7));
3587 }
3588 class_charcount++;
3589 class_lastchar = c;
3590 }
3591 }
3592
3593 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3594
3595 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3596
3597 if (c == 0) /* Missing terminating ']' */
3598 {
3599 *errorcodeptr = ERR6;
3600 goto FAILED;
3601 }
3602
3603
3604 /* This code has been disabled because it would mean that \s counts as
3605 an explicit \r or \n reference, and that's not really what is wanted. Now
3606 we set the flag only if there is a literal "\r" or "\n" in the class. */
3607
3608 #if 0
3609 /* Remember whether \r or \n are in this class */
3610
3611 if (negate_class)
3612 {
3613 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3614 }
3615 else
3616 {
3617 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3618 }
3619 #endif
3620
3621
3622 /* If class_charcount is 1, we saw precisely one character whose value is
3623 less than 256. As long as there were no characters >= 128 and there was no
3624 use of \p or \P, in other words, no use of any XCLASS features, we can
3625 optimize.
3626
3627 In UTF-8 mode, we can optimize the negative case only if there were no
3628 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3629 operate on single-bytes only. This is an historical hangover. Maybe one day
3630 we can tidy these opcodes to handle multi-byte characters.
3631
3632 The optimization throws away the bit map. We turn the item into a
3633 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3634 that OP_NOT does not support multibyte characters. In the positive case, it
3635 can cause firstbyte to be set. Otherwise, there can be no first char if
3636 this item is first, whatever repeat count may follow. In the case of
3637 reqbyte, save the previous value for reinstating. */
3638
3639 #ifdef SUPPORT_UTF8
3640 if (class_charcount == 1 && !class_utf8 &&
3641 (!utf8 || !negate_class || class_lastchar < 128))
3642 #else
3643 if (class_charcount == 1)
3644 #endif
3645 {
3646 zeroreqbyte = reqbyte;
3647
3648 /* The OP_NOT opcode works on one-byte characters only. */
3649
3650 if (negate_class)
3651 {
3652 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3653 zerofirstbyte = firstbyte;
3654 *code++ = OP_NOT;
3655 *code++ = class_lastchar;
3656 break;
3657 }
3658
3659 /* For a single, positive character, get the value into mcbuffer, and
3660 then we can handle this with the normal one-character code. */
3661
3662 #ifdef SUPPORT_UTF8
3663 if (utf8 && class_lastchar > 127)
3664 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3665 else
3666 #endif
3667 {
3668 mcbuffer[0] = class_lastchar;
3669 mclength = 1;
3670 }
3671 goto ONE_CHAR;
3672 } /* End of 1-char optimization */
3673
3674 /* The general case - not the one-char optimization. If this is the first
3675 thing in the branch, there can be no first char setting, whatever the
3676 repeat count. Any reqbyte setting must remain unchanged after any kind of
3677 repeat. */
3678
3679 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3680 zerofirstbyte = firstbyte;
3681 zeroreqbyte = reqbyte;
3682
3683 /* If there are characters with values > 255, we have to compile an
3684 extended class, with its own opcode, unless there was a negated special
3685 such as \S in the class, because in that case all characters > 255 are in
3686 the class, so any that were explicitly given as well can be ignored. If
3687 (when there are explicit characters > 255 that must be listed) there are no
3688 characters < 256, we can omit the bitmap in the actual compiled code. */
3689
3690 #ifdef SUPPORT_UTF8
3691 if (class_utf8 && !should_flip_negation)
3692 {
3693 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3694 *code++ = OP_XCLASS;
3695 code += LINK_SIZE;
3696 *code = negate_class? XCL_NOT : 0;
3697
3698 /* If the map is required, move up the extra data to make room for it;
3699 otherwise just move the code pointer to the end of the extra data. */
3700
3701 if (class_charcount > 0)
3702 {
3703 *code++ |= XCL_MAP;
3704 memmove(code + 32, code, class_utf8data - code);
3705 memcpy(code, classbits, 32);
3706 code = class_utf8data + 32;
3707 }
3708 else code = class_utf8data;
3709
3710 /* Now fill in the complete length of the item */
3711
3712 PUT(previous, 1, code - previous);
3713 break; /* End of class handling */
3714 }
3715 #endif
3716
3717 /* If there are no characters > 255, set the opcode to OP_CLASS or
3718 OP_NCLASS, depending on whether the whole class was negated and whether
3719 there were negative specials such as \S in the class. Then copy the 32-byte
3720 map into the code vector, negating it if necessary. */
3721
3722 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3723 if (negate_class)
3724 {
3725 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3726 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3727 }
3728 else
3729 {
3730 memcpy(code, classbits, 32);
3731 }
3732 code += 32;
3733 break;
3734
3735
3736 /* ===================================================================*/
3737 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3738 has been tested above. */
3739
3740 case CHAR_LEFT_CURLY_BRACKET:
3741 if (!is_quantifier) goto NORMAL_CHAR;
3742 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3743 if (*errorcodeptr != 0) goto FAILED;
3744 goto REPEAT;
3745
3746 case CHAR_ASTERISK:
3747 repeat_min = 0;
3748 repeat_max = -1;
3749 goto REPEAT;
3750
3751 case CHAR_PLUS:
3752 repeat_min = 1;
3753 repeat_max = -1;
3754 goto REPEAT;
3755
3756 case CHAR_QUESTION_MARK:
3757 repeat_min = 0;
3758 repeat_max = 1;
3759
3760 REPEAT:
3761 if (previous == NULL)
3762 {
3763 *errorcodeptr = ERR9;
3764 goto FAILED;
3765 }
3766
3767 if (repeat_min == 0)
3768 {
3769 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3770 reqbyte = zeroreqbyte; /* Ditto */
3771 }
3772
3773 /* Remember whether this is a variable length repeat */
3774
3775 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3776
3777 op_type = 0; /* Default single-char op codes */
3778 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3779
3780 /* Save start of previous item, in case we have to move it up to make space
3781 for an inserted OP_ONCE for the additional '+' extension. */
3782
3783 tempcode = previous;
3784
3785 /* If the next character is '+', we have a possessive quantifier. This
3786 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3787 If the next character is '?' this is a minimizing repeat, by default,
3788 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3789 repeat type to the non-default. */
3790
3791 if (ptr[1] == CHAR_PLUS)
3792 {
3793 repeat_type = 0; /* Force greedy */
3794 possessive_quantifier = TRUE;
3795 ptr++;
3796 }
3797 else if (ptr[1] == CHAR_QUESTION_MARK)
3798 {
3799 repeat_type = greedy_non_default;
3800 ptr++;
3801 }
3802 else repeat_type = greedy_default;
3803
3804 /* If previous was a character match, abolish the item and generate a
3805 repeat item instead. If a char item has a minumum of more than one, ensure
3806 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3807 the first thing in a branch because the x will have gone into firstbyte
3808 instead. */
3809
3810 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3811 {
3812 /* Deal with UTF-8 characters that take up more than one byte. It's
3813 easier to write this out separately than try to macrify it. Use c to
3814 hold the length of the character in bytes, plus 0x80 to flag that it's a
3815 length rather than a small character. */
3816
3817 #ifdef SUPPORT_UTF8
3818 if (utf8 && (code[-1] & 0x80) != 0)
3819 {
3820 uschar *lastchar = code - 1;
3821 while((*lastchar & 0xc0) == 0x80) lastchar--;
3822 c = code - lastchar; /* Length of UTF-8 character */
3823 memcpy(utf8_char, lastchar, c); /* Save the char */
3824 c |= 0x80; /* Flag c as a length */
3825 }
3826 else
3827 #endif
3828
3829 /* Handle the case of a single byte - either with no UTF8 support, or
3830 with UTF-8 disabled, or for a UTF-8 character < 128. */
3831
3832 {
3833 c = code[-1];
3834 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3835 }
3836
3837 /* If the repetition is unlimited, it pays to see if the next thing on
3838 the line is something that cannot possibly match this character. If so,
3839 automatically possessifying this item gains some performance in the case
3840 where the match fails. */
3841
3842 if (!possessive_quantifier &&
3843 repeat_max < 0 &&
3844 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3845 options, cd))
3846 {
3847 repeat_type = 0; /* Force greedy */
3848 possessive_quantifier = TRUE;
3849 }
3850
3851 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3852 }
3853
3854 /* If previous was a single negated character ([^a] or similar), we use
3855 one of the special opcodes, replacing it. The code is shared with single-
3856 character repeats by setting opt_type to add a suitable offset into
3857 repeat_type. We can also test for auto-possessification. OP_NOT is
3858 currently used only for single-byte chars. */
3859
3860 else if (*previous == OP_NOT)
3861 {
3862 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3863 c = previous[1];
3864 if (!possessive_quantifier &&
3865 repeat_max < 0 &&
3866 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3867 {
3868 repeat_type = 0; /* Force greedy */
3869 possessive_quantifier = TRUE;
3870 }
3871 goto OUTPUT_SINGLE_REPEAT;
3872 }
3873
3874 /* If previous was a character type match (\d or similar), abolish it and
3875 create a suitable repeat item. The code is shared with single-character
3876 repeats by setting op_type to add a suitable offset into repeat_type. Note
3877 the the Unicode property types will be present only when SUPPORT_UCP is
3878 defined, but we don't wrap the little bits of code here because it just
3879 makes it horribly messy. */
3880
3881 else if (*previous < OP_EODN)
3882 {
3883 uschar *oldcode;
3884 int prop_type, prop_value;
3885 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3886 c = *previous;
3887
3888 if (!possessive_quantifier &&
3889 repeat_max < 0 &&
3890 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3891 {
3892 repeat_type = 0; /* Force greedy */
3893 possessive_quantifier = TRUE;
3894 }
3895
3896 OUTPUT_SINGLE_REPEAT:
3897 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3898 {
3899 prop_type = previous[1];
3900 prop_value = previous[2];
3901 }
3902 else prop_type = prop_value = -1;
3903
3904 oldcode = code;
3905 code = previous; /* Usually overwrite previous item */
3906
3907 /* If the maximum is zero then the minimum must also be zero; Perl allows
3908 this case, so we do too - by simply omitting the item altogether. */
3909
3910 if (repeat_max == 0) goto END_REPEAT;
3911
3912 /*--------------------------------------------------------------------*/
3913 /* This code is obsolete from release 8.00; the restriction was finally
3914 removed: */
3915
3916 /* All real repeats make it impossible to handle partial matching (maybe
3917 one day we will be able to remove this restriction). */
3918
3919 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3920 /*--------------------------------------------------------------------*/
3921
3922 /* Combine the op_type with the repeat_type */
3923
3924 repeat_type += op_type;
3925
3926 /* A minimum of zero is handled either as the special case * or ?, or as
3927 an UPTO, with the maximum given. */
3928
3929 if (repeat_min == 0)
3930 {
3931 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3932 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3933 else
3934 {
3935 *code++ = OP_UPTO + repeat_type;
3936 PUT2INC(code, 0, repeat_max);
3937 }
3938 }
3939
3940 /* A repeat minimum of 1 is optimized into some special cases. If the
3941 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3942 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3943 one less than the maximum. */
3944
3945 else if (repeat_min == 1)
3946 {
3947 if (repeat_max == -1)
3948 *code++ = OP_PLUS + repeat_type;
3949 else
3950 {
3951 code = oldcode; /* leave previous item in place */
3952 if (repeat_max == 1) goto END_REPEAT;
3953 *code++ = OP_UPTO + repeat_type;
3954 PUT2INC(code, 0, repeat_max - 1);
3955 }
3956 }
3957
3958 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3959 handled as an EXACT followed by an UPTO. */
3960
3961 else
3962 {
3963 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3964 PUT2INC(code, 0, repeat_min);
3965
3966 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3967 we have to insert the character for the previous code. For a repeated
3968 Unicode property match, there are two extra bytes that define the
3969 required property. In UTF-8 mode, long characters have their length in
3970 c, with the 0x80 bit as a flag. */
3971
3972 if (repeat_max < 0)
3973 {
3974 #ifdef SUPPORT_UTF8
3975 if (utf8 && c >= 128)
3976 {
3977 memcpy(code, utf8_char, c & 7);
3978 code += c & 7;
3979 }
3980 else
3981 #endif
3982 {
3983 *code++ = c;
3984 if (prop_type >= 0)
3985 {
3986 *code++ = prop_type;
3987 *code++ = prop_value;
3988 }
3989 }
3990 *code++ = OP_STAR + repeat_type;
3991 }
3992
3993 /* Else insert an UPTO if the max is greater than the min, again
3994 preceded by the character, for the previously inserted code. If the
3995 UPTO is just for 1 instance, we can use QUERY instead. */
3996
3997 else if (repeat_max != repeat_min)
3998 {
3999 #ifdef SUPPORT_UTF8
4000 if (utf8 && c >= 128)
4001 {
4002 memcpy(code, utf8_char, c & 7);
4003 code += c & 7;
4004 }
4005 else
4006 #endif
4007 *code++ = c;
4008 if (prop_type >= 0)
4009 {
4010 *code++ = prop_type;
4011 *code++ = prop_value;
4012 }
4013 repeat_max -= repeat_min;
4014
4015 if (repeat_max == 1)
4016 {
4017 *code++ = OP_QUERY + repeat_type;
4018 }
4019 else
4020 {
4021 *code++ = OP_UPTO + repeat_type;
4022 PUT2INC(code, 0, repeat_max);
4023 }
4024 }
4025 }
4026
4027 /* The character or character type itself comes last in all cases. */
4028
4029 #ifdef SUPPORT_UTF8
4030 if (utf8 && c >= 128)
4031 {
4032 memcpy(code, utf8_char, c & 7);
4033 code += c & 7;
4034 }
4035 else
4036 #endif
4037 *code++ = c;
4038
4039 /* For a repeated Unicode property match, there are two extra bytes that
4040 define the required property. */
4041
4042 #ifdef SUPPORT_UCP
4043 if (prop_type >= 0)
4044 {
4045 *code++ = prop_type;
4046 *code++ = prop_value;
4047 }
4048 #endif
4049 }
4050
4051 /* If previous was a character class or a back reference, we put the repeat
4052 stuff after it, but just skip the item if the repeat was {0,0}. */
4053
4054 else if (*previous == OP_CLASS ||
4055 *previous == OP_NCLASS ||
4056 #ifdef SUPPORT_UTF8
4057 *previous == OP_XCLASS ||
4058 #endif
4059 *previous == OP_REF)
4060 {
4061 if (repeat_max == 0)
4062 {
4063 code = previous;
4064 goto END_REPEAT;
4065 }
4066
4067 /*--------------------------------------------------------------------*/
4068 /* This code is obsolete from release 8.00; the restriction was finally
4069 removed: */
4070
4071 /* All real repeats make it impossible to handle partial matching (maybe
4072 one day we will be able to remove this restriction). */
4073
4074 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4075 /*--------------------------------------------------------------------*/
4076
4077 if (repeat_min == 0 && repeat_max == -1)
4078 *code++ = OP_CRSTAR + repeat_type;
4079 else if (repeat_min == 1 && repeat_max == -1)
4080 *code++ = OP_CRPLUS + repeat_type;
4081 else if (repeat_min == 0 && repeat_max == 1)
4082 *code++ = OP_CRQUERY + repeat_type;
4083 else
4084 {
4085 *code++ = OP_CRRANGE + repeat_type;
4086 PUT2INC(code, 0, repeat_min);
4087 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4088 PUT2INC(code, 0, repeat_max);
4089 }
4090 }
4091
4092 /* If previous was a bracket group, we may have to replicate it in certain
4093 cases. */
4094
4095 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4096 *previous == OP_ONCE || *previous == OP_COND)
4097 {
4098 register int i;
4099 int ketoffset = 0;
4100 int len = code - previous;
4101 uschar *bralink = NULL;
4102
4103 /* Repeating a DEFINE group is pointless */
4104
4105 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4106 {
4107 *errorcodeptr = ERR55;
4108 goto FAILED;
4109 }
4110
4111 /* If the maximum repeat count is unlimited, find the end of the bracket
4112 by scanning through from the start, and compute the offset back to it
4113 from the current code pointer. There may be an OP_OPT setting following
4114 the final KET, so we can't find the end just by going back from the code
4115 pointer. */
4116
4117 if (repeat_max == -1)
4118 {
4119 register uschar *ket = previous;
4120 do ket += GET(ket, 1); while (*ket != OP_KET);
4121 ketoffset = code - ket;
4122 }
4123
4124 /* The case of a zero minimum is special because of the need to stick
4125 OP_BRAZERO in front of it, and because the group appears once in the
4126 data, whereas in other cases it appears the minimum number of times. For
4127 this reason, it is simplest to treat this case separately, as otherwise
4128 the code gets far too messy. There are several special subcases when the
4129 minimum is zero. */
4130
4131 if (repeat_min == 0)
4132 {
4133 /* If the maximum is also zero, we used to just omit the group from the
4134 output altogether, like this:
4135
4136 ** if (repeat_max == 0)
4137 ** {
4138 ** code = previous;
4139 ** goto END_REPEAT;
4140 ** }
4141
4142 However, that fails when a group is referenced as a subroutine from
4143 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4144 so that it is skipped on execution. As we don't have a list of which
4145 groups are referenced, we cannot do this selectively.
4146
4147 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4148 and do no more at this point. However, we do need to adjust any
4149 OP_RECURSE calls inside the group that refer to the group itself or any
4150 internal or forward referenced group, because the offset is from the
4151 start of the whole regex. Temporarily terminate the pattern while doing
4152 this. */
4153
4154 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4155 {
4156 *code = OP_END;
4157 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4158 memmove(previous+1, previous, len);
4159 code++;
4160 if (repeat_max == 0)
4161 {
4162 *previous++ = OP_SKIPZERO;
4163 goto END_REPEAT;
4164 }
4165 *previous++ = OP_BRAZERO + repeat_type;
4166 }
4167
4168 /* If the maximum is greater than 1 and limited, we have to replicate
4169 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4170 The first one has to be handled carefully because it's the original
4171 copy, which has to be moved up. The remainder can be handled by code
4172 that is common with the non-zero minimum case below. We have to
4173 adjust the value or repeat_max, since one less copy is required. Once
4174 again, we may have to adjust any OP_RECURSE calls inside the group. */
4175
4176 else
4177 {
4178 int offset;
4179 *code = OP_END;
4180 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4181 memmove(previous + 2 + LINK_SIZE, previous, len);
4182 code += 2 + LINK_SIZE;
4183 *previous++ = OP_BRAZERO + repeat_type;
4184 *previous++ = OP_BRA;
4185
4186 /* We chain together the bracket offset fields that have to be
4187 filled in later when the ends of the brackets are reached. */
4188
4189 offset = (bralink == NULL)? 0 : previous - bralink;
4190 bralink = previous;
4191 PUTINC(previous, 0, offset);
4192 }
4193
4194 repeat_max--;
4195 }
4196
4197 /* If the minimum is greater than zero, replicate the group as many
4198 times as necessary, and adjust the maximum to the number of subsequent
4199 copies that we need. If we set a first char from the group, and didn't
4200 set a required char, copy the latter from the former. If there are any
4201 forward reference subroutine calls in the group, there will be entries on
4202 the workspace list; replicate these with an appropriate increment. */
4203
4204 else
4205 {
4206 if (repeat_min > 1)
4207 {
4208 /* In the pre-compile phase, we don't actually do the replication. We
4209 just adjust the length as if we had. Do some paranoid checks for
4210 potential integer overflow. */
4211
4212 if (lengthptr != NULL)
4213 {
4214 int delta = (repeat_min - 1)*length_prevgroup;
4215 if ((double)(repeat_min - 1)*(double)length_prevgroup >
4216 (double)INT_MAX ||
4217 OFLOW_MAX - *lengthptr < delta)
4218 {
4219 *errorcodeptr = ERR20;
4220 goto FAILED;
4221 }
4222 *lengthptr += delta;
4223 }
4224
4225 /* This is compiling for real */
4226
4227 else
4228 {
4229 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4230 for (i = 1; i < repeat_min; i++)
4231 {
4232 uschar *hc;
4233 uschar *this_hwm = cd->hwm;
4234 memcpy(code, previous, len);
4235 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4236 {
4237 PUT(cd->hwm, 0, GET(hc, 0) + len);
4238 cd->hwm += LINK_SIZE;
4239 }
4240 save_hwm = this_hwm;
4241 code += len;
4242 }
4243 }
4244 }
4245
4246 if (repeat_max > 0) repeat_max -= repeat_min;
4247 }
4248
4249 /* This code is common to both the zero and non-zero minimum cases. If
4250 the maximum is limited, it replicates the group in a nested fashion,
4251 remembering the bracket starts on a stack. In the case of a zero minimum,
4252 the first one was set up above. In all cases the repeat_max now specifies
4253 the number of additional copies needed. Again, we must remember to
4254 replicate entries on the forward reference list. */
4255
4256 if (repeat_max >= 0)
4257 {
4258 /* In the pre-compile phase, we don't actually do the replication. We
4259 just adjust the length as if we had. For each repetition we must add 1
4260 to the length for BRAZERO and for all but the last repetition we must
4261 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4262 paranoid checks to avoid integer overflow. */
4263
4264 if (lengthptr != NULL && repeat_max > 0)
4265 {
4266 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4267 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4268 if ((double)repeat_max *
4269 (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4270 > (double)INT_MAX ||
4271 OFLOW_MAX - *lengthptr < delta)
4272 {
4273 *errorcodeptr = ERR20;
4274 goto FAILED;
4275 }
4276 *lengthptr += delta;
4277 }
4278
4279 /* This is compiling for real */
4280
4281 else for (i = repeat_max - 1; i >= 0; i--)
4282 {
4283 uschar *hc;
4284 uschar *this_hwm = cd->hwm;
4285
4286 *code++ = OP_BRAZERO + repeat_type;
4287
4288 /* All but the final copy start a new nesting, maintaining the
4289 chain of brackets outstanding. */
4290
4291 if (i != 0)
4292 {
4293 int offset;
4294 *code++ = OP_BRA;
4295 offset = (bralink == NULL)? 0 : code - bralink;
4296 bralink = code;
4297 PUTINC(code, 0, offset);
4298 }
4299
4300 memcpy(code, previous, len);
4301 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4302 {
4303 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4304 cd->hwm += LINK_SIZE;
4305 }
4306 save_hwm = this_hwm;
4307 code += len;
4308 }
4309
4310 /* Now chain through the pending brackets, and fill in their length
4311 fields (which are holding the chain links pro tem). */
4312
4313 while (bralink != NULL)
4314 {
4315 int oldlinkoffset;
4316 int offset = code - bralink + 1;
4317 uschar *bra = code - offset;
4318 oldlinkoffset = GET(bra, 1);
4319 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4320 *code++ = OP_KET;
4321 PUTINC(code, 0, offset);
4322 PUT(bra, 1, offset);
4323 }
4324 }
4325
4326 /* If the maximum is unlimited, set a repeater in the final copy. We
4327 can't just offset backwards from the current code point, because we
4328 don't know if there's been an options resetting after the ket. The
4329 correct offset was computed above.
4330
4331 Then, when we are doing the actual compile phase, check to see whether
4332 this group is a non-atomic one that could match an empty string. If so,
4333 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4334 that runtime checking can be done. [This check is also applied to
4335 atomic groups at runtime, but in a different way.] */
4336
4337 else
4338 {
4339 uschar *ketcode = code - ketoffset;
4340 uschar *bracode = ketcode - GET(ketcode, 1);
4341 *ketcode = OP_KETRMAX + repeat_type;
4342 if (lengthptr == NULL && *bracode != OP_ONCE)
4343 {
4344 uschar *scode = bracode;
4345 do
4346 {
4347 if (could_be_empty_branch(scode, ketcode, utf8))
4348 {
4349 *bracode += OP_SBRA - OP_BRA;
4350 break;
4351 }
4352 scode += GET(scode, 1);
4353 }
4354 while (*scode == OP_ALT);
4355 }
4356 }
4357 }
4358
4359 /* If previous is OP_FAIL, it was generated by an empty class [] in
4360 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4361 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4362 error above. We can just ignore the repeat in JS case. */
4363
4364 else if (*previous == OP_FAIL) goto END_REPEAT;
4365
4366 /* Else there's some kind of shambles */
4367
4368 else
4369 {
4370 *errorcodeptr = ERR11;
4371 goto FAILED;
4372 }
4373
4374 /* If the character following a repeat is '+', or if certain optimization
4375 tests above succeeded, possessive_quantifier is TRUE. For some of the
4376 simpler opcodes, there is an special alternative opcode for this. For
4377 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4378 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4379 but the special opcodes can optimize it a bit. The repeated item starts at
4380 tempcode, not at previous, which might be the first part of a string whose
4381 (former) last char we repeated.
4382
4383 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4384 an 'upto' may follow. We skip over an 'exact' item, and then test the
4385 length of what remains before proceeding. */
4386
4387 if (possessive_quantifier)
4388 {
4389 int len;
4390
4391 if (*tempcode == OP_TYPEEXACT)
4392 tempcode += _pcre_OP_lengths[*tempcode] +
4393 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4394
4395 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4396 {
4397 tempcode += _pcre_OP_lengths[*tempcode];
4398 #ifdef SUPPORT_UTF8
4399 if (utf8 && tempcode[-1] >= 0xc0)
4400 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4401 #endif
4402 }
4403
4404 len = code - tempcode;
4405 if (len > 0) switch (*tempcode)
4406 {
4407 case OP_STAR: *tempcode = OP_POSSTAR; break;
4408 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4409 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4410 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4411
4412 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4413 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4414 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4415 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4416
4417 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4418 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4419 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4420 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4421
4422 default:
4423 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4424 code += 1 + LINK_SIZE;
4425 len += 1 + LINK_SIZE;
4426 tempcode[0] = OP_ONCE;
4427 *code++ = OP_KET;
4428 PUTINC(code, 0, len);
4429 PUT(tempcode, 1, len);
4430 break;
4431 }
4432 }
4433
4434 /* In all case we no longer have a previous item. We also set the
4435 "follows varying string" flag for subsequently encountered reqbytes if
4436 it isn't already set and we have just passed a varying length item. */
4437
4438 END_REPEAT:
4439 previous = NULL;
4440 cd->req_varyopt |= reqvary;
4441 break;
4442
4443
4444 /* ===================================================================*/
4445 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4446 lookbehind or option setting or condition or all the other extended
4447 parenthesis forms. */
4448
4449 case CHAR_LEFT_PARENTHESIS:
4450 newoptions = options;
4451 skipbytes = 0;
4452 bravalue = OP_CBRA;
4453 save_hwm = cd->hwm;
4454 reset_bracount = FALSE;
4455
4456 /* First deal with various "verbs" that can be introduced by '*'. */
4457
4458 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4459 {
4460 int i, namelen;
4461 const char *vn = verbnames;
4462 const uschar *name = ++ptr;
4463 previous = NULL;
4464 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4465 if (*ptr == CHAR_COLON)
4466 {
4467 *errorcodeptr = ERR59; /* Not supported */
4468 goto FAILED;
4469 }
4470 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4471 {
4472 *errorcodeptr = ERR60;
4473 goto FAILED;
4474 }
4475 namelen = ptr - name;
4476 for (i = 0; i < verbcount; i++)
4477 {
4478 if (namelen == verbs[i].len &&
4479 strncmp((char *)name, vn, namelen) == 0)
4480 {
4481 /* Check for open captures before ACCEPT */
4482
4483 if (verbs[i].op == OP_ACCEPT)
4484 {
4485 open_capitem *oc;
4486 cd->had_accept = TRUE;
4487 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4488 {
4489 *code++ = OP_CLOSE;
4490 PUT2INC(code, 0, oc->number);
4491 }
4492 }
4493 *code++ = verbs[i].op;
4494 break;
4495 }
4496 vn += verbs[i].len + 1;
4497 }
4498 if (i < verbcount) continue;
4499 *errorcodeptr = ERR60;
4500 goto FAILED;
4501 }
4502
4503 /* Deal with the extended parentheses; all are introduced by '?', and the
4504 appearance of any of them means that this is not a capturing group. */
4505
4506 else if (*ptr == CHAR_QUESTION_MARK)
4507 {
4508 int i, set, unset, namelen;
4509 int *optset;
4510 const uschar *name;
4511 uschar *slot;
4512
4513 switch (*(++ptr))
4514 {
4515 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4516 ptr++;
4517 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4518 if (*ptr == 0)
4519 {
4520 *errorcodeptr = ERR18;
4521 goto FAILED;
4522 }
4523 continue;
4524
4525
4526 /* ------------------------------------------------------------ */
4527 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4528 reset_bracount = TRUE;
4529 /* Fall through */
4530
4531 /* ------------------------------------------------------------ */
4532 case CHAR_COLON: /* Non-capturing bracket */
4533 bravalue = OP_BRA;
4534 ptr++;
4535 break;
4536
4537
4538 /* ------------------------------------------------------------ */
4539 case CHAR_LEFT_PARENTHESIS:
4540 bravalue = OP_COND; /* Conditional group */
4541
4542 /* A condition can be an assertion, a number (referring to a numbered
4543 group), a name (referring to a named group), or 'R', referring to
4544 recursion. R<digits> and R&name are also permitted for recursion tests.
4545
4546 There are several syntaxes for testing a named group: (?(name)) is used
4547 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4548
4549 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4550 be the recursive thing or the name 'R' (and similarly for 'R' followed
4551 by digits), and (b) a number could be a name that consists of digits.
4552 In both cases, we look for a name first; if not found, we try the other
4553 cases. */
4554
4555 /* For conditions that are assertions, check the syntax, and then exit
4556 the switch. This will take control down to where bracketed groups,
4557 including assertions, are processed. */
4558
4559 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4560 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4561 break;
4562
4563 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4564 below), and all need to skip 3 bytes at the start of the group. */
4565
4566 code[1+LINK_SIZE] = OP_CREF;
4567 skipbytes = 3;
4568 refsign = -1;
4569
4570 /* Check for a test for recursion in a named group. */
4571
4572 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4573 {
4574 terminator = -1;
4575 ptr += 2;
4576 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4577 }
4578
4579 /* Check for a test for a named group's having been set, using the Perl
4580 syntax (?(<name>) or (?('name') */
4581
4582 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4583 {
4584 terminator = CHAR_GREATER_THAN_SIGN;
4585 ptr++;
4586 }
4587 else if (ptr[1] == CHAR_APOSTROPHE)
4588 {
4589 terminator = CHAR_APOSTROPHE;
4590 ptr++;
4591 }
4592 else
4593 {
4594 terminator = 0;
4595 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4596 }
4597
4598 /* We now expect to read a name; any thing else is an error */
4599
4600 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4601 {
4602 ptr += 1; /* To get the right offset */
4603 *errorcodeptr = ERR28;
4604 goto FAILED;
4605 }
4606
4607 /* Read the name, but also get it as a number if it's all digits */
4608
4609 recno = 0;
4610 name = ++ptr;
4611 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4612 {
4613 if (recno >= 0)
4614 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4615 recno * 10 + *ptr - CHAR_0 : -1;
4616 ptr++;
4617 }
4618 namelen = ptr - name;
4619
4620 if ((terminator > 0 && *ptr++ != terminator) ||
4621 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4622 {
4623 ptr--; /* Error offset */
4624 *errorcodeptr = ERR26;
4625 goto FAILED;
4626 }
4627
4628 /* Do no further checking in the pre-compile phase. */
4629
4630 if (lengthptr != NULL) break;
4631
4632 /* In the real compile we do the work of looking for the actual
4633 reference. If the string started with "+" or "-" we require the rest to
4634 be digits, in which case recno will be set. */
4635
4636 if (refsign > 0)
4637 {
4638 if (recno <= 0)
4639 {
4640 *errorcodeptr = ERR58;
4641 goto FAILED;
4642 }
4643 recno = (refsign == CHAR_MINUS)?
4644 cd->bracount - recno + 1 : recno +cd->bracount;
4645 if (recno <= 0 || recno > cd->final_bracount)
4646 {
4647 *errorcodeptr = ERR15;
4648 goto FAILED;
4649 }
4650 PUT2(code, 2+LINK_SIZE, recno);
4651 break;
4652 }
4653
4654 /* Otherwise (did not start with "+" or "-"), start by looking for the
4655 name. */
4656
4657 slot = cd->name_table;
4658 for (i = 0; i < cd->names_found; i++)
4659 {
4660 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4661 slot += cd->name_entry_size;
4662 }
4663
4664 /* Found a previous named subpattern */
4665
4666 if (i < cd->names_found)
4667 {
4668 recno = GET2(slot, 0);
4669 PUT2(code, 2+LINK_SIZE, recno);
4670 }
4671
4672 /* Search the pattern for a forward reference */
4673
4674 else if ((i = find_parens(cd, name, namelen,
4675 (options & PCRE_EXTENDED) != 0)) > 0)
4676 {
4677 PUT2(code, 2+LINK_SIZE, i);
4678 }
4679
4680 /* If terminator == 0 it means that the name followed directly after
4681 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4682 some further alternatives to try. For the cases where terminator != 0
4683 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4684 now checked all the possibilities, so give an error. */
4685
4686 else if (terminator != 0)
4687 {
4688 *errorcodeptr = ERR15;
4689 goto FAILED;
4690 }
4691
4692 /* Check for (?(R) for recursion. Allow digits after R to specify a
4693 specific group number. */
4694
4695 else if (*name == CHAR_R)
4696 {
4697 recno = 0;
4698 for (i = 1; i < namelen; i++)
4699 {
4700 if ((digitab[name[i]] & ctype_digit) == 0)
4701 {
4702 *errorcodeptr = ERR15;
4703 goto FAILED;
4704 }
4705 recno = recno * 10 + name[i] - CHAR_0;
4706 }
4707 if (recno == 0) recno = RREF_ANY;
4708 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4709 PUT2(code, 2+LINK_SIZE, recno);
4710 }
4711
4712 /* Similarly, check for the (?(DEFINE) "condition", which is always
4713 false. */
4714
4715 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4716 {
4717 code[1+LINK_SIZE] = OP_DEF;
4718 skipbytes = 1;
4719 }
4720
4721 /* Check for the "name" actually being a subpattern number. We are
4722 in the second pass here, so final_bracount is set. */
4723
4724 else if (recno > 0 && recno <= cd->final_bracount)
4725 {
4726 PUT2(code, 2+LINK_SIZE, recno);
4727 }
4728
4729 /* Either an unidentified subpattern, or a reference to (?(0) */
4730
4731 else
4732 {
4733 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4734 goto FAILED;
4735 }
4736 break;
4737
4738
4739 /* ------------------------------------------------------------ */
4740 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4741 bravalue = OP_ASSERT;
4742 ptr++;
4743 break;
4744
4745
4746 /* ------------------------------------------------------------ */
4747 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4748 ptr++;
4749 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4750 {
4751 *code++ = OP_FAIL;
4752 previous = NULL;
4753 continue;
4754 }
4755 bravalue = OP_ASSERT_NOT;
4756 break;
4757
4758
4759 /* ------------------------------------------------------------ */
4760 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4761 switch (ptr[1])
4762 {
4763 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4764 bravalue = OP_ASSERTBACK;
4765 ptr += 2;
4766 break;
4767
4768 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4769 bravalue = OP_ASSERTBACK_NOT;
4770 ptr += 2;
4771 break;
4772
4773 default: /* Could be name define, else bad */
4774 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4775 ptr++; /* Correct offset for error */
4776 *errorcodeptr = ERR24;
4777 goto FAILED;
4778 }
4779 break;
4780
4781
4782 /* ------------------------------------------------------------ */
4783 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4784 bravalue = OP_ONCE;
4785 ptr++;
4786 break;
4787
4788
4789 /* ------------------------------------------------------------ */
4790 case CHAR_C: /* Callout - may be followed by digits; */
4791 previous_callout = code; /* Save for later completion */
4792 after_manual_callout = 1; /* Skip one item before completing */
4793 *code++ = OP_CALLOUT;
4794 {
4795 int n = 0;
4796 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4797 n = n * 10 + *ptr - CHAR_0;
4798 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4799 {
4800 *errorcodeptr = ERR39;
4801 goto FAILED;
4802 }
4803 if (n > 255)
4804 {
4805 *errorcodeptr = ERR38;
4806 goto FAILED;
4807 }
4808 *code++ = n;
4809 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4810 PUT(code, LINK_SIZE, 0); /* Default length */
4811 code += 2 * LINK_SIZE;
4812 }
4813 previous = NULL;
4814 continue;
4815
4816
4817 /* ------------------------------------------------------------ */
4818 case CHAR_P: /* Python-style named subpattern handling */
4819 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4820 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4821 {
4822 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4823 terminator = CHAR_RIGHT_PARENTHESIS;
4824 goto NAMED_REF_OR_RECURSE;
4825 }
4826 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4827 {
4828 *errorcodeptr = ERR41;
4829 goto FAILED;
4830 }
4831 /* Fall through to handle (?P< as (?< is handled */
4832
4833
4834 /* ------------------------------------------------------------ */
4835 DEFINE_NAME: /* Come here from (?< handling */
4836 case CHAR_APOSTROPHE:
4837 {
4838 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4839 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4840 name = ++ptr;
4841
4842 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4843 namelen = ptr - name;
4844
4845 /* In the pre-compile phase, just do a syntax check. */
4846
4847 if (lengthptr != NULL)
4848 {
4849 if (*ptr != terminator)
4850 {
4851 *errorcodeptr = ERR42;
4852 goto FAILED;
4853 }
4854 if (cd->names_found >= MAX_NAME_COUNT)
4855 {
4856 *errorcodeptr = ERR49;
4857 goto FAILED;
4858 }
4859 if (namelen + 3 > cd->name_entry_size)
4860 {
4861 cd->name_entry_size = namelen + 3;
4862 if (namelen > MAX_NAME_SIZE)
4863 {
4864 *errorcodeptr = ERR48;
4865 goto FAILED;
4866 }
4867 }
4868 }
4869
4870 /* In the real compile, create the entry in the table */
4871
4872 else
4873 {
4874 slot = cd->name_table;
4875 for (i = 0; i < cd->names_found; i++)
4876 {
4877 int crc = memcmp(name, slot+2, namelen);
4878 if (crc == 0)
4879 {
4880 if (slot[2+namelen] == 0)
4881 {
4882 if ((options & PCRE_DUPNAMES) == 0)
4883 {
4884 *errorcodeptr = ERR43;
4885 goto FAILED;
4886 }
4887 }
4888 else crc = -1; /* Current name is substring */
4889 }
4890 if (crc < 0)
4891 {
4892 memmove(slot + cd->name_entry_size, slot,
4893 (cd->names_found - i) * cd->name_entry_size);
4894 break;
4895 }
4896 slot += cd->name_entry_size;
4897 }
4898
4899 PUT2(slot, 0, cd->bracount + 1);
4900 memcpy(slot + 2, name, namelen);
4901 slot[2+namelen] = 0;
4902 }
4903 }
4904
4905 /* In both cases, count the number of names we've encountered. */
4906
4907 ptr++; /* Move past > or ' */
4908 cd->names_found++;
4909 goto NUMBERED_GROUP;
4910
4911
4912 /* ------------------------------------------------------------ */
4913 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
4914 terminator = CHAR_RIGHT_PARENTHESIS;
4915 is_recurse = TRUE;
4916 /* Fall through */
4917
4918 /* We come here from the Python syntax above that handles both
4919 references (?P=name) and recursion (?P>name), as well as falling
4920 through from the Perl recursion syntax (?&name). We also come here from
4921 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4922 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
4923
4924 NAMED_REF_OR_RECURSE:
4925 name = ++ptr;
4926 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4927 namelen = ptr - name;
4928
4929 /* In the pre-compile phase, do a syntax check and set a dummy
4930 reference number. */
4931
4932 if (lengthptr != NULL)
4933 {
4934 if (namelen == 0)
4935 {
4936 *errorcodeptr = ERR62;
4937 goto FAILED;
4938 }
4939 if (*ptr != terminator)
4940 {
4941 *errorcodeptr = ERR42;
4942 goto FAILED;
4943 }
4944 if (namelen > MAX_NAME_SIZE)
4945 {
4946 *errorcodeptr = ERR48;
4947 goto FAILED;
4948 }
4949 recno = 0;
4950 }
4951
4952 /* In the real compile, seek the name in the table. We check the name
4953 first, and then check that we have reached the end of the name in the
4954 table. That way, if the name that is longer than any in the table,
4955 the comparison will fail without reading beyond the table entry. */
4956
4957 else
4958 {
4959 slot = cd->name_table;
4960 for (i = 0; i < cd->names_found; i++)
4961 {
4962 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
4963 slot[2+namelen] == 0)
4964 break;
4965 slot += cd->name_entry_size;
4966 }
4967
4968 if (i < cd->names_found) /* Back reference */
4969 {
4970 recno = GET2(slot, 0);
4971 }
4972 else if ((recno = /* Forward back reference */
4973 find_parens(cd, name, namelen,
4974 (options & PCRE_EXTENDED) != 0)) <= 0)
4975 {
4976 *errorcodeptr = ERR15;
4977 goto FAILED;
4978 }
4979 }
4980
4981 /* In both phases, we can now go to the code than handles numerical
4982 recursion or backreferences. */
4983
4984 if (is_recurse) goto HANDLE_RECURSION;
4985 else goto HANDLE_REFERENCE;
4986
4987
4988 /* ------------------------------------------------------------ */
4989 case CHAR_R: /* Recursion */
4990 ptr++; /* Same as (?0) */
4991 /* Fall through */
4992
4993
4994 /* ------------------------------------------------------------ */
4995 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
4996 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
4997 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
4998 {
4999 const uschar *called;
5000 terminator = CHAR_RIGHT_PARENTHESIS;
5001
5002 /* Come here from the \g<...> and \g'...' code (Oniguruma
5003 compatibility). However, the syntax has been checked to ensure that
5004 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5005 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5006 ever be taken. */
5007
5008 HANDLE_NUMERICAL_RECURSION:
5009
5010 if ((refsign = *ptr) == CHAR_PLUS)
5011 {
5012 ptr++;
5013 if ((digitab[*ptr] & ctype_digit) == 0)
5014 {
5015 *errorcodeptr = ERR63;
5016 goto FAILED;
5017 }
5018 }
5019 else if (refsign == CHAR_MINUS)
5020 {
5021 if ((digitab[ptr[1]] & ctype_digit) == 0)
5022 goto OTHER_CHAR_AFTER_QUERY;
5023 ptr++;
5024 }
5025
5026 recno = 0;
5027 while((digitab[*ptr] & ctype_digit) != 0)
5028 recno = recno * 10 + *ptr++ - CHAR_0;
5029
5030 if (*ptr != terminator)
5031 {
5032 *errorcodeptr = ERR29;
5033 goto FAILED;
5034 }
5035
5036 if (refsign == CHAR_MINUS)
5037 {
5038 if (recno == 0)
5039 {
5040 *errorcodeptr = ERR58;
5041 goto FAILED;
5042 }
5043 recno = cd->bracount - recno + 1;
5044 if (recno <= 0)
5045 {
5046 *errorcodeptr = ERR15;
5047 goto FAILED;
5048 }
5049 }
5050 else if (refsign == CHAR_PLUS)
5051 {
5052 if (recno == 0)
5053 {
5054 *errorcodeptr = ERR58;
5055 goto FAILED;
5056 }
5057 recno += cd->bracount;
5058 }
5059
5060 /* Come here from code above that handles a named recursion */
5061
5062 HANDLE_RECURSION:
5063
5064 previous = code;
5065 called = cd->start_code;
5066
5067 /* When we are actually compiling, find the bracket that is being
5068 referenced. Temporarily end the regex in case it doesn't exist before
5069 this point. If we end up with a forward reference, first check that
5070 the bracket does occur later so we can give the error (and position)
5071 now. Then remember this forward reference in the workspace so it can
5072 be filled in at the end. */
5073
5074 if (lengthptr == NULL)
5075 {
5076 *code = OP_END;
5077 if (recno != 0)
5078 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5079
5080 /* Forward reference */
5081
5082 if (called == NULL)
5083 {
5084 if (find_parens(cd, NULL, recno,
5085 (options & PCRE_EXTENDED) != 0) < 0)
5086 {
5087 *errorcodeptr = ERR15;
5088 goto FAILED;
5089 }
5090 called = cd->start_code + recno;
5091 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5092 }
5093
5094 /* If not a forward reference, and the subpattern is still open,
5095 this is a recursive call. We check to see if this is a left
5096 recursion that could loop for ever, and diagnose that case. */
5097
5098 else if (GET(called, 1) == 0 &&
5099 could_be_empty(called, code, bcptr, utf8))
5100 {
5101 *errorcodeptr = ERR40;
5102 goto FAILED;
5103 }
5104 }
5105
5106 /* Insert the recursion/subroutine item, automatically wrapped inside
5107 "once" brackets. Set up a "previous group" length so that a
5108 subsequent quantifier will work. */
5109
5110 *code = OP_ONCE;
5111 PUT(code, 1, 2 + 2*LINK_SIZE);
5112 code += 1 + LINK_SIZE;
5113
5114 *code = OP_RECURSE;
5115 PUT(code, 1, called - cd->start_code);
5116 code += 1 + LINK_SIZE;
5117
5118 *code = OP_KET;
5119 PUT(code, 1, 2 + 2*LINK_SIZE);
5120 code += 1 + LINK_SIZE;
5121
5122 length_prevgroup = 3 + 3*LINK_SIZE;
5123 }
5124
5125 /* Can't determine a first byte now */
5126
5127 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5128 continue;
5129
5130
5131 /* ------------------------------------------------------------ */
5132 default: /* Other characters: check option setting */
5133 OTHER_CHAR_AFTER_QUERY:
5134 set = unset = 0;
5135 optset = &set;
5136
5137 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5138 {
5139 switch (*ptr++)
5140 {
5141 case CHAR_MINUS: optset = &unset; break;
5142
5143 case CHAR_J: /* Record that it changed in the external options */
5144 *optset |= PCRE_DUPNAMES;
5145 cd->external_flags |= PCRE_JCHANGED;
5146 break;
5147
5148 case CHAR_i: *optset |= PCRE_CASELESS; break;
5149 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5150 case CHAR_s: *optset |= PCRE_DOTALL; break;
5151 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5152 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5153 case CHAR_X: *optset |= PCRE_EXTRA; break;
5154
5155 default: *errorcodeptr = ERR12;
5156 ptr--; /* Correct the offset */
5157 goto FAILED;
5158 }
5159 }
5160
5161 /* Set up the changed option bits, but don't change anything yet. */
5162
5163 newoptions = (options | set) & (~unset);
5164
5165 /* If the options ended with ')' this is not the start of a nested
5166 group with option changes, so the options change at this level. If this
5167 item is right at the start of the pattern, the options can be
5168 abstracted and made external in the pre-compile phase, and ignored in
5169 the compile phase. This can be helpful when matching -- for instance in
5170 caseless checking of required bytes.
5171
5172 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5173 definitely *not* at the start of the pattern because something has been
5174 compiled. In the pre-compile phase, however, the code pointer can have
5175 that value after the start, because it gets reset as code is discarded
5176 during the pre-compile. However, this can happen only at top level - if
5177 we are within parentheses, the starting BRA will still be present. At
5178 any parenthesis level, the length value can be used to test if anything
5179 has been compiled at that level. Thus, a test for both these conditions
5180 is necessary to ensure we correctly detect the start of the pattern in
5181 both phases.
5182
5183 If we are not at the pattern start, compile code to change the ims
5184 options if this setting actually changes any of them, and reset the
5185 greedy defaults and the case value for firstbyte and reqbyte. */
5186
5187 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5188 {
5189 if (code == cd->start_code + 1 + LINK_SIZE &&
5190 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5191 {
5192 cd->external_options = newoptions;
5193 }
5194 else
5195 {
5196 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5197 {
5198 *code++ = OP_OPT;
5199 *code++ = newoptions & PCRE_IMS;
5200 }
5201 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5202 greedy_non_default = greedy_default ^ 1;
5203 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5204 }
5205
5206 /* Change options at this level, and pass them back for use
5207 in subsequent branches. When not at the start of the pattern, this
5208 information is also necessary so that a resetting item can be
5209 compiled at the end of a group (if we are in a group). */
5210
5211 *optionsptr = options = newoptions;
5212 previous = NULL; /* This item can't be repeated */
5213 continue; /* It is complete */
5214 }
5215
5216 /* If the options ended with ':' we are heading into a nested group
5217 with possible change of options. Such groups are non-capturing and are
5218 not assertions of any kind. All we need to do is skip over the ':';
5219 the newoptions value is handled below. */
5220
5221 bravalue = OP_BRA;
5222 ptr++;
5223 } /* End of switch for character following (? */
5224 } /* End of (? handling */
5225
5226 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5227 all unadorned brackets become non-capturing and behave like (?:...)
5228 brackets. */
5229
5230 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5231 {
5232 bravalue = OP_BRA;
5233 }
5234
5235 /* Else we have a capturing group. */
5236
5237 else
5238 {
5239 NUMBERED_GROUP:
5240 cd->bracount += 1;
5241 PUT2(code, 1+LINK_SIZE, cd->bracount);
5242 skipbytes = 2;
5243 }
5244
5245 /* Process nested bracketed regex. Assertions may not be repeated, but
5246 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5247 non-register variable in order to be able to pass its address because some
5248 compilers complain otherwise. Pass in a new setting for the ims options if
5249 they have changed. */
5250
5251 previous = (bravalue >= OP_ONCE)? code : NULL;
5252 *code = bravalue;
5253 tempcode = code;
5254 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5255 length_prevgroup = 0; /* Initialize for pre-compile phase */
5256
5257 if (!compile_regex(
5258 newoptions, /* The complete new option state */
5259 options & PCRE_IMS, /* The previous ims option state */
5260 &tempcode, /* Where to put code (updated) */
5261 &ptr, /* Input pointer (updated) */
5262 errorcodeptr, /* Where to put an error message */
5263 (bravalue == OP_ASSERTBACK ||
5264 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5265 reset_bracount, /* True if (?| group */
5266 skipbytes, /* Skip over bracket number */
5267 &subfirstbyte, /* For possible first char */
5268 &subreqbyte, /* For possible last char */
5269 bcptr, /* Current branch chain */
5270 cd, /* Tables block */
5271 (lengthptr == NULL)? NULL : /* Actual compile phase */
5272 &length_prevgroup /* Pre-compile phase */
5273 ))
5274 goto FAILED;
5275
5276 /* At the end of compiling, code is still pointing to the start of the
5277 group, while tempcode has been updated to point past the end of the group
5278 and any option resetting that may follow it. The pattern pointer (ptr)
5279 is on the bracket. */
5280
5281 /* If this is a conditional bracket, check that there are no more than
5282 two branches in the group, or just one if it's a DEFINE group. We do this
5283 in the real compile phase, not in the pre-pass, where the whole group may
5284 not be available. */
5285
5286 if (bravalue == OP_COND && lengthptr == NULL)
5287 {
5288 uschar *tc = code;
5289 int condcount = 0;
5290
5291 do {
5292 condcount++;
5293 tc += GET(tc,1);
5294 }
5295 while (*tc != OP_KET);
5296
5297 /* A DEFINE group is never obeyed inline (the "condition" is always
5298 false). It must have only one branch. */
5299
5300 if (code[LINK_SIZE+1] == OP_DEF)
5301 {
5302 if (condcount > 1)
5303 {
5304 *errorcodeptr = ERR54;
5305 goto FAILED;
5306 }
5307 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5308 }
5309
5310 /* A "normal" conditional group. If there is just one branch, we must not
5311 make use of its firstbyte or reqbyte, because this is equivalent to an
5312 empty second branch. */
5313
5314 else
5315 {
5316 if (condcount > 2)
5317 {
5318 *errorcodeptr = ERR27;
5319 goto FAILED;
5320 }
5321 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5322 }
5323 }
5324
5325 /* Error if hit end of pattern */
5326
5327 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5328 {
5329 *errorcodeptr = ERR14;
5330 goto FAILED;
5331 }
5332
5333 /* In the pre-compile phase, update the length by the length of the group,
5334 less the brackets at either end. Then reduce the compiled code to just a
5335 set of non-capturing brackets so that it doesn't use much memory if it is
5336 duplicated by a quantifier.*/
5337
5338 if (lengthptr != NULL)
5339 {
5340 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5341 {
5342 *errorcodeptr = ERR20;
5343 goto FAILED;
5344 }
5345 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5346 *code++ = OP_BRA;
5347 PUTINC(code, 0, 1 + LINK_SIZE);
5348 *code++ = OP_KET;
5349 PUTINC(code, 0, 1 + LINK_SIZE);
5350 break; /* No need to waste time with special character handling */
5351 }
5352
5353 /* Otherwise update the main code pointer to the end of the group. */
5354
5355 code = tempcode;
5356
5357 /* For a DEFINE group, required and first character settings are not
5358 relevant. */
5359
5360 if (bravalue == OP_DEF) break;
5361
5362 /* Handle updating of the required and first characters for other types of
5363 group. Update for normal brackets of all kinds, and conditions with two
5364 branches (see code above). If the bracket is followed by a quantifier with
5365 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5366 zerofirstbyte outside the main loop so that they can be accessed for the
5367 back off. */
5368
5369 zeroreqbyte = reqbyte;
5370 zerofirstbyte = firstbyte;
5371 groupsetfirstbyte = FALSE;
5372
5373 if (bravalue >= OP_ONCE)
5374 {
5375 /* If we have not yet set a firstbyte in this branch, take it from the
5376 subpattern, remembering that it was set here so that a repeat of more
5377 than one can replicate it as reqbyte if necessary. If the subpattern has
5378 no firstbyte, set "none" for the whole branch. In both cases, a zero
5379 repeat forces firstbyte to "none". */
5380
5381 if (firstbyte == REQ_UNSET)
5382 {
5383 if (subfirstbyte >= 0)
5384 {
5385 firstbyte = subfirstbyte;
5386 groupsetfirstbyte = TRUE;
5387 }
5388 else firstbyte = REQ_NONE;
5389 zerofirstbyte = REQ_NONE;
5390 }
5391
5392 /* If firstbyte was previously set, convert the subpattern's firstbyte
5393 into reqbyte if there wasn't one, using the vary flag that was in
5394 existence beforehand. */
5395
5396 else if (subfirstbyte >= 0 && subreqbyte < 0)
5397 subreqbyte = subfirstbyte | tempreqvary;
5398
5399 /* If the subpattern set a required byte (or set a first byte that isn't
5400 really the first byte - see above), set it. */
5401
5402 if (subreqbyte >= 0) reqbyte = subreqbyte;
5403 }
5404
5405 /* For a forward assertion, we take the reqbyte, if set. This can be
5406 helpful if the pattern that follows the assertion doesn't set a different
5407 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5408 for an assertion, however because it leads to incorrect effect for patterns
5409 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5410 of a firstbyte. This is overcome by a scan at the end if there's no
5411 firstbyte, looking for an asserted first char. */
5412
5413 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5414 break; /* End of processing '(' */
5415
5416
5417 /* ===================================================================*/
5418 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5419 are arranged to be the negation of the corresponding OP_values. For the
5420 back references, the values are ESC_REF plus the reference number. Only
5421 back references and those types that consume a character may be repeated.
5422 We can test for values between ESC_b and ESC_Z for the latter; this may
5423 have to change if any new ones are ever created. */
5424
5425 case CHAR_BACKSLASH:
5426 tempptr = ptr;
5427 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5428 if (*errorcodeptr != 0) goto FAILED;
5429
5430 if (c < 0)
5431 {
5432 if (-c == ESC_Q) /* Handle start of quoted string */
5433 {
5434 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5435 ptr += 2; /* avoid empty string */
5436 else inescq = TRUE;
5437 continue;
5438 }
5439
5440 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5441
5442 /* For metasequences that actually match a character, we disable the
5443 setting of a first character if it hasn't already been set. */
5444
5445 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5446 firstbyte = REQ_NONE;
5447
5448 /* Set values to reset to if this is followed by a zero repeat. */
5449
5450 zerofirstbyte = firstbyte;
5451 zeroreqbyte = reqbyte;
5452
5453 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5454 is a subroutine call by number (Oniguruma syntax). In fact, the value
5455 -ESC_g is returned only for these cases. So we don't need to check for <
5456 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5457 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5458 that is a synonym for a named back reference). */
5459
5460 if (-c == ESC_g)
5461 {
5462 const uschar *p;
5463 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5464 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5465 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5466
5467 /* These two statements stop the compiler for warning about possibly
5468 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5469 fact, because we actually check for a number below, the paths that
5470 would actually be in error are never taken. */
5471
5472 skipbytes = 0;
5473 reset_bracount = FALSE;
5474
5475 /* Test for a name */
5476
5477 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5478 {
5479 BOOL isnumber = TRUE;
5480 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5481 {
5482 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5483 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5484 }
5485 if (*p != terminator)
5486 {
5487 *errorcodeptr = ERR57;
5488 break;
5489 }
5490 if (isnumber)
5491 {
5492 ptr++;
5493 goto HANDLE_NUMERICAL_RECURSION;
5494 }
5495 is_recurse = TRUE;
5496 goto NAMED_REF_OR_RECURSE;
5497 }
5498
5499 /* Test a signed number in angle brackets or quotes. */
5500
5501 p = ptr + 2;
5502 while ((digitab[*p] & ctype_digit) != 0) p++;
5503 if (*p != terminator)
5504 {
5505 *errorcodeptr = ERR57;
5506 break;
5507 }
5508 ptr++;
5509 goto HANDLE_NUMERICAL_RECURSION;
5510 }
5511
5512 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5513 We also support \k{name} (.NET syntax) */
5514
5515 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5516 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5517 {
5518 is_recurse = FALSE;
5519 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5520 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5521 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5522 goto NAMED_REF_OR_RECURSE;
5523 }
5524
5525 /* Back references are handled specially; must disable firstbyte if
5526 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5527 ':' later. */
5528
5529 if (-c >= ESC_REF)
5530 {
5531 recno = -c - ESC_REF;
5532
5533 HANDLE_REFERENCE: /* Come here from named backref handling */
5534 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5535 previous = code;
5536 *code++ = OP_REF;
5537 PUT2INC(code, 0, recno);
5538 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5539 if (recno > cd->top_backref) cd->top_backref = recno;
5540 }
5541
5542 /* So are Unicode property matches, if supported. */
5543
5544 #ifdef SUPPORT_UCP
5545 else if (-c == ESC_P || -c == ESC_p)
5546 {
5547 BOOL negated;
5548 int pdata;
5549 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5550 if (ptype < 0) goto FAILED;
5551 previous = code;
5552 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5553 *code++ = ptype;
5554 *code++ = pdata;
5555 }
5556 #else
5557
5558 /* If Unicode properties are not supported, \X, \P, and \p are not
5559 allowed. */
5560
5561 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5562 {
5563 *errorcodeptr = ERR45;
5564 goto FAILED;
5565 }
5566 #endif
5567
5568 /* For the rest (including \X when Unicode properties are supported), we
5569 can obtain the OP value by negating the escape value. */
5570
5571 else
5572 {
5573 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5574 *code++ = -c;
5575 }
5576 continue;
5577 }
5578
5579 /* We have a data character whose value is in c. In UTF-8 mode it may have
5580 a value > 127. We set its representation in the length/buffer, and then
5581 handle it as a data character. */
5582
5583 #ifdef SUPPORT_UTF8
5584 if (utf8 && c > 127)
5585 mclength = _pcre_ord2utf8(c, mcbuffer);
5586 else
5587 #endif
5588
5589 {
5590 mcbuffer[0] = c;
5591 mclength = 1;
5592 }
5593 goto ONE_CHAR;
5594
5595
5596 /* ===================================================================*/
5597 /* Handle a literal character. It is guaranteed not to be whitespace or #
5598 when the extended flag is set. If we are in UTF-8 mode, it may be a
5599 multi-byte literal character. */
5600
5601 default:
5602 NORMAL_CHAR:
5603 mclength = 1;
5604 mcbuffer[0] = c;
5605
5606 #ifdef SUPPORT_UTF8
5607 if (utf8 && c >= 0xc0)
5608 {
5609 while ((ptr[1] & 0xc0) == 0x80)
5610 mcbuffer[mclength++] = *(++ptr);
5611 }
5612 #endif
5613
5614 /* At this point we have the character's bytes in mcbuffer, and the length
5615 in mclength. When not in UTF-8 mode, the length is always 1. */
5616
5617 ONE_CHAR:
5618 previous = code;
5619 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5620 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5621
5622 /* Remember if \r or \n were seen */
5623
5624 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5625 cd->external_flags |= PCRE_HASCRORLF;
5626
5627 /* Set the first and required bytes appropriately. If no previous first
5628 byte, set it from this character, but revert to none on a zero repeat.
5629 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5630 repeat. */
5631
5632 if (firstbyte == REQ_UNSET)
5633 {
5634 zerofirstbyte = REQ_NONE;
5635 zeroreqbyte = reqbyte;
5636
5637 /* If the character is more than one byte long, we can set firstbyte
5638 only if it is not to be matched caselessly. */
5639
5640 if (mclength == 1 || req_caseopt == 0)
5641 {
5642 firstbyte = mcbuffer[0] | req_caseopt;
5643 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5644 }
5645 else firstbyte = reqbyte = REQ_NONE;
5646 }
5647
5648 /* firstbyte was previously set; we can set reqbyte only the length is
5649 1 or the matching is caseful. */
5650
5651 else
5652 {
5653 zerofirstbyte = firstbyte;
5654 zeroreqbyte = reqbyte;
5655 if (mclength == 1 || req_caseopt == 0)
5656 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5657 }
5658
5659 break; /* End of literal character handling */
5660 }
5661 } /* end of big loop */
5662
5663
5664 /* Control never reaches here by falling through, only by a goto for all the
5665 error states. Pass back the position in the pattern so that it can be displayed
5666 to the user for diagnosing the error. */
5667
5668 FAILED:
5669 *ptrptr = ptr;
5670 return FALSE;
5671 }
5672
5673
5674
5675
5676 /*************************************************
5677 * Compile sequence of alternatives *
5678 *************************************************/
5679
5680 /* On entry, ptr is pointing past the bracket character, but on return it
5681 points to the closing bracket, or vertical bar, or end of string. The code
5682 variable is pointing at the byte into which the BRA operator has been stored.
5683 If the ims options are changed at the start (for a (?ims: group) or during any
5684 branch, we need to insert an OP_OPT item at the start of every following branch
5685 to ensure they get set correctly at run time, and also pass the new options
5686 into every subsequent branch compile.
5687
5688 This function is used during the pre-compile phase when we are trying to find
5689 out the amount of memory needed, as well as during the real compile phase. The
5690 value of lengthptr distinguishes the two phases.
5691
5692 Arguments:
5693 options option bits, including any changes for this subpattern
5694 oldims previous settings of ims option bits
5695 codeptr -> the address of the current code pointer
5696 ptrptr -> the address of the current pattern pointer
5697 errorcodeptr -> pointer to error code variable
5698 lookbehind TRUE if this is a lookbehind assertion
5699 reset_bracount TRUE to reset the count for each branch
5700 skipbytes skip this many bytes at start (for brackets and OP_COND)
5701 firstbyteptr place to put the first required character, or a negative number
5702 reqbyteptr place to put the last required character, or a negative number
5703 bcptr pointer to the chain of currently open branches
5704 cd points to the data block with tables pointers etc.
5705 lengthptr NULL during the real compile phase
5706 points to length accumulator during pre-compile phase
5707
5708 Returns: TRUE on success
5709 */
5710
5711 static BOOL
5712 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5713 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5714 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5715 int *lengthptr)
5716 {
5717 const uschar *ptr = *ptrptr;
5718 uschar *code = *codeptr;
5719 uschar *last_branch = code;
5720 uschar *start_bracket = code;
5721 uschar *reverse_count = NULL;
5722 open_capitem capitem;
5723 int capnumber = 0;
5724 int firstbyte, reqbyte;
5725 int branchfirstbyte, branchreqbyte;
5726 int length;
5727 int orig_bracount;
5728 int max_bracount;
5729 branch_chain bc;
5730
5731 bc.outer = bcptr;
5732 bc.current = code;
5733
5734 firstbyte = reqbyte = REQ_UNSET;
5735
5736 /* Accumulate the length for use in the pre-compile phase. Start with the
5737 length of the BRA and KET and any extra bytes that are required at the
5738 beginning. We accumulate in a local variable to save frequent testing of
5739 lenthptr for NULL. We cannot do this by looking at the value of code at the
5740 start and end of each alternative, because compiled items are discarded during
5741 the pre-compile phase so that the work space is not exceeded. */
5742
5743 length = 2 + 2*LINK_SIZE + skipbytes;
5744
5745 /* WARNING: If the above line is changed for any reason, you must also change
5746 the code that abstracts option settings at the start of the pattern and makes
5747 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5748 pre-compile phase to find out whether anything has yet been compiled or not. */
5749
5750 /* If this is a capturing subpattern, add to the chain of open capturing items
5751 so that we can detect them if (*ACCEPT) is encountered. */
5752
5753 if (*code == OP_CBRA)
5754 {
5755 capnumber = GET2(code, 1 + LINK_SIZE);
5756 capitem.number = capnumber;
5757 capitem.next = cd->open_caps;
5758 cd->open_caps = &capitem;
5759 }
5760
5761 /* Offset is set zero to mark that this bracket is still open */
5762
5763 PUT(code, 1, 0);
5764 code += 1 + LINK_SIZE + skipbytes;
5765
5766 /* Loop for each alternative branch */
5767
5768 orig_bracount = max_bracount = cd->bracount;
5769 for (;;)
5770 {
5771 /* For a (?| group, reset the capturing bracket count so that each branch
5772 uses the same numbers. */
5773
5774 if (reset_bracount) cd->bracount = orig_bracount;
5775
5776 /* Handle a change of ims options at the start of the branch */
5777
5778 if ((options & PCRE_IMS) != oldims)
5779 {
5780 *code++ = OP_OPT;
5781 *code++ = options & PCRE_IMS;
5782 length += 2;
5783 }
5784
5785 /* Set up dummy OP_REVERSE if lookbehind assertion */
5786
5787 if (lookbehind)
5788 {
5789 *code++ = OP_REVERSE;
5790 reverse_count = code;
5791 PUTINC(code, 0, 0);
5792 length += 1 + LINK_SIZE;
5793 }
5794
5795 /* Now compile the branch; in the pre-compile phase its length gets added
5796 into the length. */
5797
5798 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5799 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5800 {
5801 *ptrptr = ptr;
5802 return FALSE;
5803 }
5804
5805 /* Keep the highest bracket count in case (?| was used and some branch
5806 has fewer than the rest. */
5807
5808 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5809
5810 /* In the real compile phase, there is some post-processing to be done. */
5811
5812 if (lengthptr == NULL)
5813 {
5814 /* If this is the first branch, the firstbyte and reqbyte values for the
5815 branch become the values for the regex. */
5816
5817 if (*last_branch != OP_ALT)
5818 {
5819 firstbyte = branchfirstbyte;
5820 reqbyte = branchreqbyte;
5821 }
5822
5823 /* If this is not the first branch, the first char and reqbyte have to
5824 match the values from all the previous branches, except that if the
5825 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5826 and we set REQ_VARY for the regex. */
5827
5828 else
5829 {
5830 /* If we previously had a firstbyte, but it doesn't match the new branch,
5831 we have to abandon the firstbyte for the regex, but if there was
5832 previously no reqbyte, it takes on the value of the old firstbyte. */
5833
5834 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5835 {
5836 if (reqbyte < 0) reqbyte = firstbyte;
5837 firstbyte = REQ_NONE;
5838 }
5839
5840 /* If we (now or from before) have no firstbyte, a firstbyte from the
5841 branch becomes a reqbyte if there isn't a branch reqbyte. */
5842
5843 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5844 branchreqbyte = branchfirstbyte;
5845
5846 /* Now ensure that the reqbytes match */
5847
5848 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5849 reqbyte = REQ_NONE;
5850 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5851 }
5852
5853 /* If lookbehind, check that this branch matches a fixed-length string, and
5854 put the length into the OP_REVERSE item. Temporarily mark the end of the
5855 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
5856 because there may be forward references that we can't check here. Set a
5857 flag to cause another lookbehind check at the end. Why not do it all at the
5858 end? Because common, erroneous checks are picked up here and the offset of
5859 the problem can be shown. */
5860
5861 if (lookbehind)
5862 {
5863 int fixed_length;
5864 *code = OP_END;
5865 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
5866 DPRINTF(("fixed length = %d\n", fixed_length));
5867 if (fixed_length == -3)
5868 {
5869 cd->check_lookbehind = TRUE;
5870 }
5871 else if (fixed_length < 0)
5872 {
5873 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5874 *ptrptr = ptr;
5875 return FALSE;
5876 }
5877 else { PUT(reverse_count, 0, fixed_length); }
5878 }
5879 }
5880
5881 /* Reached end of expression, either ')' or end of pattern. In the real
5882 compile phase, go back through the alternative branches and reverse the chain
5883 of offsets, with the field in the BRA item now becoming an offset to the
5884 first alternative. If there are no alternatives, it points to the end of the
5885 group. The length in the terminating ket is always the length of the whole
5886 bracketed item. If any of the ims options were changed inside the group,
5887 compile a resetting op-code following, except at the very end of the pattern.
5888 Return leaving the pointer at the terminating char. */
5889
5890 if (*ptr != CHAR_VERTICAL_LINE)
5891 {
5892 if (lengthptr == NULL)
5893 {
5894 int branch_length = code - last_branch;
5895 do
5896 {
5897 int prev_length = GET(last_branch, 1);
5898 PUT(last_branch, 1, branch_length);
5899 branch_length = prev_length;
5900 last_branch -= branch_length;
5901 }
5902 while (branch_length > 0);
5903 }
5904
5905 /* If it was a capturing subpattern, remove it from the chain. */
5906
5907 if (capnumber > 0) cd->open_caps = cd->open_caps->next;
5908
5909 /* Fill in the ket */
5910
5911 *code = OP_KET;
5912 PUT(code, 1, code - start_bracket);
5913 code += 1 + LINK_SIZE;
5914
5915 /* Resetting option if needed */
5916
5917 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
5918 {
5919 *code++ = OP_OPT;
5920 *code++ = oldims;
5921 length += 2;
5922 }
5923
5924 /* Retain the highest bracket number, in case resetting was used. */
5925
5926 cd->bracount = max_bracount;
5927
5928 /* Set values to pass back */
5929
5930 *codeptr = code;
5931 *ptrptr = ptr;
5932 *firstbyteptr = firstbyte;
5933 *reqbyteptr = reqbyte;
5934 if (lengthptr != NULL)
5935 {
5936 if (OFLOW_MAX - *lengthptr < length)
5937 {
5938 *errorcodeptr = ERR20;
5939 return FALSE;
5940 }
5941 *lengthptr += length;
5942 }
5943 return TRUE;
5944 }
5945
5946 /* Another branch follows. In the pre-compile phase, we can move the code
5947 pointer back to where it was for the start of the first branch. (That is,
5948 pretend that each branch is the only one.)
5949
5950 In the real compile phase, insert an ALT node. Its length field points back
5951 to the previous branch while the bracket remains open. At the end the chain
5952 is reversed. It's done like this so that the start of the bracket has a
5953 zero offset until it is closed, making it possible to detect recursion. */
5954
5955 if (lengthptr != NULL)
5956 {
5957 code = *codeptr + 1 + LINK_SIZE + skipbytes;
5958 length += 1 + LINK_SIZE;
5959 }
5960 else
5961 {
5962 *code = OP_ALT;
5963 PUT(code, 1, code - last_branch);
5964 bc.current = last_branch = code;
5965 code += 1 + LINK_SIZE;
5966 }
5967
5968 ptr++;
5969 }
5970 /* Control never reaches here */
5971 }
5972
5973
5974
5975
5976 /*************************************************
5977 * Check for anchored expression *
5978 *************************************************/
5979
5980 /* Try to find out if this is an anchored regular expression. Consider each
5981 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
5982 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
5983 it's anchored. However, if this is a multiline pattern, then only OP_SOD
5984 counts, since OP_CIRC can match in the middle.
5985
5986 We can also consider a regex to be anchored if OP_SOM starts all its branches.
5987 This is the code for \G, which means "match at start of match position, taking
5988 into account the match offset".
5989
5990 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
5991 because that will try the rest of the pattern at all possible matching points,
5992 so there is no point trying again.... er ....
5993
5994 .... except when the .* appears inside capturing parentheses, and there is a
5995 subsequent back reference to those parentheses. We haven't enough information
5996 to catch that case precisely.
5997
5998 At first, the best we could do was to detect when .* was in capturing brackets
5999 and the highest back reference was greater than or equal to that level.
6000 However, by keeping a bitmap of the first 31 back references, we can catch some
6001 of the more common cases more precisely.
6002
6003 Arguments:
6004 code points to start of expression (the bracket)
6005 options points to the options setting
6006 bracket_map a bitmap of which brackets we are inside while testing; this
6007 handles up to substring 31; after that we just have to take
6008 the less precise approach
6009 backref_map the back reference bitmap
6010
6011 Returns: TRUE or FALSE
6012 */
6013
6014 static BOOL
6015 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6016 unsigned int backref_map)
6017 {
6018 do {
6019 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6020 options, PCRE_MULTILINE, FALSE);
6021 register int op = *scode;
6022
6023 /* Non-capturing brackets */
6024
6025 if (op == OP_BRA)
6026 {
6027 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6028 }
6029
6030 /* Capturing brackets */
6031
6032 else if (op == OP_CBRA)
6033 {
6034 int n = GET2(scode, 1+LINK_SIZE);
6035 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6036 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6037 }
6038
6039 /* Other brackets */
6040
6041 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6042 {
6043 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6044 }
6045
6046 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6047 it isn't in brackets that are or may be referenced. */
6048
6049 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6050 op == OP_TYPEPOSSTAR))
6051 {
6052 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6053 return FALSE;
6054 }
6055
6056 /* Check for explicit anchoring */
6057
6058 else if (op != OP_SOD && op != OP_SOM &&
6059 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6060 return FALSE;
6061 code += GET(code, 1);
6062 }
6063 while (*code == OP_ALT); /* Loop for each alternative */
6064 return TRUE;
6065 }
6066
6067
6068
6069 /*************************************************
6070 * Check for starting with ^ or .* *
6071 *************************************************/
6072
6073 /* This is called to find out if every branch starts with ^ or .* so that
6074 "first char" processing can be done to speed things up in multiline
6075 matching and for non-DOTALL patterns that start with .* (which must start at
6076 the beginning or after \n). As in the case of is_anchored() (see above), we
6077 have to take account of back references to capturing brackets that contain .*
6078 because in that case we can't make the assumption.
6079
6080 Arguments:
6081 code points to start of expression (the bracket)
6082 bracket_map a bitmap of which brackets we are inside while testing; this
6083 handles up to substring 31; after that we just have to take
6084 the less precise approach
6085 backref_map the back reference bitmap
6086
6087 Returns: TRUE or FALSE
6088 */
6089
6090 static BOOL
6091 is_startline(const uschar *code, unsigned int bracket_map,
6092 unsigned int backref_map)
6093 {
6094 do {
6095 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6096 NULL, 0, FALSE);
6097 register int op = *scode;
6098
6099 /* If we are at the start of a conditional assertion group, *both* the
6100 conditional assertion *and* what follows the condition must satisfy the test
6101 for start of line. Other kinds of condition fail. Note that there may be an
6102 auto-callout at the start of a condition. */
6103
6104 if (op == OP_COND)
6105 {
6106 scode += 1 + LINK_SIZE;
6107 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6108 switch (*scode)
6109 {
6110 case OP_CREF:
6111 case OP_RREF:
6112 case OP_DEF:
6113 return FALSE;
6114
6115 default: /* Assertion */
6116 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6117 do scode += GET(scode, 1); while (*scode == OP_ALT);
6118 scode += 1 + LINK_SIZE;
6119 break;
6120 }
6121 scode = first_significant_code(scode, NULL, 0, FALSE);
6122 op = *scode;
6123 }
6124
6125 /* Non-capturing brackets */
6126
6127 if (op == OP_BRA)
6128 {
6129 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6130 }
6131
6132 /* Capturing brackets */
6133
6134 else if (op == OP_CBRA)
6135 {
6136 int n = GET2(scode, 1+LINK_SIZE);
6137 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6138 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6139 }
6140
6141 /* Other brackets */
6142
6143 else if (op == OP_ASSERT || op == OP_ONCE)
6144 {
6145 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6146 }
6147
6148 /* .* means "start at start or after \n" if it isn't in brackets that
6149 may be referenced. */
6150
6151 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6152 {
6153 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6154 }
6155
6156 /* Check for explicit circumflex */
6157
6158 else if (op != OP_CIRC) return FALSE;
6159
6160 /* Move on to the next alternative */
6161
6162 code += GET(code, 1);
6163 }
6164 while (*code == OP_ALT); /* Loop for each alternative */
6165 return TRUE;
6166 }
6167
6168
6169
6170 /*************************************************
6171 * Check for asserted fixed first char *
6172 *************************************************/
6173
6174 /* During compilation, the "first char" settings from forward assertions are
6175 discarded, because they can cause conflicts with actual literals that follow.
6176 However, if we end up without a first char setting for an unanchored pattern,
6177 it is worth scanning the regex to see if there is an initial asserted first
6178 char. If all branches start with the same asserted char, or with a bracket all
6179 of whose alternatives start with the same asserted char (recurse ad lib), then
6180 we return that char, otherwise -1.
6181
6182 Arguments:
6183 code points to start of expression (the bracket)
6184 options pointer to the options (used to check casing changes)
6185 inassert TRUE if in an assertion
6186
6187 Returns: -1 or the fixed first char
6188 */
6189
6190 static int
6191 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6192 {
6193 register int c = -1;
6194 do {
6195 int d;
6196 const uschar *scode =
6197 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6198 register int op = *scode;
6199
6200 switch(op)
6201 {
6202 default:
6203 return -1;
6204
6205 case OP_BRA:
6206 case OP_CBRA:
6207 case OP_ASSERT:
6208 case OP_ONCE:
6209 case OP_COND:
6210 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6211 return -1;
6212 if (c < 0) c = d; else if (c != d) return -1;
6213 break;
6214
6215 case OP_EXACT: /* Fall through */
6216 scode += 2;
6217
6218 case OP_CHAR:
6219 case OP_CHARNC:
6220 case OP_PLUS:
6221 case OP_MINPLUS:
6222 case OP_POSPLUS:
6223 if (!inassert) return -1;
6224 if (c < 0)
6225 {
6226 c = scode[1];
6227 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6228 }
6229 else if (c != scode[1]) return -1;
6230 break;
6231 }
6232
6233 code += GET(code, 1);
6234 }
6235 while (*code == OP_ALT);
6236 return c;
6237 }
6238
6239
6240
6241 /*************************************************
6242 * Compile a Regular Expression *
6243 *************************************************/
6244
6245 /* This function takes a string and returns a pointer to a block of store
6246 holding a compiled version of the expression. The original API for this
6247 function had no error code return variable; it is retained for backwards
6248 compatibility. The new function is given a new name.
6249
6250 Arguments:
6251 pattern the regular expression
6252 options various option bits
6253 errorcodeptr pointer to error code variable (pcre_compile2() only)
6254 can be NULL if you don't want a code value
6255 errorptr pointer to pointer to error text
6256 erroroffset ptr offset in pattern where error was detected
6257 tables pointer to character tables or NULL
6258
6259 Returns: pointer to compiled data block, or NULL on error,
6260 with errorptr and erroroffset set
6261 */
6262
6263 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6264 pcre_compile(const char *pattern, int options, const char **errorptr,
6265 int *erroroffset, const unsigned char *tables)
6266 {
6267 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6268 }
6269
6270
6271 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6272 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6273 const char **errorptr, int *erroroffset, const unsigned char *tables)
6274 {
6275 real_pcre *re;
6276 int length = 1; /* For final END opcode */
6277 int firstbyte, reqbyte, newline;
6278 int errorcode = 0;
6279 int skipatstart = 0;
6280 BOOL utf8 = (options & PCRE_UTF8) != 0;
6281 size_t size;
6282 uschar *code;
6283 const uschar *codestart;
6284 const uschar *ptr;
6285 compile_data compile_block;
6286 compile_data *cd = &compile_block;
6287
6288 /* This space is used for "compiling" into during the first phase, when we are
6289 computing the amount of memory that is needed. Compiled items are thrown away
6290 as soon as possible, so that a fairly large buffer should be sufficient for
6291 this purpose. The same space is used in the second phase for remembering where
6292 to fill in forward references to subpatterns. */
6293
6294 uschar cworkspace[COMPILE_WORK_SIZE];
6295
6296 /* Set this early so that early errors get offset 0. */
6297
6298 ptr = (const uschar *)pattern;
6299
6300 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6301 can do is just return NULL, but we can set a code value if there is a code
6302 pointer. */
6303
6304 if (errorptr == NULL)
6305 {
6306 if (errorcodeptr != NULL) *errorcodeptr = 99;
6307 return NULL;
6308 }
6309
6310 *errorptr = NULL;
6311 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6312
6313 /* However, we can give a message for this error */
6314
6315 if (erroroffset == NULL)
6316 {
6317 errorcode = ERR16;
6318 goto PCRE_EARLY_ERROR_RETURN2;
6319 }
6320
6321 *erroroffset = 0;
6322
6323 /* Set up pointers to the individual character tables */
6324
6325 if (tables == NULL) tables = _pcre_default_tables;
6326 cd->lcc = tables + lcc_offset;
6327 cd->fcc = tables + fcc_offset;
6328 cd->cbits = tables + cbits_offset;
6329 cd->ctypes = tables + ctypes_offset;
6330
6331 /* Check that all undefined public option bits are zero */
6332
6333 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6334 {
6335 errorcode = ERR17;
6336 goto PCRE_EARLY_ERROR_RETURN;
6337 }
6338
6339 /* Check for global one-time settings at the start of the pattern, and remember
6340 the offset for later. */
6341
6342 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6343 ptr[skipatstart+1] == CHAR_ASTERISK)
6344 {
6345 int newnl = 0;
6346 int newbsr = 0;
6347
6348 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6349 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6350
6351 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6352 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6353 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6354 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6355 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6356 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6357 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6358 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6359 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6360 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6361
6362 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6363 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6364 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6365 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6366
6367 if (newnl != 0)
6368 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6369 else if (newbsr != 0)
6370 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6371 else break;
6372 }
6373
6374 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6375
6376 #ifdef SUPPORT_UTF8
6377 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6378 (*erroroffset = _pcre_valid_utf8((uschar *)pattern, -1)) >= 0)
6379 {
6380 errorcode = ERR44;
6381 goto PCRE_EARLY_ERROR_RETURN2;
6382 }
6383 #else
6384 if (utf8)
6385 {
6386 errorcode = ERR32;
6387 goto PCRE_EARLY_ERROR_RETURN;
6388 }
6389 #endif
6390
6391 /* Check validity of \R options. */
6392
6393 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6394 {
6395 case 0:
6396 case PCRE_BSR_ANYCRLF:
6397 case PCRE_BSR_UNICODE:
6398 break;
6399 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6400 }
6401
6402 /* Handle different types of newline. The three bits give seven cases. The
6403 current code allows for fixed one- or two-byte sequences, plus "any" and
6404 "anycrlf". */
6405
6406 switch (options & PCRE_NEWLINE_BITS)
6407 {
6408 case 0: newline = NEWLINE; break; /* Build-time default */
6409 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6410 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6411 case PCRE_NEWLINE_CR+
6412 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6413 case PCRE_NEWLINE_ANY: newline = -1; break;
6414 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6415 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6416 }
6417
6418 if (newline == -2)
6419 {
6420 cd->nltype = NLTYPE_ANYCRLF;
6421 }
6422 else if (newline < 0)
6423 {
6424 cd->nltype = NLTYPE_ANY;
6425 }
6426 else
6427 {
6428 cd->nltype = NLTYPE_FIXED;
6429 if (newline > 255)
6430 {
6431 cd->nllen = 2;
6432 cd->nl[0] = (newline >> 8) & 255;
6433 cd->nl[1] = newline & 255;
6434 }
6435 else
6436 {
6437 cd->nllen = 1;
6438 cd->nl[0] = newline;
6439 }
6440 }
6441
6442 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6443 references to help in deciding whether (.*) can be treated as anchored or not.
6444 */
6445
6446 cd->top_backref = 0;
6447 cd->backref_map = 0;
6448
6449 /* Reflect pattern for debugging output */
6450
6451 DPRINTF(("------------------------------------------------------------------\n"));
6452 DPRINTF(("%s\n", pattern));
6453
6454 /* Pretend to compile the pattern while actually just accumulating the length
6455 of memory required. This behaviour is triggered by passing a non-NULL final
6456 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6457 to compile parts of the pattern into; the compiled code is discarded when it is
6458 no longer needed, so hopefully this workspace will never overflow, though there
6459 is a test for its doing so. */
6460
6461 cd->bracount = cd->final_bracount = 0;
6462 cd->names_found = 0;
6463 cd->name_entry_size = 0;
6464 cd->name_table = NULL;
6465 cd->start_workspace = cworkspace;
6466 cd->start_code = cworkspace;
6467 cd->hwm = cworkspace;
6468 cd->start_pattern = (const uschar *)pattern;
6469 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6470 cd->req_varyopt = 0;
6471 cd->external_options = options;
6472 cd->external_flags = 0;
6473 cd->open_caps = NULL;
6474
6475 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6476 don't need to look at the result of the function here. The initial options have
6477 been put into the cd block so that they can be changed if an option setting is
6478 found within the regex right at the beginning. Bringing initial option settings
6479 outside can help speed up starting point checks. */
6480
6481 ptr += skipatstart;
6482 code = cworkspace;
6483 *code = OP_BRA;
6484 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6485 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6486 &length);
6487 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6488
6489 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6490 cd->hwm - cworkspace));
6491
6492 if (length > MAX_PATTERN_SIZE)
6493 {
6494 errorcode = ERR20;
6495 goto PCRE_EARLY_ERROR_RETURN;
6496 }
6497
6498 /* Compute the size of data block needed and get it, either from malloc or
6499 externally provided function. Integer overflow should no longer be possible
6500 because nowadays we limit the maximum value of cd->names_found and
6501 cd->name_entry_size. */
6502
6503 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6504 re = (real_pcre *)(pcre_malloc)(size);
6505
6506 if (re == NULL)
6507 {
6508 errorcode = ERR21;
6509 goto PCRE_EARLY_ERROR_RETURN;
6510 }
6511
6512 /* Put in the magic number, and save the sizes, initial options, internal
6513 flags, and character table pointer. NULL is used for the default character
6514 tables. The nullpad field is at the end; it's there to help in the case when a
6515 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6516 pointers. */
6517
6518 re->magic_number = MAGIC_NUMBER;
6519 re->size = size;
6520 re->options = cd->external_options;
6521 re->flags = cd->external_flags;
6522 re->dummy1 = 0;
6523 re->first_byte = 0;
6524 re->req_byte = 0;
6525 re->name_table_offset = sizeof(real_pcre);
6526 re->name_entry_size = cd->name_entry_size;
6527 re->name_count = cd->names_found;
6528 re->ref_count = 0;
6529 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6530 re->nullpad = NULL;
6531
6532 /* The starting points of the name/number translation table and of the code are
6533 passed around in the compile data block. The start/end pattern and initial
6534 options are already set from the pre-compile phase, as is the name_entry_size
6535 field. Reset the bracket count and the names_found field. Also reset the hwm
6536 field; this time it's used for remembering forward references to subpatterns.
6537 */
6538
6539 cd->final_bracount = cd->bracount; /* Save for checking forward references */
6540 cd->bracount = 0;
6541 cd->names_found = 0;
6542 cd->name_table = (uschar *)re + re->name_table_offset;
6543 codestart = cd->name_table + re->name_entry_size * re->name_count;
6544 cd->start_code = codestart;
6545 cd->hwm = cworkspace;
6546 cd->req_varyopt = 0;
6547 cd->had_accept = FALSE;
6548 cd->check_lookbehind = FALSE;
6549 cd->open_caps = NULL;
6550
6551 /* Set up a starting, non-extracting bracket, then compile the expression. On
6552 error, errorcode will be set non-zero, so we don't need to look at the result
6553 of the function here. */
6554
6555 ptr = (const uschar *)pattern + skipatstart;
6556 code = (uschar *)codestart;
6557 *code = OP_BRA;
6558 (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
6559 &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
6560 re->top_bracket = cd->bracount;
6561 re->top_backref = cd->top_backref;
6562 re->flags = cd->external_flags;
6563
6564 if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
6565
6566 /* If not reached end of pattern on success, there's an excess bracket. */
6567
6568 if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
6569
6570 /* Fill in the terminating state and check for disastrous overflow, but
6571 if debugging, leave the test till after things are printed out. */
6572
6573 *code++ = OP_END;
6574
6575 #ifndef DEBUG
6576 if (code - codestart > length) errorcode = ERR23;
6577 #endif
6578
6579 /* Fill in any forward references that are required. */
6580
6581 while (errorcode == 0 && cd->hwm > cworkspace)
6582 {
6583 int offset, recno;
6584 const uschar *groupptr;
6585 cd->hwm -= LINK_SIZE;
6586 offset = GET(cd->hwm, 0);
6587 recno = GET(codestart, offset);
6588 groupptr = _pcre_find_bracket(codestart, utf8, recno);
6589 if (groupptr == NULL) errorcode = ERR53;
6590 else PUT(((uschar *)codestart), offset, groupptr - codestart);
6591 }
6592
6593 /* Give an error if there's back reference to a non-existent capturing
6594 subpattern. */
6595
6596 if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
6597
6598 /* If there were any lookbehind assertions that contained OP_RECURSE
6599 (recursions or subroutine calls), a flag is set for them to be checked here,
6600 because they may contain forward references. Actual recursions can't be fixed
6601 length, but subroutine calls can. It is done like this so that those without
6602 OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
6603 exceptional ones forgo this. We scan the pattern to check that they are fixed
6604 length, and set their lengths. */
6605
6606 if (cd->check_lookbehind)
6607 {
6608 uschar *cc = (uschar *)codestart;
6609
6610 /* Loop, searching for OP_REVERSE items, and process those that do not have
6611 their length set. (Actually, it will also re-process any that have a length
6612 of zero, but that is a pathological case, and it does no harm.) When we find
6613 one, we temporarily terminate the branch it is in while we scan it. */
6614
6615 for (cc = (uschar *)_pcre_find_bracket(codestart, utf8, -1);
6616 cc != NULL;
6617 cc = (uschar *)_pcre_find_bracket(cc, utf8, -1))
6618 {
6619 if (GET(cc, 1) == 0)
6620 {
6621 int fixed_length;
6622 uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
6623 int end_op = *be;
6624 *be = OP_END;
6625 fixed_length = find_fixedlength(cc, re->options, TRUE, cd);
6626 *be = end_op;
6627 DPRINTF(("fixed length = %d\n", fixed_length));
6628 if (fixed_length < 0)
6629 {
6630 errorcode = (fixed_length == -2)? ERR36 : ERR25;
6631 break;
6632 }
6633 PUT(cc, 1, fixed_length);
6634 }
6635 cc += 1 + LINK_SIZE;
6636 }
6637 }
6638
6639 /* Failed to compile, or error while post-processing */
6640
6641 if (errorcode != 0)
6642 {
6643 (pcre_free)(re);
6644 PCRE_EARLY_ERROR_RETURN:
6645 *erroroffset = ptr - (const uschar *)pattern;
6646 PCRE_EARLY_ERROR_RETURN2: