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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 499 - (show annotations) (download)
Wed Mar 3 20:09:39 2010 UTC (4 years, 4 months ago) by ph10
File MIME type: text/plain
File size: 222228 byte(s)
Add some checks for the eint vector size and the list of compile-time error 
texts.

1 /*************************************************
2 * Perl-Compatible Regular Expressions *
3 *************************************************/
4
5 /* PCRE is a library of functions to support regular expressions whose syntax
6 and semantics are as close as possible to those of the Perl 5 language.
7
8 Written by Philip Hazel
9 Copyright (c) 1997-2010 University of Cambridge
10
11 -----------------------------------------------------------------------------
12 Redistribution and use in source and binary forms, with or without
13 modification, are permitted provided that the following conditions are met:
14
15 * Redistributions of source code must retain the above copyright notice,
16 this list of conditions and the following disclaimer.
17
18 * Redistributions in binary form must reproduce the above copyright
19 notice, this list of conditions and the following disclaimer in the
20 documentation and/or other materials provided with the distribution.
21
22 * Neither the name of the University of Cambridge nor the names of its
23 contributors may be used to endorse or promote products derived from
24 this software without specific prior written permission.
25
26 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27 AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30 LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 POSSIBILITY OF SUCH DAMAGE.
37 -----------------------------------------------------------------------------
38 */
39
40
41 /* This module contains the external function pcre_compile(), along with
42 supporting internal functions that are not used by other modules. */
43
44
45 #ifdef HAVE_CONFIG_H
46 #include "config.h"
47 #endif
48
49 #define NLBLOCK cd /* Block containing newline information */
50 #define PSSTART start_pattern /* Field containing processed string start */
51 #define PSEND end_pattern /* Field containing processed string end */
52
53 #include "pcre_internal.h"
54
55
56 /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is
57 also used by pcretest. PCRE_DEBUG is not defined when building a production
58 library. */
59
60 #ifdef PCRE_DEBUG
61 #include "pcre_printint.src"
62 #endif
63
64
65 /* Macro for setting individual bits in class bitmaps. */
66
67 #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68
69 /* Maximum length value to check against when making sure that the integer that
70 holds the compiled pattern length does not overflow. We make it a bit less than
71 INT_MAX to allow for adding in group terminating bytes, so that we don't have
72 to check them every time. */
73
74 #define OFLOW_MAX (INT_MAX - 20)
75
76
77 /*************************************************
78 * Code parameters and static tables *
79 *************************************************/
80
81 /* This value specifies the size of stack workspace that is used during the
82 first pre-compile phase that determines how much memory is required. The regex
83 is partly compiled into this space, but the compiled parts are discarded as
84 soon as they can be, so that hopefully there will never be an overrun. The code
85 does, however, check for an overrun. The largest amount I've seen used is 218,
86 so this number is very generous.
87
88 The same workspace is used during the second, actual compile phase for
89 remembering forward references to groups so that they can be filled in at the
90 end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
91 is 4 there is plenty of room. */
92
93 #define COMPILE_WORK_SIZE (4096)
94
95
96 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
97 are simple data values; negative values are for special things like \d and so
98 on. Zero means further processing is needed (for things like \x), or the escape
99 is invalid. */
100
101 #ifndef EBCDIC
102
103 /* This is the "normal" table for ASCII systems or for EBCDIC systems running
104 in UTF-8 mode. */
105
106 static const short int escapes[] = {
107 0, 0,
108 0, 0,
109 0, 0,
110 0, 0,
111 0, 0,
112 CHAR_COLON, CHAR_SEMICOLON,
113 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
114 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
115 CHAR_COMMERCIAL_AT, -ESC_A,
116 -ESC_B, -ESC_C,
117 -ESC_D, -ESC_E,
118 0, -ESC_G,
119 -ESC_H, 0,
120 0, -ESC_K,
121 0, 0,
122 0, 0,
123 -ESC_P, -ESC_Q,
124 -ESC_R, -ESC_S,
125 0, 0,
126 -ESC_V, -ESC_W,
127 -ESC_X, 0,
128 -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
129 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
130 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
131 CHAR_GRAVE_ACCENT, 7,
132 -ESC_b, 0,
133 -ESC_d, ESC_e,
134 ESC_f, 0,
135 -ESC_h, 0,
136 0, -ESC_k,
137 0, 0,
138 ESC_n, 0,
139 -ESC_p, 0,
140 ESC_r, -ESC_s,
141 ESC_tee, 0,
142 -ESC_v, -ESC_w,
143 0, 0,
144 -ESC_z
145 };
146
147 #else
148
149 /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
150
151 static const short int escapes[] = {
152 /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
153 /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
154 /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
155 /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
156 /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
157 /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
158 /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
159 /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
160 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
161 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
162 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
163 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
164 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
165 /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
166 /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
167 /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
168 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
169 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
170 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
171 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
172 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
173 /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
174 /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
175 };
176 #endif
177
178
179 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
180 searched linearly. Put all the names into a single string, in order to reduce
181 the number of relocations when a shared library is dynamically linked. The
182 string is built from string macros so that it works in UTF-8 mode on EBCDIC
183 platforms. */
184
185 typedef struct verbitem {
186 int len;
187 int op;
188 } verbitem;
189
190 static const char verbnames[] =
191 STRING_ACCEPT0
192 STRING_COMMIT0
193 STRING_F0
194 STRING_FAIL0
195 STRING_PRUNE0
196 STRING_SKIP0
197 STRING_THEN;
198
199 static const verbitem verbs[] = {
200 { 6, OP_ACCEPT },
201 { 6, OP_COMMIT },
202 { 1, OP_FAIL },
203 { 4, OP_FAIL },
204 { 5, OP_PRUNE },
205 { 4, OP_SKIP },
206 { 4, OP_THEN }
207 };
208
209 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
210
211
212 /* Tables of names of POSIX character classes and their lengths. The names are
213 now all in a single string, to reduce the number of relocations when a shared
214 library is dynamically loaded. The list of lengths is terminated by a zero
215 length entry. The first three must be alpha, lower, upper, as this is assumed
216 for handling case independence. */
217
218 static const char posix_names[] =
219 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
220 STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
221 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
222 STRING_word0 STRING_xdigit;
223
224 static const uschar posix_name_lengths[] = {
225 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
226
227 /* Table of class bit maps for each POSIX class. Each class is formed from a
228 base map, with an optional addition or removal of another map. Then, for some
229 classes, there is some additional tweaking: for [:blank:] the vertical space
230 characters are removed, and for [:alpha:] and [:alnum:] the underscore
231 character is removed. The triples in the table consist of the base map offset,
232 second map offset or -1 if no second map, and a non-negative value for map
233 addition or a negative value for map subtraction (if there are two maps). The
234 absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
235 remove vertical space characters, 2 => remove underscore. */
236
237 static const int posix_class_maps[] = {
238 cbit_word, cbit_digit, -2, /* alpha */
239 cbit_lower, -1, 0, /* lower */
240 cbit_upper, -1, 0, /* upper */
241 cbit_word, -1, 2, /* alnum - word without underscore */
242 cbit_print, cbit_cntrl, 0, /* ascii */
243 cbit_space, -1, 1, /* blank - a GNU extension */
244 cbit_cntrl, -1, 0, /* cntrl */
245 cbit_digit, -1, 0, /* digit */
246 cbit_graph, -1, 0, /* graph */
247 cbit_print, -1, 0, /* print */
248 cbit_punct, -1, 0, /* punct */
249 cbit_space, -1, 0, /* space */
250 cbit_word, -1, 0, /* word - a Perl extension */
251 cbit_xdigit,-1, 0 /* xdigit */
252 };
253
254
255 #define STRING(a) # a
256 #define XSTRING(s) STRING(s)
257
258 /* The texts of compile-time error messages. These are "char *" because they
259 are passed to the outside world. Do not ever re-use any error number, because
260 they are documented. Always add a new error instead. Messages marked DEAD below
261 are no longer used. This used to be a table of strings, but in order to reduce
262 the number of relocations needed when a shared library is loaded dynamically,
263 it is now one long string. We cannot use a table of offsets, because the
264 lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
265 simply count through to the one we want - this isn't a performance issue
266 because these strings are used only when there is a compilation error.
267
268 Each substring ends with \0 to insert a null character. This includes the final
269 substring, so that the whole string ends with \0\0, which can be detected when
270 counting through. */
271
272 static const char error_texts[] =
273 "no error\0"
274 "\\ at end of pattern\0"
275 "\\c at end of pattern\0"
276 "unrecognized character follows \\\0"
277 "numbers out of order in {} quantifier\0"
278 /* 5 */
279 "number too big in {} quantifier\0"
280 "missing terminating ] for character class\0"
281 "invalid escape sequence in character class\0"
282 "range out of order in character class\0"
283 "nothing to repeat\0"
284 /* 10 */
285 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
286 "internal error: unexpected repeat\0"
287 "unrecognized character after (? or (?-\0"
288 "POSIX named classes are supported only within a class\0"
289 "missing )\0"
290 /* 15 */
291 "reference to non-existent subpattern\0"
292 "erroffset passed as NULL\0"
293 "unknown option bit(s) set\0"
294 "missing ) after comment\0"
295 "parentheses nested too deeply\0" /** DEAD **/
296 /* 20 */
297 "regular expression is too large\0"
298 "failed to get memory\0"
299 "unmatched parentheses\0"
300 "internal error: code overflow\0"
301 "unrecognized character after (?<\0"
302 /* 25 */
303 "lookbehind assertion is not fixed length\0"
304 "malformed number or name after (?(\0"
305 "conditional group contains more than two branches\0"
306 "assertion expected after (?(\0"
307 "(?R or (?[+-]digits must be followed by )\0"
308 /* 30 */
309 "unknown POSIX class name\0"
310 "POSIX collating elements are not supported\0"
311 "this version of PCRE is not compiled with PCRE_UTF8 support\0"
312 "spare error\0" /** DEAD **/
313 "character value in \\x{...} sequence is too large\0"
314 /* 35 */
315 "invalid condition (?(0)\0"
316 "\\C not allowed in lookbehind assertion\0"
317 "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
318 "number after (?C is > 255\0"
319 "closing ) for (?C expected\0"
320 /* 40 */
321 "recursive call could loop indefinitely\0"
322 "unrecognized character after (?P\0"
323 "syntax error in subpattern name (missing terminator)\0"
324 "two named subpatterns have the same name\0"
325 "invalid UTF-8 string\0"
326 /* 45 */
327 "support for \\P, \\p, and \\X has not been compiled\0"
328 "malformed \\P or \\p sequence\0"
329 "unknown property name after \\P or \\p\0"
330 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
331 "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
332 /* 50 */
333 "repeated subpattern is too long\0" /** DEAD **/
334 "octal value is greater than \\377 (not in UTF-8 mode)\0"
335 "internal error: overran compiling workspace\0"
336 "internal error: previously-checked referenced subpattern not found\0"
337 "DEFINE group contains more than one branch\0"
338 /* 55 */
339 "repeating a DEFINE group is not allowed\0"
340 "inconsistent NEWLINE options\0"
341 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
342 "a numbered reference must not be zero\0"
343 "(*VERB) with an argument is not supported\0"
344 /* 60 */
345 "(*VERB) not recognized\0"
346 "number is too big\0"
347 "subpattern name expected\0"
348 "digit expected after (?+\0"
349 "] is an invalid data character in JavaScript compatibility mode\0"
350 /* 65 */
351 "different names for subpatterns of the same number are not allowed\0";
352
353 /* Table to identify digits and hex digits. This is used when compiling
354 patterns. Note that the tables in chartables are dependent on the locale, and
355 may mark arbitrary characters as digits - but the PCRE compiling code expects
356 to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
357 a private table here. It costs 256 bytes, but it is a lot faster than doing
358 character value tests (at least in some simple cases I timed), and in some
359 applications one wants PCRE to compile efficiently as well as match
360 efficiently.
361
362 For convenience, we use the same bit definitions as in chartables:
363
364 0x04 decimal digit
365 0x08 hexadecimal digit
366
367 Then we can use ctype_digit and ctype_xdigit in the code. */
368
369 #ifndef EBCDIC
370
371 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
372 UTF-8 mode. */
373
374 static const unsigned char digitab[] =
375 {
376 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
377 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
378 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
379 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
380 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
381 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
382 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
383 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
384 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
385 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
386 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
387 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
388 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
389 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
390 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
391 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
392 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
393 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
394 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
395 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
396 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
397 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
398 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
399 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
400 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
401 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
402 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
403 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
404 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
405 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
406 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
407 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
408
409 #else
410
411 /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
412
413 static const unsigned char digitab[] =
414 {
415 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
416 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
417 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
418 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
419 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
420 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
421 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
422 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
423 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
424 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
425 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
426 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
427 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
428 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
429 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
430 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
431 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
432 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
433 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
434 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
435 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
436 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
437 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
438 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
439 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
440 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
441 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
442 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
443 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
444 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
445 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
446 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
447
448 static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
449 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
450 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
451 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
452 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
453 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
454 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
455 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
456 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
457 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
458 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
459 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
460 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
461 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
462 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
463 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
464 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
465 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
466 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
467 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
468 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
469 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
470 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
471 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
472 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
473 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
474 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
475 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
476 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
477 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
478 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
479 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
480 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
481 #endif
482
483
484 /* Definition to allow mutual recursion */
485
486 static BOOL
487 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
488 int *, int *, branch_chain *, compile_data *, int *);
489
490
491
492 /*************************************************
493 * Find an error text *
494 *************************************************/
495
496 /* The error texts are now all in one long string, to save on relocations. As
497 some of the text is of unknown length, we can't use a table of offsets.
498 Instead, just count through the strings. This is not a performance issue
499 because it happens only when there has been a compilation error.
500
501 Argument: the error number
502 Returns: pointer to the error string
503 */
504
505 static const char *
506 find_error_text(int n)
507 {
508 const char *s = error_texts;
509 for (; n > 0; n--)
510 {
511 while (*s++ != 0) {};
512 if (*s == 0) return "Error text not found (please report)";
513 }
514 return s;
515 }
516
517
518 /*************************************************
519 * Handle escapes *
520 *************************************************/
521
522 /* This function is called when a \ has been encountered. It either returns a
523 positive value for a simple escape such as \n, or a negative value which
524 encodes one of the more complicated things such as \d. A backreference to group
525 n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
526 UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
527 ptr is pointing at the \. On exit, it is on the final character of the escape
528 sequence.
529
530 Arguments:
531 ptrptr points to the pattern position pointer
532 errorcodeptr points to the errorcode variable
533 bracount number of previous extracting brackets
534 options the options bits
535 isclass TRUE if inside a character class
536
537 Returns: zero or positive => a data character
538 negative => a special escape sequence
539 on error, errorcodeptr is set
540 */
541
542 static int
543 check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
544 int options, BOOL isclass)
545 {
546 BOOL utf8 = (options & PCRE_UTF8) != 0;
547 const uschar *ptr = *ptrptr + 1;
548 int c, i;
549
550 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
551 ptr--; /* Set pointer back to the last byte */
552
553 /* If backslash is at the end of the pattern, it's an error. */
554
555 if (c == 0) *errorcodeptr = ERR1;
556
557 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
558 in a table. A non-zero result is something that can be returned immediately.
559 Otherwise further processing may be required. */
560
561 #ifndef EBCDIC /* ASCII/UTF-8 coding */
562 else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
563 else if ((i = escapes[c - CHAR_0]) != 0) c = i;
564
565 #else /* EBCDIC coding */
566 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
567 else if ((i = escapes[c - 0x48]) != 0) c = i;
568 #endif
569
570 /* Escapes that need further processing, or are illegal. */
571
572 else
573 {
574 const uschar *oldptr;
575 BOOL braced, negated;
576
577 switch (c)
578 {
579 /* A number of Perl escapes are not handled by PCRE. We give an explicit
580 error. */
581
582 case CHAR_l:
583 case CHAR_L:
584 case CHAR_N:
585 case CHAR_u:
586 case CHAR_U:
587 *errorcodeptr = ERR37;
588 break;
589
590 /* \g must be followed by one of a number of specific things:
591
592 (1) A number, either plain or braced. If positive, it is an absolute
593 backreference. If negative, it is a relative backreference. This is a Perl
594 5.10 feature.
595
596 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
597 is part of Perl's movement towards a unified syntax for back references. As
598 this is synonymous with \k{name}, we fudge it up by pretending it really
599 was \k.
600
601 (3) For Oniguruma compatibility we also support \g followed by a name or a
602 number either in angle brackets or in single quotes. However, these are
603 (possibly recursive) subroutine calls, _not_ backreferences. Just return
604 the -ESC_g code (cf \k). */
605
606 case CHAR_g:
607 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
608 {
609 c = -ESC_g;
610 break;
611 }
612
613 /* Handle the Perl-compatible cases */
614
615 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
616 {
617 const uschar *p;
618 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
619 if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
620 if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
621 {
622 c = -ESC_k;
623 break;
624 }
625 braced = TRUE;
626 ptr++;
627 }
628 else braced = FALSE;
629
630 if (ptr[1] == CHAR_MINUS)
631 {
632 negated = TRUE;
633 ptr++;
634 }
635 else negated = FALSE;
636
637 c = 0;
638 while ((digitab[ptr[1]] & ctype_digit) != 0)
639 c = c * 10 + *(++ptr) - CHAR_0;
640
641 if (c < 0) /* Integer overflow */
642 {
643 *errorcodeptr = ERR61;
644 break;
645 }
646
647 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
648 {
649 *errorcodeptr = ERR57;
650 break;
651 }
652
653 if (c == 0)
654 {
655 *errorcodeptr = ERR58;
656 break;
657 }
658
659 if (negated)
660 {
661 if (c > bracount)
662 {
663 *errorcodeptr = ERR15;
664 break;
665 }
666 c = bracount - (c - 1);
667 }
668
669 c = -(ESC_REF + c);
670 break;
671
672 /* The handling of escape sequences consisting of a string of digits
673 starting with one that is not zero is not straightforward. By experiment,
674 the way Perl works seems to be as follows:
675
676 Outside a character class, the digits are read as a decimal number. If the
677 number is less than 10, or if there are that many previous extracting
678 left brackets, then it is a back reference. Otherwise, up to three octal
679 digits are read to form an escaped byte. Thus \123 is likely to be octal
680 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
681 value is greater than 377, the least significant 8 bits are taken. Inside a
682 character class, \ followed by a digit is always an octal number. */
683
684 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
685 case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
686
687 if (!isclass)
688 {
689 oldptr = ptr;
690 c -= CHAR_0;
691 while ((digitab[ptr[1]] & ctype_digit) != 0)
692 c = c * 10 + *(++ptr) - CHAR_0;
693 if (c < 0) /* Integer overflow */
694 {
695 *errorcodeptr = ERR61;
696 break;
697 }
698 if (c < 10 || c <= bracount)
699 {
700 c = -(ESC_REF + c);
701 break;
702 }
703 ptr = oldptr; /* Put the pointer back and fall through */
704 }
705
706 /* Handle an octal number following \. If the first digit is 8 or 9, Perl
707 generates a binary zero byte and treats the digit as a following literal.
708 Thus we have to pull back the pointer by one. */
709
710 if ((c = *ptr) >= CHAR_8)
711 {
712 ptr--;
713 c = 0;
714 break;
715 }
716
717 /* \0 always starts an octal number, but we may drop through to here with a
718 larger first octal digit. The original code used just to take the least
719 significant 8 bits of octal numbers (I think this is what early Perls used
720 to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
721 than 3 octal digits. */
722
723 case CHAR_0:
724 c -= CHAR_0;
725 while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
726 c = c * 8 + *(++ptr) - CHAR_0;
727 if (!utf8 && c > 255) *errorcodeptr = ERR51;
728 break;
729
730 /* \x is complicated. \x{ddd} is a character number which can be greater
731 than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
732 treated as a data character. */
733
734 case CHAR_x:
735 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
736 {
737 const uschar *pt = ptr + 2;
738 int count = 0;
739
740 c = 0;
741 while ((digitab[*pt] & ctype_xdigit) != 0)
742 {
743 register int cc = *pt++;
744 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
745 count++;
746
747 #ifndef EBCDIC /* ASCII/UTF-8 coding */
748 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
749 c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
750 #else /* EBCDIC coding */
751 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
752 c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
753 #endif
754 }
755
756 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
757 {
758 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
759 ptr = pt;
760 break;
761 }
762
763 /* If the sequence of hex digits does not end with '}', then we don't
764 recognize this construct; fall through to the normal \x handling. */
765 }
766
767 /* Read just a single-byte hex-defined char */
768
769 c = 0;
770 while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
771 {
772 int cc; /* Some compilers don't like */
773 cc = *(++ptr); /* ++ in initializers */
774 #ifndef EBCDIC /* ASCII/UTF-8 coding */
775 if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
776 c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
777 #else /* EBCDIC coding */
778 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
779 c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
780 #endif
781 }
782 break;
783
784 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
785 This coding is ASCII-specific, but then the whole concept of \cx is
786 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
787
788 case CHAR_c:
789 c = *(++ptr);
790 if (c == 0)
791 {
792 *errorcodeptr = ERR2;
793 break;
794 }
795
796 #ifndef EBCDIC /* ASCII/UTF-8 coding */
797 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
798 c ^= 0x40;
799 #else /* EBCDIC coding */
800 if (c >= CHAR_a && c <= CHAR_z) c += 64;
801 c ^= 0xC0;
802 #endif
803 break;
804
805 /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
806 other alphanumeric following \ is an error if PCRE_EXTRA was set;
807 otherwise, for Perl compatibility, it is a literal. This code looks a bit
808 odd, but there used to be some cases other than the default, and there may
809 be again in future, so I haven't "optimized" it. */
810
811 default:
812 if ((options & PCRE_EXTRA) != 0) switch(c)
813 {
814 default:
815 *errorcodeptr = ERR3;
816 break;
817 }
818 break;
819 }
820 }
821
822 *ptrptr = ptr;
823 return c;
824 }
825
826
827
828 #ifdef SUPPORT_UCP
829 /*************************************************
830 * Handle \P and \p *
831 *************************************************/
832
833 /* This function is called after \P or \p has been encountered, provided that
834 PCRE is compiled with support for Unicode properties. On entry, ptrptr is
835 pointing at the P or p. On exit, it is pointing at the final character of the
836 escape sequence.
837
838 Argument:
839 ptrptr points to the pattern position pointer
840 negptr points to a boolean that is set TRUE for negation else FALSE
841 dptr points to an int that is set to the detailed property value
842 errorcodeptr points to the error code variable
843
844 Returns: type value from ucp_type_table, or -1 for an invalid type
845 */
846
847 static int
848 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
849 {
850 int c, i, bot, top;
851 const uschar *ptr = *ptrptr;
852 char name[32];
853
854 c = *(++ptr);
855 if (c == 0) goto ERROR_RETURN;
856
857 *negptr = FALSE;
858
859 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
860 negation. */
861
862 if (c == CHAR_LEFT_CURLY_BRACKET)
863 {
864 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
865 {
866 *negptr = TRUE;
867 ptr++;
868 }
869 for (i = 0; i < (int)sizeof(name) - 1; i++)
870 {
871 c = *(++ptr);
872 if (c == 0) goto ERROR_RETURN;
873 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
874 name[i] = c;
875 }
876 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
877 name[i] = 0;
878 }
879
880 /* Otherwise there is just one following character */
881
882 else
883 {
884 name[0] = c;
885 name[1] = 0;
886 }
887
888 *ptrptr = ptr;
889
890 /* Search for a recognized property name using binary chop */
891
892 bot = 0;
893 top = _pcre_utt_size;
894
895 while (bot < top)
896 {
897 i = (bot + top) >> 1;
898 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
899 if (c == 0)
900 {
901 *dptr = _pcre_utt[i].value;
902 return _pcre_utt[i].type;
903 }
904 if (c > 0) bot = i + 1; else top = i;
905 }
906
907 *errorcodeptr = ERR47;
908 *ptrptr = ptr;
909 return -1;
910
911 ERROR_RETURN:
912 *errorcodeptr = ERR46;
913 *ptrptr = ptr;
914 return -1;
915 }
916 #endif
917
918
919
920
921 /*************************************************
922 * Check for counted repeat *
923 *************************************************/
924
925 /* This function is called when a '{' is encountered in a place where it might
926 start a quantifier. It looks ahead to see if it really is a quantifier or not.
927 It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
928 where the ddds are digits.
929
930 Arguments:
931 p pointer to the first char after '{'
932
933 Returns: TRUE or FALSE
934 */
935
936 static BOOL
937 is_counted_repeat(const uschar *p)
938 {
939 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
940 while ((digitab[*p] & ctype_digit) != 0) p++;
941 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
942
943 if (*p++ != CHAR_COMMA) return FALSE;
944 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
945
946 if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
947 while ((digitab[*p] & ctype_digit) != 0) p++;
948
949 return (*p == CHAR_RIGHT_CURLY_BRACKET);
950 }
951
952
953
954 /*************************************************
955 * Read repeat counts *
956 *************************************************/
957
958 /* Read an item of the form {n,m} and return the values. This is called only
959 after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
960 so the syntax is guaranteed to be correct, but we need to check the values.
961
962 Arguments:
963 p pointer to first char after '{'
964 minp pointer to int for min
965 maxp pointer to int for max
966 returned as -1 if no max
967 errorcodeptr points to error code variable
968
969 Returns: pointer to '}' on success;
970 current ptr on error, with errorcodeptr set non-zero
971 */
972
973 static const uschar *
974 read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
975 {
976 int min = 0;
977 int max = -1;
978
979 /* Read the minimum value and do a paranoid check: a negative value indicates
980 an integer overflow. */
981
982 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
983 if (min < 0 || min > 65535)
984 {
985 *errorcodeptr = ERR5;
986 return p;
987 }
988
989 /* Read the maximum value if there is one, and again do a paranoid on its size.
990 Also, max must not be less than min. */
991
992 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
993 {
994 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
995 {
996 max = 0;
997 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
998 if (max < 0 || max > 65535)
999 {
1000 *errorcodeptr = ERR5;
1001 return p;
1002 }
1003 if (max < min)
1004 {
1005 *errorcodeptr = ERR4;
1006 return p;
1007 }
1008 }
1009 }
1010
1011 /* Fill in the required variables, and pass back the pointer to the terminating
1012 '}'. */
1013
1014 *minp = min;
1015 *maxp = max;
1016 return p;
1017 }
1018
1019
1020
1021 /*************************************************
1022 * Subroutine for finding forward reference *
1023 *************************************************/
1024
1025 /* This recursive function is called only from find_parens() below. The
1026 top-level call starts at the beginning of the pattern. All other calls must
1027 start at a parenthesis. It scans along a pattern's text looking for capturing
1028 subpatterns, and counting them. If it finds a named pattern that matches the
1029 name it is given, it returns its number. Alternatively, if the name is NULL, it
1030 returns when it reaches a given numbered subpattern. We know that if (?P< is
1031 encountered, the name will be terminated by '>' because that is checked in the
1032 first pass. Recursion is used to keep track of subpatterns that reset the
1033 capturing group numbers - the (?| feature.
1034
1035 Arguments:
1036 ptrptr address of the current character pointer (updated)
1037 cd compile background data
1038 name name to seek, or NULL if seeking a numbered subpattern
1039 lorn name length, or subpattern number if name is NULL
1040 xmode TRUE if we are in /x mode
1041 count pointer to the current capturing subpattern number (updated)
1042
1043 Returns: the number of the named subpattern, or -1 if not found
1044 */
1045
1046 static int
1047 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1048 BOOL xmode, int *count)
1049 {
1050 uschar *ptr = *ptrptr;
1051 int start_count = *count;
1052 int hwm_count = start_count;
1053 BOOL dup_parens = FALSE;
1054
1055 /* If the first character is a parenthesis, check on the type of group we are
1056 dealing with. The very first call may not start with a parenthesis. */
1057
1058 if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1059 {
1060 if (ptr[1] == CHAR_QUESTION_MARK &&
1061 ptr[2] == CHAR_VERTICAL_LINE)
1062 {
1063 ptr += 3;
1064 dup_parens = TRUE;
1065 }
1066
1067 /* Handle a normal, unnamed capturing parenthesis */
1068
1069 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1070 {
1071 *count += 1;
1072 if (name == NULL && *count == lorn) return *count;
1073 ptr++;
1074 }
1075
1076 /* Handle a condition. If it is an assertion, just carry on so that it
1077 is processed as normal. If not, skip to the closing parenthesis of the
1078 condition (there can't be any nested parens. */
1079
1080 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1081 {
1082 ptr += 2;
1083 if (ptr[1] != CHAR_QUESTION_MARK)
1084 {
1085 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1086 if (*ptr != 0) ptr++;
1087 }
1088 }
1089
1090 /* We have either (? or (* and not a condition */
1091
1092 else
1093 {
1094 ptr += 2;
1095 if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1096
1097 /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1098
1099 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1100 ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1101 {
1102 int term;
1103 const uschar *thisname;
1104 *count += 1;
1105 if (name == NULL && *count == lorn) return *count;
1106 term = *ptr++;
1107 if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1108 thisname = ptr;
1109 while (*ptr != term) ptr++;
1110 if (name != NULL && lorn == ptr - thisname &&
1111 strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1112 return *count;
1113 term++;
1114 }
1115 }
1116 }
1117
1118 /* Past any initial parenthesis handling, scan for parentheses or vertical
1119 bars. */
1120
1121 for (; *ptr != 0; ptr++)
1122 {
1123 /* Skip over backslashed characters and also entire \Q...\E */
1124
1125 if (*ptr == CHAR_BACKSLASH)
1126 {
1127 if (*(++ptr) == 0) goto FAIL_EXIT;
1128 if (*ptr == CHAR_Q) for (;;)
1129 {
1130 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1131 if (*ptr == 0) goto FAIL_EXIT;
1132 if (*(++ptr) == CHAR_E) break;
1133 }
1134 continue;
1135 }
1136
1137 /* Skip over character classes; this logic must be similar to the way they
1138 are handled for real. If the first character is '^', skip it. Also, if the
1139 first few characters (either before or after ^) are \Q\E or \E we skip them
1140 too. This makes for compatibility with Perl. Note the use of STR macros to
1141 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1142
1143 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1144 {
1145 BOOL negate_class = FALSE;
1146 for (;;)
1147 {
1148 if (ptr[1] == CHAR_BACKSLASH)
1149 {
1150 if (ptr[2] == CHAR_E)
1151 ptr+= 2;
1152 else if (strncmp((const char *)ptr+2,
1153 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1154 ptr += 4;
1155 else
1156 break;
1157 }
1158 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1159 {
1160 negate_class = TRUE;
1161 ptr++;
1162 }
1163 else break;
1164 }
1165
1166 /* If the next character is ']', it is a data character that must be
1167 skipped, except in JavaScript compatibility mode. */
1168
1169 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1170 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1171 ptr++;
1172
1173 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1174 {
1175 if (*ptr == 0) return -1;
1176 if (*ptr == CHAR_BACKSLASH)
1177 {
1178 if (*(++ptr) == 0) goto FAIL_EXIT;
1179 if (*ptr == CHAR_Q) for (;;)
1180 {
1181 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1182 if (*ptr == 0) goto FAIL_EXIT;
1183 if (*(++ptr) == CHAR_E) break;
1184 }
1185 continue;
1186 }
1187 }
1188 continue;
1189 }
1190
1191 /* Skip comments in /x mode */
1192
1193 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1194 {
1195 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1196 if (*ptr == 0) goto FAIL_EXIT;
1197 continue;
1198 }
1199
1200 /* Check for the special metacharacters */
1201
1202 if (*ptr == CHAR_LEFT_PARENTHESIS)
1203 {
1204 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1205 if (rc > 0) return rc;
1206 if (*ptr == 0) goto FAIL_EXIT;
1207 }
1208
1209 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1210 {
1211 if (dup_parens && *count < hwm_count) *count = hwm_count;
1212 *ptrptr = ptr;
1213 return -1;
1214 }
1215
1216 else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1217 {
1218 if (*count > hwm_count) hwm_count = *count;
1219 *count = start_count;
1220 }
1221 }
1222
1223 FAIL_EXIT:
1224 *ptrptr = ptr;
1225 return -1;
1226 }
1227
1228
1229
1230
1231 /*************************************************
1232 * Find forward referenced subpattern *
1233 *************************************************/
1234
1235 /* This function scans along a pattern's text looking for capturing
1236 subpatterns, and counting them. If it finds a named pattern that matches the
1237 name it is given, it returns its number. Alternatively, if the name is NULL, it
1238 returns when it reaches a given numbered subpattern. This is used for forward
1239 references to subpatterns. We used to be able to start this scan from the
1240 current compiling point, using the current count value from cd->bracount, and
1241 do it all in a single loop, but the addition of the possibility of duplicate
1242 subpattern numbers means that we have to scan from the very start, in order to
1243 take account of such duplicates, and to use a recursive function to keep track
1244 of the different types of group.
1245
1246 Arguments:
1247 cd compile background data
1248 name name to seek, or NULL if seeking a numbered subpattern
1249 lorn name length, or subpattern number if name is NULL
1250 xmode TRUE if we are in /x mode
1251
1252 Returns: the number of the found subpattern, or -1 if not found
1253 */
1254
1255 static int
1256 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1257 {
1258 uschar *ptr = (uschar *)cd->start_pattern;
1259 int count = 0;
1260 int rc;
1261
1262 /* If the pattern does not start with an opening parenthesis, the first call
1263 to find_parens_sub() will scan right to the end (if necessary). However, if it
1264 does start with a parenthesis, find_parens_sub() will return when it hits the
1265 matching closing parens. That is why we have to have a loop. */
1266
1267 for (;;)
1268 {
1269 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1270 if (rc > 0 || *ptr++ == 0) break;
1271 }
1272
1273 return rc;
1274 }
1275
1276
1277
1278
1279 /*************************************************
1280 * Find first significant op code *
1281 *************************************************/
1282
1283 /* This is called by several functions that scan a compiled expression looking
1284 for a fixed first character, or an anchoring op code etc. It skips over things
1285 that do not influence this. For some calls, a change of option is important.
1286 For some calls, it makes sense to skip negative forward and all backward
1287 assertions, and also the \b assertion; for others it does not.
1288
1289 Arguments:
1290 code pointer to the start of the group
1291 options pointer to external options
1292 optbit the option bit whose changing is significant, or
1293 zero if none are
1294 skipassert TRUE if certain assertions are to be skipped
1295
1296 Returns: pointer to the first significant opcode
1297 */
1298
1299 static const uschar*
1300 first_significant_code(const uschar *code, int *options, int optbit,
1301 BOOL skipassert)
1302 {
1303 for (;;)
1304 {
1305 switch ((int)*code)
1306 {
1307 case OP_OPT:
1308 if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1309 *options = (int)code[1];
1310 code += 2;
1311 break;
1312
1313 case OP_ASSERT_NOT:
1314 case OP_ASSERTBACK:
1315 case OP_ASSERTBACK_NOT:
1316 if (!skipassert) return code;
1317 do code += GET(code, 1); while (*code == OP_ALT);
1318 code += _pcre_OP_lengths[*code];
1319 break;
1320
1321 case OP_WORD_BOUNDARY:
1322 case OP_NOT_WORD_BOUNDARY:
1323 if (!skipassert) return code;
1324 /* Fall through */
1325
1326 case OP_CALLOUT:
1327 case OP_CREF:
1328 case OP_NCREF:
1329 case OP_RREF:
1330 case OP_NRREF:
1331 case OP_DEF:
1332 code += _pcre_OP_lengths[*code];
1333 break;
1334
1335 default:
1336 return code;
1337 }
1338 }
1339 /* Control never reaches here */
1340 }
1341
1342
1343
1344
1345 /*************************************************
1346 * Find the fixed length of a branch *
1347 *************************************************/
1348
1349 /* Scan a branch and compute the fixed length of subject that will match it,
1350 if the length is fixed. This is needed for dealing with backward assertions.
1351 In UTF8 mode, the result is in characters rather than bytes. The branch is
1352 temporarily terminated with OP_END when this function is called.
1353
1354 This function is called when a backward assertion is encountered, so that if it
1355 fails, the error message can point to the correct place in the pattern.
1356 However, we cannot do this when the assertion contains subroutine calls,
1357 because they can be forward references. We solve this by remembering this case
1358 and doing the check at the end; a flag specifies which mode we are running in.
1359
1360 Arguments:
1361 code points to the start of the pattern (the bracket)
1362 options the compiling options
1363 atend TRUE if called when the pattern is complete
1364 cd the "compile data" structure
1365
1366 Returns: the fixed length,
1367 or -1 if there is no fixed length,
1368 or -2 if \C was encountered
1369 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1370 */
1371
1372 static int
1373 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1374 {
1375 int length = -1;
1376
1377 register int branchlength = 0;
1378 register uschar *cc = code + 1 + LINK_SIZE;
1379
1380 /* Scan along the opcodes for this branch. If we get to the end of the
1381 branch, check the length against that of the other branches. */
1382
1383 for (;;)
1384 {
1385 int d;
1386 uschar *ce, *cs;
1387 register int op = *cc;
1388 switch (op)
1389 {
1390 case OP_CBRA:
1391 case OP_BRA:
1392 case OP_ONCE:
1393 case OP_COND:
1394 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1395 if (d < 0) return d;
1396 branchlength += d;
1397 do cc += GET(cc, 1); while (*cc == OP_ALT);
1398 cc += 1 + LINK_SIZE;
1399 break;
1400
1401 /* Reached end of a branch; if it's a ket it is the end of a nested
1402 call. If it's ALT it is an alternation in a nested call. If it is
1403 END it's the end of the outer call. All can be handled by the same code. */
1404
1405 case OP_ALT:
1406 case OP_KET:
1407 case OP_KETRMAX:
1408 case OP_KETRMIN:
1409 case OP_END:
1410 if (length < 0) length = branchlength;
1411 else if (length != branchlength) return -1;
1412 if (*cc != OP_ALT) return length;
1413 cc += 1 + LINK_SIZE;
1414 branchlength = 0;
1415 break;
1416
1417 /* A true recursion implies not fixed length, but a subroutine call may
1418 be OK. If the subroutine is a forward reference, we can't deal with
1419 it until the end of the pattern, so return -3. */
1420
1421 case OP_RECURSE:
1422 if (!atend) return -3;
1423 cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1424 do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1425 if (cc > cs && cc < ce) return -1; /* Recursion */
1426 d = find_fixedlength(cs + 2, options, atend, cd);
1427 if (d < 0) return d;
1428 branchlength += d;
1429 cc += 1 + LINK_SIZE;
1430 break;
1431
1432 /* Skip over assertive subpatterns */
1433
1434 case OP_ASSERT:
1435 case OP_ASSERT_NOT:
1436 case OP_ASSERTBACK:
1437 case OP_ASSERTBACK_NOT:
1438 do cc += GET(cc, 1); while (*cc == OP_ALT);
1439 /* Fall through */
1440
1441 /* Skip over things that don't match chars */
1442
1443 case OP_REVERSE:
1444 case OP_CREF:
1445 case OP_NCREF:
1446 case OP_RREF:
1447 case OP_NRREF:
1448 case OP_DEF:
1449 case OP_OPT:
1450 case OP_CALLOUT:
1451 case OP_SOD:
1452 case OP_SOM:
1453 case OP_EOD:
1454 case OP_EODN:
1455 case OP_CIRC:
1456 case OP_DOLL:
1457 case OP_NOT_WORD_BOUNDARY:
1458 case OP_WORD_BOUNDARY:
1459 cc += _pcre_OP_lengths[*cc];
1460 break;
1461
1462 /* Handle literal characters */
1463
1464 case OP_CHAR:
1465 case OP_CHARNC:
1466 case OP_NOT:
1467 branchlength++;
1468 cc += 2;
1469 #ifdef SUPPORT_UTF8
1470 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1471 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1472 #endif
1473 break;
1474
1475 /* Handle exact repetitions. The count is already in characters, but we
1476 need to skip over a multibyte character in UTF8 mode. */
1477
1478 case OP_EXACT:
1479 branchlength += GET2(cc,1);
1480 cc += 4;
1481 #ifdef SUPPORT_UTF8
1482 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1483 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1484 #endif
1485 break;
1486
1487 case OP_TYPEEXACT:
1488 branchlength += GET2(cc,1);
1489 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1490 cc += 4;
1491 break;
1492
1493 /* Handle single-char matchers */
1494
1495 case OP_PROP:
1496 case OP_NOTPROP:
1497 cc += 2;
1498 /* Fall through */
1499
1500 case OP_NOT_DIGIT:
1501 case OP_DIGIT:
1502 case OP_NOT_WHITESPACE:
1503 case OP_WHITESPACE:
1504 case OP_NOT_WORDCHAR:
1505 case OP_WORDCHAR:
1506 case OP_ANY:
1507 case OP_ALLANY:
1508 branchlength++;
1509 cc++;
1510 break;
1511
1512 /* The single-byte matcher isn't allowed */
1513
1514 case OP_ANYBYTE:
1515 return -2;
1516
1517 /* Check a class for variable quantification */
1518
1519 #ifdef SUPPORT_UTF8
1520 case OP_XCLASS:
1521 cc += GET(cc, 1) - 33;
1522 /* Fall through */
1523 #endif
1524
1525 case OP_CLASS:
1526 case OP_NCLASS:
1527 cc += 33;
1528
1529 switch (*cc)
1530 {
1531 case OP_CRSTAR:
1532 case OP_CRMINSTAR:
1533 case OP_CRQUERY:
1534 case OP_CRMINQUERY:
1535 return -1;
1536
1537 case OP_CRRANGE:
1538 case OP_CRMINRANGE:
1539 if (GET2(cc,1) != GET2(cc,3)) return -1;
1540 branchlength += GET2(cc,1);
1541 cc += 5;
1542 break;
1543
1544 default:
1545 branchlength++;
1546 }
1547 break;
1548
1549 /* Anything else is variable length */
1550
1551 default:
1552 return -1;
1553 }
1554 }
1555 /* Control never gets here */
1556 }
1557
1558
1559
1560
1561 /*************************************************
1562 * Scan compiled regex for specific bracket *
1563 *************************************************/
1564
1565 /* This little function scans through a compiled pattern until it finds a
1566 capturing bracket with the given number, or, if the number is negative, an
1567 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1568 so that it can be called from pcre_study() when finding the minimum matching
1569 length.
1570
1571 Arguments:
1572 code points to start of expression
1573 utf8 TRUE in UTF-8 mode
1574 number the required bracket number or negative to find a lookbehind
1575
1576 Returns: pointer to the opcode for the bracket, or NULL if not found
1577 */
1578
1579 const uschar *
1580 _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1581 {
1582 for (;;)
1583 {
1584 register int c = *code;
1585 if (c == OP_END) return NULL;
1586
1587 /* XCLASS is used for classes that cannot be represented just by a bit
1588 map. This includes negated single high-valued characters. The length in
1589 the table is zero; the actual length is stored in the compiled code. */
1590
1591 if (c == OP_XCLASS) code += GET(code, 1);
1592
1593 /* Handle recursion */
1594
1595 else if (c == OP_REVERSE)
1596 {
1597 if (number < 0) return (uschar *)code;
1598 code += _pcre_OP_lengths[c];
1599 }
1600
1601 /* Handle capturing bracket */
1602
1603 else if (c == OP_CBRA)
1604 {
1605 int n = GET2(code, 1+LINK_SIZE);
1606 if (n == number) return (uschar *)code;
1607 code += _pcre_OP_lengths[c];
1608 }
1609
1610 /* Otherwise, we can get the item's length from the table, except that for
1611 repeated character types, we have to test for \p and \P, which have an extra
1612 two bytes of parameters. */
1613
1614 else
1615 {
1616 switch(c)
1617 {
1618 case OP_TYPESTAR:
1619 case OP_TYPEMINSTAR:
1620 case OP_TYPEPLUS:
1621 case OP_TYPEMINPLUS:
1622 case OP_TYPEQUERY:
1623 case OP_TYPEMINQUERY:
1624 case OP_TYPEPOSSTAR:
1625 case OP_TYPEPOSPLUS:
1626 case OP_TYPEPOSQUERY:
1627 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1628 break;
1629
1630 case OP_TYPEUPTO:
1631 case OP_TYPEMINUPTO:
1632 case OP_TYPEEXACT:
1633 case OP_TYPEPOSUPTO:
1634 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1635 break;
1636 }
1637
1638 /* Add in the fixed length from the table */
1639
1640 code += _pcre_OP_lengths[c];
1641
1642 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1643 a multi-byte character. The length in the table is a minimum, so we have to
1644 arrange to skip the extra bytes. */
1645
1646 #ifdef SUPPORT_UTF8
1647 if (utf8) switch(c)
1648 {
1649 case OP_CHAR:
1650 case OP_CHARNC:
1651 case OP_EXACT:
1652 case OP_UPTO:
1653 case OP_MINUPTO:
1654 case OP_POSUPTO:
1655 case OP_STAR:
1656 case OP_MINSTAR:
1657 case OP_POSSTAR:
1658 case OP_PLUS:
1659 case OP_MINPLUS:
1660 case OP_POSPLUS:
1661 case OP_QUERY:
1662 case OP_MINQUERY:
1663 case OP_POSQUERY:
1664 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1665 break;
1666 }
1667 #else
1668 (void)(utf8); /* Keep compiler happy by referencing function argument */
1669 #endif
1670 }
1671 }
1672 }
1673
1674
1675
1676 /*************************************************
1677 * Scan compiled regex for recursion reference *
1678 *************************************************/
1679
1680 /* This little function scans through a compiled pattern until it finds an
1681 instance of OP_RECURSE.
1682
1683 Arguments:
1684 code points to start of expression
1685 utf8 TRUE in UTF-8 mode
1686
1687 Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1688 */
1689
1690 static const uschar *
1691 find_recurse(const uschar *code, BOOL utf8)
1692 {
1693 for (;;)
1694 {
1695 register int c = *code;
1696 if (c == OP_END) return NULL;
1697 if (c == OP_RECURSE) return code;
1698
1699 /* XCLASS is used for classes that cannot be represented just by a bit
1700 map. This includes negated single high-valued characters. The length in
1701 the table is zero; the actual length is stored in the compiled code. */
1702
1703 if (c == OP_XCLASS) code += GET(code, 1);
1704
1705 /* Otherwise, we can get the item's length from the table, except that for
1706 repeated character types, we have to test for \p and \P, which have an extra
1707 two bytes of parameters. */
1708
1709 else
1710 {
1711 switch(c)
1712 {
1713 case OP_TYPESTAR:
1714 case OP_TYPEMINSTAR:
1715 case OP_TYPEPLUS:
1716 case OP_TYPEMINPLUS:
1717 case OP_TYPEQUERY:
1718 case OP_TYPEMINQUERY:
1719 case OP_TYPEPOSSTAR:
1720 case OP_TYPEPOSPLUS:
1721 case OP_TYPEPOSQUERY:
1722 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1723 break;
1724
1725 case OP_TYPEPOSUPTO:
1726 case OP_TYPEUPTO:
1727 case OP_TYPEMINUPTO:
1728 case OP_TYPEEXACT:
1729 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1730 break;
1731 }
1732
1733 /* Add in the fixed length from the table */
1734
1735 code += _pcre_OP_lengths[c];
1736
1737 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1738 by a multi-byte character. The length in the table is a minimum, so we have
1739 to arrange to skip the extra bytes. */
1740
1741 #ifdef SUPPORT_UTF8
1742 if (utf8) switch(c)
1743 {
1744 case OP_CHAR:
1745 case OP_CHARNC:
1746 case OP_EXACT:
1747 case OP_UPTO:
1748 case OP_MINUPTO:
1749 case OP_POSUPTO:
1750 case OP_STAR:
1751 case OP_MINSTAR:
1752 case OP_POSSTAR:
1753 case OP_PLUS:
1754 case OP_MINPLUS:
1755 case OP_POSPLUS:
1756 case OP_QUERY:
1757 case OP_MINQUERY:
1758 case OP_POSQUERY:
1759 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1760 break;
1761 }
1762 #else
1763 (void)(utf8); /* Keep compiler happy by referencing function argument */
1764 #endif
1765 }
1766 }
1767 }
1768
1769
1770
1771 /*************************************************
1772 * Scan compiled branch for non-emptiness *
1773 *************************************************/
1774
1775 /* This function scans through a branch of a compiled pattern to see whether it
1776 can match the empty string or not. It is called from could_be_empty()
1777 below and from compile_branch() when checking for an unlimited repeat of a
1778 group that can match nothing. Note that first_significant_code() skips over
1779 backward and negative forward assertions when its final argument is TRUE. If we
1780 hit an unclosed bracket, we return "empty" - this means we've struck an inner
1781 bracket whose current branch will already have been scanned.
1782
1783 Arguments:
1784 code points to start of search
1785 endcode points to where to stop
1786 utf8 TRUE if in UTF8 mode
1787
1788 Returns: TRUE if what is matched could be empty
1789 */
1790
1791 static BOOL
1792 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1793 {
1794 register int c;
1795 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1796 code < endcode;
1797 code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1798 {
1799 const uschar *ccode;
1800
1801 c = *code;
1802
1803 /* Skip over forward assertions; the other assertions are skipped by
1804 first_significant_code() with a TRUE final argument. */
1805
1806 if (c == OP_ASSERT)
1807 {
1808 do code += GET(code, 1); while (*code == OP_ALT);
1809 c = *code;
1810 continue;
1811 }
1812
1813 /* Groups with zero repeats can of course be empty; skip them. */
1814
1815 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1816 {
1817 code += _pcre_OP_lengths[c];
1818 do code += GET(code, 1); while (*code == OP_ALT);
1819 c = *code;
1820 continue;
1821 }
1822
1823 /* For other groups, scan the branches. */
1824
1825 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1826 {
1827 BOOL empty_branch;
1828 if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1829
1830 /* If a conditional group has only one branch, there is a second, implied,
1831 empty branch, so just skip over the conditional, because it could be empty.
1832 Otherwise, scan the individual branches of the group. */
1833
1834 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1835 code += GET(code, 1);
1836 else
1837 {
1838 empty_branch = FALSE;
1839 do
1840 {
1841 if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1842 empty_branch = TRUE;
1843 code += GET(code, 1);
1844 }
1845 while (*code == OP_ALT);
1846 if (!empty_branch) return FALSE; /* All branches are non-empty */
1847 }
1848
1849 c = *code;
1850 continue;
1851 }
1852
1853 /* Handle the other opcodes */
1854
1855 switch (c)
1856 {
1857 /* Check for quantifiers after a class. XCLASS is used for classes that
1858 cannot be represented just by a bit map. This includes negated single
1859 high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1860 actual length is stored in the compiled code, so we must update "code"
1861 here. */
1862
1863 #ifdef SUPPORT_UTF8
1864 case OP_XCLASS:
1865 ccode = code += GET(code, 1);
1866 goto CHECK_CLASS_REPEAT;
1867 #endif
1868
1869 case OP_CLASS:
1870 case OP_NCLASS:
1871 ccode = code + 33;
1872
1873 #ifdef SUPPORT_UTF8
1874 CHECK_CLASS_REPEAT:
1875 #endif
1876
1877 switch (*ccode)
1878 {
1879 case OP_CRSTAR: /* These could be empty; continue */
1880 case OP_CRMINSTAR:
1881 case OP_CRQUERY:
1882 case OP_CRMINQUERY:
1883 break;
1884
1885 default: /* Non-repeat => class must match */
1886 case OP_CRPLUS: /* These repeats aren't empty */
1887 case OP_CRMINPLUS:
1888 return FALSE;
1889
1890 case OP_CRRANGE:
1891 case OP_CRMINRANGE:
1892 if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1893 break;
1894 }
1895 break;
1896
1897 /* Opcodes that must match a character */
1898
1899 case OP_PROP:
1900 case OP_NOTPROP:
1901 case OP_EXTUNI:
1902 case OP_NOT_DIGIT:
1903 case OP_DIGIT:
1904 case OP_NOT_WHITESPACE:
1905 case OP_WHITESPACE:
1906 case OP_NOT_WORDCHAR:
1907 case OP_WORDCHAR:
1908 case OP_ANY:
1909 case OP_ALLANY:
1910 case OP_ANYBYTE:
1911 case OP_CHAR:
1912 case OP_CHARNC:
1913 case OP_NOT:
1914 case OP_PLUS:
1915 case OP_MINPLUS:
1916 case OP_POSPLUS:
1917 case OP_EXACT:
1918 case OP_NOTPLUS:
1919 case OP_NOTMINPLUS:
1920 case OP_NOTPOSPLUS:
1921 case OP_NOTEXACT:
1922 case OP_TYPEPLUS:
1923 case OP_TYPEMINPLUS:
1924 case OP_TYPEPOSPLUS:
1925 case OP_TYPEEXACT:
1926 return FALSE;
1927
1928 /* These are going to continue, as they may be empty, but we have to
1929 fudge the length for the \p and \P cases. */
1930
1931 case OP_TYPESTAR:
1932 case OP_TYPEMINSTAR:
1933 case OP_TYPEPOSSTAR:
1934 case OP_TYPEQUERY:
1935 case OP_TYPEMINQUERY:
1936 case OP_TYPEPOSQUERY:
1937 if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1938 break;
1939
1940 /* Same for these */
1941
1942 case OP_TYPEUPTO:
1943 case OP_TYPEMINUPTO:
1944 case OP_TYPEPOSUPTO:
1945 if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1946 break;
1947
1948 /* End of branch */
1949
1950 case OP_KET:
1951 case OP_KETRMAX:
1952 case OP_KETRMIN:
1953 case OP_ALT:
1954 return TRUE;
1955
1956 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1957 MINUPTO, and POSUPTO may be followed by a multibyte character */
1958
1959 #ifdef SUPPORT_UTF8
1960 case OP_STAR:
1961 case OP_MINSTAR:
1962 case OP_POSSTAR:
1963 case OP_QUERY:
1964 case OP_MINQUERY:
1965 case OP_POSQUERY:
1966 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1967 break;
1968
1969 case OP_UPTO:
1970 case OP_MINUPTO:
1971 case OP_POSUPTO:
1972 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1973 break;
1974 #endif
1975 }
1976 }
1977
1978 return TRUE;
1979 }
1980
1981
1982
1983 /*************************************************
1984 * Scan compiled regex for non-emptiness *
1985 *************************************************/
1986
1987 /* This function is called to check for left recursive calls. We want to check
1988 the current branch of the current pattern to see if it could match the empty
1989 string. If it could, we must look outwards for branches at other levels,
1990 stopping when we pass beyond the bracket which is the subject of the recursion.
1991
1992 Arguments:
1993 code points to start of the recursion
1994 endcode points to where to stop (current RECURSE item)
1995 bcptr points to the chain of current (unclosed) branch starts
1996 utf8 TRUE if in UTF-8 mode
1997
1998 Returns: TRUE if what is matched could be empty
1999 */
2000
2001 static BOOL
2002 could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2003 BOOL utf8)
2004 {
2005 while (bcptr != NULL && bcptr->current_branch >= code)
2006 {
2007 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8))
2008 return FALSE;
2009 bcptr = bcptr->outer;
2010 }
2011 return TRUE;
2012 }
2013
2014
2015
2016 /*************************************************
2017 * Check for POSIX class syntax *
2018 *************************************************/
2019
2020 /* This function is called when the sequence "[:" or "[." or "[=" is
2021 encountered in a character class. It checks whether this is followed by a
2022 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2023 reach an unescaped ']' without the special preceding character, return FALSE.
2024
2025 Originally, this function only recognized a sequence of letters between the
2026 terminators, but it seems that Perl recognizes any sequence of characters,
2027 though of course unknown POSIX names are subsequently rejected. Perl gives an
2028 "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2029 didn't consider this to be a POSIX class. Likewise for [:1234:].
2030
2031 The problem in trying to be exactly like Perl is in the handling of escapes. We
2032 have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2033 class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2034 below handles the special case of \], but does not try to do any other escape
2035 processing. This makes it different from Perl for cases such as [:l\ower:]
2036 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2037 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2038 I think.
2039
2040 Arguments:
2041 ptr pointer to the initial [
2042 endptr where to return the end pointer
2043
2044 Returns: TRUE or FALSE
2045 */
2046
2047 static BOOL
2048 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2049 {
2050 int terminator; /* Don't combine these lines; the Solaris cc */
2051 terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2052 for (++ptr; *ptr != 0; ptr++)
2053 {
2054 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2055 {
2056 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2057 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2058 {
2059 *endptr = ptr;
2060 return TRUE;
2061 }
2062 }
2063 }
2064 return FALSE;
2065 }
2066
2067
2068
2069
2070 /*************************************************
2071 * Check POSIX class name *
2072 *************************************************/
2073
2074 /* This function is called to check the name given in a POSIX-style class entry
2075 such as [:alnum:].
2076
2077 Arguments:
2078 ptr points to the first letter
2079 len the length of the name
2080
2081 Returns: a value representing the name, or -1 if unknown
2082 */
2083
2084 static int
2085 check_posix_name(const uschar *ptr, int len)
2086 {
2087 const char *pn = posix_names;
2088 register int yield = 0;
2089 while (posix_name_lengths[yield] != 0)
2090 {
2091 if (len == posix_name_lengths[yield] &&
2092 strncmp((const char *)ptr, pn, len) == 0) return yield;
2093 pn += posix_name_lengths[yield] + 1;
2094 yield++;
2095 }
2096 return -1;
2097 }
2098
2099
2100 /*************************************************
2101 * Adjust OP_RECURSE items in repeated group *
2102 *************************************************/
2103
2104 /* OP_RECURSE items contain an offset from the start of the regex to the group
2105 that is referenced. This means that groups can be replicated for fixed
2106 repetition simply by copying (because the recursion is allowed to refer to
2107 earlier groups that are outside the current group). However, when a group is
2108 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2109 inserted before it, after it has been compiled. This means that any OP_RECURSE
2110 items within it that refer to the group itself or any contained groups have to
2111 have their offsets adjusted. That one of the jobs of this function. Before it
2112 is called, the partially compiled regex must be temporarily terminated with
2113 OP_END.
2114
2115 This function has been extended with the possibility of forward references for
2116 recursions and subroutine calls. It must also check the list of such references
2117 for the group we are dealing with. If it finds that one of the recursions in
2118 the current group is on this list, it adjusts the offset in the list, not the
2119 value in the reference (which is a group number).
2120
2121 Arguments:
2122 group points to the start of the group
2123 adjust the amount by which the group is to be moved
2124 utf8 TRUE in UTF-8 mode
2125 cd contains pointers to tables etc.
2126 save_hwm the hwm forward reference pointer at the start of the group
2127
2128 Returns: nothing
2129 */
2130
2131 static void
2132 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2133 uschar *save_hwm)
2134 {
2135 uschar *ptr = group;
2136
2137 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2138 {
2139 int offset;
2140 uschar *hc;
2141
2142 /* See if this recursion is on the forward reference list. If so, adjust the
2143 reference. */
2144
2145 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2146 {
2147 offset = GET(hc, 0);
2148 if (cd->start_code + offset == ptr + 1)
2149 {
2150 PUT(hc, 0, offset + adjust);
2151 break;
2152 }
2153 }
2154
2155 /* Otherwise, adjust the recursion offset if it's after the start of this
2156 group. */
2157
2158 if (hc >= cd->hwm)
2159 {
2160 offset = GET(ptr, 1);
2161 if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2162 }
2163
2164 ptr += 1 + LINK_SIZE;
2165 }
2166 }
2167
2168
2169
2170 /*************************************************
2171 * Insert an automatic callout point *
2172 *************************************************/
2173
2174 /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2175 callout points before each pattern item.
2176
2177 Arguments:
2178 code current code pointer
2179 ptr current pattern pointer
2180 cd pointers to tables etc
2181
2182 Returns: new code pointer
2183 */
2184
2185 static uschar *
2186 auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2187 {
2188 *code++ = OP_CALLOUT;
2189 *code++ = 255;
2190 PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2191 PUT(code, LINK_SIZE, 0); /* Default length */
2192 return code + 2*LINK_SIZE;
2193 }
2194
2195
2196
2197 /*************************************************
2198 * Complete a callout item *
2199 *************************************************/
2200
2201 /* A callout item contains the length of the next item in the pattern, which
2202 we can't fill in till after we have reached the relevant point. This is used
2203 for both automatic and manual callouts.
2204
2205 Arguments:
2206 previous_callout points to previous callout item
2207 ptr current pattern pointer
2208 cd pointers to tables etc
2209
2210 Returns: nothing
2211 */
2212
2213 static void
2214 complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2215 {
2216 int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2217 PUT(previous_callout, 2 + LINK_SIZE, length);
2218 }
2219
2220
2221
2222 #ifdef SUPPORT_UCP
2223 /*************************************************
2224 * Get othercase range *
2225 *************************************************/
2226
2227 /* This function is passed the start and end of a class range, in UTF-8 mode
2228 with UCP support. It searches up the characters, looking for internal ranges of
2229 characters in the "other" case. Each call returns the next one, updating the
2230 start address.
2231
2232 Arguments:
2233 cptr points to starting character value; updated
2234 d end value
2235 ocptr where to put start of othercase range
2236 odptr where to put end of othercase range
2237
2238 Yield: TRUE when range returned; FALSE when no more
2239 */
2240
2241 static BOOL
2242 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2243 unsigned int *odptr)
2244 {
2245 unsigned int c, othercase, next;
2246
2247 for (c = *cptr; c <= d; c++)
2248 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2249
2250 if (c > d) return FALSE;
2251
2252 *ocptr = othercase;
2253 next = othercase + 1;
2254
2255 for (++c; c <= d; c++)
2256 {
2257 if (UCD_OTHERCASE(c) != next) break;
2258 next++;
2259 }
2260
2261 *odptr = next - 1;
2262 *cptr = c;
2263
2264 return TRUE;
2265 }
2266 #endif /* SUPPORT_UCP */
2267
2268
2269
2270 /*************************************************
2271 * Check if auto-possessifying is possible *
2272 *************************************************/
2273
2274 /* This function is called for unlimited repeats of certain items, to see
2275 whether the next thing could possibly match the repeated item. If not, it makes
2276 sense to automatically possessify the repeated item.
2277
2278 Arguments:
2279 op_code the repeated op code
2280 this data for this item, depends on the opcode
2281 utf8 TRUE in UTF-8 mode
2282 utf8_char used for utf8 character bytes, NULL if not relevant
2283 ptr next character in pattern
2284 options options bits
2285 cd contains pointers to tables etc.
2286
2287 Returns: TRUE if possessifying is wanted
2288 */
2289
2290 static BOOL
2291 check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2292 const uschar *ptr, int options, compile_data *cd)
2293 {
2294 int next;
2295
2296 /* Skip whitespace and comments in extended mode */
2297
2298 if ((options & PCRE_EXTENDED) != 0)
2299 {
2300 for (;;)
2301 {
2302 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2303 if (*ptr == CHAR_NUMBER_SIGN)
2304 {
2305 while (*(++ptr) != 0)
2306 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2307 }
2308 else break;
2309 }
2310 }
2311
2312 /* If the next item is one that we can handle, get its value. A non-negative
2313 value is a character, a negative value is an escape value. */
2314
2315 if (*ptr == CHAR_BACKSLASH)
2316 {
2317 int temperrorcode = 0;
2318 next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2319 if (temperrorcode != 0) return FALSE;
2320 ptr++; /* Point after the escape sequence */
2321 }
2322
2323 else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2324 {
2325 #ifdef SUPPORT_UTF8
2326 if (utf8) { GETCHARINC(next, ptr); } else
2327 #endif
2328 next = *ptr++;
2329 }
2330
2331 else return FALSE;
2332
2333 /* Skip whitespace and comments in extended mode */
2334
2335 if ((options & PCRE_EXTENDED) != 0)
2336 {
2337 for (;;)
2338 {
2339 while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2340 if (*ptr == CHAR_NUMBER_SIGN)
2341 {
2342 while (*(++ptr) != 0)
2343 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2344 }
2345 else break;
2346 }
2347 }
2348
2349 /* If the next thing is itself optional, we have to give up. */
2350
2351 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2352 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2353 return FALSE;
2354
2355 /* Now compare the next item with the previous opcode. If the previous is a
2356 positive single character match, "item" either contains the character or, if
2357 "item" is greater than 127 in utf8 mode, the character's bytes are in
2358 utf8_char. */
2359
2360
2361 /* Handle cases when the next item is a character. */
2362
2363 if (next >= 0) switch(op_code)
2364 {
2365 case OP_CHAR:
2366 #ifdef SUPPORT_UTF8
2367 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2368 #else
2369 (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2370 #endif
2371 return item != next;
2372
2373 /* For CHARNC (caseless character) we must check the other case. If we have
2374 Unicode property support, we can use it to test the other case of
2375 high-valued characters. */
2376
2377 case OP_CHARNC:
2378 #ifdef SUPPORT_UTF8
2379 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2380 #endif
2381 if (item == next) return FALSE;
2382 #ifdef SUPPORT_UTF8
2383 if (utf8)
2384 {
2385 unsigned int othercase;
2386 if (next < 128) othercase = cd->fcc[next]; else
2387 #ifdef SUPPORT_UCP
2388 othercase = UCD_OTHERCASE((unsigned int)next);
2389 #else
2390 othercase = NOTACHAR;
2391 #endif
2392 return (unsigned int)item != othercase;
2393 }
2394 else
2395 #endif /* SUPPORT_UTF8 */
2396 return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2397
2398 /* For OP_NOT, "item" must be a single-byte character. */
2399
2400 case OP_NOT:
2401 if (item == next) return TRUE;
2402 if ((options & PCRE_CASELESS) == 0) return FALSE;
2403 #ifdef SUPPORT_UTF8
2404 if (utf8)
2405 {
2406 unsigned int othercase;
2407 if (next < 128) othercase = cd->fcc[next]; else
2408 #ifdef SUPPORT_UCP
2409 othercase = UCD_OTHERCASE(next);
2410 #else
2411 othercase = NOTACHAR;
2412 #endif
2413 return (unsigned int)item == othercase;
2414 }
2415 else
2416 #endif /* SUPPORT_UTF8 */
2417 return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2418
2419 case OP_DIGIT:
2420 return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2421
2422 case OP_NOT_DIGIT:
2423 return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2424
2425 case OP_WHITESPACE:
2426 return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2427
2428 case OP_NOT_WHITESPACE:
2429 return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2430
2431 case OP_WORDCHAR:
2432 return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2433
2434 case OP_NOT_WORDCHAR:
2435 return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2436
2437 case OP_HSPACE:
2438 case OP_NOT_HSPACE:
2439 switch(next)
2440 {
2441 case 0x09:
2442 case 0x20:
2443 case 0xa0:
2444 case 0x1680:
2445 case 0x180e:
2446 case 0x2000:
2447 case 0x2001:
2448 case 0x2002:
2449 case 0x2003:
2450 case 0x2004:
2451 case 0x2005:
2452 case 0x2006:
2453 case 0x2007:
2454 case 0x2008:
2455 case 0x2009:
2456 case 0x200A:
2457 case 0x202f:
2458 case 0x205f:
2459 case 0x3000:
2460 return op_code != OP_HSPACE;
2461 default:
2462 return op_code == OP_HSPACE;
2463 }
2464
2465 case OP_VSPACE:
2466 case OP_NOT_VSPACE:
2467 switch(next)
2468 {
2469 case 0x0a:
2470 case 0x0b:
2471 case 0x0c:
2472 case 0x0d:
2473 case 0x85:
2474 case 0x2028:
2475 case 0x2029:
2476 return op_code != OP_VSPACE;
2477 default:
2478 return op_code == OP_VSPACE;
2479 }
2480
2481 default:
2482 return FALSE;
2483 }
2484
2485
2486 /* Handle the case when the next item is \d, \s, etc. */
2487
2488 switch(op_code)
2489 {
2490 case OP_CHAR:
2491 case OP_CHARNC:
2492 #ifdef SUPPORT_UTF8
2493 if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2494 #endif
2495 switch(-next)
2496 {
2497 case ESC_d:
2498 return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2499
2500 case ESC_D:
2501 return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2502
2503 case ESC_s:
2504 return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2505
2506 case ESC_S:
2507 return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2508
2509 case ESC_w:
2510 return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2511
2512 case ESC_W:
2513 return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2514
2515 case ESC_h:
2516 case ESC_H:
2517 switch(item)
2518 {
2519 case 0x09:
2520 case 0x20:
2521 case 0xa0:
2522 case 0x1680:
2523 case 0x180e:
2524 case 0x2000:
2525 case 0x2001:
2526 case 0x2002:
2527 case 0x2003:
2528 case 0x2004:
2529 case 0x2005:
2530 case 0x2006:
2531 case 0x2007:
2532 case 0x2008:
2533 case 0x2009:
2534 case 0x200A:
2535 case 0x202f:
2536 case 0x205f:
2537 case 0x3000:
2538 return -next != ESC_h;
2539 default:
2540 return -next == ESC_h;
2541 }
2542
2543 case ESC_v:
2544 case ESC_V:
2545 switch(item)
2546 {
2547 case 0x0a:
2548 case 0x0b:
2549 case 0x0c:
2550 case 0x0d:
2551 case 0x85:
2552 case 0x2028:
2553 case 0x2029:
2554 return -next != ESC_v;
2555 default:
2556 return -next == ESC_v;
2557 }
2558
2559 default:
2560 return FALSE;
2561 }
2562
2563 case OP_DIGIT:
2564 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2565 next == -ESC_h || next == -ESC_v;
2566
2567 case OP_NOT_DIGIT:
2568 return next == -ESC_d;
2569
2570 case OP_WHITESPACE:
2571 return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2572
2573 case OP_NOT_WHITESPACE:
2574 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2575
2576 case OP_HSPACE:
2577 return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2578
2579 case OP_NOT_HSPACE:
2580 return next == -ESC_h;
2581
2582 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2583 case OP_VSPACE:
2584 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2585
2586 case OP_NOT_VSPACE:
2587 return next == -ESC_v;
2588
2589 case OP_WORDCHAR:
2590 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2591
2592 case OP_NOT_WORDCHAR:
2593 return next == -ESC_w || next == -ESC_d;
2594
2595 default:
2596 return FALSE;
2597 }
2598
2599 /* Control does not reach here */
2600 }
2601
2602
2603
2604 /*************************************************
2605 * Compile one branch *
2606 *************************************************/
2607
2608 /* Scan the pattern, compiling it into the a vector. If the options are
2609 changed during the branch, the pointer is used to change the external options
2610 bits. This function is used during the pre-compile phase when we are trying
2611 to find out the amount of memory needed, as well as during the real compile
2612 phase. The value of lengthptr distinguishes the two phases.
2613
2614 Arguments:
2615 optionsptr pointer to the option bits
2616 codeptr points to the pointer to the current code point
2617 ptrptr points to the current pattern pointer
2618 errorcodeptr points to error code variable
2619 firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2620 reqbyteptr set to the last literal character required, else < 0
2621 bcptr points to current branch chain
2622 cd contains pointers to tables etc.
2623 lengthptr NULL during the real compile phase
2624 points to length accumulator during pre-compile phase
2625
2626 Returns: TRUE on success
2627 FALSE, with *errorcodeptr set non-zero on error
2628 */
2629
2630 static BOOL
2631 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2632 int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2633 compile_data *cd, int *lengthptr)
2634 {
2635 int repeat_type, op_type;
2636 int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2637 int bravalue = 0;
2638 int greedy_default, greedy_non_default;
2639 int firstbyte, reqbyte;
2640 int zeroreqbyte, zerofirstbyte;
2641 int req_caseopt, reqvary, tempreqvary;
2642 int options = *optionsptr;
2643 int after_manual_callout = 0;
2644 int length_prevgroup = 0;
2645 register int c;
2646 register uschar *code = *codeptr;
2647 uschar *last_code = code;
2648 uschar *orig_code = code;
2649 uschar *tempcode;
2650 BOOL inescq = FALSE;
2651 BOOL groupsetfirstbyte = FALSE;
2652 const uschar *ptr = *ptrptr;
2653 const uschar *tempptr;
2654 uschar *previous = NULL;
2655 uschar *previous_callout = NULL;
2656 uschar *save_hwm = NULL;
2657 uschar classbits[32];
2658
2659 #ifdef SUPPORT_UTF8
2660 BOOL class_utf8;
2661 BOOL utf8 = (options & PCRE_UTF8) != 0;
2662 uschar *class_utf8data;
2663 uschar *class_utf8data_base;
2664 uschar utf8_char[6];
2665 #else
2666 BOOL utf8 = FALSE;
2667 uschar *utf8_char = NULL;
2668 #endif
2669
2670 #ifdef PCRE_DEBUG
2671 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2672 #endif
2673
2674 /* Set up the default and non-default settings for greediness */
2675
2676 greedy_default = ((options & PCRE_UNGREEDY) != 0);
2677 greedy_non_default = greedy_default ^ 1;
2678
2679 /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2680 matching encountered yet". It gets changed to REQ_NONE if we hit something that
2681 matches a non-fixed char first char; reqbyte just remains unset if we never
2682 find one.
2683
2684 When we hit a repeat whose minimum is zero, we may have to adjust these values
2685 to take the zero repeat into account. This is implemented by setting them to
2686 zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2687 item types that can be repeated set these backoff variables appropriately. */
2688
2689 firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2690
2691 /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2692 according to the current setting of the caseless flag. REQ_CASELESS is a bit
2693 value > 255. It is added into the firstbyte or reqbyte variables to record the
2694 case status of the value. This is used only for ASCII characters. */
2695
2696 req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2697
2698 /* Switch on next character until the end of the branch */
2699
2700 for (;; ptr++)
2701 {
2702 BOOL negate_class;
2703 BOOL should_flip_negation;
2704 BOOL possessive_quantifier;
2705 BOOL is_quantifier;
2706 BOOL is_recurse;
2707 BOOL reset_bracount;
2708 int class_charcount;
2709 int class_lastchar;
2710 int newoptions;
2711 int recno;
2712 int refsign;
2713 int skipbytes;
2714 int subreqbyte;
2715 int subfirstbyte;
2716 int terminator;
2717 int mclength;
2718 uschar mcbuffer[8];
2719
2720 /* Get next byte in the pattern */
2721
2722 c = *ptr;
2723
2724 /* If we are in the pre-compile phase, accumulate the length used for the
2725 previous cycle of this loop. */
2726
2727 if (lengthptr != NULL)
2728 {
2729 #ifdef PCRE_DEBUG
2730 if (code > cd->hwm) cd->hwm = code; /* High water info */
2731 #endif
2732 if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2733 {
2734 *errorcodeptr = ERR52;
2735 goto FAILED;
2736 }
2737
2738 /* There is at least one situation where code goes backwards: this is the
2739 case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2740 the class is simply eliminated. However, it is created first, so we have to
2741 allow memory for it. Therefore, don't ever reduce the length at this point.
2742 */
2743
2744 if (code < last_code) code = last_code;
2745
2746 /* Paranoid check for integer overflow */
2747
2748 if (OFLOW_MAX - *lengthptr < code - last_code)
2749 {
2750 *errorcodeptr = ERR20;
2751 goto FAILED;
2752 }
2753
2754 *lengthptr += code - last_code;
2755 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2756
2757 /* If "previous" is set and it is not at the start of the work space, move
2758 it back to there, in order to avoid filling up the work space. Otherwise,
2759 if "previous" is NULL, reset the current code pointer to the start. */
2760
2761 if (previous != NULL)
2762 {
2763 if (previous > orig_code)
2764 {
2765 memmove(orig_code, previous, code - previous);
2766 code -= previous - orig_code;
2767 previous = orig_code;
2768 }
2769 }
2770 else code = orig_code;
2771
2772 /* Remember where this code item starts so we can pick up the length
2773 next time round. */
2774
2775 last_code = code;
2776 }
2777
2778 /* In the real compile phase, just check the workspace used by the forward
2779 reference list. */
2780
2781 else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2782 {
2783 *errorcodeptr = ERR52;
2784 goto FAILED;
2785 }
2786
2787 /* If in \Q...\E, check for the end; if not, we have a literal */
2788
2789 if (inescq && c != 0)
2790 {
2791 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2792 {
2793 inescq = FALSE;
2794 ptr++;
2795 continue;
2796 }
2797 else
2798 {
2799 if (previous_callout != NULL)
2800 {
2801 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2802 complete_callout(previous_callout, ptr, cd);
2803 previous_callout = NULL;
2804 }
2805 if ((options & PCRE_AUTO_CALLOUT) != 0)
2806 {
2807 previous_callout = code;
2808 code = auto_callout(code, ptr, cd);
2809 }
2810 goto NORMAL_CHAR;
2811 }
2812 }
2813
2814 /* Fill in length of a previous callout, except when the next thing is
2815 a quantifier. */
2816
2817 is_quantifier =
2818 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2819 (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2820
2821 if (!is_quantifier && previous_callout != NULL &&
2822 after_manual_callout-- <= 0)
2823 {
2824 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2825 complete_callout(previous_callout, ptr, cd);
2826 previous_callout = NULL;
2827 }
2828
2829 /* In extended mode, skip white space and comments */
2830
2831 if ((options & PCRE_EXTENDED) != 0)
2832 {
2833 if ((cd->ctypes[c] & ctype_space) != 0) continue;
2834 if (c == CHAR_NUMBER_SIGN)
2835 {
2836 while (*(++ptr) != 0)
2837 {
2838 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2839 }
2840 if (*ptr != 0) continue;
2841
2842 /* Else fall through to handle end of string */
2843 c = 0;
2844 }
2845 }
2846
2847 /* No auto callout for quantifiers. */
2848
2849 if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2850 {
2851 previous_callout = code;
2852 code = auto_callout(code, ptr, cd);
2853 }
2854
2855 switch(c)
2856 {
2857 /* ===================================================================*/
2858 case 0: /* The branch terminates at string end */
2859 case CHAR_VERTICAL_LINE: /* or | or ) */
2860 case CHAR_RIGHT_PARENTHESIS:
2861 *firstbyteptr = firstbyte;
2862 *reqbyteptr = reqbyte;
2863 *codeptr = code;
2864 *ptrptr = ptr;
2865 if (lengthptr != NULL)
2866 {
2867 if (OFLOW_MAX - *lengthptr < code - last_code)
2868 {
2869 *errorcodeptr = ERR20;
2870 goto FAILED;
2871 }
2872 *lengthptr += code - last_code; /* To include callout length */
2873 DPRINTF((">> end branch\n"));
2874 }
2875 return TRUE;
2876
2877
2878 /* ===================================================================*/
2879 /* Handle single-character metacharacters. In multiline mode, ^ disables
2880 the setting of any following char as a first character. */
2881
2882 case CHAR_CIRCUMFLEX_ACCENT:
2883 if ((options & PCRE_MULTILINE) != 0)
2884 {
2885 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2886 }
2887 previous = NULL;
2888 *code++ = OP_CIRC;
2889 break;
2890
2891 case CHAR_DOLLAR_SIGN:
2892 previous = NULL;
2893 *code++ = OP_DOLL;
2894 break;
2895
2896 /* There can never be a first char if '.' is first, whatever happens about
2897 repeats. The value of reqbyte doesn't change either. */
2898
2899 case CHAR_DOT:
2900 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2901 zerofirstbyte = firstbyte;
2902 zeroreqbyte = reqbyte;
2903 previous = code;
2904 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2905 break;
2906
2907
2908 /* ===================================================================*/
2909 /* Character classes. If the included characters are all < 256, we build a
2910 32-byte bitmap of the permitted characters, except in the special case
2911 where there is only one such character. For negated classes, we build the
2912 map as usual, then invert it at the end. However, we use a different opcode
2913 so that data characters > 255 can be handled correctly.
2914
2915 If the class contains characters outside the 0-255 range, a different
2916 opcode is compiled. It may optionally have a bit map for characters < 256,
2917 but those above are are explicitly listed afterwards. A flag byte tells
2918 whether the bitmap is present, and whether this is a negated class or not.
2919
2920 In JavaScript compatibility mode, an isolated ']' causes an error. In
2921 default (Perl) mode, it is treated as a data character. */
2922
2923 case CHAR_RIGHT_SQUARE_BRACKET:
2924 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2925 {
2926 *errorcodeptr = ERR64;
2927 goto FAILED;
2928 }
2929 goto NORMAL_CHAR;
2930
2931 case CHAR_LEFT_SQUARE_BRACKET:
2932 previous = code;
2933
2934 /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2935 they are encountered at the top level, so we'll do that too. */
2936
2937 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2938 ptr[1] == CHAR_EQUALS_SIGN) &&
2939 check_posix_syntax(ptr, &tempptr))
2940 {
2941 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2942 goto FAILED;
2943 }
2944
2945 /* If the first character is '^', set the negation flag and skip it. Also,
2946 if the first few characters (either before or after ^) are \Q\E or \E we
2947 skip them too. This makes for compatibility with Perl. */
2948
2949 negate_class = FALSE;
2950 for (;;)
2951 {
2952 c = *(++ptr);
2953 if (c == CHAR_BACKSLASH)
2954 {
2955 if (ptr[1] == CHAR_E)
2956 ptr++;
2957 else if (strncmp((const char *)ptr+1,
2958 STR_Q STR_BACKSLASH STR_E, 3) == 0)
2959 ptr += 3;
2960 else
2961 break;
2962 }
2963 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2964 negate_class = TRUE;
2965 else break;
2966 }
2967
2968 /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2969 an initial ']' is taken as a data character -- the code below handles
2970 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2971 [^] must match any character, so generate OP_ALLANY. */
2972
2973 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2974 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2975 {
2976 *code++ = negate_class? OP_ALLANY : OP_FAIL;
2977 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2978 zerofirstbyte = firstbyte;
2979 break;
2980 }
2981
2982 /* If a class contains a negative special such as \S, we need to flip the
2983 negation flag at the end, so that support for characters > 255 works
2984 correctly (they are all included in the class). */
2985
2986 should_flip_negation = FALSE;
2987
2988 /* Keep a count of chars with values < 256 so that we can optimize the case
2989 of just a single character (as long as it's < 256). However, For higher
2990 valued UTF-8 characters, we don't yet do any optimization. */
2991
2992 class_charcount = 0;
2993 class_lastchar = -1;
2994
2995 /* Initialize the 32-char bit map to all zeros. We build the map in a
2996 temporary bit of memory, in case the class contains only 1 character (less
2997 than 256), because in that case the compiled code doesn't use the bit map.
2998 */
2999
3000 memset(classbits, 0, 32 * sizeof(uschar));
3001
3002 #ifdef SUPPORT_UTF8
3003 class_utf8 = FALSE; /* No chars >= 256 */
3004 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3005 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3006 #endif
3007
3008 /* Process characters until ] is reached. By writing this as a "do" it
3009 means that an initial ] is taken as a data character. At the start of the
3010 loop, c contains the first byte of the character. */
3011
3012 if (c != 0) do
3013 {
3014 const uschar *oldptr;
3015
3016 #ifdef SUPPORT_UTF8
3017 if (utf8 && c > 127)
3018 { /* Braces are required because the */
3019 GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3020 }
3021
3022 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3023 data and reset the pointer. This is so that very large classes that
3024 contain a zillion UTF-8 characters no longer overwrite the work space
3025 (which is on the stack). */
3026
3027 if (lengthptr != NULL)
3028 {
3029 *lengthptr += class_utf8data - class_utf8data_base;
3030 class_utf8data = class_utf8data_base;
3031 }
3032
3033 #endif
3034
3035 /* Inside \Q...\E everything is literal except \E */
3036
3037 if (inescq)
3038 {
3039 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3040 {
3041 inescq = FALSE; /* Reset literal state */
3042 ptr++; /* Skip the 'E' */
3043 continue; /* Carry on with next */
3044 }
3045 goto CHECK_RANGE; /* Could be range if \E follows */
3046 }
3047
3048 /* Handle POSIX class names. Perl allows a negation extension of the
3049 form [:^name:]. A square bracket that doesn't match the syntax is
3050 treated as a literal. We also recognize the POSIX constructions
3051 [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3052 5.6 and 5.8 do. */
3053
3054 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3055 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3056 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3057 {
3058 BOOL local_negate = FALSE;
3059 int posix_class, taboffset, tabopt;
3060 register const uschar *cbits = cd->cbits;
3061 uschar pbits[32];
3062
3063 if (ptr[1] != CHAR_COLON)
3064 {
3065 *errorcodeptr = ERR31;
3066 goto FAILED;
3067 }
3068
3069 ptr += 2;
3070 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3071 {
3072 local_negate = TRUE;
3073 should_flip_negation = TRUE; /* Note negative special */
3074 ptr++;
3075 }
3076
3077 posix_class = check_posix_name(ptr, tempptr - ptr);
3078 if (posix_class < 0)
3079 {
3080 *errorcodeptr = ERR30;
3081 goto FAILED;
3082 }
3083
3084 /* If matching is caseless, upper and lower are converted to
3085 alpha. This relies on the fact that the class table starts with
3086 alpha, lower, upper as the first 3 entries. */
3087
3088 if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3089 posix_class = 0;
3090
3091 /* We build the bit map for the POSIX class in a chunk of local store
3092 because we may be adding and subtracting from it, and we don't want to
3093 subtract bits that may be in the main map already. At the end we or the
3094 result into the bit map that is being built. */
3095
3096 posix_class *= 3;
3097
3098 /* Copy in the first table (always present) */
3099
3100 memcpy(pbits, cbits + posix_class_maps[posix_class],
3101 32 * sizeof(uschar));
3102
3103 /* If there is a second table, add or remove it as required. */
3104
3105 taboffset = posix_class_maps[posix_class + 1];
3106 tabopt = posix_class_maps[posix_class + 2];
3107
3108 if (taboffset >= 0)
3109 {
3110 if (tabopt >= 0)
3111 for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3112 else
3113 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3114 }
3115
3116 /* Not see if we need to remove any special characters. An option
3117 value of 1 removes vertical space and 2 removes underscore. */
3118
3119 if (tabopt < 0) tabopt = -tabopt;
3120 if (tabopt == 1) pbits[1] &= ~0x3c;
3121 else if (tabopt == 2) pbits[11] &= 0x7f;
3122
3123 /* Add the POSIX table or its complement into the main table that is
3124 being built and we are done. */
3125
3126 if (local_negate)
3127 for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3128 else
3129 for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3130
3131 ptr = tempptr + 1;
3132 class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3133 continue; /* End of POSIX syntax handling */
3134 }
3135
3136 /* Backslash may introduce a single character, or it may introduce one
3137 of the specials, which just set a flag. The sequence \b is a special
3138 case. Inside a class (and only there) it is treated as backspace.
3139 Elsewhere it marks a word boundary. Other escapes have preset maps ready
3140 to 'or' into the one we are building. We assume they have more than one
3141 character in them, so set class_charcount bigger than one. */
3142
3143 if (c == CHAR_BACKSLASH)
3144 {
3145 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3146 if (*errorcodeptr != 0) goto FAILED;
3147
3148 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3149 else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3150 else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3151 else if (-c == ESC_Q) /* Handle start of quoted string */
3152 {
3153 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3154 {
3155 ptr += 2; /* avoid empty string */
3156 }
3157 else inescq = TRUE;
3158 continue;
3159 }
3160 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3161
3162 if (c < 0)
3163 {
3164 register const uschar *cbits = cd->cbits;
3165 class_charcount += 2; /* Greater than 1 is what matters */
3166
3167 /* Save time by not doing this in the pre-compile phase. */
3168
3169 if (lengthptr == NULL) switch (-c)
3170 {
3171 case ESC_d:
3172 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3173 continue;
3174
3175 case ESC_D:
3176 should_flip_negation = TRUE;
3177 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3178 continue;
3179
3180 case ESC_w:
3181 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3182 continue;
3183
3184 case ESC_W:
3185 should_flip_negation = TRUE;
3186 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3187 continue;
3188
3189 case ESC_s:
3190 for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3191 classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3192 continue;
3193
3194 case ESC_S:
3195 should_flip_negation = TRUE;
3196 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3197 classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3198 continue;
3199
3200 default: /* Not recognized; fall through */
3201 break; /* Need "default" setting to stop compiler warning. */
3202 }
3203
3204 /* In the pre-compile phase, just do the recognition. */
3205
3206 else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3207 c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3208
3209 /* We need to deal with \H, \h, \V, and \v in both phases because
3210 they use extra memory. */
3211
3212 if (-c == ESC_h)
3213 {
3214 SETBIT(classbits, 0x09); /* VT */
3215 SETBIT(classbits, 0x20); /* SPACE */
3216 SETBIT(classbits, 0xa0); /* NSBP */
3217 #ifdef SUPPORT_UTF8
3218 if (utf8)
3219 {
3220 class_utf8 = TRUE;
3221 *class_utf8data++ = XCL_SINGLE;
3222 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3223 *class_utf8data++ = XCL_SINGLE;
3224 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3225 *class_utf8data++ = XCL_RANGE;
3226 class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3227 class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3228 *class_utf8data++ = XCL_SINGLE;
3229 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3230 *class_utf8data++ = XCL_SINGLE;
3231 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3232 *class_utf8data++ = XCL_SINGLE;
3233 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3234 }
3235 #endif
3236 continue;
3237 }
3238
3239 if (-c == ESC_H)
3240 {
3241 for (c = 0; c < 32; c++)
3242 {
3243 int x = 0xff;
3244 switch (c)
3245 {
3246 case 0x09/8: x ^= 1 << (0x09%8); break;
3247 case 0x20/8: x ^= 1 << (0x20%8); break;
3248 case 0xa0/8: x ^= 1 << (0xa0%8); break;
3249 default: break;
3250 }
3251 classbits[c] |= x;
3252 }
3253
3254 #ifdef SUPPORT_UTF8
3255 if (utf8)
3256 {
3257 class_utf8 = TRUE;
3258 *class_utf8data++ = XCL_RANGE;
3259 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3260 class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3261 *class_utf8data++ = XCL_RANGE;
3262 class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3263 class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3264 *class_utf8data++ = XCL_RANGE;
3265 class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3266 class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3267 *class_utf8data++ = XCL_RANGE;
3268 class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3269 class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3270 *class_utf8data++ = XCL_RANGE;
3271 class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3272 class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3273 *class_utf8data++ = XCL_RANGE;
3274 class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3275 class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3276 *class_utf8data++ = XCL_RANGE;
3277 class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3278 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3279 }
3280 #endif
3281 continue;
3282 }
3283
3284 if (-c == ESC_v)
3285 {
3286 SETBIT(classbits, 0x0a); /* LF */
3287 SETBIT(classbits, 0x0b); /* VT */
3288 SETBIT(classbits, 0x0c); /* FF */
3289 SETBIT(classbits, 0x0d); /* CR */
3290 SETBIT(classbits, 0x85); /* NEL */
3291 #ifdef SUPPORT_UTF8
3292 if (utf8)
3293 {
3294 class_utf8 = TRUE;
3295 *class_utf8data++ = XCL_RANGE;
3296 class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3297 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3298 }
3299 #endif
3300 continue;
3301 }
3302
3303 if (-c == ESC_V)
3304 {
3305 for (c = 0; c < 32; c++)
3306 {
3307 int x = 0xff;
3308 switch (c)
3309 {
3310 case 0x0a/8: x ^= 1 << (0x0a%8);
3311 x ^= 1 << (0x0b%8);
3312 x ^= 1 << (0x0c%8);
3313 x ^= 1 << (0x0d%8);
3314 break;
3315 case 0x85/8: x ^= 1 << (0x85%8); break;
3316 default: break;
3317 }
3318 classbits[c] |= x;
3319 }
3320
3321 #ifdef SUPPORT_UTF8
3322 if (utf8)
3323 {
3324 class_utf8 = TRUE;
3325 *class_utf8data++ = XCL_RANGE;
3326 class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3327 class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3328 *class_utf8data++ = XCL_RANGE;
3329 class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3330 class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3331 }
3332 #endif
3333 continue;
3334 }
3335
3336 /* We need to deal with \P and \p in both phases. */
3337
3338 #ifdef SUPPORT_UCP
3339 if (-c == ESC_p || -c == ESC_P)
3340 {
3341 BOOL negated;
3342 int pdata;
3343 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3344 if (ptype < 0) goto FAILED;
3345 class_utf8 = TRUE;
3346 *class_utf8data++ = ((-c == ESC_p) != negated)?
3347 XCL_PROP : XCL_NOTPROP;
3348 *class_utf8data++ = ptype;
3349 *class_utf8data++ = pdata;
3350 class_charcount -= 2; /* Not a < 256 character */
3351 continue;
3352 }
3353 #endif
3354 /* Unrecognized escapes are faulted if PCRE is running in its
3355 strict mode. By default, for compatibility with Perl, they are
3356 treated as literals. */
3357
3358 if ((options & PCRE_EXTRA) != 0)
3359 {
3360 *errorcodeptr = ERR7;
3361 goto FAILED;
3362 }
3363
3364 class_charcount -= 2; /* Undo the default count from above */
3365 c = *ptr; /* Get the final character and fall through */
3366 }
3367
3368 /* Fall through if we have a single character (c >= 0). This may be
3369 greater than 256 in UTF-8 mode. */
3370
3371 } /* End of backslash handling */
3372
3373 /* A single character may be followed by '-' to form a range. However,
3374 Perl does not permit ']' to be the end of the range. A '-' character
3375 at the end is treated as a literal. Perl ignores orphaned \E sequences
3376 entirely. The code for handling \Q and \E is messy. */
3377
3378 CHECK_RANGE:
3379 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3380 {
3381 inescq = FALSE;
3382 ptr += 2;
3383 }
3384
3385 oldptr = ptr;
3386
3387 /* Remember \r or \n */
3388
3389 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3390
3391 /* Check for range */
3392
3393 if (!inescq && ptr[1] == CHAR_MINUS)
3394 {
3395 int d;
3396 ptr += 2;
3397 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3398
3399 /* If we hit \Q (not followed by \E) at this point, go into escaped
3400 mode. */
3401
3402 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3403 {
3404 ptr += 2;
3405 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3406 { ptr += 2; continue; }
3407 inescq = TRUE;
3408 break;
3409 }
3410
3411 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3412 {
3413 ptr = oldptr;
3414 goto LONE_SINGLE_CHARACTER;
3415 }
3416
3417 #ifdef SUPPORT_UTF8
3418 if (utf8)
3419 { /* Braces are required because the */
3420 GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3421 }
3422 else
3423 #endif
3424 d = *ptr; /* Not UTF-8 mode */
3425
3426 /* The second part of a range can be a single-character escape, but
3427 not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3428 in such circumstances. */
3429
3430 if (!inescq && d == CHAR_BACKSLASH)
3431 {
3432 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3433 if (*errorcodeptr != 0) goto FAILED;
3434
3435 /* \b is backspace; \X is literal X; \R is literal R; any other
3436 special means the '-' was literal */
3437
3438 if (d < 0)
3439 {
3440 if (d == -ESC_b) d = CHAR_BS;
3441 else if (d == -ESC_X) d = CHAR_X;
3442 else if (d == -ESC_R) d = CHAR_R; else
3443 {
3444 ptr = oldptr;
3445 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3446 }
3447 }
3448 }
3449
3450 /* Check that the two values are in the correct order. Optimize
3451 one-character ranges */
3452
3453 if (d < c)
3454 {
3455 *errorcodeptr = ERR8;
3456 goto FAILED;
3457 }
3458
3459 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3460
3461 /* Remember \r or \n */
3462
3463 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3464
3465 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3466 matching, we have to use an XCLASS with extra data items. Caseless
3467 matching for characters > 127 is available only if UCP support is
3468 available. */
3469
3470 #ifdef SUPPORT_UTF8
3471 if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3472 {
3473 class_utf8 = TRUE;
3474
3475 /* With UCP support, we can find the other case equivalents of
3476 the relevant characters. There may be several ranges. Optimize how
3477 they fit with the basic range. */
3478
3479 #ifdef SUPPORT_UCP
3480 if ((options & PCRE_CASELESS) != 0)
3481 {
3482 unsigned int occ, ocd;
3483 unsigned int cc = c;
3484 unsigned int origd = d;
3485 while (get_othercase_range(&cc, origd, &occ, &ocd))
3486 {
3487 if (occ >= (unsigned int)c &&
3488 ocd <= (unsigned int)d)
3489 continue; /* Skip embedded ranges */
3490
3491 if (occ < (unsigned int)c &&
3492 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3493 { /* if there is overlap, */
3494 c = occ; /* noting that if occ < c */
3495 continue; /* we can't have ocd > d */
3496 } /* because a subrange is */
3497 if (ocd > (unsigned int)d &&
3498 occ <= (unsigned int)d + 1) /* always shorter than */
3499 { /* the basic range. */
3500 d = ocd;
3501 continue;
3502 }
3503
3504 if (occ == ocd)
3505 {
3506 *class_utf8data++ = XCL_SINGLE;
3507 }
3508 else
3509 {
3510 *class_utf8data++ = XCL_RANGE;
3511 class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3512 }
3513 class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3514 }
3515 }
3516 #endif /* SUPPORT_UCP */
3517
3518 /* Now record the original range, possibly modified for UCP caseless
3519 overlapping ranges. */
3520
3521 *class_utf8data++ = XCL_RANGE;
3522 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3523 class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3524
3525 /* With UCP support, we are done. Without UCP support, there is no
3526 caseless matching for UTF-8 characters > 127; we can use the bit map
3527 for the smaller ones. */
3528
3529 #ifdef SUPPORT_UCP
3530 continue; /* With next character in the class */
3531 #else
3532 if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3533
3534 /* Adjust upper limit and fall through to set up the map */
3535
3536 d = 127;
3537
3538 #endif /* SUPPORT_UCP */
3539 }
3540 #endif /* SUPPORT_UTF8 */
3541
3542 /* We use the bit map for all cases when not in UTF-8 mode; else
3543 ranges that lie entirely within 0-127 when there is UCP support; else
3544 for partial ranges without UCP support. */
3545
3546 class_charcount += d - c + 1;
3547 class_lastchar = d;
3548
3549 /* We can save a bit of time by skipping this in the pre-compile. */
3550
3551 if (lengthptr == NULL) for (; c <= d; c++)
3552 {
3553 classbits[c/8] |= (1 << (c&7));
3554 if ((options & PCRE_CASELESS) != 0)
3555 {
3556 int uc = cd->fcc[c]; /* flip case */
3557 classbits[uc/8] |= (1 << (uc&7));
3558 }
3559 }
3560
3561 continue; /* Go get the next char in the class */
3562 }
3563
3564 /* Handle a lone single character - we can get here for a normal
3565 non-escape char, or after \ that introduces a single character or for an
3566 apparent range that isn't. */
3567
3568 LONE_SINGLE_CHARACTER:
3569
3570 /* Handle a character that cannot go in the bit map */
3571
3572 #ifdef SUPPORT_UTF8
3573 if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3574 {
3575 class_utf8 = TRUE;
3576 *class_utf8data++ = XCL_SINGLE;
3577 class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3578
3579 #ifdef SUPPORT_UCP
3580 if ((options & PCRE_CASELESS) != 0)
3581 {
3582 unsigned int othercase;
3583 if ((othercase = UCD_OTHERCASE(c)) != c)
3584 {
3585 *class_utf8data++ = XCL_SINGLE;
3586 class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3587 }
3588 }
3589 #endif /* SUPPORT_UCP */
3590
3591 }
3592 else
3593 #endif /* SUPPORT_UTF8 */
3594
3595 /* Handle a single-byte character */
3596 {
3597 classbits[c/8] |= (1 << (c&7));
3598 if ((options & PCRE_CASELESS) != 0)
3599 {
3600 c = cd->fcc[c]; /* flip case */
3601 classbits[c/8] |= (1 << (c&7));
3602 }
3603 class_charcount++;
3604 class_lastchar = c;
3605 }
3606 }
3607
3608 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3609
3610 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3611
3612 if (c == 0) /* Missing terminating ']' */
3613 {
3614 *errorcodeptr = ERR6;
3615 goto FAILED;
3616 }
3617
3618
3619 /* This code has been disabled because it would mean that \s counts as
3620 an explicit \r or \n reference, and that's not really what is wanted. Now
3621 we set the flag only if there is a literal "\r" or "\n" in the class. */
3622
3623 #if 0
3624 /* Remember whether \r or \n are in this class */
3625
3626 if (negate_class)
3627 {
3628 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3629 }
3630 else
3631 {
3632 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3633 }
3634 #endif
3635
3636
3637 /* If class_charcount is 1, we saw precisely one character whose value is
3638 less than 256. As long as there were no characters >= 128 and there was no
3639 use of \p or \P, in other words, no use of any XCLASS features, we can
3640 optimize.
3641
3642 In UTF-8 mode, we can optimize the negative case only if there were no
3643 characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3644 operate on single-bytes only. This is an historical hangover. Maybe one day
3645 we can tidy these opcodes to handle multi-byte characters.
3646
3647 The optimization throws away the bit map. We turn the item into a
3648 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3649 that OP_NOT does not support multibyte characters. In the positive case, it
3650 can cause firstbyte to be set. Otherwise, there can be no first char if
3651 this item is first, whatever repeat count may follow. In the case of
3652 reqbyte, save the previous value for reinstating. */
3653
3654 #ifdef SUPPORT_UTF8
3655 if (class_charcount == 1 && !class_utf8 &&
3656 (!utf8 || !negate_class || class_lastchar < 128))
3657 #else
3658 if (class_charcount == 1)
3659 #endif
3660 {
3661 zeroreqbyte = reqbyte;
3662
3663 /* The OP_NOT opcode works on one-byte characters only. */
3664
3665 if (negate_class)
3666 {
3667 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3668 zerofirstbyte = firstbyte;
3669 *code++ = OP_NOT;
3670 *code++ = class_lastchar;
3671 break;
3672 }
3673
3674 /* For a single, positive character, get the value into mcbuffer, and
3675 then we can handle this with the normal one-character code. */
3676
3677 #ifdef SUPPORT_UTF8
3678 if (utf8 && class_lastchar > 127)
3679 mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3680 else
3681 #endif
3682 {
3683 mcbuffer[0] = class_lastchar;
3684 mclength = 1;
3685 }
3686 goto ONE_CHAR;
3687 } /* End of 1-char optimization */
3688
3689 /* The general case - not the one-char optimization. If this is the first
3690 thing in the branch, there can be no first char setting, whatever the
3691 repeat count. Any reqbyte setting must remain unchanged after any kind of
3692 repeat. */
3693
3694 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3695 zerofirstbyte = firstbyte;
3696 zeroreqbyte = reqbyte;
3697
3698 /* If there are characters with values > 255, we have to compile an
3699 extended class, with its own opcode, unless there was a negated special
3700 such as \S in the class, because in that case all characters > 255 are in
3701 the class, so any that were explicitly given as well can be ignored. If
3702 (when there are explicit characters > 255 that must be listed) there are no
3703 characters < 256, we can omit the bitmap in the actual compiled code. */
3704
3705 #ifdef SUPPORT_UTF8
3706 if (class_utf8 && !should_flip_negation)
3707 {
3708 *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3709 *code++ = OP_XCLASS;
3710 code += LINK_SIZE;
3711 *code = negate_class? XCL_NOT : 0;
3712
3713 /* If the map is required, move up the extra data to make room for it;
3714 otherwise just move the code pointer to the end of the extra data. */
3715
3716 if (class_charcount > 0)
3717 {
3718 *code++ |= XCL_MAP;
3719 memmove(code + 32, code, class_utf8data - code);
3720 memcpy(code, classbits, 32);
3721 code = class_utf8data + 32;
3722 }
3723 else code = class_utf8data;
3724
3725 /* Now fill in the complete length of the item */
3726
3727 PUT(previous, 1, code - previous);
3728 break; /* End of class handling */
3729 }
3730 #endif
3731
3732 /* If there are no characters > 255, set the opcode to OP_CLASS or
3733 OP_NCLASS, depending on whether the whole class was negated and whether
3734 there were negative specials such as \S in the class. Then copy the 32-byte
3735 map into the code vector, negating it if necessary. */
3736
3737 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3738 if (negate_class)
3739 {
3740 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3741 for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3742 }
3743 else
3744 {
3745 memcpy(code, classbits, 32);
3746 }
3747 code += 32;
3748 break;
3749
3750
3751 /* ===================================================================*/
3752 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3753 has been tested above. */
3754
3755 case CHAR_LEFT_CURLY_BRACKET:
3756 if (!is_quantifier) goto NORMAL_CHAR;
3757 ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3758 if (*errorcodeptr != 0) goto FAILED;
3759 goto REPEAT;
3760
3761 case CHAR_ASTERISK:
3762 repeat_min = 0;
3763 repeat_max = -1;
3764 goto REPEAT;
3765
3766 case CHAR_PLUS:
3767 repeat_min = 1;
3768 repeat_max = -1;
3769 goto REPEAT;
3770
3771 case CHAR_QUESTION_MARK:
3772 repeat_min = 0;
3773 repeat_max = 1;
3774
3775 REPEAT:
3776 if (previous == NULL)
3777 {
3778 *errorcodeptr = ERR9;
3779 goto FAILED;
3780 }
3781
3782 if (repeat_min == 0)
3783 {
3784 firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3785 reqbyte = zeroreqbyte; /* Ditto */
3786 }
3787
3788 /* Remember whether this is a variable length repeat */
3789
3790 reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3791
3792 op_type = 0; /* Default single-char op codes */
3793 possessive_quantifier = FALSE; /* Default not possessive quantifier */
3794
3795 /* Save start of previous item, in case we have to move it up to make space
3796 for an inserted OP_ONCE for the additional '+' extension. */
3797
3798 tempcode = previous;
3799
3800 /* If the next character is '+', we have a possessive quantifier. This
3801 implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3802 If the next character is '?' this is a minimizing repeat, by default,
3803 but if PCRE_UNGREEDY is set, it works the other way round. We change the
3804 repeat type to the non-default. */
3805
3806 if (ptr[1] == CHAR_PLUS)
3807 {
3808 repeat_type = 0; /* Force greedy */
3809 possessive_quantifier = TRUE;
3810 ptr++;
3811 }
3812 else if (ptr[1] == CHAR_QUESTION_MARK)
3813 {
3814 repeat_type = greedy_non_default;
3815 ptr++;
3816 }
3817 else repeat_type = greedy_default;
3818
3819 /* If previous was a character match, abolish the item and generate a
3820 repeat item instead. If a char item has a minumum of more than one, ensure
3821 that it is set in reqbyte - it might not be if a sequence such as x{3} is
3822 the first thing in a branch because the x will have gone into firstbyte
3823 instead. */
3824
3825 if (*previous == OP_CHAR || *previous == OP_CHARNC)
3826 {
3827 /* Deal with UTF-8 characters that take up more than one byte. It's
3828 easier to write this out separately than try to macrify it. Use c to
3829 hold the length of the character in bytes, plus 0x80 to flag that it's a
3830 length rather than a small character. */
3831
3832 #ifdef SUPPORT_UTF8
3833 if (utf8 && (code[-1] & 0x80) != 0)
3834 {
3835 uschar *lastchar = code - 1;
3836 while((*lastchar & 0xc0) == 0x80) lastchar--;
3837 c = code - lastchar; /* Length of UTF-8 character */
3838 memcpy(utf8_char, lastchar, c); /* Save the char */
3839 c |= 0x80; /* Flag c as a length */
3840 }
3841 else
3842 #endif
3843
3844 /* Handle the case of a single byte - either with no UTF8 support, or
3845 with UTF-8 disabled, or for a UTF-8 character < 128. */
3846
3847 {
3848 c = code[-1];
3849 if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3850 }
3851
3852 /* If the repetition is unlimited, it pays to see if the next thing on
3853 the line is something that cannot possibly match this character. If so,
3854 automatically possessifying this item gains some performance in the case
3855 where the match fails. */
3856
3857 if (!possessive_quantifier &&
3858 repeat_max < 0 &&
3859 check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3860 options, cd))
3861 {
3862 repeat_type = 0; /* Force greedy */
3863 possessive_quantifier = TRUE;
3864 }
3865
3866 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3867 }
3868
3869 /* If previous was a single negated character ([^a] or similar), we use
3870 one of the special opcodes, replacing it. The code is shared with single-
3871 character repeats by setting opt_type to add a suitable offset into
3872 repeat_type. We can also test for auto-possessification. OP_NOT is
3873 currently used only for single-byte chars. */
3874
3875 else if (*previous == OP_NOT)
3876 {
3877 op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3878 c = previous[1];
3879 if (!possessive_quantifier &&
3880 repeat_max < 0 &&
3881 check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3882 {
3883 repeat_type = 0; /* Force greedy */
3884 possessive_quantifier = TRUE;
3885 }
3886 goto OUTPUT_SINGLE_REPEAT;
3887 }
3888
3889 /* If previous was a character type match (\d or similar), abolish it and
3890 create a suitable repeat item. The code is shared with single-character
3891 repeats by setting op_type to add a suitable offset into repeat_type. Note
3892 the the Unicode property types will be present only when SUPPORT_UCP is
3893 defined, but we don't wrap the little bits of code here because it just
3894 makes it horribly messy. */
3895
3896 else if (*previous < OP_EODN)
3897 {
3898 uschar *oldcode;
3899 int prop_type, prop_value;
3900 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3901 c = *previous;
3902
3903 if (!possessive_quantifier &&
3904 repeat_max < 0 &&
3905 check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3906 {
3907 repeat_type = 0; /* Force greedy */
3908 possessive_quantifier = TRUE;
3909 }
3910
3911 OUTPUT_SINGLE_REPEAT:
3912 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3913 {
3914 prop_type = previous[1];
3915 prop_value = previous[2];
3916 }
3917 else prop_type = prop_value = -1;
3918
3919 oldcode = code;
3920 code = previous; /* Usually overwrite previous item */
3921
3922 /* If the maximum is zero then the minimum must also be zero; Perl allows
3923 this case, so we do too - by simply omitting the item altogether. */
3924
3925 if (repeat_max == 0) goto END_REPEAT;
3926
3927 /*--------------------------------------------------------------------*/
3928 /* This code is obsolete from release 8.00; the restriction was finally
3929 removed: */
3930
3931 /* All real repeats make it impossible to handle partial matching (maybe
3932 one day we will be able to remove this restriction). */
3933
3934 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3935 /*--------------------------------------------------------------------*/
3936
3937 /* Combine the op_type with the repeat_type */
3938
3939 repeat_type += op_type;
3940
3941 /* A minimum of zero is handled either as the special case * or ?, or as
3942 an UPTO, with the maximum given. */
3943
3944 if (repeat_min == 0)
3945 {
3946 if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3947 else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3948 else
3949 {
3950 *code++ = OP_UPTO + repeat_type;
3951 PUT2INC(code, 0, repeat_max);
3952 }
3953 }
3954
3955 /* A repeat minimum of 1 is optimized into some special cases. If the
3956 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3957 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3958 one less than the maximum. */
3959
3960 else if (repeat_min == 1)
3961 {
3962 if (repeat_max == -1)
3963 *code++ = OP_PLUS + repeat_type;
3964 else
3965 {
3966 code = oldcode; /* leave previous item in place */
3967 if (repeat_max == 1) goto END_REPEAT;
3968 *code++ = OP_UPTO + repeat_type;
3969 PUT2INC(code, 0, repeat_max - 1);
3970 }
3971 }
3972
3973 /* The case {n,n} is just an EXACT, while the general case {n,m} is
3974 handled as an EXACT followed by an UPTO. */
3975
3976 else
3977 {
3978 *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3979 PUT2INC(code, 0, repeat_min);
3980
3981 /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3982 we have to insert the character for the previous code. For a repeated
3983 Unicode property match, there are two extra bytes that define the
3984 required property. In UTF-8 mode, long characters have their length in
3985 c, with the 0x80 bit as a flag. */
3986
3987 if (repeat_max < 0)
3988 {
3989 #ifdef SUPPORT_UTF8
3990 if (utf8 && c >= 128)
3991 {
3992 memcpy(code, utf8_char, c & 7);
3993 code += c & 7;
3994 }
3995 else
3996 #endif
3997 {
3998 *code++ = c;
3999 if (prop_type >= 0)
4000 {
4001 *code++ = prop_type;
4002 *code++ = prop_value;
4003 }
4004 }
4005 *code++ = OP_STAR + repeat_type;
4006 }
4007
4008 /* Else insert an UPTO if the max is greater than the min, again
4009 preceded by the character, for the previously inserted code. If the
4010 UPTO is just for 1 instance, we can use QUERY instead. */
4011
4012 else if (repeat_max != repeat_min)
4013 {
4014 #ifdef SUPPORT_UTF8
4015 if (utf8 && c >= 128)
4016 {
4017 memcpy(code, utf8_char, c & 7);
4018 code += c & 7;
4019 }
4020 else
4021 #endif
4022 *code++ = c;
4023 if (prop_type >= 0)
4024 {
4025 *code++ = prop_type;
4026 *code++ = prop_value;
4027 }
4028 repeat_max -= repeat_min;
4029
4030 if (repeat_max == 1)
4031 {
4032 *code++ = OP_QUERY + repeat_type;
4033 }
4034 else
4035 {
4036 *code++ = OP_UPTO + repeat_type;
4037 PUT2INC(code, 0, repeat_max);
4038 }
4039 }
4040 }
4041
4042 /* The character or character type itself comes last in all cases. */
4043
4044 #ifdef SUPPORT_UTF8
4045 if (utf8 && c >= 128)
4046 {
4047 memcpy(code, utf8_char, c & 7);
4048 code += c & 7;
4049 }
4050 else
4051 #endif
4052 *code++ = c;
4053
4054 /* For a repeated Unicode property match, there are two extra bytes that
4055 define the required property. */
4056
4057 #ifdef SUPPORT_UCP
4058 if (prop_type >= 0)
4059 {
4060 *code++ = prop_type;
4061 *code++ = prop_value;
4062 }
4063 #endif
4064 }
4065
4066 /* If previous was a character class or a back reference, we put the repeat
4067 stuff after it, but just skip the item if the repeat was {0,0}. */
4068
4069 else if (*previous == OP_CLASS ||
4070 *previous == OP_NCLASS ||
4071 #ifdef SUPPORT_UTF8
4072 *previous == OP_XCLASS ||
4073 #endif
4074 *previous == OP_REF)
4075 {
4076 if (repeat_max == 0)
4077 {
4078 code = previous;
4079 goto END_REPEAT;
4080 }
4081
4082 /*--------------------------------------------------------------------*/
4083 /* This code is obsolete from release 8.00; the restriction was finally
4084 removed: */
4085
4086 /* All real repeats make it impossible to handle partial matching (maybe
4087 one day we will be able to remove this restriction). */
4088
4089 /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
4090 /*--------------------------------------------------------------------*/
4091
4092 if (repeat_min == 0 && repeat_max == -1)
4093 *code++ = OP_CRSTAR + repeat_type;
4094 else if (repeat_min == 1 && repeat_max == -1)
4095 *code++ = OP_CRPLUS + repeat_type;
4096 else if (repeat_min == 0 && repeat_max == 1)
4097 *code++ = OP_CRQUERY + repeat_type;
4098 else
4099 {
4100 *code++ = OP_CRRANGE + repeat_type;
4101 PUT2INC(code, 0, repeat_min);
4102 if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4103 PUT2INC(code, 0, repeat_max);
4104 }
4105 }
4106
4107 /* If previous was a bracket group, we may have to replicate it in certain
4108 cases. */
4109
4110 else if (*previous == OP_BRA || *previous == OP_CBRA ||
4111 *previous == OP_ONCE || *previous == OP_COND)
4112 {
4113 register int i;
4114 int ketoffset = 0;
4115 int len = code - previous;
4116 uschar *bralink = NULL;
4117
4118 /* Repeating a DEFINE group is pointless */
4119
4120 if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4121 {
4122 *errorcodeptr = ERR55;
4123 goto FAILED;
4124 }
4125
4126 /* If the maximum repeat count is unlimited, find the end of the bracket
4127 by scanning through from the start, and compute the offset back to it
4128 from the current code pointer. There may be an OP_OPT setting following
4129 the final KET, so we can't find the end just by going back from the code
4130 pointer. */
4131
4132 if (repeat_max == -1)
4133 {
4134 register uschar *ket = previous;
4135 do ket += GET(ket, 1); while (*ket != OP_KET);
4136 ketoffset = code - ket;
4137 }
4138
4139 /* The case of a zero minimum is special because of the need to stick
4140 OP_BRAZERO in front of it, and because the group appears once in the
4141 data, whereas in other cases it appears the minimum number of times. For
4142 this reason, it is simplest to treat this case separately, as otherwise
4143 the code gets far too messy. There are several special subcases when the
4144 minimum is zero. */
4145
4146 if (repeat_min == 0)
4147 {
4148 /* If the maximum is also zero, we used to just omit the group from the
4149 output altogether, like this:
4150
4151 ** if (repeat_max == 0)
4152 ** {
4153 ** code = previous;
4154 ** goto END_REPEAT;
4155 ** }
4156
4157 However, that fails when a group is referenced as a subroutine from
4158 elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4159 so that it is skipped on execution. As we don't have a list of which
4160 groups are referenced, we cannot do this selectively.
4161
4162 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4163 and do no more at this point. However, we do need to adjust any
4164 OP_RECURSE calls inside the group that refer to the group itself or any
4165 internal or forward referenced group, because the offset is from the
4166 start of the whole regex. Temporarily terminate the pattern while doing
4167 this. */
4168
4169 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4170 {
4171 *code = OP_END;
4172 adjust_recurse(previous, 1, utf8, cd, save_hwm);
4173 memmove(previous+1, previous, len);
4174 code++;
4175 if (repeat_max == 0)
4176 {
4177 *previous++ = OP_SKIPZERO;
4178 goto END_REPEAT;
4179 }
4180 *previous++ = OP_BRAZERO + repeat_type;
4181 }
4182
4183 /* If the maximum is greater than 1 and limited, we have to replicate
4184 in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4185 The first one has to be handled carefully because it's the original
4186 copy, which has to be moved up. The remainder can be handled by code
4187 that is common with the non-zero minimum case below. We have to
4188 adjust the value or repeat_max, since one less copy is required. Once
4189 again, we may have to adjust any OP_RECURSE calls inside the group. */
4190
4191 else
4192 {
4193 int offset;
4194 *code = OP_END;
4195 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4196 memmove(previous + 2 + LINK_SIZE, previous, len);
4197 code += 2 + LINK_SIZE;
4198 *previous++ = OP_BRAZERO + repeat_type;
4199 *previous++ = OP_BRA;
4200
4201 /* We chain together the bracket offset fields that have to be
4202 filled in later when the ends of the brackets are reached. */
4203
4204 offset = (bralink == NULL)? 0 : previous - bralink;
4205 bralink = previous;
4206 PUTINC(previous, 0, offset);
4207 }
4208
4209 repeat_max--;
4210 }
4211
4212 /* If the minimum is greater than zero, replicate the group as many
4213 times as necessary, and adjust the maximum to the number of subsequent
4214 copies that we need. If we set a first char from the group, and didn't
4215 set a required char, copy the latter from the former. If there are any
4216 forward reference subroutine calls in the group, there will be entries on
4217 the workspace list; replicate these with an appropriate increment. */
4218
4219 else
4220 {
4221 if (repeat_min > 1)
4222 {
4223 /* In the pre-compile phase, we don't actually do the replication. We
4224 just adjust the length as if we had. Do some paranoid checks for
4225 potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
4226 integer type when available, otherwise double. */
4227
4228 if (lengthptr != NULL)
4229 {
4230 int delta = (repeat_min - 1)*length_prevgroup;
4231 if ((INT64_OR_DOUBLE)(repeat_min - 1)*
4232 (INT64_OR_DOUBLE)length_prevgroup >
4233 (INT64_OR_DOUBLE)INT_MAX ||
4234 OFLOW_MAX - *lengthptr < delta)
4235 {
4236 *errorcodeptr = ERR20;
4237 goto FAILED;
4238 }
4239 *lengthptr += delta;
4240 }
4241
4242 /* This is compiling for real */
4243
4244 else
4245 {
4246 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4247 for (i = 1; i < repeat_min; i++)
4248 {
4249 uschar *hc;
4250 uschar *this_hwm = cd->hwm;
4251 memcpy(code, previous, len);
4252 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4253 {
4254 PUT(cd->hwm, 0, GET(hc, 0) + len);
4255 cd->hwm += LINK_SIZE;
4256 }
4257 save_hwm = this_hwm;
4258 code += len;
4259 }
4260 }
4261 }
4262
4263 if (repeat_max > 0) repeat_max -= repeat_min;
4264 }
4265
4266 /* This code is common to both the zero and non-zero minimum cases. If
4267 the maximum is limited, it replicates the group in a nested fashion,
4268 remembering the bracket starts on a stack. In the case of a zero minimum,
4269 the first one was set up above. In all cases the repeat_max now specifies
4270 the number of additional copies needed. Again, we must remember to
4271 replicate entries on the forward reference list. */
4272
4273 if (repeat_max >= 0)
4274 {
4275 /* In the pre-compile phase, we don't actually do the replication. We
4276 just adjust the length as if we had. For each repetition we must add 1
4277 to the length for BRAZERO and for all but the last repetition we must
4278 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4279 paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
4280 a 64-bit integer type when available, otherwise double. */
4281
4282 if (lengthptr != NULL && repeat_max > 0)
4283 {
4284 int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4285 2 - 2*LINK_SIZE; /* Last one doesn't nest */
4286 if ((INT64_OR_DOUBLE)repeat_max *
4287 (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4288 > (INT64_OR_DOUBLE)INT_MAX ||
4289 OFLOW_MAX - *lengthptr < delta)
4290 {
4291 *errorcodeptr = ERR20;
4292 goto FAILED;
4293 }
4294 *lengthptr += delta;
4295 }
4296
4297 /* This is compiling for real */
4298
4299 else for (i = repeat_max - 1; i >= 0; i--)
4300 {
4301 uschar *hc;
4302 uschar *this_hwm = cd->hwm;
4303
4304 *code++ = OP_BRAZERO + repeat_type;
4305
4306 /* All but the final copy start a new nesting, maintaining the
4307 chain of brackets outstanding. */
4308
4309 if (i != 0)
4310 {
4311 int offset;
4312 *code++ = OP_BRA;
4313 offset = (bralink == NULL)? 0 : code - bralink;
4314 bralink = code;
4315 PUTINC(code, 0, offset);
4316 }
4317
4318 memcpy(code, previous, len);
4319 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4320 {
4321 PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4322 cd->hwm += LINK_SIZE;
4323 }
4324 save_hwm = this_hwm;
4325 code += len;
4326 }
4327
4328 /* Now chain through the pending brackets, and fill in their length
4329 fields (which are holding the chain links pro tem). */
4330
4331 while (bralink != NULL)
4332 {
4333 int oldlinkoffset;
4334 int offset = code - bralink + 1;
4335 uschar *bra = code - offset;
4336 oldlinkoffset = GET(bra, 1);
4337 bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4338 *code++ = OP_KET;
4339 PUTINC(code, 0, offset);
4340 PUT(bra, 1, offset);
4341 }
4342 }
4343
4344 /* If the maximum is unlimited, set a repeater in the final copy. We
4345 can't just offset backwards from the current code point, because we
4346 don't know if there's been an options resetting after the ket. The
4347 correct offset was computed above.
4348
4349 Then, when we are doing the actual compile phase, check to see whether
4350 this group is a non-atomic one that could match an empty string. If so,
4351 convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4352 that runtime checking can be done. [This check is also applied to
4353 atomic groups at runtime, but in a different way.] */
4354
4355 else
4356 {
4357 uschar *ketcode = code - ketoffset;
4358 uschar *bracode = ketcode - GET(ketcode, 1);
4359 *ketcode = OP_KETRMAX + repeat_type;
4360 if (lengthptr == NULL && *bracode != OP_ONCE)
4361 {
4362 uschar *scode = bracode;
4363 do
4364 {
4365 if (could_be_empty_branch(scode, ketcode, utf8))
4366 {
4367 *bracode += OP_SBRA - OP_BRA;
4368 break;
4369 }
4370 scode += GET(scode, 1);
4371 }
4372 while (*scode == OP_ALT);
4373 }
4374 }
4375 }
4376
4377 /* If previous is OP_FAIL, it was generated by an empty class [] in
4378 JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4379 by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4380 error above. We can just ignore the repeat in JS case. */
4381
4382 else if (*previous == OP_FAIL) goto END_REPEAT;
4383
4384 /* Else there's some kind of shambles */
4385
4386 else
4387 {
4388 *errorcodeptr = ERR11;
4389 goto FAILED;
4390 }
4391
4392 /* If the character following a repeat is '+', or if certain optimization
4393 tests above succeeded, possessive_quantifier is TRUE. For some of the
4394 simpler opcodes, there is an special alternative opcode for this. For
4395 anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4396 The '+' notation is just syntactic sugar, taken from Sun's Java package,
4397 but the special opcodes can optimize it a bit. The repeated item starts at
4398 tempcode, not at previous, which might be the first part of a string whose
4399 (former) last char we repeated.
4400
4401 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4402 an 'upto' may follow. We skip over an 'exact' item, and then test the
4403 length of what remains before proceeding. */
4404
4405 if (possessive_quantifier)
4406 {
4407 int len;
4408
4409 if (*tempcode == OP_TYPEEXACT)
4410 tempcode += _pcre_OP_lengths[*tempcode] +
4411 ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
4412
4413 else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
4414 {
4415 tempcode += _pcre_OP_lengths[*tempcode];
4416 #ifdef SUPPORT_UTF8
4417 if (utf8 && tempcode[-1] >= 0xc0)
4418 tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
4419 #endif
4420 }
4421
4422 len = code - tempcode;
4423 if (len > 0) switch (*tempcode)
4424 {
4425 case OP_STAR: *tempcode = OP_POSSTAR; break;
4426 case OP_PLUS: *tempcode = OP_POSPLUS; break;
4427 case OP_QUERY: *tempcode = OP_POSQUERY; break;
4428 case OP_UPTO: *tempcode = OP_POSUPTO; break;
4429
4430 case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4431 case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4432 case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4433 case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4434
4435 case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4436 case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4437 case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4438 case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4439
4440 /* Because we are moving code along, we must ensure that any
4441 pending recursive references are updated. */
4442
4443 default:
4444 *code = OP_END;
4445 adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
4446 memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4447 code += 1 + LINK_SIZE;
4448 len += 1 + LINK_SIZE;
4449 tempcode[0] = OP_ONCE;
4450 *code++ = OP_KET;
4451 PUTINC(code, 0, len);
4452 PUT(tempcode, 1, len);
4453 break;
4454 }
4455 }
4456
4457 /* In all case we no longer have a previous item. We also set the
4458 "follows varying string" flag for subsequently encountered reqbytes if
4459 it isn't already set and we have just passed a varying length item. */
4460
4461 END_REPEAT:
4462 previous = NULL;
4463 cd->req_varyopt |= reqvary;
4464 break;
4465
4466
4467 /* ===================================================================*/
4468 /* Start of nested parenthesized sub-expression, or comment or lookahead or
4469 lookbehind or option setting or condition or all the other extended
4470 parenthesis forms. */
4471
4472 case CHAR_LEFT_PARENTHESIS:
4473 newoptions = options;
4474 skipbytes = 0;
4475 bravalue = OP_CBRA;
4476 save_hwm = cd->hwm;
4477 reset_bracount = FALSE;
4478
4479 /* First deal with various "verbs" that can be introduced by '*'. */
4480
4481 if (*(++ptr) == CHAR_ASTERISK && (cd->ctypes[ptr[1]] & ctype_letter) != 0)
4482 {
4483 int i, namelen;
4484 const char *vn = verbnames;
4485 const uschar *name = ++ptr;
4486 previous = NULL;
4487 while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
4488 if (*ptr == CHAR_COLON)
4489 {
4490 *errorcodeptr = ERR59; /* Not supported */
4491 goto FAILED;
4492 }
4493 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4494 {
4495 *errorcodeptr = ERR60;
4496 goto FAILED;
4497 }
4498 namelen = ptr - name;
4499 for (i = 0; i < verbcount; i++)
4500 {
4501 if (namelen == verbs[i].len &&
4502 strncmp((char *)name, vn, namelen) == 0)
4503 {
4504 /* Check for open captures before ACCEPT */
4505
4506 if (verbs[i].op == OP_ACCEPT)
4507 {
4508 open_capitem *oc;
4509 cd->had_accept = TRUE;
4510 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
4511 {
4512 *code++ = OP_CLOSE;
4513 PUT2INC(code, 0, oc->number);
4514 }
4515 }
4516 *code++ = verbs[i].op;
4517 break;
4518 }
4519 vn += verbs[i].len + 1;
4520 }
4521 if (i < verbcount) continue;
4522 *errorcodeptr = ERR60;
4523 goto FAILED;
4524 }
4525
4526 /* Deal with the extended parentheses; all are introduced by '?', and the
4527 appearance of any of them means that this is not a capturing group. */
4528
4529 else if (*ptr == CHAR_QUESTION_MARK)
4530 {
4531 int i, set, unset, namelen;
4532 int *optset;
4533 const uschar *name;
4534 uschar *slot;
4535
4536 switch (*(++ptr))
4537 {
4538 case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
4539 ptr++;
4540 while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
4541 if (*ptr == 0)
4542 {
4543 *errorcodeptr = ERR18;
4544 goto FAILED;
4545 }
4546 continue;
4547
4548
4549 /* ------------------------------------------------------------ */
4550 case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
4551 reset_bracount = TRUE;
4552 /* Fall through */
4553
4554 /* ------------------------------------------------------------ */
4555 case CHAR_COLON: /* Non-capturing bracket */
4556 bravalue = OP_BRA;
4557 ptr++;
4558 break;
4559
4560
4561 /* ------------------------------------------------------------ */
4562 case CHAR_LEFT_PARENTHESIS:
4563 bravalue = OP_COND; /* Conditional group */
4564
4565 /* A condition can be an assertion, a number (referring to a numbered
4566 group), a name (referring to a named group), or 'R', referring to
4567 recursion. R<digits> and R&name are also permitted for recursion tests.
4568
4569 There are several syntaxes for testing a named group: (?(name)) is used
4570 by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4571
4572 There are two unfortunate ambiguities, caused by history. (a) 'R' can
4573 be the recursive thing or the name 'R' (and similarly for 'R' followed
4574 by digits), and (b) a number could be a name that consists of digits.
4575 In both cases, we look for a name first; if not found, we try the other
4576 cases. */
4577
4578 /* For conditions that are assertions, check the syntax, and then exit
4579 the switch. This will take control down to where bracketed groups,
4580 including assertions, are processed. */
4581
4582 if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4583 ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
4584 break;
4585
4586 /* Most other conditions use OP_CREF (a couple change to OP_RREF
4587 below), and all need to skip 3 bytes at the start of the group. */
4588
4589 code[1+LINK_SIZE] = OP_CREF;
4590 skipbytes = 3;
4591 refsign = -1;
4592
4593 /* Check for a test for recursion in a named group. */
4594
4595 if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
4596 {
4597 terminator = -1;
4598 ptr += 2;
4599 code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4600 }
4601
4602 /* Check for a test for a named group's having been set, using the Perl
4603 syntax (?(<name>) or (?('name') */
4604
4605 else if (ptr[1] == CHAR_LESS_THAN_SIGN)
4606 {
4607 terminator = CHAR_GREATER_THAN_SIGN;
4608 ptr++;
4609 }
4610 else if (ptr[1] == CHAR_APOSTROPHE)
4611 {
4612 terminator = CHAR_APOSTROPHE;
4613 ptr++;
4614 }
4615 else
4616 {
4617 terminator = 0;
4618 if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
4619 }
4620
4621 /* We now expect to read a name; any thing else is an error */
4622
4623 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4624 {
4625 ptr += 1; /* To get the right offset */
4626 *errorcodeptr = ERR28;
4627 goto FAILED;
4628 }
4629
4630 /* Read the name, but also get it as a number if it's all digits */
4631
4632 recno = 0;
4633 name = ++ptr;
4634 while ((cd->ctypes[*ptr] & ctype_word) != 0)
4635 {
4636 if (recno >= 0)
4637 recno = ((digitab[*ptr] & ctype_digit) != 0)?
4638 recno * 10 + *ptr - CHAR_0 : -1;
4639 ptr++;
4640 }
4641 namelen = ptr - name;
4642
4643 if ((terminator > 0 && *ptr++ != terminator) ||
4644 *ptr++ != CHAR_RIGHT_PARENTHESIS)
4645 {
4646 ptr--; /* Error offset */
4647 *errorcodeptr = ERR26;
4648 goto FAILED;
4649 }
4650
4651 /* Do no further checking in the pre-compile phase. */
4652
4653 if (lengthptr != NULL) break;
4654
4655 /* In the real compile we do the work of looking for the actual
4656 reference. If the string started with "+" or "-" we require the rest to
4657 be digits, in which case recno will be set. */
4658
4659 if (refsign > 0)
4660 {
4661 if (recno <= 0)
4662 {
4663 *errorcodeptr = ERR58;
4664 goto FAILED;
4665 }
4666 recno = (refsign == CHAR_MINUS)?
4667 cd->bracount - recno + 1 : recno +cd->bracount;
4668 if (recno <= 0 || recno > cd->final_bracount)
4669 {
4670 *errorcodeptr = ERR15;
4671 goto FAILED;
4672 }
4673 PUT2(code, 2+LINK_SIZE, recno);
4674 break;
4675 }
4676
4677 /* Otherwise (did not start with "+" or "-"), start by looking for the
4678 name. If we find a name, add one to the opcode to change OP_CREF or
4679 OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
4680 except they record that the reference was originally to a name. The
4681 information is used to check duplicate names. */
4682
4683 slot = cd->name_table;
4684 for (i = 0; i < cd->names_found; i++)
4685 {
4686 if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4687 slot += cd->name_entry_size;
4688 }
4689
4690 /* Found a previous named subpattern */
4691
4692 if (i < cd->names_found)
4693 {
4694 recno = GET2(slot, 0);
4695 PUT2(code, 2+LINK_SIZE, recno);
4696 code[1+LINK_SIZE]++;
4697 }
4698
4699 /* Search the pattern for a forward reference */
4700
4701 else if ((i = find_parens(cd, name, namelen,
4702 (options & PCRE_EXTENDED) != 0)) > 0)
4703 {
4704 PUT2(code, 2+LINK_SIZE, i);
4705 code[1+LINK_SIZE]++;
4706 }
4707
4708 /* If terminator == 0 it means that the name followed directly after
4709 the opening parenthesis [e.g. (?(abc)...] and in this case there are
4710 some further alternatives to try. For the cases where terminator != 0
4711 [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
4712 now checked all the possibilities, so give an error. */
4713
4714 else if (terminator != 0)
4715 {
4716 *errorcodeptr = ERR15;
4717 goto FAILED;
4718 }
4719
4720 /* Check for (?(R) for recursion. Allow digits after R to specify a
4721 specific group number. */
4722
4723 else if (*name == CHAR_R)
4724 {
4725 recno = 0;
4726 for (i = 1; i < namelen; i++)
4727 {
4728 if ((digitab[name[i]] & ctype_digit) == 0)
4729 {
4730 *errorcodeptr = ERR15;
4731 goto FAILED;
4732 }
4733 recno = recno * 10 + name[i] - CHAR_0;
4734 }
4735 if (recno == 0) recno = RREF_ANY;
4736 code[1+LINK_SIZE] = OP_RREF; /* Change test type */
4737 PUT2(code, 2+LINK_SIZE, recno);
4738 }
4739
4740 /* Similarly, check for the (?(DEFINE) "condition", which is always
4741 false. */
4742
4743 else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
4744 {
4745 code[1+LINK_SIZE] = OP_DEF;
4746 skipbytes = 1;
4747 }
4748
4749 /* Check for the "name" actually being a subpattern number. We are
4750 in the second pass here, so final_bracount is set. */
4751
4752 else if (recno > 0 && recno <= cd->final_bracount)
4753 {
4754 PUT2(code, 2+LINK_SIZE, recno);
4755 }
4756
4757 /* Either an unidentified subpattern, or a reference to (?(0) */
4758
4759 else
4760 {
4761 *errorcodeptr = (recno == 0)? ERR35: ERR15;
4762 goto FAILED;
4763 }
4764 break;
4765
4766
4767 /* ------------------------------------------------------------ */
4768 case CHAR_EQUALS_SIGN: /* Positive lookahead */
4769 bravalue = OP_ASSERT;
4770 ptr++;
4771 break;
4772
4773
4774 /* ------------------------------------------------------------ */
4775 case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
4776 ptr++;
4777 if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
4778 {
4779 *code++ = OP_FAIL;
4780 previous = NULL;
4781 continue;
4782 }
4783 bravalue = OP_ASSERT_NOT;
4784 break;
4785
4786
4787 /* ------------------------------------------------------------ */
4788 case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
4789 switch (ptr[1])
4790 {
4791 case CHAR_EQUALS_SIGN: /* Positive lookbehind */
4792 bravalue = OP_ASSERTBACK;
4793 ptr += 2;
4794 break;
4795
4796 case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
4797 bravalue = OP_ASSERTBACK_NOT;
4798 ptr += 2;
4799 break;
4800
4801 default: /* Could be name define, else bad */
4802 if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
4803 ptr++; /* Correct offset for error */
4804 *errorcodeptr = ERR24;
4805 goto FAILED;
4806 }
4807 break;
4808
4809
4810 /* ------------------------------------------------------------ */
4811 case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
4812 bravalue = OP_ONCE;
4813 ptr++;
4814 break;
4815
4816
4817 /* ------------------------------------------------------------ */
4818 case CHAR_C: /* Callout - may be followed by digits; */
4819 previous_callout = code; /* Save for later completion */
4820 after_manual_callout = 1; /* Skip one item before completing */
4821 *code++ = OP_CALLOUT;
4822 {
4823 int n = 0;
4824 while ((digitab[*(++ptr)] & ctype_digit) != 0)
4825 n = n * 10 + *ptr - CHAR_0;
4826 if (*ptr != CHAR_RIGHT_PARENTHESIS)
4827 {
4828 *errorcodeptr = ERR39;
4829 goto FAILED;
4830 }
4831 if (n > 255)
4832 {
4833 *errorcodeptr = ERR38;
4834 goto FAILED;
4835 }
4836 *code++ = n;
4837 PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
4838 PUT(code, LINK_SIZE, 0); /* Default length */
4839 code += 2 * LINK_SIZE;
4840 }
4841 previous = NULL;
4842 continue;
4843
4844
4845 /* ------------------------------------------------------------ */
4846 case CHAR_P: /* Python-style named subpattern handling */
4847 if (*(++ptr) == CHAR_EQUALS_SIGN ||
4848 *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
4849 {
4850 is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
4851 terminator = CHAR_RIGHT_PARENTHESIS;
4852 goto NAMED_REF_OR_RECURSE;
4853 }
4854 else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
4855 {
4856 *errorcodeptr = ERR41;
4857 goto FAILED;
4858 }
4859 /* Fall through to handle (?P< as (?< is handled */
4860
4861
4862 /* ------------------------------------------------------------ */
4863 DEFINE_NAME: /* Come here from (?< handling */
4864 case CHAR_APOSTROPHE:
4865 {
4866 terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
4867 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
4868 name = ++ptr;
4869
4870 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
4871 namelen = ptr - name;
4872
4873 /* In the pre-compile phase, just do a syntax check. */
4874
4875 if (lengthptr != NULL)
4876 {
4877 if (*ptr != terminator)
4878 {
4879 *errorcodeptr = ERR42;
4880 goto FAILED;
4881 }
4882 if (cd->names_found >= MAX_NAME_COUNT)
4883 {
4884 *errorcodeptr = ERR49;
4885 goto FAILED;
4886 }
4887 if (namelen + 3 > cd->name_entry_size)
4888 {
4889 cd->name_entry_size = namelen + 3;
4890 if (namelen > MAX_NAME_SIZE)
4891 {
4892 *errorcodeptr = ERR48;
4893 goto FAILED;
4894 }
4895 }
4896 }
4897
4898 /* In the real compile, create the entry in the table, maintaining
4899 alphabetical order. Duplicate names for different numbers are
4900 permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
4901 number are always OK. (An existing number can be re-used if (?|
4902 appears in the pattern.) In either event, a duplicate name results in
4903 a duplicate entry in the table, even if the number is the same. This
4904 is because the number of names, and hence the table size, is computed
4905 in the pre-compile, and it affects various numbers and pointers which
4906 would all have to be modified, and the compiled code moved down, if
4907 duplicates with the same number were omitted from the table. This
4908 doesn't seem worth the hassle. However, *different* names for the
4909 same number are not permitted. */
4910
4911 else
4912 {
4913 BOOL dupname = FALSE;
4914 slot = cd->name_table;
4915
4916 for (i = 0; i < cd->names_found; i++)
4917 {
4918 int crc = memcmp(name, slot+2, namelen);
4919 if (crc == 0)
4920 {
4921 if (slot[2+namelen] == 0)
4922 {
4923 if (GET2(slot, 0) != cd->bracount + 1 &&
4924 (options & PCRE_DUPNAMES) == 0)
4925 {
4926 *errorcodeptr = ERR43;
4927 goto FAILED;
4928 }
4929 else dupname = TRUE;
4930 }
4931 else crc = -1; /* Current name is a substring */
4932 }
4933
4934 /* Make space in the table and break the loop for an earlier
4935 name. For a duplicate or later name, carry on. We do this for
4936 duplicates so that in the simple case (when ?(| is not used) they
4937 are in order of their numbers. */
4938
4939 if (crc < 0)
4940 {
4941 memmove(slot + cd->name_entry_size, slot,
4942 (cd->names_found - i) * cd->name_entry_size);
4943 break;
4944 }
4945
4946 /* Continue the loop for a later or duplicate name */
4947
4948 slot += cd->name_entry_size;
4949 }
4950
4951 /* For non-duplicate names, check for a duplicate number before
4952 adding the new name. */
4953
4954 if (!dupname)
4955 {
4956 uschar *cslot = cd->name_table;
4957 for (i = 0; i < cd->names_found; i++)
4958 {
4959 if (cslot != slot)
4960 {
4961 if (GET2(cslot, 0) == cd->bracount + 1)
4962 {
4963 *errorcodeptr = ERR65;
4964 goto FAILED;
4965 }
4966 }
4967 else i--;
4968 cslot += cd->name_entry_size;
4969 }
4970 }
4971
4972 PUT2(slot, 0, cd->bracount + 1);
4973 memcpy(slot + 2, name, namelen);
4974 slot[2+namelen] = 0;
4975 }
4976 }
4977
4978 /* In both pre-compile and compile, count the number of names we've
4979 encountered. */
4980
4981 cd->names_found++;
4982 ptr++; /* Move past > or ' */
4983 goto NUMBERED_GROUP;
4984
4985
4986 /* ------------------------------------------------------------ */
4987 case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
4988 terminator = CHAR_RIGHT_PARENTHESIS;
4989 is_recurse = TRUE;
4990 /* Fall through */
4991
4992 /* We come here from the Python syntax above that handles both
4993 references (?P=name) and recursion (?P>name), as well as falling
4994 through from the Perl recursion syntax (?&name). We also come here from
4995 the Perl \k<name> or \k'name' back reference syntax and the \k{name}
4996 .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
4997
4998 NAMED_REF_OR_RECURSE:
4999 name = ++ptr;
5000 while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
5001 namelen = ptr - name;
5002
5003 /* In the pre-compile phase, do a syntax check and set a dummy
5004 reference number. */
5005
5006 if (lengthptr != NULL)
5007 {
5008 if (namelen == 0)
5009 {
5010 *errorcodeptr = ERR62;
5011 goto FAILED;
5012 }
5013 if (*ptr != terminator)
5014 {
5015 *errorcodeptr = ERR42;
5016 goto FAILED;
5017 }
5018 if (namelen > MAX_NAME_SIZE)
5019 {
5020 *errorcodeptr = ERR48;
5021 goto FAILED;
5022 }
5023 recno = 0;
5024 }
5025
5026 /* In the real compile, seek the name in the table. We check the name
5027 first, and then check that we have reached the end of the name in the
5028 table. That way, if the name that is longer than any in the table,
5029 the comparison will fail without reading beyond the table entry. */
5030
5031 else
5032 {
5033 slot = cd->name_table;
5034 for (i = 0; i < cd->names_found; i++)
5035 {
5036 if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5037 slot[2+namelen] == 0)
5038 break;
5039 slot += cd->name_entry_size;
5040 }
5041
5042 if (i < cd->names_found) /* Back reference */
5043 {
5044 recno = GET2(slot, 0);
5045 }
5046 else if ((recno = /* Forward back reference */
5047 find_parens(cd, name, namelen,
5048 (options & PCRE_EXTENDED) != 0)) <= 0)
5049 {
5050 *errorcodeptr = ERR15;
5051 goto FAILED;
5052 }
5053 }
5054
5055 /* In both phases, we can now go to the code than handles numerical
5056 recursion or backreferences. */
5057
5058 if (is_recurse) goto HANDLE_RECURSION;
5059 else goto HANDLE_REFERENCE;
5060
5061
5062 /* ------------------------------------------------------------ */
5063 case CHAR_R: /* Recursion */
5064 ptr++; /* Same as (?0) */
5065 /* Fall through */
5066
5067
5068 /* ------------------------------------------------------------ */
5069 case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5070 case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5071 case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
5072 {
5073 const uschar *called;
5074 terminator = CHAR_RIGHT_PARENTHESIS;
5075
5076 /* Come here from the \g<...> and \g'...' code (Oniguruma
5077 compatibility). However, the syntax has been checked to ensure that
5078 the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5079 be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5080 ever be taken. */
5081
5082 HANDLE_NUMERICAL_RECURSION:
5083
5084 if ((refsign = *ptr) == CHAR_PLUS)
5085 {
5086 ptr++;
5087 if ((digitab[*ptr] & ctype_digit) == 0)
5088 {
5089 *errorcodeptr = ERR63;
5090 goto FAILED;
5091 }
5092 }
5093 else if (refsign == CHAR_MINUS)
5094 {
5095 if ((digitab[ptr[1]] & ctype_digit) == 0)
5096 goto OTHER_CHAR_AFTER_QUERY;
5097 ptr++;
5098 }
5099
5100 recno = 0;
5101 while((digitab[*ptr] & ctype_digit) != 0)
5102 recno = recno * 10 + *ptr++ - CHAR_0;
5103
5104 if (*ptr != terminator)
5105 {
5106 *errorcodeptr = ERR29;
5107 goto FAILED;
5108 }
5109
5110 if (refsign == CHAR_MINUS)
5111 {
5112 if (recno == 0)
5113 {
5114 *errorcodeptr = ERR58;
5115 goto FAILED;
5116 }
5117 recno = cd->bracount - recno + 1;
5118 if (recno <= 0)
5119 {
5120 *errorcodeptr = ERR15;
5121 goto FAILED;
5122 }
5123 }
5124 else if (refsign == CHAR_PLUS)
5125 {
5126 if (recno == 0)
5127 {
5128 *errorcodeptr = ERR58;
5129 goto FAILED;
5130 }
5131 recno += cd->bracount;
5132 }
5133
5134 /* Come here from code above that handles a named recursion */
5135
5136 HANDLE_RECURSION:
5137
5138 previous = code;
5139 called = cd->start_code;
5140
5141 /* When we are actually compiling, find the bracket that is being
5142 referenced. Temporarily end the regex in case it doesn't exist before
5143 this point. If we end up with a forward reference, first check that
5144 the bracket does occur later so we can give the error (and position)
5145 now. Then remember this forward reference in the workspace so it can
5146 be filled in at the end. */
5147
5148 if (lengthptr == NULL)
5149 {
5150 *code = OP_END;
5151 if (recno != 0)
5152 called = _pcre_find_bracket(cd->start_code, utf8, recno);
5153
5154 /* Forward reference */
5155
5156 if (called == NULL)
5157 {
5158 if (find_parens(cd, NULL, recno,
5159 (options & PCRE_EXTENDED) != 0) < 0)
5160 {
5161 *errorcodeptr = ERR15;
5162 goto FAILED;
5163 }
5164
5165 /* Fudge the value of "called" so that when it is inserted as an
5166 offset below, what it actually inserted is the reference number
5167 of the group. */
5168
5169 called = cd->start_code + recno;
5170 PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
5171 }
5172
5173 /* If not a forward reference, and the subpattern is still open,
5174 this is a recursive call. We check to see if this is a left
5175 recursion that could loop for ever, and diagnose that case. */
5176
5177 else if (GET(called, 1) == 0 &&
5178 could_be_empty(called, code, bcptr, utf8))
5179 {
5180 *errorcodeptr = ERR40;
5181 goto FAILED;
5182 }
5183 }
5184
5185 /* Insert the recursion/subroutine item, automatically wrapped inside
5186 "once" brackets. Set up a "previous group" length so that a
5187 subsequent quantifier will work. */
5188
5189 *code = OP_ONCE;
5190 PUT(code, 1, 2 + 2*LINK_SIZE);
5191 code += 1 + LINK_SIZE;
5192
5193 *code = OP_RECURSE;
5194 PUT(code, 1, called - cd->start_code);
5195 code += 1 + LINK_SIZE;
5196
5197 *code = OP_KET;
5198 PUT(code, 1, 2 + 2*LINK_SIZE);
5199 code += 1 + LINK_SIZE;
5200
5201 length_prevgroup = 3 + 3*LINK_SIZE;
5202 }
5203
5204 /* Can't determine a first byte now */
5205
5206 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5207 continue;
5208
5209
5210 /* ------------------------------------------------------------ */
5211 default: /* Other characters: check option setting */
5212 OTHER_CHAR_AFTER_QUERY:
5213 set = unset = 0;
5214 optset = &set;
5215
5216 while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
5217 {
5218 switch (*ptr++)
5219 {
5220 case CHAR_MINUS: optset = &unset; break;
5221
5222 case CHAR_J: /* Record that it changed in the external options */
5223 *optset |= PCRE_DUPNAMES;
5224 cd->external_flags |= PCRE_JCHANGED;
5225 break;
5226
5227 case CHAR_i: *optset |= PCRE_CASELESS; break;
5228 case CHAR_m: *optset |= PCRE_MULTILINE; break;
5229 case CHAR_s: *optset |= PCRE_DOTALL; break;
5230 case CHAR_x: *optset |= PCRE_EXTENDED; break;
5231 case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5232 case CHAR_X: *optset |= PCRE_EXTRA; break;
5233
5234 default: *errorcodeptr = ERR12;
5235 ptr--; /* Correct the offset */
5236 goto FAILED;
5237 }
5238 }
5239
5240 /* Set up the changed option bits, but don't change anything yet. */
5241
5242 newoptions = (options | set) & (~unset);
5243
5244 /* If the options ended with ')' this is not the start of a nested
5245 group with option changes, so the options change at this level. If this
5246 item is right at the start of the pattern, the options can be
5247 abstracted and made external in the pre-compile phase, and ignored in
5248 the compile phase. This can be helpful when matching -- for instance in
5249 caseless checking of required bytes.
5250
5251 If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5252 definitely *not* at the start of the pattern because something has been
5253 compiled. In the pre-compile phase, however, the code pointer can have
5254 that value after the start, because it gets reset as code is discarded
5255 during the pre-compile. However, this can happen only at top level - if
5256 we are within parentheses, the starting BRA will still be present. At
5257 any parenthesis level, the length value can be used to test if anything
5258 has been compiled at that level. Thus, a test for both these conditions
5259 is necessary to ensure we correctly detect the start of the pattern in
5260 both phases.
5261
5262 If we are not at the pattern start, compile code to change the ims
5263 options if this setting actually changes any of them, and reset the
5264 greedy defaults and the case value for firstbyte and reqbyte. */
5265
5266 if (*ptr == CHAR_RIGHT_PARENTHESIS)
5267 {
5268 if (code == cd->start_code + 1 + LINK_SIZE &&
5269 (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5270 {
5271 cd->external_options = newoptions;
5272 }
5273 else
5274 {
5275 if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5276 {
5277 *code++ = OP_OPT;
5278 *code++ = newoptions & PCRE_IMS;
5279 }
5280 greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5281 greedy_non_default = greedy_default ^ 1;
5282 req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
5283 }
5284
5285 /* Change options at this level, and pass them back for use
5286 in subsequent branches. When not at the start of the pattern, this
5287 information is also necessary so that a resetting item can be
5288 compiled at the end of a group (if we are in a group). */
5289
5290 *optionsptr = options = newoptions;
5291 previous = NULL; /* This item can't be repeated */
5292 continue; /* It is complete */
5293 }
5294
5295 /* If the options ended with ':' we are heading into a nested group
5296 with possible change of options. Such groups are non-capturing and are
5297 not assertions of any kind. All we need to do is skip over the ':';
5298 the newoptions value is handled below. */
5299
5300 bravalue = OP_BRA;
5301 ptr++;
5302 } /* End of switch for character following (? */
5303 } /* End of (? handling */
5304
5305 /* Opening parenthesis not followed by '?'. If PCRE_NO_AUTO_CAPTURE is set,
5306 all unadorned brackets become non-capturing and behave like (?:...)
5307 brackets. */
5308
5309 else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5310 {
5311 bravalue = OP_BRA;
5312 }
5313
5314 /* Else we have a capturing group. */
5315
5316 else
5317 {
5318 NUMBERED_GROUP:
5319 cd->bracount += 1;
5320 PUT2(code, 1+LINK_SIZE, cd->bracount);
5321 skipbytes = 2;
5322 }
5323
5324 /* Process nested bracketed regex. Assertions may not be repeated, but
5325 other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5326 non-register variable in order to be able to pass its address because some
5327 compilers complain otherwise. Pass in a new setting for the ims options if
5328 they have changed. */
5329
5330 previous = (bravalue >= OP_ONCE)? code : NULL;
5331 *code = bravalue;
5332 tempcode = code;
5333 tempreqvary = cd->req_varyopt; /* Save value before bracket */
5334 length_prevgroup = 0; /* Initialize for pre-compile phase */
5335
5336 if (!compile_regex(
5337 newoptions, /* The complete new option state */
5338 options & PCRE_IMS, /* The previous ims option state */
5339 &tempcode, /* Where to put code (updated) */
5340 &ptr, /* Input pointer (updated) */
5341 errorcodeptr, /* Where to put an error message */
5342 (bravalue == OP_ASSERTBACK ||
5343 bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5344 reset_bracount, /* True if (?| group */
5345 skipbytes, /* Skip over bracket number */
5346 &subfirstbyte, /* For possible first char */
5347 &subreqbyte, /* For possible last char */
5348 bcptr, /* Current branch chain */
5349 cd, /* Tables block */
5350 (lengthptr == NULL)? NULL : /* Actual compile phase */
5351 &length_prevgroup /* Pre-compile phase */
5352 ))
5353 goto FAILED;
5354
5355 /* At the end of compiling, code is still pointing to the start of the
5356 group, while tempcode has been updated to point past the end of the group
5357 and any option resetting that may follow it. The pattern pointer (ptr)
5358 is on the bracket. */
5359
5360 /* If this is a conditional bracket, check that there are no more than
5361 two branches in the group, or just one if it's a DEFINE group. We do this
5362 in the real compile phase, not in the pre-pass, where the whole group may
5363 not be available. */
5364
5365 if (bravalue == OP_COND && lengthptr == NULL)
5366 {
5367 uschar *tc = code;
5368 int condcount = 0;
5369
5370 do {
5371 condcount++;
5372 tc += GET(tc,1);
5373 }
5374 while (*tc != OP_KET);
5375
5376 /* A DEFINE group is never obeyed inline (the "condition" is always
5377 false). It must have only one branch. */
5378
5379 if (code[LINK_SIZE+1] == OP_DEF)
5380 {
5381 if (condcount > 1)
5382 {
5383 *errorcodeptr = ERR54;
5384 goto FAILED;
5385 }
5386 bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5387 }
5388
5389 /* A "normal" conditional group. If there is just one branch, we must not
5390 make use of its firstbyte or reqbyte, because this is equivalent to an
5391 empty second branch. */
5392
5393 else
5394 {
5395 if (condcount > 2)
5396 {
5397 *errorcodeptr = ERR27;
5398 goto FAILED;
5399 }
5400 if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5401 }
5402 }
5403
5404 /* Error if hit end of pattern */
5405
5406 if (*ptr != CHAR_RIGHT_PARENTHESIS)
5407 {
5408 *errorcodeptr = ERR14;
5409 goto FAILED;
5410 }
5411
5412 /* In the pre-compile phase, update the length by the length of the group,
5413 less the brackets at either end. Then reduce the compiled code to just a
5414 set of non-capturing brackets so that it doesn't use much memory if it is
5415 duplicated by a quantifier.*/
5416
5417 if (lengthptr != NULL)
5418 {
5419 if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5420 {
5421 *errorcodeptr = ERR20;
5422 goto FAILED;
5423 }
5424 *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5425 *code++ = OP_BRA;
5426 PUTINC(code, 0, 1 + LINK_SIZE);
5427 *code++ = OP_KET;
5428 PUTINC(code, 0, 1 + LINK_SIZE);
5429 break; /* No need to waste time with special character handling */
5430 }
5431
5432 /* Otherwise update the main code pointer to the end of the group. */
5433
5434 code = tempcode;
5435
5436 /* For a DEFINE group, required and first character settings are not
5437 relevant. */
5438
5439 if (bravalue == OP_DEF) break;
5440
5441 /* Handle updating of the required and first characters for other types of
5442 group. Update for normal brackets of all kinds, and conditions with two
5443 branches (see code above). If the bracket is followed by a quantifier with
5444 zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5445 zerofirstbyte outside the main loop so that they can be accessed for the
5446 back off. */
5447
5448 zeroreqbyte = reqbyte;
5449 zerofirstbyte = firstbyte;
5450 groupsetfirstbyte = FALSE;
5451
5452 if (bravalue >= OP_ONCE)
5453 {
5454 /* If we have not yet set a firstbyte in this branch, take it from the
5455 subpattern, remembering that it was set here so that a repeat of more
5456 than one can replicate it as reqbyte if necessary. If the subpattern has
5457 no firstbyte, set "none" for the whole branch. In both cases, a zero
5458 repeat forces firstbyte to "none". */
5459
5460 if (firstbyte == REQ_UNSET)
5461 {
5462 if (subfirstbyte >= 0)
5463 {
5464 firstbyte = subfirstbyte;
5465 groupsetfirstbyte = TRUE;
5466 }
5467 else firstbyte = REQ_NONE;
5468 zerofirstbyte = REQ_NONE;
5469 }
5470
5471 /* If firstbyte was previously set, convert the subpattern's firstbyte
5472 into reqbyte if there wasn't one, using the vary flag that was in
5473 existence beforehand. */
5474
5475 else if (subfirstbyte >= 0 && subreqbyte < 0)
5476 subreqbyte = subfirstbyte | tempreqvary;
5477
5478 /* If the subpattern set a required byte (or set a first byte that isn't
5479 really the first byte - see above), set it. */
5480
5481 if (subreqbyte >= 0) reqbyte = subreqbyte;
5482 }
5483
5484 /* For a forward assertion, we take the reqbyte, if set. This can be
5485 helpful if the pattern that follows the assertion doesn't set a different
5486 char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5487 for an assertion, however because it leads to incorrect effect for patterns
5488 such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5489 of a firstbyte. This is overcome by a scan at the end if there's no
5490 firstbyte, looking for an asserted first char. */
5491
5492 else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5493 break; /* End of processing '(' */
5494
5495
5496 /* ===================================================================*/
5497 /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
5498 are arranged to be the negation of the corresponding OP_values. For the
5499 back references, the values are ESC_REF plus the reference number. Only
5500 back references and those types that consume a character may be repeated.
5501 We can test for values between ESC_b and ESC_Z for the latter; this may
5502 have to change if any new ones are ever created. */
5503
5504 case CHAR_BACKSLASH:
5505 tempptr = ptr;
5506 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5507 if (*errorcodeptr != 0) goto FAILED;
5508
5509 if (c < 0)
5510 {
5511 if (-c == ESC_Q) /* Handle start of quoted string */
5512 {
5513 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5514 ptr += 2; /* avoid empty string */
5515 else inescq = TRUE;
5516 continue;
5517 }
5518
5519 if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5520
5521 /* For metasequences that actually match a character, we disable the
5522 setting of a first character if it hasn't already been set. */
5523
5524 if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5525 firstbyte = REQ_NONE;
5526
5527 /* Set values to reset to if this is followed by a zero repeat. */
5528
5529 zerofirstbyte = firstbyte;
5530 zeroreqbyte = reqbyte;
5531
5532 /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5533 is a subroutine call by number (Oniguruma syntax). In fact, the value
5534 -ESC_g is returned only for these cases. So we don't need to check for <
5535 or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5536 -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5537 that is a synonym for a named back reference). */
5538
5539 if (-c == ESC_g)
5540 {
5541 const uschar *p;
5542 save_hwm = cd->hwm; /* Normally this is set when '(' is read */
5543 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5544 CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
5545
5546 /* These two statements stop the compiler for warning about possibly
5547 unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5548 fact, because we actually check for a number below, the paths that
5549 would actually be in error are never taken. */
5550
5551 skipbytes = 0;
5552 reset_bracount = FALSE;
5553
5554 /* Test for a name */
5555
5556 if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
5557 {
5558 BOOL isnumber = TRUE;
5559 for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5560 {
5561 if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5562 if ((cd->ctypes[*p] & ctype_word) == 0) break;
5563 }
5564 if (*p != terminator)
5565 {
5566 *errorcodeptr = ERR57;
5567 break;
5568 }
5569 if (isnumber)
5570 {
5571 ptr++;
5572 goto HANDLE_NUMERICAL_RECURSION;
5573 }
5574 is_recurse = TRUE;
5575 goto NAMED_REF_OR_RECURSE;
5576 }
5577
5578 /* Test a signed number in angle brackets or quotes. */
5579
5580 p = ptr + 2;
5581 while ((digitab[*p] & ctype_digit) != 0) p++;
5582 if (*p != terminator)
5583 {
5584 *errorcodeptr = ERR57;
5585 break;
5586 }
5587 ptr++;
5588 goto HANDLE_NUMERICAL_RECURSION;
5589 }
5590
5591 /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5592 We also support \k{name} (.NET syntax) */
5593
5594 if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5595 ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
5596 {
5597 is_recurse = FALSE;
5598 terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5599 CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5600 CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
5601 goto NAMED_REF_OR_RECURSE;
5602 }
5603
5604 /* Back references are handled specially; must disable firstbyte if
5605 not set to cope with cases like (?=(\w+))\1: which would otherwise set
5606 ':' later. */
5607
5608 if (-c >= ESC_REF)
5609 {
5610 open_capitem *oc;
5611 recno = -c - ESC_REF;
5612
5613 HANDLE_REFERENCE: /* Come here from named backref handling */
5614 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5615 previous = code;
5616 *code++ = OP_REF;
5617 PUT2INC(code, 0, recno);
5618 cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5619 if (recno > cd->top_backref) cd->top_backref = recno;
5620
5621 /* Check to see if this back reference is recursive, that it, it
5622 is inside the group that it references. A flag is set so that the
5623 group can be made atomic. */
5624
5625 for (oc = cd->open_caps; oc != NULL; oc = oc->next)
5626 {
5627 if (oc->number == recno)
5628 {
5629 oc->flag = TRUE;
5630 break;
5631 }
5632 }
5633 }
5634
5635 /* So are Unicode property matches, if supported. */
5636
5637 #ifdef SUPPORT_UCP
5638 else if (-c == ESC_P || -c == ESC_p)
5639 {
5640 BOOL negated;
5641 int pdata;
5642 int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5643 if (ptype < 0) goto FAILED;
5644 previous = code;
5645 *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5646 *code++ = ptype;
5647 *code++ = pdata;
5648 }
5649 #else
5650
5651 /* If Unicode properties are not supported, \X, \P, and \p are not
5652 allowed. */
5653
5654 else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5655 {
5656 *errorcodeptr = ERR45;
5657 goto FAILED;
5658 }
5659 #endif
5660
5661 /* For the rest (including \X when Unicode properties are supported), we
5662 can obtain the OP value by negating the escape value. */
5663
5664 else
5665 {
5666 previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
5667 *code++ = -c;
5668 }
5669 continue;
5670 }
5671
5672 /* We have a data character whose value is in c. In UTF-8 mode it may have
5673 a value > 127. We set its representation in the length/buffer, and then
5674 handle it as a data character. */
5675
5676 #ifdef SUPPORT_UTF8
5677 if (utf8 && c > 127)
5678 mclength = _pcre_ord2utf8(c, mcbuffer);
5679 else
5680 #endif
5681
5682 {
5683 mcbuffer[0] = c;
5684 mclength = 1;
5685 }
5686 goto ONE_CHAR;
5687
5688
5689 /* ===================================================================*/
5690 /* Handle a literal character. It is guaranteed not to be whitespace or #
5691 when the extended flag is set. If we are in UTF-8 mode, it may be a
5692 multi-byte literal character. */
5693
5694 default:
5695 NORMAL_CHAR:
5696 mclength = 1;
5697 mcbuffer[0] = c;
5698
5699 #ifdef SUPPORT_UTF8
5700 if (utf8 && c >= 0xc0)
5701 {
5702 while ((ptr[1] & 0xc0) == 0x80)
5703 mcbuffer[mclength++] = *(++ptr);
5704 }
5705 #endif
5706
5707 /* At this point we have the character's bytes in mcbuffer, and the length
5708 in mclength. When not in UTF-8 mode, the length is always 1. */
5709
5710 ONE_CHAR:
5711 previous = code;
5712 *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
5713 for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
5714
5715 /* Remember if \r or \n were seen */
5716
5717 if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
5718 cd->external_flags |= PCRE_HASCRORLF;
5719
5720 /* Set the first and required bytes appropriately. If no previous first
5721 byte, set it from this character, but revert to none on a zero repeat.
5722 Otherwise, leave the firstbyte value alone, and don't change it on a zero
5723 repeat. */
5724
5725 if (firstbyte == REQ_UNSET)
5726 {
5727 zerofirstbyte = REQ_NONE;
5728 zeroreqbyte = reqbyte;
5729
5730 /* If the character is more than one byte long, we can set firstbyte
5731 only if it is not to be matched caselessly. */
5732
5733 if (mclength == 1 || req_caseopt == 0)
5734 {
5735 firstbyte = mcbuffer[0] | req_caseopt;
5736 if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
5737 }
5738 else firstbyte = reqbyte = REQ_NONE;
5739 }
5740
5741 /* firstbyte was previously set; we can set reqbyte only the length is
5742 1 or the matching is caseful. */
5743
5744 else
5745 {
5746 zerofirstbyte = firstbyte;
5747 zeroreqbyte = reqbyte;
5748 if (mclength == 1 || req_caseopt == 0)
5749 reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
5750 }
5751
5752 break; /* End of literal character handling */
5753 }
5754 } /* end of big loop */
5755
5756
5757 /* Control never reaches here by falling through, only by a goto for all the
5758 error states. Pass back the position in the pattern so that it can be displayed
5759 to the user for diagnosing the error. */
5760
5761 FAILED:
5762 *ptrptr = ptr;
5763 return FALSE;
5764 }
5765
5766
5767
5768
5769 /*************************************************
5770 * Compile sequence of alternatives *
5771 *************************************************/
5772
5773 /* On entry, ptr is pointing past the bracket character, but on return it
5774 points to the closing bracket, or vertical bar, or end of string. The code
5775 variable is pointing at the byte into which the BRA operator has been stored.
5776 If the ims options are changed at the start (for a (?ims: group) or during any
5777 branch, we need to insert an OP_OPT item at the start of every following branch
5778 to ensure they get set correctly at run time, and also pass the new options
5779 into every subsequent branch compile.
5780
5781 This function is used during the pre-compile phase when we are trying to find
5782 out the amount of memory needed, as well as during the real compile phase. The
5783 value of lengthptr distinguishes the two phases.
5784
5785 Arguments:
5786 options option bits, including any changes for this subpattern
5787 oldims previous settings of ims option bits
5788 codeptr -> the address of the current code pointer
5789 ptrptr -> the address of the current pattern pointer
5790 errorcodeptr -> pointer to error code variable
5791 lookbehind TRUE if this is a lookbehind assertion
5792 reset_bracount TRUE to reset the count for each branch
5793 skipbytes skip this many bytes at start (for brackets and OP_COND)
5794 firstbyteptr place to put the first required character, or a negative number
5795 reqbyteptr place to put the last required character, or a negative number
5796 bcptr pointer to the chain of currently open branches
5797 cd points to the data block with tables pointers etc.
5798 lengthptr NULL during the real compile phase
5799 points to length accumulator during pre-compile phase
5800
5801 Returns: TRUE on success
5802 */
5803
5804 static BOOL
5805 compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
5806 int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
5807 int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
5808 int *lengthptr)
5809 {
5810 const uschar *ptr = *ptrptr;
5811 uschar *code = *codeptr;
5812 uschar *last_branch = code;
5813 uschar *start_bracket = code;
5814 uschar *reverse_count = NULL;
5815 open_capitem capitem;
5816 int capnumber = 0;
5817 int firstbyte, reqbyte;
5818 int branchfirstbyte, branchreqbyte;
5819 int length;
5820 int orig_bracount;
5821 int max_bracount;
5822 int old_external_options = cd->external_options;
5823 branch_chain bc;
5824
5825 bc.outer = bcptr;
5826 bc.current_branch = code;
5827
5828 firstbyte = reqbyte = REQ_UNSET;
5829
5830 /* Accumulate the length for use in the pre-compile phase. Start with the
5831 length of the BRA and KET and any extra bytes that are required at the
5832 beginning. We accumulate in a local variable to save frequent testing of
5833 lenthptr for NULL. We cannot do this by looking at the value of code at the
5834 start and end of each alternative, because compiled items are discarded during
5835 the pre-compile phase so that the work space is not exceeded. */
5836
5837 length = 2 + 2*LINK_SIZE + skipbytes;
5838
5839 /* WARNING: If the above line is changed for any reason, you must also change
5840 the code that abstracts option settings at the start of the pattern and makes
5841 them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
5842 pre-compile phase to find out whether anything has yet been compiled or not. */
5843
5844 /* If this is a capturing subpattern, add to the chain of open capturing items
5845 so that we can detect them if (*ACCEPT) is encountered. This is also used to
5846 detect groups that contain recursive back references to themselves. */
5847
5848 if (*code == OP_CBRA)
5849 {
5850 capnumber = GET2(code, 1 + LINK_SIZE);
5851 capitem.number = capnumber;
5852 capitem.next = cd->open_caps;
5853 capitem.flag = FALSE;
5854 cd->open_caps = &capitem;
5855 }
5856
5857 /* Offset is set zero to mark that this bracket is still open */
5858
5859 PUT(code, 1, 0);
5860 code += 1 + LINK_SIZE + skipbytes;
5861
5862 /* Loop for each alternative branch */
5863
5864 orig_bracount = max_bracount = cd->bracount;
5865 for (;;)
5866 {
5867 /* For a (?| group, reset the capturing bracket count so that each branch
5868 uses the same numbers. */
5869
5870 if (reset_bracount) cd->bracount = orig_bracount;
5871
5872 /* Handle a change of ims options at the start of the branch */
5873
5874 if ((options & PCRE_IMS) != oldims)
5875 {
5876 *code++ = OP_OPT;
5877 *code++ = options & PCRE_IMS;
5878 length += 2;
5879 }
5880
5881 /* Set up dummy OP_REVERSE if lookbehind assertion */
5882
5883 if (lookbehind)
5884 {
5885 *code++ = OP_REVERSE;
5886 reverse_count = code;
5887 PUTINC(code, 0, 0);
5888 length += 1 + LINK_SIZE;
5889 }
5890
5891 /* Now compile the branch; in the pre-compile phase its length gets added
5892 into the length. */
5893
5894 if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
5895 &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
5896 {
5897 *ptrptr = ptr;
5898 return FALSE;
5899 }
5900
5901 /* If the external options have changed during this branch, it means that we
5902 are at the top level, and a leading option setting has been encountered. We
5903 need to re-set the original option values to take account of this so that,
5904 during the pre-compile phase, we know to allow for a re-set at the start of
5905 subsequent branches. */
5906
5907 if (old_external_options != cd->external_options)
5908 oldims = cd->external_options & PCRE_IMS;
5909
5910 /* Keep the highest bracket count in case (?| was used and some branch
5911 has fewer than the rest. */
5912
5913 if (cd->bracount > max_bracount) max_bracount = cd->bracount;
5914
5915 /* In the real compile phase, there is some post-processing to be done. */
5916
5917 if (lengthptr == NULL)
5918 {
5919 /* If this is the first branch, the firstbyte and reqbyte values for the
5920 branch become the values for the regex. */
5921
5922 if (*last_branch != OP_ALT)
5923 {
5924 firstbyte = branchfirstbyte;
5925 reqbyte = branchreqbyte;
5926 }
5927
5928 /* If this is not the first branch, the first char and reqbyte have to
5929 match the values from all the previous branches, except that if the
5930 previous value for reqbyte didn't have REQ_VARY set, it can still match,
5931 and we set REQ_VARY for the regex. */
5932
5933 else
5934 {
5935 /* If we previously had a firstbyte, but it doesn't match the new branch,
5936 we have to abandon the firstbyte for the regex, but if there was
5937 previously no reqbyte, it takes on the value of the old firstbyte. */
5938
5939 if (firstbyte >= 0 && firstbyte != branchfirstbyte)
5940 {
5941 if (reqbyte < 0) reqbyte = firstbyte;
5942 firstbyte = REQ_NONE;
5943 }
5944
5945 /* If we (now or from before) have no firstbyte, a firstbyte from the
5946 branch becomes a reqbyte if there isn't a branch reqbyte. */
5947
5948 if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
5949 branchreqbyte = branchfirstbyte;
5950
5951 /* Now ensure that the reqbytes match */
5952
5953 if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
5954 reqbyte = REQ_NONE;
5955 else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
5956 }
5957
5958 /* If lookbehind, check that this branch matches a fixed-length string, and
5959 put the length into the OP_REVERSE item. Temporarily mark the end of the
5960 branch with OP_END. If the branch contains OP_RECURSE, the result is -3
5961 because there may be forward references that we can't check here. Set a
5962 flag to cause another lookbehind check at the end. Why not do it all at the
5963 end? Because common, erroneous checks are picked up here and the offset of
5964 the problem can be shown. */
5965
5966 if (lookbehind)
5967 {
5968 int fixed_length;
5969 *code = OP_END;
5970 fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
5971 DPRINTF(("fixed length = %d\n", fixed_length));
5972 if (fixed_length == -3)
5973 {
5974 cd->check_lookbehind = TRUE;
5975 }
5976 else if (fixed_length < 0)
5977 {
5978 *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
5979 *ptrptr = ptr;
5980 return FALSE;
5981 }
5982 else { PUT(reverse_count, 0, fixed_length); }
5983 }
5984 }
5985
5986 /* Reached end of expression, either ')' or end of pattern. In the real
5987 compile phase, go back through the alternative branches and reverse the chain
5988 of offsets, with the field in the BRA item now becoming an offset to the
5989 first alternative. If there are no alternatives, it points to the end of the
5990 group. The length in the terminating ket is always the length of the whole
5991 bracketed item. If any of the ims options were changed inside the group,
5992 compile a resetting op-code following, except at the very end of the pattern.
5993 Return leaving the pointer at the terminating char. */
5994
5995 if (*ptr != CHAR_VERTICAL_LINE)
5996 {
5997 if (lengthptr == NULL)
5998 {
5999 int branch_length = code - last_branch;
6000 do
6001 {
6002 int prev_length = GET(last_branch, 1);
6003 PUT(last_branch, 1, branch_length);
6004 branch_length = prev_length;
6005 last_branch -= branch_length;
6006 }
6007 while (branch_length > 0);
6008 }
6009
6010 /* Fill in the ket */
6011
6012 *code = OP_KET;
6013 PUT(code, 1, code - start_bracket);
6014 code += 1 + LINK_SIZE;
6015
6016 /* If it was a capturing subpattern, check to see if it contained any
6017 recursive back references. If so, we must wrap it in atomic brackets.
6018 In any event, remove the block from the chain. */
6019
6020 if (capnumber > 0)
6021 {
6022 if (cd->open_caps->flag)
6023 {
6024 memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
6025 code - start_bracket);
6026 *start_bracket = OP_ONCE;
6027 code += 1 + LINK_SIZE;
6028 PUT(start_bracket, 1, code - start_bracket);
6029 *code = OP_KET;
6030 PUT(code, 1, code - start_bracket);
6031 code += 1 + LINK_SIZE;
6032 length += 2 + 2*LINK_SIZE;
6033 }
6034 cd->open_caps = cd->open_caps->next;
6035 }
6036
6037 /* Reset options if needed. */
6038
6039 if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
6040 {
6041 *code++ = OP_OPT;
6042 *code++ = oldims;
6043 length += 2;
6044 }
6045
6046 /* Retain the highest bracket number, in case resetting was used. */
6047
6048 cd->bracount = max_bracount;
6049
6050 /* Set values to pass back */
6051
6052 *codeptr = code;
6053 *ptrptr = ptr;
6054 *firstbyteptr = firstbyte;
6055 *reqbyteptr = reqbyte;
6056 if (lengthptr != NULL)
6057 {
6058 if (OFLOW_MAX - *lengthptr < length)
6059 {
6060 *errorcodeptr = ERR20;
6061 return FALSE;
6062 }
6063 *lengthptr += length;
6064 }
6065 return TRUE;
6066 }
6067
6068 /* Another branch follows. In the pre-compile phase, we can move the code
6069 pointer back to where it was for the start of the first branch. (That is,
6070 pretend that each branch is the only one.)
6071
6072 In the real compile phase, insert an ALT node. Its length field points back
6073 to the previous branch while the bracket remains open. At the end the chain
6074 is reversed. It's done like this so that the start of the bracket has a
6075 zero offset until it is closed, making it possible to detect recursion. */
6076
6077 if (lengthptr != NULL)
6078 {
6079 code = *codeptr + 1 + LINK_SIZE + skipbytes;
6080 length += 1 + LINK_SIZE;
6081 }
6082 else
6083 {
6084 *code = OP_ALT;
6085 PUT(code, 1, code - last_branch);
6086 bc.current_branch = last_branch = code;
6087 code += 1 + LINK_SIZE;
6088 }
6089
6090 ptr++;
6091 }
6092 /* Control never reaches here */
6093 }
6094
6095
6096
6097
6098 /*************************************************
6099 * Check for anchored expression *
6100 *************************************************/
6101
6102 /* Try to find out if this is an anchored regular expression. Consider each
6103 alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6104 all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6105 it's anchored. However, if this is a multiline pattern, then only OP_SOD
6106 counts, since OP_CIRC can match in the middle.
6107
6108 We can also consider a regex to be anchored if OP_SOM starts all its branches.
6109 This is the code for \G, which means "match at start of match position, taking
6110 into account the match offset".
6111
6112 A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6113 because that will try the rest of the pattern at all possible matching points,
6114 so there is no point trying again.... er ....
6115
6116 .... except when the .* appears inside capturing parentheses, and there is a
6117 subsequent back reference to those parentheses. We haven't enough information
6118 to catch that case precisely.
6119
6120 At first, the best we could do was to detect when .* was in capturing brackets
6121 and the highest back reference was greater than or equal to that level.
6122 However, by keeping a bitmap of the first 31 back references, we can catch some
6123 of the more common cases more precisely.
6124
6125 Arguments:
6126 code points to start of expression (the bracket)
6127 options points to the options setting
6128 bracket_map a bitmap of which brackets we are inside while testing; this
6129 handles up to substring 31; after that we just have to take
6130 the less precise approach
6131 backref_map the back reference bitmap
6132
6133 Returns: TRUE or FALSE
6134 */
6135
6136 static BOOL
6137 is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6138 unsigned int backref_map)
6139 {
6140 do {
6141 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6142 options, PCRE_MULTILINE, FALSE);
6143 register int op = *scode;
6144
6145 /* Non-capturing brackets */
6146
6147 if (op == OP_BRA)
6148 {
6149 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6150 }
6151
6152 /* Capturing brackets */
6153
6154 else if (op == OP_CBRA)
6155 {
6156 int n = GET2(scode, 1+LINK_SIZE);
6157 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6158 if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6159 }
6160
6161 /* Other brackets */
6162
6163 else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6164 {
6165 if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6166 }
6167
6168 /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6169 it isn't in brackets that are or may be referenced. */
6170
6171 else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6172 op == OP_TYPEPOSSTAR))
6173 {
6174 if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6175 return FALSE;
6176 }
6177
6178 /* Check for explicit anchoring */
6179
6180 else if (op != OP_SOD && op != OP_SOM &&
6181 ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6182 return FALSE;
6183 code += GET(code, 1);
6184 }
6185 while (*code == OP_ALT); /* Loop for each alternative */
6186 return TRUE;
6187 }
6188
6189
6190
6191 /*************************************************
6192 * Check for starting with ^ or .* *
6193 *************************************************/
6194
6195 /* This is called to find out if every branch starts with ^ or .* so that
6196 "first char" processing can be done to speed things up in multiline
6197 matching and for non-DOTALL patterns that start with .* (which must start at
6198 the beginning or after \n). As in the case of is_anchored() (see above), we
6199 have to take account of back references to capturing brackets that contain .*
6200 because in that case we can't make the assumption.
6201
6202 Arguments:
6203 code points to start of expression (the bracket)
6204 bracket_map a bitmap of which brackets we are inside while testing; this
6205 handles up to substring 31; after that we just have to take
6206 the less precise approach
6207 backref_map the back reference bitmap
6208
6209 Returns: TRUE or FALSE
6210 */
6211
6212 static BOOL
6213 is_startline(const uschar *code, unsigned int bracket_map,
6214 unsigned int backref_map)
6215 {
6216 do {
6217 const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6218 NULL, 0, FALSE);
6219 register int op = *scode;
6220
6221 /* If we are at the start of a conditional assertion group, *both* the
6222 conditional assertion *and* what follows the condition must satisfy the test
6223 for start of line. Other kinds of condition fail. Note that there may be an
6224 auto-callout at the start of a condition. */
6225
6226 if (op == OP_COND)
6227 {
6228 scode += 1 + LINK_SIZE;
6229 if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6230 switch (*scode)
6231 {
6232 case OP_CREF:
6233 case OP_NCREF:
6234 case OP_RREF:
6235 case OP_NRREF:
6236 case OP_DEF:
6237 return FALSE;
6238
6239 default: /* Assertion */
6240 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6241 do scode += GET(scode, 1); while (*scode == OP_ALT);
6242 scode += 1 + LINK_SIZE;
6243 break;
6244 }
6245 scode = first_significant_code(scode, NULL, 0, FALSE);
6246 op = *scode;
6247 }
6248
6249 /* Non-capturing brackets */
6250
6251 if (op == OP_BRA)
6252 {
6253 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6254 }
6255
6256 /* Capturing brackets */
6257
6258 else if (op == OP_CBRA)
6259 {
6260 int n = GET2(scode, 1+LINK_SIZE);
6261 int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6262 if (!is_startline(scode, new_map, backref_map)) return FALSE;
6263 }
6264
6265 /* Other brackets */
6266
6267 else if (op == OP_ASSERT || op == OP_ONCE)
6268 {
6269 if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6270 }
6271
6272 /* .* means "start at start or after \n" if it isn't in brackets that
6273 may be referenced. */
6274
6275 else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6276 {
6277 if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6278 }
6279
6280 /* Check for explicit circumflex */
6281
6282 else if (op != OP_CIRC) return FALSE;
6283
6284 /* Move on to the next alternative */
6285
6286 code += GET(code, 1);
6287 }
6288 while (*code == OP_ALT); /* Loop for each alternative */
6289 return TRUE;
6290 }
6291
6292
6293
6294 /*************************************************
6295 * Check for asserted fixed first char *
6296 *************************************************/
6297
6298 /* During compilation, the "first char" settings from forward assertions are
6299 discarded, because they can cause conflicts with actual literals that follow.
6300 However, if we end up without a first char setting for an unanchored pattern,
6301 it is worth scanning the regex to see if there is an initial asserted first
6302 char. If all branches start with the same asserted char, or with a bracket all
6303 of whose alternatives start with the same asserted char (recurse ad lib), then
6304 we return that char, otherwise -1.
6305
6306 Arguments:
6307 code points to start of expression (the bracket)
6308 options pointer to the options (used to check casing changes)
6309 inassert TRUE if in an assertion
6310
6311 Returns: -1 or the fixed first char
6312 */
6313
6314 static int
6315 find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6316 {
6317 register int c = -1;
6318 do {
6319 int d;
6320 const uschar *scode =
6321 first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6322 register int op = *scode;
6323
6324 switch(op)
6325 {
6326 default:
6327 return -1;
6328
6329 case OP_BRA:
6330 case OP_CBRA:
6331 case OP_ASSERT:
6332 case OP_ONCE:
6333 case OP_COND:
6334 if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6335 return -1;
6336 if (c < 0) c = d; else if (c != d) return -1;
6337 break;
6338
6339 case OP_EXACT: /* Fall through */
6340 scode += 2;
6341
6342 case OP_CHAR:
6343 case OP_CHARNC:
6344 case OP_PLUS:
6345 case OP_MINPLUS:
6346 case OP_POSPLUS:
6347 if (!inassert) return -1;
6348 if (c < 0)
6349 {
6350 c = scode[1];
6351 if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6352 }
6353 else if (c != scode[1]) return -1;
6354 break;
6355 }
6356
6357 code += GET(code, 1);
6358 }
6359 while (*code == OP_ALT);
6360 return c;
6361 }
6362
6363
6364
6365 /*************************************************
6366 * Compile a Regular Expression *
6367 *************************************************/
6368
6369 /* This function takes a string and returns a pointer to a block of store
6370 holding a compiled version of the expression. The original API for this
6371 function had no error code return variable; it is retained for backwards
6372 compatibility. The new function is given a new name.
6373
6374 Arguments:
6375 pattern the regular expression
6376 options various option bits
6377 errorcodeptr pointer to error code variable (pcre_compile2() only)
6378 can be NULL if you don't want a code value
6379 errorptr pointer to pointer to error text
6380 erroroffset ptr offset in pattern where error was detected
6381 tables pointer to character tables or NULL
6382
6383 Returns: pointer to compiled data block, or NULL on error,
6384 with errorptr and erroroffset set
6385 */
6386
6387 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6388 pcre_compile(const char *pattern, int options, const char **errorptr,
6389 int *erroroffset, const unsigned char *tables)
6390 {
6391 return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6392 }
6393
6394
6395 PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
6396 pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6397 const char **errorptr, int *erroroffset, const unsigned char *tables)
6398 {
6399 real_pcre *re;
6400 int length = 1; /* For final END opcode */
6401 int firstbyte, reqbyte, newline;
6402 int errorcode = 0;
6403 int skipatstart = 0;
6404 BOOL utf8 = (options & PCRE_UTF8) != 0;
6405 size_t size;
6406 uschar *code;
6407 const uschar *codestart;
6408 const uschar *ptr;
6409 compile_data compile_block;
6410 compile_data *cd = &compile_block;
6411
6412 /* This space is used for "compiling" into during the first phase, when we are
6413 computing the amount of memory that is needed. Compiled items are thrown away
6414 as soon as possible, so that a fairly large buffer should be sufficient for
6415 this purpose. The same space is used in the second phase for remembering where
6416 to fill in forward references to subpatterns. */
6417
6418 uschar cworkspace[COMPILE_WORK_SIZE];
6419
6420 /* Set this early so that early errors get offset 0. */
6421
6422 ptr = (const uschar *)pattern;
6423
6424 /* We can't pass back an error message if errorptr is NULL; I guess the best we
6425 can do is just return NULL, but we can set a code value if there is a code
6426 pointer. */
6427
6428 if (errorptr == NULL)
6429 {
6430 if (errorcodeptr != NULL) *errorcodeptr = 99;
6431 return NULL;
6432 }
6433
6434 *errorptr = NULL;
6435 if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6436
6437 /* However, we can give a message for this error */
6438
6439 if (erroroffset == NULL)
6440 {
6441 errorcode = ERR16;
6442 goto PCRE_EARLY_ERROR_RETURN2;
6443 }
6444
6445 *erroroffset = 0;
6446
6447 /* Set up pointers to the individual character tables */
6448
6449 if (tables == NULL) tables = _pcre_default_tables;
6450 cd->lcc = tables + lcc_offset;
6451 cd->fcc = tables + fcc_offset;
6452 cd->cbits = tables + cbits_offset;
6453 cd->ctypes = tables + ctypes_offset;
6454
6455 /* Check that all undefined public option bits are zero */
6456
6457 if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6458 {
6459 errorcode = ERR17;
6460 goto PCRE_EARLY_ERROR_RETURN;
6461 }
6462
6463 /* Check for global one-time settings at the start of the pattern, and remember
6464 the offset for later. */
6465
6466 while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6467 ptr[skipatstart+1] == CHAR_ASTERISK)
6468 {
6469 int newnl = 0;
6470 int newbsr = 0;
6471
6472 if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6473 { skipatstart += 7; options |= PCRE_UTF8; continue; }
6474
6475 if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
6476 { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
6477 else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
6478 { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
6479 else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
6480 { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
6481 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
6482 { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
6483 else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
6484 { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6485
6486 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
6487 { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
6488 else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
6489 { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6490
6491 if (newnl != 0)
6492 options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6493 else if (newbsr != 0)
6494 options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6495 else break;
6496 }
6497
6498 /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6499
6500 #ifdef SUPPORT_UTF8
6501 if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
6502 (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
6503 {
6504 errorcode = ERR44;
6505 goto PCRE_EARLY_ERROR_RETURN2;
6506 }
6507 #else
6508 if (utf8)
6509 {
6510 errorcode = ERR32;
6511 goto PCRE_EARLY_ERROR_RETURN;
6512 }
6513 #endif
6514
6515 /* Check validity of \R options. */
6516
6517 switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6518 {
6519 case 0:
6520 case PCRE_BSR_ANYCRLF:
6521 case PCRE_BSR_UNICODE:
6522 break;
6523 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6524 }
6525
6526 /* Handle different types of newline. The three bits give seven cases. The
6527 current code allows for fixed one- or two-byte sequences, plus "any" and
6528 "anycrlf". */
6529
6530 switch (options & PCRE_NEWLINE_BITS)
6531 {
6532 case 0: newline = NEWLINE; break; /* Build-time default */
6533 case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6534 case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
6535 case PCRE_NEWLINE_CR+
6536 PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
6537 case PCRE_NEWLINE_ANY: newline = -1; break;
6538 case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6539 default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6540 }
6541
6542 if (newline == -2)
6543 {
6544 cd->nltype = NLTYPE_ANYCRLF;
6545 }
6546 else if (newline < 0)
6547 {
6548 cd->nltype = NLTYPE_ANY;
6549 }
6550 else
6551 {
6552 cd->nltype = NLTYPE_FIXED;
6553 if (newline > 255)
6554 {
6555 cd->nllen = 2;
6556 cd->nl[0] = (newline >> 8) & 255;
6557 cd->nl[1] = newline & 255;
6558 }
6559 else
6560 {
6561 cd->nllen = 1;
6562 cd->nl[0] = newline;
6563 }
6564 }
6565
6566 /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6567 references to help in deciding whether (.*) can be treated as anchored or not.
6568 */
6569
6570 cd->top_backref = 0;
6571 cd->backref_map = 0;
6572
6573 /* Reflect pattern for debugging output */
6574
6575 DPRINTF(("------------------------------------------------------------------\n"));
6576 DPRINTF(("%s\n", pattern));
6577
6578 /* Pretend to compile the pattern while actually just accumulating the length
6579 of memory required. This behaviour is triggered by passing a non-NULL final
6580 argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6581 to compile parts of the pattern into; the compiled code is discarded when it is
6582 no longer needed, so hopefully this workspace will never overflow, though there
6583 is a test for its doing so. */
6584
6585 cd->bracount = cd->final_bracount = 0;
6586 cd->names_found = 0;
6587 cd->name_entry_size = 0;
6588 cd->name_table = NULL;
6589 cd->start_workspace = cworkspace;
6590 cd->start_code = cworkspace;
6591 cd->hwm = cworkspace;
6592 cd->start_pattern = (const uschar *)pattern;
6593 cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6594 cd->req_varyopt = 0;
6595 cd->external_options = options;
6596 cd->external_flags = 0;
6597 cd->open_caps = NULL;
6598
6599 /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6600 don't need to look at the result of the function here. The initial options have
6601 been put into the cd block so that they can be changed if an option setting is
6602 found within the regex right at the beginning. Bringing initial option settings
6603 outside can help speed up starting point checks. */
6604
6605 ptr += skipatstart;
6606 code = cworkspace;
6607 *code = OP_BRA;
6608 (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6609 &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6610 &length);
6611 if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6612
6613 DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6614 cd->hwm - cworkspace));
6615
6616 if (length > MAX_PATTERN_SIZE)
6617 {
6618 errorcode = ERR20;
6619 goto PCRE_EARLY_ERROR_RETURN;
6620 }
6621
6622 /* Compute the size of data block needed and get it, either from malloc or
6623 externally provided function. Integer overflow should no longer be possible
6624 because nowadays we limit the maximum value of cd->names_found and
6625 cd->name_entry_size. */
6626
6627 size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6628 re = (real_pcre *)(pcre_malloc)(size);
6629
6630 if (re == NULL)
6631 {
6632 errorcode = ERR21;
6633 goto PCRE_EARLY_ERROR_RETURN;
6634 }
6635
6636 /* Put in the magic number, and save the sizes, initial options, internal
6637 flags, and character table pointer. NULL is used for the default character
6638 tables. The nullpad field is at the end; it's there to help in the case when a
6639 regex compiled on a system with 4-byte pointers is run on another with 8-byte
6640 pointers. */
6641
6642 re->magic_number = MAGIC_NUMBER;
6643 re->size = size;
6644 re->options = cd->external_options;
6645 re->flags = cd->external_flags;
6646 re->dummy1 = 0;
6647 re->first_byte = 0;
6648 re->req_byte = 0;
6649 re->name_table_offset = sizeof(real_pcre);
6650 re->name_entry_size = cd->name_entry_size;
6651 re->name_count = cd->names_found;
6652 re->ref_count = 0;
6653 re->tables = (tables == _pcre_default_tables)? NULL : tables;
6654 re->nullpad = NULL;
6655
6656 /* The starting points of the name/number translation table and of the code are
6657 passed around in the compile data block. The start/end pattern and initial
6658 options are already set from the pre-compile phase, as is the name_entry_size
6659 field. Reset the bracket count and the names_found field. Also reset the hwm
6660 field; this time it's u