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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 723 - (hide annotations) (download)
Sat Oct 8 15:55:23 2011 UTC (2 years, 6 months ago) by ph10
File MIME type: text/plain
File size: 244148 byte(s)
Revert handling of atomic groups that do not include captures to the old way of 
handling, thus reducing stack usage.

1 nigel 77 /*************************************************
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 ph10 598 Copyright (c) 1997-2011 University of Cambridge
10 nigel 77
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 ph10 200 #ifdef HAVE_CONFIG_H
46 ph10 236 #include "config.h"
47 ph10 200 #endif
48 ph10 199
49 nigel 93 #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 nigel 77 #include "pcre_internal.h"
54    
55    
56 ph10 475 /* 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 nigel 85
60 ph10 475 #ifdef PCRE_DEBUG
61 nigel 85 #include "pcre_printint.src"
62     #endif
63    
64    
65 ph10 178 /* Macro for setting individual bits in class bitmaps. */
66    
67     #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
68    
69 ph10 202 /* 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 ph10 178
74 ph10 202 #define OFLOW_MAX (INT_MAX - 20)
75    
76    
77 nigel 77 /*************************************************
78     * Code parameters and static tables *
79     *************************************************/
80    
81 nigel 93 /* 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 nigel 77
88 nigel 93 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 nigel 77
93 nigel 93 #define COMPILE_WORK_SIZE (4096)
94 nigel 77
95 ph10 507 /* The overrun tests check for a slightly smaller size so that they detect the
96 ph10 505 overrun before it actually does run off the end of the data block. */
97 nigel 93
98 ph10 505 #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100)
99    
100    
101 nigel 77 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
102     are simple data values; negative values are for special things like \d and so
103     on. Zero means further processing is needed (for things like \x), or the escape
104     is invalid. */
105    
106 ph10 391 #ifndef EBCDIC
107    
108     /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109 ph10 392 in UTF-8 mode. */
110 ph10 391
111 ph10 392 static const short int escapes[] = {
112 ph10 391 0, 0,
113     0, 0,
114 ph10 392 0, 0,
115     0, 0,
116     0, 0,
117 ph10 391 CHAR_COLON, CHAR_SEMICOLON,
118 ph10 392 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
119 ph10 391 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
120 ph10 392 CHAR_COMMERCIAL_AT, -ESC_A,
121     -ESC_B, -ESC_C,
122     -ESC_D, -ESC_E,
123     0, -ESC_G,
124     -ESC_H, 0,
125     0, -ESC_K,
126 ph10 391 0, 0,
127 ph10 514 -ESC_N, 0,
128 ph10 391 -ESC_P, -ESC_Q,
129     -ESC_R, -ESC_S,
130 ph10 392 0, 0,
131     -ESC_V, -ESC_W,
132     -ESC_X, 0,
133     -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
134 ph10 391 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
135 ph10 392 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
136 ph10 391 CHAR_GRAVE_ACCENT, 7,
137 ph10 392 -ESC_b, 0,
138     -ESC_d, ESC_e,
139 ph10 391 ESC_f, 0,
140     -ESC_h, 0,
141 ph10 392 0, -ESC_k,
142 ph10 391 0, 0,
143     ESC_n, 0,
144 ph10 392 -ESC_p, 0,
145     ESC_r, -ESC_s,
146 ph10 391 ESC_tee, 0,
147 ph10 392 -ESC_v, -ESC_w,
148     0, 0,
149 ph10 391 -ESC_z
150 nigel 77 };
151    
152 ph10 392 #else
153 ph10 391
154     /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155    
156 nigel 77 static const short int escapes[] = {
157     /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
158     /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
159     /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
160     /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
161     /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
162     /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
163     /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
164     /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
165 ph10 178 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
166 nigel 93 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
167 nigel 77 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
168 ph10 178 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
169 nigel 77 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
170     /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
171     /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
172     /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
173 ph10 178 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
174 ph10 514 /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
175 nigel 93 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
176 ph10 178 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
177 nigel 77 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
178     /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
179     /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
180     };
181     #endif
182    
183    
184 ph10 243 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
185     searched linearly. Put all the names into a single string, in order to reduce
186 ph10 392 the number of relocations when a shared library is dynamically linked. The
187     string is built from string macros so that it works in UTF-8 mode on EBCDIC
188 ph10 391 platforms. */
189 ph10 210
190     typedef struct verbitem {
191 ph10 510 int len; /* Length of verb name */
192     int op; /* Op when no arg, or -1 if arg mandatory */
193     int op_arg; /* Op when arg present, or -1 if not allowed */
194 ph10 211 } verbitem;
195 ph10 210
196 ph10 240 static const char verbnames[] =
197 ph10 510 "\0" /* Empty name is a shorthand for MARK */
198 ph10 512 STRING_MARK0
199 ph10 391 STRING_ACCEPT0
200     STRING_COMMIT0
201     STRING_F0
202     STRING_FAIL0
203     STRING_PRUNE0
204     STRING_SKIP0
205     STRING_THEN;
206 ph10 240
207 ph10 327 static const verbitem verbs[] = {
208 ph10 510 { 0, -1, OP_MARK },
209 ph10 512 { 4, -1, OP_MARK },
210 ph10 510 { 6, OP_ACCEPT, -1 },
211     { 6, OP_COMMIT, -1 },
212     { 1, OP_FAIL, -1 },
213     { 4, OP_FAIL, -1 },
214     { 5, OP_PRUNE, OP_PRUNE_ARG },
215     { 4, OP_SKIP, OP_SKIP_ARG },
216     { 4, OP_THEN, OP_THEN_ARG }
217 ph10 210 };
218    
219 ph10 327 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
220 ph10 210
221    
222 ph10 243 /* Tables of names of POSIX character classes and their lengths. The names are
223     now all in a single string, to reduce the number of relocations when a shared
224 ph10 240 library is dynamically loaded. The list of lengths is terminated by a zero
225     length entry. The first three must be alpha, lower, upper, as this is assumed
226     for handling case independence. */
227 nigel 77
228 ph10 240 static const char posix_names[] =
229 ph10 392 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
230     STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
231 ph10 391 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
232     STRING_word0 STRING_xdigit;
233 nigel 77
234     static const uschar posix_name_lengths[] = {
235     5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
236    
237 nigel 87 /* Table of class bit maps for each POSIX class. Each class is formed from a
238     base map, with an optional addition or removal of another map. Then, for some
239     classes, there is some additional tweaking: for [:blank:] the vertical space
240     characters are removed, and for [:alpha:] and [:alnum:] the underscore
241     character is removed. The triples in the table consist of the base map offset,
242     second map offset or -1 if no second map, and a non-negative value for map
243     addition or a negative value for map subtraction (if there are two maps). The
244     absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
245     remove vertical space characters, 2 => remove underscore. */
246 nigel 77
247     static const int posix_class_maps[] = {
248 nigel 87 cbit_word, cbit_digit, -2, /* alpha */
249     cbit_lower, -1, 0, /* lower */
250     cbit_upper, -1, 0, /* upper */
251     cbit_word, -1, 2, /* alnum - word without underscore */
252     cbit_print, cbit_cntrl, 0, /* ascii */
253     cbit_space, -1, 1, /* blank - a GNU extension */
254     cbit_cntrl, -1, 0, /* cntrl */
255     cbit_digit, -1, 0, /* digit */
256     cbit_graph, -1, 0, /* graph */
257     cbit_print, -1, 0, /* print */
258     cbit_punct, -1, 0, /* punct */
259     cbit_space, -1, 0, /* space */
260     cbit_word, -1, 0, /* word - a Perl extension */
261     cbit_xdigit,-1, 0 /* xdigit */
262 nigel 77 };
263    
264 ph10 535 /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class
265     substitutes must be in the order of the names, defined above, and there are
266 ph10 518 both positive and negative cases. NULL means no substitute. */
267 nigel 77
268 ph10 518 #ifdef SUPPORT_UCP
269     static const uschar *substitutes[] = {
270     (uschar *)"\\P{Nd}", /* \D */
271     (uschar *)"\\p{Nd}", /* \d */
272     (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */
273     (uschar *)"\\p{Xsp}", /* \s */
274     (uschar *)"\\P{Xwd}", /* \W */
275 ph10 535 (uschar *)"\\p{Xwd}" /* \w */
276 ph10 518 };
277 ph10 535
278 ph10 518 static const uschar *posix_substitutes[] = {
279     (uschar *)"\\p{L}", /* alpha */
280 ph10 535 (uschar *)"\\p{Ll}", /* lower */
281     (uschar *)"\\p{Lu}", /* upper */
282     (uschar *)"\\p{Xan}", /* alnum */
283 ph10 518 NULL, /* ascii */
284     (uschar *)"\\h", /* blank */
285     NULL, /* cntrl */
286     (uschar *)"\\p{Nd}", /* digit */
287     NULL, /* graph */
288     NULL, /* print */
289     NULL, /* punct */
290     (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */
291     (uschar *)"\\p{Xwd}", /* word */
292 ph10 535 NULL, /* xdigit */
293 ph10 518 /* Negated cases */
294     (uschar *)"\\P{L}", /* ^alpha */
295 ph10 535 (uschar *)"\\P{Ll}", /* ^lower */
296     (uschar *)"\\P{Lu}", /* ^upper */
297     (uschar *)"\\P{Xan}", /* ^alnum */
298 ph10 518 NULL, /* ^ascii */
299     (uschar *)"\\H", /* ^blank */
300     NULL, /* ^cntrl */
301     (uschar *)"\\P{Nd}", /* ^digit */
302     NULL, /* ^graph */
303     NULL, /* ^print */
304     NULL, /* ^punct */
305     (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */
306     (uschar *)"\\P{Xwd}", /* ^word */
307 ph10 535 NULL /* ^xdigit */
308 ph10 518 };
309     #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *))
310 ph10 535 #endif
311 ph10 518
312 nigel 93 #define STRING(a) # a
313     #define XSTRING(s) STRING(s)
314    
315 nigel 77 /* The texts of compile-time error messages. These are "char *" because they
316 nigel 93 are passed to the outside world. Do not ever re-use any error number, because
317     they are documented. Always add a new error instead. Messages marked DEAD below
318 ph10 243 are no longer used. This used to be a table of strings, but in order to reduce
319     the number of relocations needed when a shared library is loaded dynamically,
320     it is now one long string. We cannot use a table of offsets, because the
321     lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
322     simply count through to the one we want - this isn't a performance issue
323 ph10 507 because these strings are used only when there is a compilation error.
324 nigel 77
325 ph10 507 Each substring ends with \0 to insert a null character. This includes the final
326     substring, so that the whole string ends with \0\0, which can be detected when
327 ph10 499 counting through. */
328    
329 ph10 240 static const char error_texts[] =
330     "no error\0"
331     "\\ at end of pattern\0"
332     "\\c at end of pattern\0"
333     "unrecognized character follows \\\0"
334     "numbers out of order in {} quantifier\0"
335 nigel 77 /* 5 */
336 ph10 240 "number too big in {} quantifier\0"
337     "missing terminating ] for character class\0"
338     "invalid escape sequence in character class\0"
339     "range out of order in character class\0"
340     "nothing to repeat\0"
341 nigel 77 /* 10 */
342 ph10 240 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
343     "internal error: unexpected repeat\0"
344 ph10 269 "unrecognized character after (? or (?-\0"
345 ph10 240 "POSIX named classes are supported only within a class\0"
346     "missing )\0"
347 nigel 77 /* 15 */
348 ph10 240 "reference to non-existent subpattern\0"
349     "erroffset passed as NULL\0"
350     "unknown option bit(s) set\0"
351     "missing ) after comment\0"
352     "parentheses nested too deeply\0" /** DEAD **/
353 nigel 77 /* 20 */
354 ph10 240 "regular expression is too large\0"
355     "failed to get memory\0"
356     "unmatched parentheses\0"
357     "internal error: code overflow\0"
358     "unrecognized character after (?<\0"
359 nigel 77 /* 25 */
360 ph10 240 "lookbehind assertion is not fixed length\0"
361     "malformed number or name after (?(\0"
362     "conditional group contains more than two branches\0"
363     "assertion expected after (?(\0"
364     "(?R or (?[+-]digits must be followed by )\0"
365 nigel 77 /* 30 */
366 ph10 240 "unknown POSIX class name\0"
367     "POSIX collating elements are not supported\0"
368     "this version of PCRE is not compiled with PCRE_UTF8 support\0"
369     "spare error\0" /** DEAD **/
370     "character value in \\x{...} sequence is too large\0"
371 nigel 77 /* 35 */
372 ph10 240 "invalid condition (?(0)\0"
373     "\\C not allowed in lookbehind assertion\0"
374 ph10 514 "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
375 ph10 240 "number after (?C is > 255\0"
376     "closing ) for (?C expected\0"
377 nigel 77 /* 40 */
378 ph10 240 "recursive call could loop indefinitely\0"
379     "unrecognized character after (?P\0"
380     "syntax error in subpattern name (missing terminator)\0"
381     "two named subpatterns have the same name\0"
382     "invalid UTF-8 string\0"
383 nigel 77 /* 45 */
384 ph10 240 "support for \\P, \\p, and \\X has not been compiled\0"
385     "malformed \\P or \\p sequence\0"
386     "unknown property name after \\P or \\p\0"
387     "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
388     "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
389 nigel 91 /* 50 */
390 ph10 240 "repeated subpattern is too long\0" /** DEAD **/
391     "octal value is greater than \\377 (not in UTF-8 mode)\0"
392     "internal error: overran compiling workspace\0"
393     "internal error: previously-checked referenced subpattern not found\0"
394     "DEFINE group contains more than one branch\0"
395 nigel 93 /* 55 */
396 ph10 637 "repeating a DEFINE group is not allowed\0" /** DEAD **/
397 ph10 240 "inconsistent NEWLINE options\0"
398 ph10 333 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
399     "a numbered reference must not be zero\0"
400 ph10 510 "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
401 ph10 211 /* 60 */
402 ph10 240 "(*VERB) not recognized\0"
403 ph10 268 "number is too big\0"
404 ph10 272 "subpattern name expected\0"
405 ph10 336 "digit expected after (?+\0"
406 ph10 457 "] is an invalid data character in JavaScript compatibility mode\0"
407     /* 65 */
408 ph10 510 "different names for subpatterns of the same number are not allowed\0"
409 ph10 512 "(*MARK) must have an argument\0"
410 ph10 535 "this version of PCRE is not compiled with PCRE_UCP support\0"
411 ph10 579 "\\c must be followed by an ASCII character\0"
412 ph10 654 "\\k is not followed by a braced, angle-bracketed, or quoted name\0"
413 ph10 510 ;
414 nigel 77
415     /* Table to identify digits and hex digits. This is used when compiling
416     patterns. Note that the tables in chartables are dependent on the locale, and
417     may mark arbitrary characters as digits - but the PCRE compiling code expects
418     to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
419     a private table here. It costs 256 bytes, but it is a lot faster than doing
420     character value tests (at least in some simple cases I timed), and in some
421     applications one wants PCRE to compile efficiently as well as match
422     efficiently.
423    
424     For convenience, we use the same bit definitions as in chartables:
425    
426     0x04 decimal digit
427     0x08 hexadecimal digit
428    
429     Then we can use ctype_digit and ctype_xdigit in the code. */
430    
431 ph10 392 #ifndef EBCDIC
432 ph10 391
433 ph10 392 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
434 ph10 391 UTF-8 mode. */
435    
436 nigel 77 static const unsigned char digitab[] =
437     {
438     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
439     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
440     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
441     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
442     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
443     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
444     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
445     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
446     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
447     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
448     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
449     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
450     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
451     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
452     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
453     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
454     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
455     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
456     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
457     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
458     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
459     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
460     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
461     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
462     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
463     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
464     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
465     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
466     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
467     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
468     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
469     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
470    
471 ph10 392 #else
472 ph10 391
473     /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
474    
475 nigel 77 static const unsigned char digitab[] =
476     {
477     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
478     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
479     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
480     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
481     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
482     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
483     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
484     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
485     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
486     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
487     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
488 ph10 97 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
489 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
490     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
491     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
492     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
493     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
494     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
495     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
496     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
497     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
498     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
499     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
500     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
501     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
502     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
503     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
504     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
505     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
506     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
507     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
508     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
509    
510     static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
511     0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
512     0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
513     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
514     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
515     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
516     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
517     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
518     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
519     0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
520     0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
521     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
522 ph10 97 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
523 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
524     0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
525     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
526     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
527     0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
528     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
529     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
530     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
531     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
532     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
533     0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
534     0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
535     0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
536     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
537     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
538     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
539     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
540     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
541     0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
542     0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
543     #endif
544    
545    
546     /* Definition to allow mutual recursion */
547    
548     static BOOL
549 ph10 642 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int,
550     int *, int *, branch_chain *, compile_data *, int *);
551 nigel 77
552    
553    
554     /*************************************************
555 ph10 240 * Find an error text *
556     *************************************************/
557    
558 ph10 243 /* The error texts are now all in one long string, to save on relocations. As
559     some of the text is of unknown length, we can't use a table of offsets.
560     Instead, just count through the strings. This is not a performance issue
561 ph10 240 because it happens only when there has been a compilation error.
562    
563     Argument: the error number
564     Returns: pointer to the error string
565     */
566    
567     static const char *
568     find_error_text(int n)
569     {
570     const char *s = error_texts;
571 ph10 507 for (; n > 0; n--)
572 ph10 499 {
573     while (*s++ != 0) {};
574     if (*s == 0) return "Error text not found (please report)";
575 ph10 507 }
576 ph10 240 return s;
577     }
578    
579    
580     /*************************************************
581 ph10 640 * Check for counted repeat *
582     *************************************************/
583    
584     /* This function is called when a '{' is encountered in a place where it might
585     start a quantifier. It looks ahead to see if it really is a quantifier or not.
586     It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
587     where the ddds are digits.
588    
589     Arguments:
590     p pointer to the first char after '{'
591    
592     Returns: TRUE or FALSE
593     */
594    
595     static BOOL
596     is_counted_repeat(const uschar *p)
597     {
598     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
599     while ((digitab[*p] & ctype_digit) != 0) p++;
600     if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
601    
602     if (*p++ != CHAR_COMMA) return FALSE;
603     if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
604    
605     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
606     while ((digitab[*p] & ctype_digit) != 0) p++;
607    
608     return (*p == CHAR_RIGHT_CURLY_BRACKET);
609     }
610    
611    
612    
613     /*************************************************
614 nigel 77 * Handle escapes *
615     *************************************************/
616    
617     /* This function is called when a \ has been encountered. It either returns a
618     positive value for a simple escape such as \n, or a negative value which
619 nigel 93 encodes one of the more complicated things such as \d. A backreference to group
620     n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
621     UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
622     ptr is pointing at the \. On exit, it is on the final character of the escape
623     sequence.
624 nigel 77
625     Arguments:
626     ptrptr points to the pattern position pointer
627     errorcodeptr points to the errorcode variable
628     bracount number of previous extracting brackets
629     options the options bits
630     isclass TRUE if inside a character class
631    
632     Returns: zero or positive => a data character
633     negative => a special escape sequence
634 ph10 213 on error, errorcodeptr is set
635 nigel 77 */
636    
637     static int
638     check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
639     int options, BOOL isclass)
640     {
641 nigel 87 BOOL utf8 = (options & PCRE_UTF8) != 0;
642     const uschar *ptr = *ptrptr + 1;
643 nigel 77 int c, i;
644    
645 nigel 87 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
646     ptr--; /* Set pointer back to the last byte */
647    
648 nigel 77 /* If backslash is at the end of the pattern, it's an error. */
649    
650     if (c == 0) *errorcodeptr = ERR1;
651    
652 ph10 274 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
653     in a table. A non-zero result is something that can be returned immediately.
654 nigel 77 Otherwise further processing may be required. */
655    
656 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
657     else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
658     else if ((i = escapes[c - CHAR_0]) != 0) c = i;
659 nigel 77
660 ph10 97 #else /* EBCDIC coding */
661 ph10 274 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
662 nigel 77 else if ((i = escapes[c - 0x48]) != 0) c = i;
663     #endif
664    
665     /* Escapes that need further processing, or are illegal. */
666    
667     else
668     {
669     const uschar *oldptr;
670 nigel 93 BOOL braced, negated;
671    
672 nigel 77 switch (c)
673     {
674     /* A number of Perl escapes are not handled by PCRE. We give an explicit
675     error. */
676    
677 ph10 391 case CHAR_l:
678     case CHAR_L:
679     case CHAR_u:
680     case CHAR_U:
681 nigel 77 *errorcodeptr = ERR37;
682     break;
683    
684 ph10 654 /* In a character class, \g is just a literal "g". Outside a character
685 ph10 640 class, \g must be followed by one of a number of specific things:
686 ph10 345
687 ph10 333 (1) A number, either plain or braced. If positive, it is an absolute
688     backreference. If negative, it is a relative backreference. This is a Perl
689     5.10 feature.
690 ph10 345
691 ph10 333 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
692     is part of Perl's movement towards a unified syntax for back references. As
693     this is synonymous with \k{name}, we fudge it up by pretending it really
694     was \k.
695 ph10 345
696     (3) For Oniguruma compatibility we also support \g followed by a name or a
697     number either in angle brackets or in single quotes. However, these are
698     (possibly recursive) subroutine calls, _not_ backreferences. Just return
699 ph10 333 the -ESC_g code (cf \k). */
700 nigel 93
701 ph10 391 case CHAR_g:
702 ph10 640 if (isclass) break;
703 ph10 391 if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
704 ph10 333 {
705     c = -ESC_g;
706 ph10 345 break;
707     }
708 ph10 333
709     /* Handle the Perl-compatible cases */
710 ph10 345
711 ph10 391 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
712 nigel 93 {
713 ph10 171 const uschar *p;
714 ph10 391 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
715     if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
716     if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
717 ph10 171 {
718     c = -ESC_k;
719     break;
720 ph10 172 }
721 nigel 93 braced = TRUE;
722     ptr++;
723     }
724     else braced = FALSE;
725    
726 ph10 391 if (ptr[1] == CHAR_MINUS)
727 nigel 93 {
728     negated = TRUE;
729     ptr++;
730     }
731     else negated = FALSE;
732    
733     c = 0;
734     while ((digitab[ptr[1]] & ctype_digit) != 0)
735 ph10 391 c = c * 10 + *(++ptr) - CHAR_0;
736 ph10 220
737 ph10 333 if (c < 0) /* Integer overflow */
738 ph10 213 {
739     *errorcodeptr = ERR61;
740     break;
741 ph10 220 }
742 ph10 345
743 ph10 391 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
744 nigel 93 {
745     *errorcodeptr = ERR57;
746 ph10 213 break;
747 nigel 93 }
748 ph10 345
749 ph10 333 if (c == 0)
750     {
751     *errorcodeptr = ERR58;
752     break;
753 ph10 345 }
754 nigel 93
755     if (negated)
756     {
757     if (c > bracount)
758     {
759     *errorcodeptr = ERR15;
760 ph10 213 break;
761 nigel 93 }
762     c = bracount - (c - 1);
763     }
764    
765     c = -(ESC_REF + c);
766     break;
767    
768 nigel 77 /* The handling of escape sequences consisting of a string of digits
769     starting with one that is not zero is not straightforward. By experiment,
770     the way Perl works seems to be as follows:
771    
772     Outside a character class, the digits are read as a decimal number. If the
773     number is less than 10, or if there are that many previous extracting
774     left brackets, then it is a back reference. Otherwise, up to three octal
775     digits are read to form an escaped byte. Thus \123 is likely to be octal
776     123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
777     value is greater than 377, the least significant 8 bits are taken. Inside a
778     character class, \ followed by a digit is always an octal number. */
779    
780 ph10 391 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
781     case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
782 nigel 77
783     if (!isclass)
784     {
785     oldptr = ptr;
786 ph10 391 c -= CHAR_0;
787 nigel 77 while ((digitab[ptr[1]] & ctype_digit) != 0)
788 ph10 391 c = c * 10 + *(++ptr) - CHAR_0;
789 ph10 333 if (c < 0) /* Integer overflow */
790 ph10 213 {
791     *errorcodeptr = ERR61;
792 ph10 220 break;
793     }
794 nigel 77 if (c < 10 || c <= bracount)
795     {
796     c = -(ESC_REF + c);
797     break;
798     }
799     ptr = oldptr; /* Put the pointer back and fall through */
800     }
801    
802     /* Handle an octal number following \. If the first digit is 8 or 9, Perl
803     generates a binary zero byte and treats the digit as a following literal.
804     Thus we have to pull back the pointer by one. */
805    
806 ph10 391 if ((c = *ptr) >= CHAR_8)
807 nigel 77 {
808     ptr--;
809     c = 0;
810     break;
811     }
812    
813     /* \0 always starts an octal number, but we may drop through to here with a
814 nigel 91 larger first octal digit. The original code used just to take the least
815     significant 8 bits of octal numbers (I think this is what early Perls used
816     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
817     than 3 octal digits. */
818 nigel 77
819 ph10 391 case CHAR_0:
820     c -= CHAR_0;
821     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
822     c = c * 8 + *(++ptr) - CHAR_0;
823 nigel 91 if (!utf8 && c > 255) *errorcodeptr = ERR51;
824 nigel 77 break;
825    
826 nigel 87 /* \x is complicated. \x{ddd} is a character number which can be greater
827     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
828     treated as a data character. */
829 nigel 77
830 ph10 391 case CHAR_x:
831     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
832 nigel 77 {
833     const uschar *pt = ptr + 2;
834 nigel 87 int count = 0;
835    
836 nigel 77 c = 0;
837     while ((digitab[*pt] & ctype_xdigit) != 0)
838     {
839 nigel 87 register int cc = *pt++;
840 ph10 391 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
841 nigel 77 count++;
842 nigel 87
843 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
844     if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
845     c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
846 ph10 97 #else /* EBCDIC coding */
847 ph10 391 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
848     c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
849 nigel 77 #endif
850     }
851 nigel 87
852 ph10 391 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
853 nigel 77 {
854 nigel 87 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
855 nigel 77 ptr = pt;
856     break;
857     }
858 nigel 87
859 nigel 77 /* If the sequence of hex digits does not end with '}', then we don't
860     recognize this construct; fall through to the normal \x handling. */
861     }
862    
863 nigel 87 /* Read just a single-byte hex-defined char */
864 nigel 77
865     c = 0;
866     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
867     {
868 ph10 391 int cc; /* Some compilers don't like */
869     cc = *(++ptr); /* ++ in initializers */
870     #ifndef EBCDIC /* ASCII/UTF-8 coding */
871     if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
872     c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
873 ph10 97 #else /* EBCDIC coding */
874 ph10 391 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
875     c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
876 nigel 77 #endif
877     }
878     break;
879    
880 nigel 93 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
881 ph10 574 An error is given if the byte following \c is not an ASCII character. This
882     coding is ASCII-specific, but then the whole concept of \cx is
883 nigel 93 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
884 nigel 77
885 ph10 391 case CHAR_c:
886 nigel 77 c = *(++ptr);
887     if (c == 0)
888     {
889     *errorcodeptr = ERR2;
890 ph10 213 break;
891 nigel 77 }
892 ph10 574 #ifndef EBCDIC /* ASCII/UTF-8 coding */
893     if (c > 127) /* Excludes all non-ASCII in either mode */
894     {
895     *errorcodeptr = ERR68;
896 ph10 579 break;
897     }
898 ph10 391 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
899 nigel 77 c ^= 0x40;
900 ph10 574 #else /* EBCDIC coding */
901 ph10 391 if (c >= CHAR_a && c <= CHAR_z) c += 64;
902 nigel 77 c ^= 0xC0;
903     #endif
904     break;
905    
906     /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
907 ph10 274 other alphanumeric following \ is an error if PCRE_EXTRA was set;
908     otherwise, for Perl compatibility, it is a literal. This code looks a bit
909     odd, but there used to be some cases other than the default, and there may
910     be again in future, so I haven't "optimized" it. */
911 nigel 77
912     default:
913     if ((options & PCRE_EXTRA) != 0) switch(c)
914     {
915     default:
916     *errorcodeptr = ERR3;
917     break;
918     }
919     break;
920     }
921     }
922 ph10 518
923     /* Perl supports \N{name} for character names, as well as plain \N for "not
924 ph10 654 newline". PCRE does not support \N{name}. However, it does support
925 ph10 640 quantification such as \N{2,3}. */
926 nigel 77
927 ph10 640 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET &&
928     !is_counted_repeat(ptr+2))
929 ph10 518 *errorcodeptr = ERR37;
930 ph10 514
931 ph10 518 /* If PCRE_UCP is set, we change the values for \d etc. */
932    
933     if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
934     c -= (ESC_DU - ESC_D);
935    
936     /* Set the pointer to the final character before returning. */
937    
938 nigel 77 *ptrptr = ptr;
939     return c;
940     }
941    
942    
943    
944     #ifdef SUPPORT_UCP
945     /*************************************************
946     * Handle \P and \p *
947     *************************************************/
948    
949     /* This function is called after \P or \p has been encountered, provided that
950     PCRE is compiled with support for Unicode properties. On entry, ptrptr is
951     pointing at the P or p. On exit, it is pointing at the final character of the
952     escape sequence.
953    
954     Argument:
955     ptrptr points to the pattern position pointer
956     negptr points to a boolean that is set TRUE for negation else FALSE
957 nigel 87 dptr points to an int that is set to the detailed property value
958 nigel 77 errorcodeptr points to the error code variable
959    
960 nigel 87 Returns: type value from ucp_type_table, or -1 for an invalid type
961 nigel 77 */
962    
963     static int
964 nigel 87 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
965 nigel 77 {
966     int c, i, bot, top;
967     const uschar *ptr = *ptrptr;
968 nigel 87 char name[32];
969 nigel 77
970     c = *(++ptr);
971     if (c == 0) goto ERROR_RETURN;
972    
973     *negptr = FALSE;
974    
975 nigel 87 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
976     negation. */
977 nigel 77
978 ph10 391 if (c == CHAR_LEFT_CURLY_BRACKET)
979 nigel 77 {
980 ph10 391 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
981 nigel 77 {
982     *negptr = TRUE;
983     ptr++;
984     }
985 ph10 199 for (i = 0; i < (int)sizeof(name) - 1; i++)
986 nigel 77 {
987     c = *(++ptr);
988     if (c == 0) goto ERROR_RETURN;
989 ph10 391 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
990 nigel 77 name[i] = c;
991     }
992 ph10 391 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
993 nigel 77 name[i] = 0;
994     }
995    
996     /* Otherwise there is just one following character */
997    
998     else
999     {
1000     name[0] = c;
1001     name[1] = 0;
1002     }
1003    
1004     *ptrptr = ptr;
1005    
1006     /* Search for a recognized property name using binary chop */
1007    
1008     bot = 0;
1009     top = _pcre_utt_size;
1010    
1011     while (bot < top)
1012     {
1013 nigel 87 i = (bot + top) >> 1;
1014 ph10 240 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
1015 nigel 87 if (c == 0)
1016     {
1017     *dptr = _pcre_utt[i].value;
1018     return _pcre_utt[i].type;
1019     }
1020 nigel 77 if (c > 0) bot = i + 1; else top = i;
1021     }
1022    
1023     *errorcodeptr = ERR47;
1024     *ptrptr = ptr;
1025     return -1;
1026    
1027     ERROR_RETURN:
1028     *errorcodeptr = ERR46;
1029     *ptrptr = ptr;
1030     return -1;
1031     }
1032     #endif
1033    
1034    
1035    
1036    
1037     /*************************************************
1038     * Read repeat counts *
1039     *************************************************/
1040    
1041     /* Read an item of the form {n,m} and return the values. This is called only
1042     after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1043     so the syntax is guaranteed to be correct, but we need to check the values.
1044    
1045     Arguments:
1046     p pointer to first char after '{'
1047     minp pointer to int for min
1048     maxp pointer to int for max
1049     returned as -1 if no max
1050     errorcodeptr points to error code variable
1051    
1052     Returns: pointer to '}' on success;
1053     current ptr on error, with errorcodeptr set non-zero
1054     */
1055    
1056     static const uschar *
1057     read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1058     {
1059     int min = 0;
1060     int max = -1;
1061    
1062 nigel 81 /* Read the minimum value and do a paranoid check: a negative value indicates
1063     an integer overflow. */
1064    
1065 ph10 391 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1066 nigel 81 if (min < 0 || min > 65535)
1067     {
1068     *errorcodeptr = ERR5;
1069     return p;
1070     }
1071 nigel 77
1072 nigel 81 /* Read the maximum value if there is one, and again do a paranoid on its size.
1073     Also, max must not be less than min. */
1074    
1075 ph10 391 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1076 nigel 77 {
1077 ph10 391 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1078 nigel 77 {
1079     max = 0;
1080 ph10 391 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1081 nigel 81 if (max < 0 || max > 65535)
1082     {
1083     *errorcodeptr = ERR5;
1084     return p;
1085     }
1086 nigel 77 if (max < min)
1087     {
1088     *errorcodeptr = ERR4;
1089     return p;
1090     }
1091     }
1092     }
1093    
1094 nigel 81 /* Fill in the required variables, and pass back the pointer to the terminating
1095     '}'. */
1096 nigel 77
1097 nigel 81 *minp = min;
1098     *maxp = max;
1099 nigel 77 return p;
1100     }
1101    
1102    
1103    
1104     /*************************************************
1105 ph10 408 * Subroutine for finding forward reference *
1106 nigel 91 *************************************************/
1107    
1108 ph10 408 /* This recursive function is called only from find_parens() below. The
1109     top-level call starts at the beginning of the pattern. All other calls must
1110     start at a parenthesis. It scans along a pattern's text looking for capturing
1111 nigel 93 subpatterns, and counting them. If it finds a named pattern that matches the
1112     name it is given, it returns its number. Alternatively, if the name is NULL, it
1113 ph10 578 returns when it reaches a given numbered subpattern. Recursion is used to keep
1114     track of subpatterns that reset the capturing group numbers - the (?| feature.
1115 nigel 91
1116 ph10 578 This function was originally called only from the second pass, in which we know
1117     that if (?< or (?' or (?P< is encountered, the name will be correctly
1118     terminated because that is checked in the first pass. There is now one call to
1119     this function in the first pass, to check for a recursive back reference by
1120     name (so that we can make the whole group atomic). In this case, we need check
1121 ph10 579 only up to the current position in the pattern, and that is still OK because
1122     and previous occurrences will have been checked. To make this work, the test
1123     for "end of pattern" is a check against cd->end_pattern in the main loop,
1124 ph10 578 instead of looking for a binary zero. This means that the special first-pass
1125 ph10 579 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1126     processing items within the loop are OK, because afterwards the main loop will
1127 ph10 578 terminate.)
1128    
1129 nigel 91 Arguments:
1130 ph10 408 ptrptr address of the current character pointer (updated)
1131 ph10 345 cd compile background data
1132 nigel 93 name name to seek, or NULL if seeking a numbered subpattern
1133     lorn name length, or subpattern number if name is NULL
1134     xmode TRUE if we are in /x mode
1135 ph10 579 utf8 TRUE if we are in UTF-8 mode
1136 ph10 411 count pointer to the current capturing subpattern number (updated)
1137 nigel 91
1138     Returns: the number of the named subpattern, or -1 if not found
1139     */
1140    
1141     static int
1142 ph10 408 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1143 ph10 556 BOOL xmode, BOOL utf8, int *count)
1144 nigel 91 {
1145 ph10 408 uschar *ptr = *ptrptr;
1146     int start_count = *count;
1147     int hwm_count = start_count;
1148     BOOL dup_parens = FALSE;
1149 nigel 93
1150 ph10 411 /* If the first character is a parenthesis, check on the type of group we are
1151 ph10 408 dealing with. The very first call may not start with a parenthesis. */
1152    
1153     if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1154     {
1155 ph10 544 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1156 ph10 545
1157 ph10 544 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1158 ph10 545
1159 ph10 544 /* Handle a normal, unnamed capturing parenthesis. */
1160 ph10 408
1161 ph10 544 else if (ptr[1] != CHAR_QUESTION_MARK)
1162 ph10 408 {
1163     *count += 1;
1164     if (name == NULL && *count == lorn) return *count;
1165 ph10 411 ptr++;
1166 ph10 408 }
1167    
1168 ph10 544 /* All cases now have (? at the start. Remember when we are in a group
1169     where the parenthesis numbers are duplicated. */
1170    
1171     else if (ptr[2] == CHAR_VERTICAL_LINE)
1172     {
1173     ptr += 3;
1174     dup_parens = TRUE;
1175     }
1176 ph10 545
1177 ph10 544 /* Handle comments; all characters are allowed until a ket is reached. */
1178    
1179     else if (ptr[2] == CHAR_NUMBER_SIGN)
1180     {
1181     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1182     goto FAIL_EXIT;
1183 ph10 545 }
1184 ph10 544
1185 ph10 408 /* Handle a condition. If it is an assertion, just carry on so that it
1186     is processed as normal. If not, skip to the closing parenthesis of the
1187 ph10 544 condition (there can't be any nested parens). */
1188 ph10 411
1189 ph10 408 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1190     {
1191 ph10 411 ptr += 2;
1192 ph10 408 if (ptr[1] != CHAR_QUESTION_MARK)
1193     {
1194     while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1195 ph10 411 if (*ptr != 0) ptr++;
1196 ph10 408 }
1197 ph10 411 }
1198    
1199 ph10 544 /* Start with (? but not a condition. */
1200 ph10 408
1201     else
1202 ph10 411 {
1203 ph10 408 ptr += 2;
1204     if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1205    
1206     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1207 ph10 411
1208 ph10 408 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1209     ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1210     {
1211     int term;
1212     const uschar *thisname;
1213     *count += 1;
1214     if (name == NULL && *count == lorn) return *count;
1215     term = *ptr++;
1216     if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1217     thisname = ptr;
1218     while (*ptr != term) ptr++;
1219     if (name != NULL && lorn == ptr - thisname &&
1220     strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1221     return *count;
1222 ph10 461 term++;
1223 ph10 411 }
1224 ph10 408 }
1225 ph10 411 }
1226 ph10 408
1227 ph10 411 /* Past any initial parenthesis handling, scan for parentheses or vertical
1228 ph10 579 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1229     first-pass call when this value is temporarily adjusted to stop at the current
1230 ph10 578 position. So DO NOT change this to a test for binary zero. */
1231 ph10 408
1232 ph10 578 for (; ptr < cd->end_pattern; ptr++)
1233 nigel 91 {
1234 nigel 93 /* Skip over backslashed characters and also entire \Q...\E */
1235    
1236 ph10 391 if (*ptr == CHAR_BACKSLASH)
1237 nigel 93 {
1238 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1239 ph10 391 if (*ptr == CHAR_Q) for (;;)
1240 nigel 93 {
1241 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1242 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1243 ph10 391 if (*(++ptr) == CHAR_E) break;
1244 nigel 93 }
1245     continue;
1246     }
1247    
1248 ph10 340 /* Skip over character classes; this logic must be similar to the way they
1249     are handled for real. If the first character is '^', skip it. Also, if the
1250     first few characters (either before or after ^) are \Q\E or \E we skip them
1251 ph10 392 too. This makes for compatibility with Perl. Note the use of STR macros to
1252 ph10 391 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1253 nigel 93
1254 ph10 391 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1255 nigel 93 {
1256 ph10 340 BOOL negate_class = FALSE;
1257     for (;;)
1258     {
1259 ph10 438 if (ptr[1] == CHAR_BACKSLASH)
1260 ph10 340 {
1261 ph10 438 if (ptr[2] == CHAR_E)
1262     ptr+= 2;
1263     else if (strncmp((const char *)ptr+2,
1264 ph10 392 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1265 ph10 438 ptr += 4;
1266 ph10 392 else
1267 ph10 391 break;
1268 ph10 340 }
1269 ph10 438 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1270 ph10 461 {
1271 ph10 340 negate_class = TRUE;
1272 ph10 438 ptr++;
1273 ph10 461 }
1274 ph10 340 else break;
1275     }
1276    
1277     /* If the next character is ']', it is a data character that must be
1278 ph10 341 skipped, except in JavaScript compatibility mode. */
1279 ph10 345
1280 ph10 392 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1281 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1282 ph10 345 ptr++;
1283    
1284 ph10 391 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1285 nigel 93 {
1286 ph10 220 if (*ptr == 0) return -1;
1287 ph10 391 if (*ptr == CHAR_BACKSLASH)
1288 nigel 93 {
1289 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1290 ph10 391 if (*ptr == CHAR_Q) for (;;)
1291 nigel 93 {
1292 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1293 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1294 ph10 391 if (*(++ptr) == CHAR_E) break;
1295 nigel 93 }
1296     continue;
1297     }
1298     }
1299     continue;
1300     }
1301    
1302     /* Skip comments in /x mode */
1303    
1304 ph10 391 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1305 nigel 93 {
1306 ph10 579 ptr++;
1307 ph10 556 while (*ptr != 0)
1308     {
1309     if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1310     ptr++;
1311 ph10 579 #ifdef SUPPORT_UTF8
1312 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1313     #endif
1314     }
1315 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1316 nigel 93 continue;
1317     }
1318    
1319 ph10 408 /* Check for the special metacharacters */
1320 ph10 411
1321 ph10 408 if (*ptr == CHAR_LEFT_PARENTHESIS)
1322 nigel 93 {
1323 ph10 556 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1324 ph10 408 if (rc > 0) return rc;
1325     if (*ptr == 0) goto FAIL_EXIT;
1326 nigel 93 }
1327 ph10 411
1328 ph10 408 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1329     {
1330     if (dup_parens && *count < hwm_count) *count = hwm_count;
1331 ph10 545 goto FAIL_EXIT;
1332 ph10 408 }
1333 ph10 411
1334     else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1335 ph10 408 {
1336     if (*count > hwm_count) hwm_count = *count;
1337     *count = start_count;
1338 ph10 411 }
1339 ph10 408 }
1340 nigel 93
1341 ph10 408 FAIL_EXIT:
1342     *ptrptr = ptr;
1343     return -1;
1344     }
1345 nigel 93
1346    
1347    
1348    
1349 ph10 408 /*************************************************
1350     * Find forward referenced subpattern *
1351     *************************************************/
1352 nigel 93
1353 ph10 408 /* This function scans along a pattern's text looking for capturing
1354     subpatterns, and counting them. If it finds a named pattern that matches the
1355     name it is given, it returns its number. Alternatively, if the name is NULL, it
1356     returns when it reaches a given numbered subpattern. This is used for forward
1357     references to subpatterns. We used to be able to start this scan from the
1358     current compiling point, using the current count value from cd->bracount, and
1359     do it all in a single loop, but the addition of the possibility of duplicate
1360     subpattern numbers means that we have to scan from the very start, in order to
1361     take account of such duplicates, and to use a recursive function to keep track
1362     of the different types of group.
1363    
1364     Arguments:
1365     cd compile background data
1366     name name to seek, or NULL if seeking a numbered subpattern
1367     lorn name length, or subpattern number if name is NULL
1368     xmode TRUE if we are in /x mode
1369 ph10 579 utf8 TRUE if we are in UTF-8 mode
1370 ph10 408
1371     Returns: the number of the found subpattern, or -1 if not found
1372     */
1373    
1374     static int
1375 ph10 556 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1376     BOOL utf8)
1377 ph10 408 {
1378     uschar *ptr = (uschar *)cd->start_pattern;
1379     int count = 0;
1380     int rc;
1381    
1382     /* If the pattern does not start with an opening parenthesis, the first call
1383     to find_parens_sub() will scan right to the end (if necessary). However, if it
1384     does start with a parenthesis, find_parens_sub() will return when it hits the
1385     matching closing parens. That is why we have to have a loop. */
1386    
1387 ph10 411 for (;;)
1388     {
1389 ph10 556 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1390 ph10 411 if (rc > 0 || *ptr++ == 0) break;
1391     }
1392    
1393 ph10 408 return rc;
1394 nigel 91 }
1395    
1396    
1397    
1398 ph10 408
1399 nigel 91 /*************************************************
1400 nigel 77 * Find first significant op code *
1401     *************************************************/
1402    
1403     /* This is called by several functions that scan a compiled expression looking
1404     for a fixed first character, or an anchoring op code etc. It skips over things
1405 ph10 602 that do not influence this. For some calls, it makes sense to skip negative
1406     forward and all backward assertions, and also the \b assertion; for others it
1407     does not.
1408 nigel 77
1409     Arguments:
1410     code pointer to the start of the group
1411     skipassert TRUE if certain assertions are to be skipped
1412    
1413     Returns: pointer to the first significant opcode
1414     */
1415    
1416     static const uschar*
1417 ph10 604 first_significant_code(const uschar *code, BOOL skipassert)
1418 nigel 77 {
1419     for (;;)
1420     {
1421     switch ((int)*code)
1422     {
1423     case OP_ASSERT_NOT:
1424     case OP_ASSERTBACK:
1425     case OP_ASSERTBACK_NOT:
1426     if (!skipassert) return code;
1427     do code += GET(code, 1); while (*code == OP_ALT);
1428     code += _pcre_OP_lengths[*code];
1429     break;
1430    
1431     case OP_WORD_BOUNDARY:
1432     case OP_NOT_WORD_BOUNDARY:
1433     if (!skipassert) return code;
1434     /* Fall through */
1435    
1436     case OP_CALLOUT:
1437     case OP_CREF:
1438 ph10 459 case OP_NCREF:
1439 nigel 93 case OP_RREF:
1440 ph10 459 case OP_NRREF:
1441 nigel 93 case OP_DEF:
1442 nigel 77 code += _pcre_OP_lengths[*code];
1443     break;
1444    
1445     default:
1446     return code;
1447     }
1448     }
1449     /* Control never reaches here */
1450     }
1451    
1452    
1453    
1454    
1455     /*************************************************
1456 ph10 454 * Find the fixed length of a branch *
1457 nigel 77 *************************************************/
1458    
1459 ph10 454 /* Scan a branch and compute the fixed length of subject that will match it,
1460 nigel 77 if the length is fixed. This is needed for dealing with backward assertions.
1461 ph10 461 In UTF8 mode, the result is in characters rather than bytes. The branch is
1462 ph10 454 temporarily terminated with OP_END when this function is called.
1463 nigel 77
1464 ph10 461 This function is called when a backward assertion is encountered, so that if it
1465     fails, the error message can point to the correct place in the pattern.
1466 ph10 454 However, we cannot do this when the assertion contains subroutine calls,
1467 ph10 461 because they can be forward references. We solve this by remembering this case
1468 ph10 454 and doing the check at the end; a flag specifies which mode we are running in.
1469    
1470 nigel 77 Arguments:
1471     code points to the start of the pattern (the bracket)
1472 ph10 604 utf8 TRUE in UTF-8 mode
1473 ph10 461 atend TRUE if called when the pattern is complete
1474     cd the "compile data" structure
1475 nigel 77
1476 ph10 461 Returns: the fixed length,
1477 ph10 454 or -1 if there is no fixed length,
1478 nigel 77 or -2 if \C was encountered
1479 ph10 454 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1480 nigel 77 */
1481    
1482     static int
1483 ph10 604 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1484 nigel 77 {
1485     int length = -1;
1486    
1487     register int branchlength = 0;
1488     register uschar *cc = code + 1 + LINK_SIZE;
1489    
1490     /* Scan along the opcodes for this branch. If we get to the end of the
1491     branch, check the length against that of the other branches. */
1492    
1493     for (;;)
1494     {
1495     int d;
1496 ph10 454 uschar *ce, *cs;
1497 nigel 77 register int op = *cc;
1498     switch (op)
1499     {
1500 ph10 604 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1501     OP_BRA (normal non-capturing bracket) because the other variants of these
1502     opcodes are all concerned with unlimited repeated groups, which of course
1503     are not of fixed length. They will cause a -1 response from the default
1504     case of this switch. */
1505    
1506 nigel 93 case OP_CBRA:
1507 nigel 77 case OP_BRA:
1508     case OP_ONCE:
1509 ph10 723 case OP_ONCE_NC:
1510 nigel 77 case OP_COND:
1511 ph10 604 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1512 nigel 77 if (d < 0) return d;
1513     branchlength += d;
1514     do cc += GET(cc, 1); while (*cc == OP_ALT);
1515     cc += 1 + LINK_SIZE;
1516     break;
1517    
1518     /* Reached end of a branch; if it's a ket it is the end of a nested
1519     call. If it's ALT it is an alternation in a nested call. If it is
1520 ph10 604 END it's the end of the outer call. All can be handled by the same code.
1521     Note that we must not include the OP_KETRxxx opcodes here, because they
1522     all imply an unlimited repeat. */
1523 nigel 77
1524     case OP_ALT:
1525     case OP_KET:
1526     case OP_END:
1527     if (length < 0) length = branchlength;
1528     else if (length != branchlength) return -1;
1529     if (*cc != OP_ALT) return length;
1530     cc += 1 + LINK_SIZE;
1531     branchlength = 0;
1532     break;
1533 ph10 461
1534 ph10 454 /* A true recursion implies not fixed length, but a subroutine call may
1535     be OK. If the subroutine is a forward reference, we can't deal with
1536     it until the end of the pattern, so return -3. */
1537 ph10 461
1538 ph10 454 case OP_RECURSE:
1539     if (!atend) return -3;
1540     cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1541     do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1542     if (cc > cs && cc < ce) return -1; /* Recursion */
1543 ph10 604 d = find_fixedlength(cs + 2, utf8, atend, cd);
1544 ph10 461 if (d < 0) return d;
1545 ph10 454 branchlength += d;
1546     cc += 1 + LINK_SIZE;
1547 ph10 461 break;
1548 nigel 77
1549     /* Skip over assertive subpatterns */
1550    
1551     case OP_ASSERT:
1552     case OP_ASSERT_NOT:
1553     case OP_ASSERTBACK:
1554     case OP_ASSERTBACK_NOT:
1555     do cc += GET(cc, 1); while (*cc == OP_ALT);
1556     /* Fall through */
1557    
1558     /* Skip over things that don't match chars */
1559    
1560     case OP_REVERSE:
1561     case OP_CREF:
1562 ph10 459 case OP_NCREF:
1563 nigel 93 case OP_RREF:
1564 ph10 459 case OP_NRREF:
1565 nigel 93 case OP_DEF:
1566 nigel 77 case OP_CALLOUT:
1567     case OP_SOD:
1568     case OP_SOM:
1569 ph10 500 case OP_SET_SOM:
1570 nigel 77 case OP_EOD:
1571     case OP_EODN:
1572     case OP_CIRC:
1573 ph10 602 case OP_CIRCM:
1574 nigel 77 case OP_DOLL:
1575 ph10 602 case OP_DOLLM:
1576 nigel 77 case OP_NOT_WORD_BOUNDARY:
1577     case OP_WORD_BOUNDARY:
1578     cc += _pcre_OP_lengths[*cc];
1579     break;
1580    
1581     /* Handle literal characters */
1582    
1583     case OP_CHAR:
1584 ph10 602 case OP_CHARI:
1585 nigel 91 case OP_NOT:
1586 ph10 604 case OP_NOTI:
1587 nigel 77 branchlength++;
1588     cc += 2;
1589     #ifdef SUPPORT_UTF8
1590 ph10 604 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1591 nigel 77 #endif
1592     break;
1593    
1594     /* Handle exact repetitions. The count is already in characters, but we
1595     need to skip over a multibyte character in UTF8 mode. */
1596    
1597     case OP_EXACT:
1598     branchlength += GET2(cc,1);
1599     cc += 4;
1600     #ifdef SUPPORT_UTF8
1601 ph10 604 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1602 nigel 77 #endif
1603     break;
1604    
1605     case OP_TYPEEXACT:
1606     branchlength += GET2(cc,1);
1607 ph10 220 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1608 nigel 77 cc += 4;
1609     break;
1610    
1611     /* Handle single-char matchers */
1612    
1613     case OP_PROP:
1614     case OP_NOTPROP:
1615 nigel 87 cc += 2;
1616 nigel 77 /* Fall through */
1617    
1618     case OP_NOT_DIGIT:
1619     case OP_DIGIT:
1620     case OP_NOT_WHITESPACE:
1621     case OP_WHITESPACE:
1622     case OP_NOT_WORDCHAR:
1623     case OP_WORDCHAR:
1624     case OP_ANY:
1625 ph10 342 case OP_ALLANY:
1626 nigel 77 branchlength++;
1627     cc++;
1628     break;
1629    
1630     /* The single-byte matcher isn't allowed */
1631    
1632     case OP_ANYBYTE:
1633     return -2;
1634    
1635     /* Check a class for variable quantification */
1636    
1637     #ifdef SUPPORT_UTF8
1638     case OP_XCLASS:
1639     cc += GET(cc, 1) - 33;
1640     /* Fall through */
1641     #endif
1642    
1643     case OP_CLASS:
1644     case OP_NCLASS:
1645     cc += 33;
1646    
1647     switch (*cc)
1648     {
1649     case OP_CRSTAR:
1650     case OP_CRMINSTAR:
1651     case OP_CRQUERY:
1652     case OP_CRMINQUERY:
1653     return -1;
1654    
1655     case OP_CRRANGE:
1656     case OP_CRMINRANGE:
1657     if (GET2(cc,1) != GET2(cc,3)) return -1;
1658     branchlength += GET2(cc,1);
1659     cc += 5;
1660     break;
1661    
1662     default:
1663     branchlength++;
1664     }
1665     break;
1666    
1667     /* Anything else is variable length */
1668    
1669     default:
1670     return -1;
1671     }
1672     }
1673     /* Control never gets here */
1674     }
1675    
1676    
1677    
1678    
1679     /*************************************************
1680 ph10 454 * Scan compiled regex for specific bracket *
1681 nigel 77 *************************************************/
1682    
1683     /* This little function scans through a compiled pattern until it finds a
1684 ph10 454 capturing bracket with the given number, or, if the number is negative, an
1685 ph10 461 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1686     so that it can be called from pcre_study() when finding the minimum matching
1687 ph10 455 length.
1688 nigel 77
1689     Arguments:
1690     code points to start of expression
1691     utf8 TRUE in UTF-8 mode
1692 ph10 454 number the required bracket number or negative to find a lookbehind
1693 nigel 77
1694     Returns: pointer to the opcode for the bracket, or NULL if not found
1695     */
1696    
1697 ph10 455 const uschar *
1698     _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1699 nigel 77 {
1700     for (;;)
1701     {
1702     register int c = *code;
1703 ph10 618
1704 nigel 77 if (c == OP_END) return NULL;
1705 nigel 91
1706     /* XCLASS is used for classes that cannot be represented just by a bit
1707     map. This includes negated single high-valued characters. The length in
1708     the table is zero; the actual length is stored in the compiled code. */
1709    
1710     if (c == OP_XCLASS) code += GET(code, 1);
1711 ph10 461
1712 ph10 454 /* Handle recursion */
1713 ph10 461
1714 ph10 454 else if (c == OP_REVERSE)
1715     {
1716 ph10 461 if (number < 0) return (uschar *)code;
1717 ph10 454 code += _pcre_OP_lengths[c];
1718     }
1719 nigel 91
1720 nigel 93 /* Handle capturing bracket */
1721 nigel 91
1722 ph10 604 else if (c == OP_CBRA || c == OP_SCBRA ||
1723     c == OP_CBRAPOS || c == OP_SCBRAPOS)
1724 nigel 77 {
1725 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1726 nigel 77 if (n == number) return (uschar *)code;
1727 nigel 93 code += _pcre_OP_lengths[c];
1728 nigel 77 }
1729 nigel 91
1730 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1731     repeated character types, we have to test for \p and \P, which have an extra
1732 ph10 512 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1733 ph10 510 must add in its length. */
1734 nigel 91
1735 nigel 77 else
1736     {
1737 ph10 218 switch(c)
1738     {
1739     case OP_TYPESTAR:
1740     case OP_TYPEMINSTAR:
1741     case OP_TYPEPLUS:
1742     case OP_TYPEMINPLUS:
1743     case OP_TYPEQUERY:
1744     case OP_TYPEMINQUERY:
1745     case OP_TYPEPOSSTAR:
1746     case OP_TYPEPOSPLUS:
1747     case OP_TYPEPOSQUERY:
1748     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1749 ph10 220 break;
1750 ph10 221
1751     case OP_TYPEUPTO:
1752     case OP_TYPEMINUPTO:
1753     case OP_TYPEEXACT:
1754     case OP_TYPEPOSUPTO:
1755     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1756     break;
1757 ph10 512
1758 ph10 510 case OP_MARK:
1759     case OP_PRUNE_ARG:
1760     case OP_SKIP_ARG:
1761     code += code[1];
1762 ph10 512 break;
1763 ph10 550
1764     case OP_THEN_ARG:
1765 ph10 716 code += code[1];
1766 ph10 550 break;
1767 ph10 220 }
1768    
1769 ph10 218 /* Add in the fixed length from the table */
1770 ph10 220
1771 nigel 77 code += _pcre_OP_lengths[c];
1772 ph10 220
1773 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1774     a multi-byte character. The length in the table is a minimum, so we have to
1775     arrange to skip the extra bytes. */
1776 ph10 220
1777 ph10 107 #ifdef SUPPORT_UTF8
1778 nigel 77 if (utf8) switch(c)
1779     {
1780     case OP_CHAR:
1781 ph10 602 case OP_CHARI:
1782 nigel 77 case OP_EXACT:
1783 ph10 602 case OP_EXACTI:
1784 nigel 77 case OP_UPTO:
1785 ph10 602 case OP_UPTOI:
1786 nigel 77 case OP_MINUPTO:
1787 ph10 602 case OP_MINUPTOI:
1788 nigel 93 case OP_POSUPTO:
1789 ph10 602 case OP_POSUPTOI:
1790 nigel 77 case OP_STAR:
1791 ph10 602 case OP_STARI:
1792 nigel 77 case OP_MINSTAR:
1793 ph10 602 case OP_MINSTARI:
1794 nigel 93 case OP_POSSTAR:
1795 ph10 602 case OP_POSSTARI:
1796 nigel 77 case OP_PLUS:
1797 ph10 602 case OP_PLUSI:
1798 nigel 77 case OP_MINPLUS:
1799 ph10 602 case OP_MINPLUSI:
1800 nigel 93 case OP_POSPLUS:
1801 ph10 602 case OP_POSPLUSI:
1802 nigel 77 case OP_QUERY:
1803 ph10 602 case OP_QUERYI:
1804 nigel 77 case OP_MINQUERY:
1805 ph10 602 case OP_MINQUERYI:
1806 nigel 93 case OP_POSQUERY:
1807 ph10 602 case OP_POSQUERYI:
1808 nigel 93 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1809 nigel 77 break;
1810     }
1811 ph10 369 #else
1812     (void)(utf8); /* Keep compiler happy by referencing function argument */
1813 ph10 111 #endif
1814 nigel 77 }
1815     }
1816     }
1817    
1818    
1819    
1820     /*************************************************
1821     * Scan compiled regex for recursion reference *
1822     *************************************************/
1823    
1824     /* This little function scans through a compiled pattern until it finds an
1825     instance of OP_RECURSE.
1826    
1827     Arguments:
1828     code points to start of expression
1829     utf8 TRUE in UTF-8 mode
1830    
1831     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1832     */
1833    
1834     static const uschar *
1835     find_recurse(const uschar *code, BOOL utf8)
1836     {
1837     for (;;)
1838     {
1839     register int c = *code;
1840     if (c == OP_END) return NULL;
1841 nigel 91 if (c == OP_RECURSE) return code;
1842 ph10 220
1843 nigel 91 /* XCLASS is used for classes that cannot be represented just by a bit
1844     map. This includes negated single high-valued characters. The length in
1845     the table is zero; the actual length is stored in the compiled code. */
1846    
1847     if (c == OP_XCLASS) code += GET(code, 1);
1848    
1849 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1850     repeated character types, we have to test for \p and \P, which have an extra
1851 ph10 512 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1852 ph10 510 must add in its length. */
1853 nigel 91
1854 nigel 77 else
1855     {
1856 ph10 218 switch(c)
1857     {
1858     case OP_TYPESTAR:
1859     case OP_TYPEMINSTAR:
1860     case OP_TYPEPLUS:
1861     case OP_TYPEMINPLUS:
1862     case OP_TYPEQUERY:
1863     case OP_TYPEMINQUERY:
1864     case OP_TYPEPOSSTAR:
1865     case OP_TYPEPOSPLUS:
1866     case OP_TYPEPOSQUERY:
1867     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1868 ph10 220 break;
1869 ph10 221
1870     case OP_TYPEPOSUPTO:
1871     case OP_TYPEUPTO:
1872     case OP_TYPEMINUPTO:
1873     case OP_TYPEEXACT:
1874     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1875     break;
1876 ph10 512
1877 ph10 510 case OP_MARK:
1878     case OP_PRUNE_ARG:
1879     case OP_SKIP_ARG:
1880     code += code[1];
1881 ph10 512 break;
1882 ph10 550
1883     case OP_THEN_ARG:
1884 ph10 716 code += code[1];
1885 ph10 550 break;
1886 ph10 220 }
1887    
1888 ph10 218 /* Add in the fixed length from the table */
1889    
1890 nigel 77 code += _pcre_OP_lengths[c];
1891 ph10 220
1892 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1893     by a multi-byte character. The length in the table is a minimum, so we have
1894     to arrange to skip the extra bytes. */
1895 ph10 220
1896 ph10 107 #ifdef SUPPORT_UTF8
1897 nigel 77 if (utf8) switch(c)
1898     {
1899     case OP_CHAR:
1900 ph10 602 case OP_CHARI:
1901 nigel 77 case OP_EXACT:
1902 ph10 602 case OP_EXACTI:
1903 nigel 77 case OP_UPTO:
1904 ph10 602 case OP_UPTOI:
1905 nigel 77 case OP_MINUPTO:
1906 ph10 602 case OP_MINUPTOI:
1907 nigel 93 case OP_POSUPTO:
1908 ph10 602 case OP_POSUPTOI:
1909 nigel 77 case OP_STAR:
1910 ph10 602 case OP_STARI:
1911 nigel 77 case OP_MINSTAR:
1912 ph10 602 case OP_MINSTARI:
1913 nigel 93 case OP_POSSTAR:
1914 ph10 602 case OP_POSSTARI:
1915 nigel 77 case OP_PLUS:
1916 ph10 602 case OP_PLUSI:
1917 nigel 77 case OP_MINPLUS:
1918 ph10 602 case OP_MINPLUSI:
1919 nigel 93 case OP_POSPLUS:
1920 ph10 602 case OP_POSPLUSI:
1921 nigel 77 case OP_QUERY:
1922 ph10 602 case OP_QUERYI:
1923 nigel 77 case OP_MINQUERY:
1924 ph10 602 case OP_MINQUERYI:
1925 nigel 93 case OP_POSQUERY:
1926 ph10 602 case OP_POSQUERYI:
1927 nigel 93 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1928 nigel 77 break;
1929     }
1930 ph10 369 #else
1931     (void)(utf8); /* Keep compiler happy by referencing function argument */
1932 ph10 111 #endif
1933 nigel 77 }
1934     }
1935     }
1936    
1937    
1938    
1939     /*************************************************
1940     * Scan compiled branch for non-emptiness *
1941     *************************************************/
1942    
1943     /* This function scans through a branch of a compiled pattern to see whether it
1944 nigel 93 can match the empty string or not. It is called from could_be_empty()
1945     below and from compile_branch() when checking for an unlimited repeat of a
1946     group that can match nothing. Note that first_significant_code() skips over
1947 ph10 282 backward and negative forward assertions when its final argument is TRUE. If we
1948     hit an unclosed bracket, we return "empty" - this means we've struck an inner
1949     bracket whose current branch will already have been scanned.
1950 nigel 77
1951     Arguments:
1952     code points to start of search
1953     endcode points to where to stop
1954     utf8 TRUE if in UTF8 mode
1955 ph10 503 cd contains pointers to tables etc.
1956 nigel 77
1957     Returns: TRUE if what is matched could be empty
1958     */
1959    
1960     static BOOL
1961 ph10 503 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1962     compile_data *cd)
1963 nigel 77 {
1964     register int c;
1965 ph10 604 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1966 nigel 77 code < endcode;
1967 ph10 604 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1968 nigel 77 {
1969     const uschar *ccode;
1970    
1971     c = *code;
1972 ph10 507
1973 ph10 286 /* Skip over forward assertions; the other assertions are skipped by
1974 ph10 282 first_significant_code() with a TRUE final argument. */
1975 ph10 286
1976 ph10 282 if (c == OP_ASSERT)
1977 ph10 286 {
1978 ph10 282 do code += GET(code, 1); while (*code == OP_ALT);
1979     c = *code;
1980     continue;
1981 ph10 286 }
1982 ph10 172
1983 ph10 503 /* For a recursion/subroutine call, if its end has been reached, which
1984 ph10 624 implies a backward reference subroutine call, we can scan it. If it's a
1985     forward reference subroutine call, we can't. To detect forward reference
1986 ph10 654 we have to scan up the list that is kept in the workspace. This function is
1987     called only when doing the real compile, not during the pre-compile that
1988 ph10 624 measures the size of the compiled pattern. */
1989 ph10 507
1990 ph10 503 if (c == OP_RECURSE)
1991     {
1992 ph10 624 const uschar *scode;
1993     BOOL empty_branch;
1994 ph10 654
1995 ph10 624 /* Test for forward reference */
1996 ph10 654
1997 ph10 624 for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1998 ph10 654 if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1999 ph10 624
2000     /* Not a forward reference, test for completed backward reference */
2001 ph10 654
2002 ph10 624 empty_branch = FALSE;
2003     scode = cd->start_code + GET(code, 1);
2004 ph10 503 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
2005 ph10 654
2006 ph10 624 /* Completed backwards reference */
2007 ph10 654
2008 ph10 503 do
2009     {
2010 ph10 504 if (could_be_empty_branch(scode, endcode, utf8, cd))
2011     {
2012     empty_branch = TRUE;
2013 ph10 507 break;
2014     }
2015 ph10 503 scode += GET(scode, 1);
2016     }
2017     while (*scode == OP_ALT);
2018 ph10 654
2019 ph10 504 if (!empty_branch) return FALSE; /* All branches are non-empty */
2020 ph10 503 continue;
2021 ph10 507 }
2022 ph10 170
2023 ph10 604 /* Groups with zero repeats can of course be empty; skip them. */
2024    
2025     if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2026     c == OP_BRAPOSZERO)
2027     {
2028     code += _pcre_OP_lengths[c];
2029     do code += GET(code, 1); while (*code == OP_ALT);
2030     c = *code;
2031     continue;
2032     }
2033    
2034     /* A nested group that is already marked as "could be empty" can just be
2035     skipped. */
2036    
2037     if (c == OP_SBRA || c == OP_SBRAPOS ||
2038     c == OP_SCBRA || c == OP_SCBRAPOS)
2039     {
2040     do code += GET(code, 1); while (*code == OP_ALT);
2041     c = *code;
2042     continue;
2043     }
2044    
2045 ph10 170 /* For other groups, scan the branches. */
2046 ph10 172
2047 ph10 604 if (c == OP_BRA || c == OP_BRAPOS ||
2048     c == OP_CBRA || c == OP_CBRAPOS ||
2049 ph10 723 c == OP_ONCE || c == OP_ONCE_NC ||
2050     c == OP_COND)
2051 nigel 77 {
2052     BOOL empty_branch;
2053     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2054 ph10 406
2055     /* If a conditional group has only one branch, there is a second, implied,
2056 ph10 395 empty branch, so just skip over the conditional, because it could be empty.
2057     Otherwise, scan the individual branches of the group. */
2058 ph10 406
2059 ph10 395 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2060 nigel 77 code += GET(code, 1);
2061 ph10 395 else
2062 ph10 406 {
2063 ph10 395 empty_branch = FALSE;
2064     do
2065     {
2066 ph10 503 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2067 ph10 395 empty_branch = TRUE;
2068     code += GET(code, 1);
2069     }
2070     while (*code == OP_ALT);
2071     if (!empty_branch) return FALSE; /* All branches are non-empty */
2072 nigel 77 }
2073 ph10 406
2074 ph10 172 c = *code;
2075 nigel 93 continue;
2076 nigel 77 }
2077    
2078 nigel 93 /* Handle the other opcodes */
2079    
2080     switch (c)
2081 nigel 77 {
2082 ph10 216 /* Check for quantifiers after a class. XCLASS is used for classes that
2083     cannot be represented just by a bit map. This includes negated single
2084     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2085 ph10 220 actual length is stored in the compiled code, so we must update "code"
2086 ph10 216 here. */
2087 nigel 77
2088     #ifdef SUPPORT_UTF8
2089     case OP_XCLASS:
2090 ph10 216 ccode = code += GET(code, 1);
2091 nigel 77 goto CHECK_CLASS_REPEAT;
2092     #endif
2093    
2094     case OP_CLASS:
2095     case OP_NCLASS:
2096     ccode = code + 33;
2097    
2098     #ifdef SUPPORT_UTF8
2099     CHECK_CLASS_REPEAT:
2100     #endif
2101    
2102     switch (*ccode)
2103     {
2104     case OP_CRSTAR: /* These could be empty; continue */
2105     case OP_CRMINSTAR:
2106     case OP_CRQUERY:
2107     case OP_CRMINQUERY:
2108     break;
2109    
2110     default: /* Non-repeat => class must match */
2111     case OP_CRPLUS: /* These repeats aren't empty */
2112     case OP_CRMINPLUS:
2113     return FALSE;
2114    
2115     case OP_CRRANGE:
2116     case OP_CRMINRANGE:
2117     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2118     break;
2119     }
2120     break;
2121    
2122     /* Opcodes that must match a character */
2123    
2124     case OP_PROP:
2125     case OP_NOTPROP:
2126     case OP_EXTUNI:
2127     case OP_NOT_DIGIT:
2128     case OP_DIGIT:
2129     case OP_NOT_WHITESPACE:
2130     case OP_WHITESPACE:
2131     case OP_NOT_WORDCHAR:
2132     case OP_WORDCHAR:
2133     case OP_ANY:
2134 ph10 345 case OP_ALLANY:
2135 nigel 77 case OP_ANYBYTE:
2136     case OP_CHAR:
2137 ph10 602 case OP_CHARI:
2138 nigel 77 case OP_NOT:
2139 ph10 602 case OP_NOTI:
2140 nigel 77 case OP_PLUS:
2141     case OP_MINPLUS:
2142 nigel 93 case OP_POSPLUS:
2143 nigel 77 case OP_EXACT:
2144     case OP_NOTPLUS:
2145     case OP_NOTMINPLUS:
2146 nigel 93 case OP_NOTPOSPLUS:
2147 nigel 77 case OP_NOTEXACT:
2148     case OP_TYPEPLUS:
2149     case OP_TYPEMINPLUS:
2150 nigel 93 case OP_TYPEPOSPLUS:
2151 nigel 77 case OP_TYPEEXACT:
2152     return FALSE;
2153 ph10 227
2154     /* These are going to continue, as they may be empty, but we have to
2155     fudge the length for the \p and \P cases. */
2156    
2157 ph10 224 case OP_TYPESTAR:
2158     case OP_TYPEMINSTAR:
2159     case OP_TYPEPOSSTAR:
2160     case OP_TYPEQUERY:
2161     case OP_TYPEMINQUERY:
2162     case OP_TYPEPOSQUERY:
2163     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2164 ph10 227 break;
2165    
2166 ph10 224 /* Same for these */
2167 ph10 227
2168 ph10 224 case OP_TYPEUPTO:
2169     case OP_TYPEMINUPTO:
2170     case OP_TYPEPOSUPTO:
2171     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2172     break;
2173 nigel 77
2174     /* End of branch */
2175    
2176     case OP_KET:
2177     case OP_KETRMAX:
2178     case OP_KETRMIN:
2179 ph10 604 case OP_KETRPOS:
2180 nigel 77 case OP_ALT:
2181     return TRUE;
2182    
2183 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2184     MINUPTO, and POSUPTO may be followed by a multibyte character */
2185 nigel 77
2186     #ifdef SUPPORT_UTF8
2187     case OP_STAR:
2188 ph10 602 case OP_STARI:
2189 nigel 77 case OP_MINSTAR:
2190 ph10 602 case OP_MINSTARI:
2191 nigel 93 case OP_POSSTAR:
2192 ph10 602 case OP_POSSTARI:
2193 nigel 77 case OP_QUERY:
2194 ph10 602 case OP_QUERYI:
2195 nigel 77 case OP_MINQUERY:
2196 ph10 602 case OP_MINQUERYI:
2197 nigel 93 case OP_POSQUERY:
2198 ph10 602 case OP_POSQUERYI:
2199 ph10 426 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2200     break;
2201 ph10 461
2202 nigel 77 case OP_UPTO:
2203 ph10 602 case OP_UPTOI:
2204 nigel 77 case OP_MINUPTO:
2205 ph10 602 case OP_MINUPTOI:
2206 nigel 93 case OP_POSUPTO:
2207 ph10 602 case OP_POSUPTOI:
2208 ph10 426 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2209 nigel 77 break;
2210     #endif
2211 ph10 503
2212 ph10 510 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2213     string. */
2214    
2215     case OP_MARK:
2216     case OP_PRUNE_ARG:
2217     case OP_SKIP_ARG:
2218     code += code[1];
2219 ph10 512 break;
2220 ph10 510
2221 ph10 550 case OP_THEN_ARG:
2222 ph10 716 code += code[1];
2223 ph10 550 break;
2224    
2225 ph10 503 /* None of the remaining opcodes are required to match a character. */
2226 ph10 507
2227 ph10 503 default:
2228 ph10 507 break;
2229 nigel 77 }
2230     }
2231    
2232     return TRUE;
2233     }
2234    
2235    
2236    
2237     /*************************************************
2238     * Scan compiled regex for non-emptiness *
2239     *************************************************/
2240    
2241     /* This function is called to check for left recursive calls. We want to check
2242     the current branch of the current pattern to see if it could match the empty
2243     string. If it could, we must look outwards for branches at other levels,
2244     stopping when we pass beyond the bracket which is the subject of the recursion.
2245 ph10 654 This function is called only during the real compile, not during the
2246 ph10 624 pre-compile.
2247 nigel 77
2248     Arguments:
2249     code points to start of the recursion
2250     endcode points to where to stop (current RECURSE item)
2251     bcptr points to the chain of current (unclosed) branch starts
2252     utf8 TRUE if in UTF-8 mode
2253 ph10 507 cd pointers to tables etc
2254 nigel 77
2255     Returns: TRUE if what is matched could be empty
2256     */
2257    
2258     static BOOL
2259     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2260 ph10 503 BOOL utf8, compile_data *cd)
2261 nigel 77 {
2262 ph10 475 while (bcptr != NULL && bcptr->current_branch >= code)
2263 nigel 77 {
2264 ph10 503 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2265 ph10 475 return FALSE;
2266 nigel 77 bcptr = bcptr->outer;
2267     }
2268     return TRUE;
2269     }
2270    
2271    
2272    
2273     /*************************************************
2274     * Check for POSIX class syntax *
2275     *************************************************/
2276    
2277     /* This function is called when the sequence "[:" or "[." or "[=" is
2278 ph10 295 encountered in a character class. It checks whether this is followed by a
2279 ph10 298 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2280 ph10 295 reach an unescaped ']' without the special preceding character, return FALSE.
2281 nigel 77
2282 ph10 298 Originally, this function only recognized a sequence of letters between the
2283     terminators, but it seems that Perl recognizes any sequence of characters,
2284     though of course unknown POSIX names are subsequently rejected. Perl gives an
2285     "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2286     didn't consider this to be a POSIX class. Likewise for [:1234:].
2287 ph10 295
2288 ph10 298 The problem in trying to be exactly like Perl is in the handling of escapes. We
2289     have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2290     class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2291     below handles the special case of \], but does not try to do any other escape
2292     processing. This makes it different from Perl for cases such as [:l\ower:]
2293 ph10 295 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2294 ph10 298 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2295 ph10 295 I think.
2296    
2297 ph10 640 A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not.
2298     It seems that the appearance of a nested POSIX class supersedes an apparent
2299     external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or
2300 ph10 691 a digit.
2301 ph10 640
2302 ph10 661 In Perl, unescaped square brackets may also appear as part of class names. For
2303     example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for
2304     [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not
2305 ph10 691 seem right at all. PCRE does not allow closing square brackets in POSIX class
2306 ph10 661 names.
2307    
2308 ph10 295 Arguments:
2309 nigel 77 ptr pointer to the initial [
2310     endptr where to return the end pointer
2311    
2312     Returns: TRUE or FALSE
2313     */
2314    
2315     static BOOL
2316 ph10 295 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2317 nigel 77 {
2318     int terminator; /* Don't combine these lines; the Solaris cc */
2319     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2320 ph10 295 for (++ptr; *ptr != 0; ptr++)
2321 nigel 77 {
2322 ph10 654 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2323     ptr++;
2324 ph10 691 else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2325 ph10 640 else
2326 ph10 298 {
2327 ph10 391 if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2328 ph10 295 {
2329     *endptr = ptr;
2330     return TRUE;
2331 ph10 298 }
2332 ph10 640 if (*ptr == CHAR_LEFT_SQUARE_BRACKET &&
2333     (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2334     ptr[1] == CHAR_EQUALS_SIGN) &&
2335     check_posix_syntax(ptr, endptr))
2336 ph10 654 return FALSE;
2337 ph10 298 }
2338     }
2339 nigel 77 return FALSE;
2340     }
2341    
2342    
2343    
2344    
2345     /*************************************************
2346     * Check POSIX class name *
2347     *************************************************/
2348    
2349     /* This function is called to check the name given in a POSIX-style class entry
2350     such as [:alnum:].
2351    
2352     Arguments:
2353     ptr points to the first letter
2354     len the length of the name
2355    
2356     Returns: a value representing the name, or -1 if unknown
2357     */
2358    
2359     static int
2360     check_posix_name(const uschar *ptr, int len)
2361     {
2362 ph10 240 const char *pn = posix_names;
2363 nigel 77 register int yield = 0;
2364     while (posix_name_lengths[yield] != 0)
2365     {
2366     if (len == posix_name_lengths[yield] &&
2367 ph10 240 strncmp((const char *)ptr, pn, len) == 0) return yield;
2368 ph10 243 pn += posix_name_lengths[yield] + 1;
2369 nigel 77 yield++;
2370     }
2371     return -1;
2372     }
2373    
2374    
2375     /*************************************************
2376     * Adjust OP_RECURSE items in repeated group *
2377     *************************************************/
2378    
2379     /* OP_RECURSE items contain an offset from the start of the regex to the group
2380     that is referenced. This means that groups can be replicated for fixed
2381     repetition simply by copying (because the recursion is allowed to refer to
2382     earlier groups that are outside the current group). However, when a group is
2383 ph10 335 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2384     inserted before it, after it has been compiled. This means that any OP_RECURSE
2385     items within it that refer to the group itself or any contained groups have to
2386     have their offsets adjusted. That one of the jobs of this function. Before it
2387     is called, the partially compiled regex must be temporarily terminated with
2388     OP_END.
2389 nigel 77
2390 nigel 93 This function has been extended with the possibility of forward references for
2391     recursions and subroutine calls. It must also check the list of such references
2392     for the group we are dealing with. If it finds that one of the recursions in
2393     the current group is on this list, it adjusts the offset in the list, not the
2394     value in the reference (which is a group number).
2395    
2396 nigel 77 Arguments:
2397     group points to the start of the group
2398     adjust the amount by which the group is to be moved
2399     utf8 TRUE in UTF-8 mode
2400     cd contains pointers to tables etc.
2401 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
2402 nigel 77
2403     Returns: nothing
2404     */
2405    
2406     static void
2407 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2408     uschar *save_hwm)
2409 nigel 77 {
2410     uschar *ptr = group;
2411 ph10 224
2412 nigel 77 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2413     {
2414 nigel 93 int offset;
2415     uschar *hc;
2416    
2417     /* See if this recursion is on the forward reference list. If so, adjust the
2418     reference. */
2419 ph10 345
2420 nigel 93 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2421     {
2422     offset = GET(hc, 0);
2423     if (cd->start_code + offset == ptr + 1)
2424     {
2425     PUT(hc, 0, offset + adjust);
2426     break;
2427     }
2428     }
2429    
2430     /* Otherwise, adjust the recursion offset if it's after the start of this
2431     group. */
2432    
2433     if (hc >= cd->hwm)
2434     {
2435     offset = GET(ptr, 1);
2436     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2437     }
2438    
2439 nigel 77 ptr += 1 + LINK_SIZE;
2440     }
2441     }
2442    
2443    
2444    
2445     /*************************************************
2446     * Insert an automatic callout point *
2447     *************************************************/
2448    
2449     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2450     callout points before each pattern item.
2451    
2452     Arguments:
2453     code current code pointer
2454     ptr current pattern pointer
2455     cd pointers to tables etc
2456    
2457     Returns: new code pointer
2458     */
2459    
2460     static uschar *
2461     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2462     {
2463     *code++ = OP_CALLOUT;
2464     *code++ = 255;
2465 ph10 530 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2466     PUT(code, LINK_SIZE, 0); /* Default length */
2467 nigel 77 return code + 2*LINK_SIZE;
2468     }
2469    
2470    
2471    
2472     /*************************************************
2473     * Complete a callout item *
2474     *************************************************/
2475    
2476     /* A callout item contains the length of the next item in the pattern, which
2477     we can't fill in till after we have reached the relevant point. This is used
2478     for both automatic and manual callouts.
2479    
2480     Arguments:
2481     previous_callout points to previous callout item
2482     ptr current pattern pointer
2483     cd pointers to tables etc
2484    
2485     Returns: nothing
2486     */
2487    
2488     static void
2489     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2490     {
2491 ph10 530 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2492 nigel 77 PUT(previous_callout, 2 + LINK_SIZE, length);
2493     }
2494    
2495    
2496    
2497     #ifdef SUPPORT_UCP
2498     /*************************************************
2499     * Get othercase range *
2500     *************************************************/
2501    
2502     /* This function is passed the start and end of a class range, in UTF-8 mode
2503     with UCP support. It searches up the characters, looking for internal ranges of
2504     characters in the "other" case. Each call returns the next one, updating the
2505     start address.
2506    
2507     Arguments:
2508     cptr points to starting character value; updated
2509     d end value
2510     ocptr where to put start of othercase range
2511     odptr where to put end of othercase range
2512    
2513     Yield: TRUE when range returned; FALSE when no more
2514     */
2515    
2516     static BOOL
2517 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2518     unsigned int *odptr)
2519 nigel 77 {
2520 nigel 93 unsigned int c, othercase, next;
2521 nigel 77
2522     for (c = *cptr; c <= d; c++)
2523 ph10 349 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2524 nigel 77
2525     if (c > d) return FALSE;
2526    
2527     *ocptr = othercase;
2528     next = othercase + 1;
2529    
2530     for (++c; c <= d; c++)
2531     {
2532 ph10 349 if (UCD_OTHERCASE(c) != next) break;
2533 nigel 77 next++;
2534     }
2535    
2536     *odptr = next - 1;
2537     *cptr = c;
2538    
2539     return TRUE;
2540     }
2541 ph10 532
2542    
2543    
2544     /*************************************************
2545     * Check a character and a property *
2546     *************************************************/
2547    
2548     /* This function is called by check_auto_possessive() when a property item
2549     is adjacent to a fixed character.
2550    
2551     Arguments:
2552     c the character
2553     ptype the property type
2554     pdata the data for the type
2555     negated TRUE if it's a negated property (\P or \p{^)
2556 ph10 535
2557 ph10 532 Returns: TRUE if auto-possessifying is OK
2558 ph10 535 */
2559 ph10 532
2560     static BOOL
2561     check_char_prop(int c, int ptype, int pdata, BOOL negated)
2562     {
2563     const ucd_record *prop = GET_UCD(c);
2564     switch(ptype)
2565     {
2566     case PT_LAMP:
2567     return (prop->chartype == ucp_Lu ||
2568     prop->chartype == ucp_Ll ||
2569     prop->chartype == ucp_Lt) == negated;
2570    
2571     case PT_GC:
2572     return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2573    
2574     case PT_PC:
2575     return (pdata == prop->chartype) == negated;
2576    
2577     case PT_SC:
2578     return (pdata == prop->script) == negated;
2579    
2580     /* These are specials */
2581    
2582     case PT_ALNUM:
2583     return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2584     _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2585    
2586     case PT_SPACE: /* Perl space */
2587     return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2588     c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2589     == negated;
2590    
2591     case PT_PXSPACE: /* POSIX space */
2592     return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2593     c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2594     c == CHAR_FF || c == CHAR_CR)
2595     == negated;
2596    
2597     case PT_WORD:
2598     return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2599     _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2600     c == CHAR_UNDERSCORE) == negated;
2601     }
2602 ph10 535 return FALSE;
2603 ph10 532 }
2604 nigel 77 #endif /* SUPPORT_UCP */
2605    
2606    
2607 nigel 93
2608 nigel 77 /*************************************************
2609 nigel 93 * Check if auto-possessifying is possible *
2610     *************************************************/
2611    
2612     /* This function is called for unlimited repeats of certain items, to see
2613     whether the next thing could possibly match the repeated item. If not, it makes
2614     sense to automatically possessify the repeated item.
2615    
2616     Arguments:
2617 ph10 532 previous pointer to the repeated opcode
2618 nigel 93 utf8 TRUE in UTF-8 mode
2619     ptr next character in pattern
2620     options options bits
2621     cd contains pointers to tables etc.
2622    
2623     Returns: TRUE if possessifying is wanted
2624     */
2625    
2626     static BOOL
2627 ph10 535 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2628 ph10 532 int options, compile_data *cd)
2629 nigel 93 {
2630 ph10 532 int c, next;
2631     int op_code = *previous++;
2632 nigel 93
2633     /* Skip whitespace and comments in extended mode */
2634    
2635     if ((options & PCRE_EXTENDED) != 0)
2636     {
2637     for (;;)
2638     {
2639     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2640 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2641 nigel 93 {
2642 ph10 579 ptr++;
2643 ph10 556 while (*ptr != 0)
2644     {
2645 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2646 ph10 556 ptr++;
2647 ph10 579 #ifdef SUPPORT_UTF8
2648 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2649     #endif
2650     }
2651 nigel 93 }
2652     else break;
2653     }
2654     }
2655    
2656     /* If the next item is one that we can handle, get its value. A non-negative
2657     value is a character, a negative value is an escape value. */
2658    
2659 ph10 391 if (*ptr == CHAR_BACKSLASH)
2660 nigel 93 {
2661     int temperrorcode = 0;
2662     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2663     if (temperrorcode != 0) return FALSE;
2664     ptr++; /* Point after the escape sequence */
2665     }
2666    
2667     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2668     {
2669     #ifdef SUPPORT_UTF8
2670     if (utf8) { GETCHARINC(next, ptr); } else
2671     #endif
2672     next = *ptr++;
2673     }
2674    
2675     else return FALSE;
2676    
2677     /* Skip whitespace and comments in extended mode */
2678    
2679     if ((options & PCRE_EXTENDED) != 0)
2680     {
2681     for (;;)
2682     {
2683     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2684 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2685 nigel 93 {
2686 ph10 579 ptr++;
2687 ph10 556 while (*ptr != 0)
2688     {
2689 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2690 ph10 556 ptr++;
2691 ph10 579 #ifdef SUPPORT_UTF8
2692 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2693     #endif
2694     }
2695 nigel 93 }
2696     else break;
2697     }
2698     }
2699    
2700     /* If the next thing is itself optional, we have to give up. */
2701    
2702 ph10 392 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2703 ph10 391 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2704     return FALSE;
2705 nigel 93
2706 ph10 532 /* Now compare the next item with the previous opcode. First, handle cases when
2707     the next item is a character. */
2708 nigel 93
2709     if (next >= 0) switch(op_code)
2710     {
2711     case OP_CHAR:
2712 ph10 535 #ifdef SUPPORT_UTF8
2713 ph10 532 GETCHARTEST(c, previous);
2714 ph10 369 #else
2715 ph10 532 c = *previous;
2716 ph10 535 #endif
2717     return c != next;
2718 nigel 93
2719 ph10 602 /* For CHARI (caseless character) we must check the other case. If we have
2720 nigel 93 Unicode property support, we can use it to test the other case of
2721     high-valued characters. */
2722    
2723 ph10 602 case OP_CHARI:
2724 ph10 535 #ifdef SUPPORT_UTF8
2725 ph10 532 GETCHARTEST(c, previous);
2726     #else
2727     c = *previous;
2728 ph10 535 #endif
2729 ph10 532 if (c == next) return FALSE;
2730 nigel 93 #ifdef SUPPORT_UTF8
2731     if (utf8)
2732     {
2733     unsigned int othercase;
2734     if (next < 128) othercase = cd->fcc[next]; else
2735     #ifdef SUPPORT_UCP
2736 ph10 349 othercase = UCD_OTHERCASE((unsigned int)next);
2737 nigel 93 #else
2738     othercase = NOTACHAR;
2739     #endif
2740 ph10 532 return (unsigned int)c != othercase;
2741 nigel 93 }
2742     else
2743     #endif /* SUPPORT_UTF8 */
2744 ph10 532 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2745 nigel 93
2746 ph10 602 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2747 ph10 604 opcodes are not used for multi-byte characters, because they are coded using
2748 ph10 602 an XCLASS instead. */
2749 nigel 93
2750     case OP_NOT:
2751 ph10 602 return (c = *previous) == next;
2752 ph10 604
2753     case OP_NOTI:
2754 ph10 532 if ((c = *previous) == next) return TRUE;
2755 nigel 93 #ifdef SUPPORT_UTF8
2756     if (utf8)
2757     {
2758     unsigned int othercase;
2759     if (next < 128) othercase = cd->fcc[next]; else
2760     #ifdef SUPPORT_UCP
2761 ph10 349 othercase = UCD_OTHERCASE(next);
2762 nigel 93 #else
2763     othercase = NOTACHAR;
2764     #endif
2765 ph10 532 return (unsigned int)c == othercase;
2766 nigel 93 }
2767     else
2768     #endif /* SUPPORT_UTF8 */
2769 ph10 532 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2770 nigel 93
2771 ph10 535 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2772     When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2773    
2774 nigel 93 case OP_DIGIT:
2775     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2776    
2777     case OP_NOT_DIGIT:
2778     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2779    
2780     case OP_WHITESPACE:
2781     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2782    
2783     case OP_NOT_WHITESPACE:
2784     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2785    
2786     case OP_WORDCHAR:
2787     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2788    
2789     case OP_NOT_WORDCHAR:
2790     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2791    
2792 ph10 180 case OP_HSPACE:
2793     case OP_NOT_HSPACE:
2794     switch(next)
2795     {
2796     case 0x09:
2797     case 0x20:
2798     case 0xa0:
2799     case 0x1680:
2800     case 0x180e:
2801     case 0x2000:
2802     case 0x2001:
2803     case 0x2002:
2804     case 0x2003:
2805     case 0x2004:
2806     case 0x2005:
2807     case 0x2006:
2808     case 0x2007:
2809     case 0x2008:
2810     case 0x2009:
2811     case 0x200A:
2812     case 0x202f:
2813     case 0x205f:
2814     case 0x3000:
2815 ph10 528 return op_code == OP_NOT_HSPACE;
2816 ph10 180 default:
2817 ph10 528 return op_code != OP_NOT_HSPACE;
2818 ph10 180 }
2819    
2820 ph10 528 case OP_ANYNL:
2821 ph10 180 case OP_VSPACE:
2822     case OP_NOT_VSPACE:
2823     switch(next)
2824     {
2825     case 0x0a:
2826     case 0x0b:
2827     case 0x0c:
2828     case 0x0d:
2829     case 0x85:
2830     case 0x2028:
2831     case 0x2029:
2832 ph10 528 return op_code == OP_NOT_VSPACE;
2833 ph10 180 default:
2834 ph10 528 return op_code != OP_NOT_VSPACE;
2835 ph10 180 }
2836    
2837 ph10 532 #ifdef SUPPORT_UCP
2838     case OP_PROP:
2839     return check_char_prop(next, previous[0], previous[1], FALSE);
2840 ph10 535
2841 ph10 532 case OP_NOTPROP:
2842     return check_char_prop(next, previous[0], previous[1], TRUE);
2843     #endif
2844    
2845 nigel 93 default:
2846     return FALSE;
2847     }
2848    
2849    
2850 ph10 535 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2851     is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2852     generated only when PCRE_UCP is *not* set, that is, when only ASCII
2853     characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2854 ph10 532 replaced by OP_PROP codes when PCRE_UCP is set. */
2855 nigel 93
2856     switch(op_code)
2857     {
2858     case OP_CHAR:
2859 ph10 602 case OP_CHARI:
2860 ph10 535 #ifdef SUPPORT_UTF8
2861 ph10 532 GETCHARTEST(c, previous);
2862     #else
2863     c = *previous;
2864 ph10 535 #endif
2865 nigel 93 switch(-next)
2866     {
2867     case ESC_d:
2868 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2869 nigel 93
2870     case ESC_D:
2871 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2872 nigel 93
2873     case ESC_s:
2874 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2875 nigel 93
2876     case ESC_S:
2877 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2878 nigel 93
2879     case ESC_w:
2880 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2881 nigel 93
2882     case ESC_W:
2883 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2884 ph10 182
2885 ph10 180 case ESC_h:
2886     case ESC_H:
2887 ph10 532 switch(c)
2888 ph10 180 {
2889     case 0x09:
2890     case 0x20:
2891     case 0xa0:
2892     case 0x1680:
2893     case 0x180e:
2894     case 0x2000:
2895     case 0x2001:
2896     case 0x2002:
2897     case 0x2003:
2898     case 0x2004:
2899     case 0x2005:
2900     case 0x2006:
2901     case 0x2007:
2902     case 0x2008:
2903     case 0x2009:
2904     case 0x200A:
2905     case 0x202f:
2906     case 0x205f:
2907     case 0x3000:
2908     return -next != ESC_h;
2909     default:
2910     return -next == ESC_h;
2911 ph10 182 }
2912    
2913 ph10 180 case ESC_v:
2914     case ESC_V:
2915 ph10 532 switch(c)
2916 ph10 180 {
2917     case 0x0a:
2918     case 0x0b:
2919     case 0x0c:
2920     case 0x0d:
2921     case 0x85:
2922     case 0x2028:
2923     case 0x2029:
2924     return -next != ESC_v;
2925     default:
2926     return -next == ESC_v;
2927 ph10 182 }
2928 ph10 535
2929     /* When PCRE_UCP is set, these values get generated for \d etc. Find
2930     their substitutions and process them. The result will always be either
2931 ph10 532 -ESC_p or -ESC_P. Then fall through to process those values. */
2932 ph10 535
2933 ph10 532 #ifdef SUPPORT_UCP
2934     case ESC_du:
2935     case ESC_DU:
2936     case ESC_wu:
2937     case ESC_WU:
2938     case ESC_su:
2939     case ESC_SU:
2940     {
2941     int temperrorcode = 0;
2942     ptr = substitutes[-next - ESC_DU];
2943     next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2944     if (temperrorcode != 0) return FALSE;
2945     ptr++; /* For compatibility */
2946     }
2947 ph10 535 /* Fall through */
2948 nigel 93
2949 ph10 532 case ESC_p:
2950     case ESC_P:
2951     {
2952     int ptype, pdata, errorcodeptr;
2953 ph10 535 BOOL negated;
2954    
2955 ph10 532 ptr--; /* Make ptr point at the p or P */
2956     ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2957     if (ptype < 0) return FALSE;
2958     ptr++; /* Point past the final curly ket */
2959 ph10 535
2960 ph10 532 /* If the property item is optional, we have to give up. (When generated
2961     from \d etc by PCRE_UCP, this test will have been applied much earlier,
2962     to the original \d etc. At this point, ptr will point to a zero byte. */
2963 ph10 535
2964 ph10 532 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2965     strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2966     return FALSE;
2967 ph10 535
2968 ph10 532 /* Do the property check. */
2969 ph10 535
2970 ph10 532 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2971 ph10 535 }
2972 ph10 532 #endif
2973    
2974 nigel 93 default:
2975     return FALSE;
2976     }
2977    
2978 ph10 535 /* In principle, support for Unicode properties should be integrated here as
2979     well. It means re-organizing the above code so as to get hold of the property
2980     values before switching on the op-code. However, I wonder how many patterns
2981     combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2982     these op-codes are never generated.) */
2983    
2984 nigel 93 case OP_DIGIT:
2985 ph10 180 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2986 ph10 528 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2987 nigel 93
2988     case OP_NOT_DIGIT:
2989     return next == -ESC_d;
2990    
2991     case OP_WHITESPACE:
2992 ph10 528 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2993 nigel 93
2994     case OP_NOT_WHITESPACE:
2995 ph10 180 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2996 nigel 93
2997 ph10 180 case OP_HSPACE:
2998 ph10 535 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2999 ph10 528 next == -ESC_w || next == -ESC_v || next == -ESC_R;
3000 ph10 180
3001     case OP_NOT_HSPACE:
3002     return next == -ESC_h;
3003 ph10 182
3004 ph10 180 /* Can't have \S in here because VT matches \S (Perl anomaly) */
3005 ph10 535 case OP_ANYNL:
3006 ph10 182 case OP_VSPACE:
3007 ph10 180 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
3008    
3009     case OP_NOT_VSPACE:
3010 ph10 528 return next == -ESC_v || next == -ESC_R;
3011 ph10 180
3012 nigel 93 case OP_WORDCHAR:
3013 ph10 535 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
3014 ph10 528 next == -ESC_v || next == -ESC_R;
3015 nigel 93
3016     case OP_NOT_WORDCHAR:
3017     return next == -ESC_w || next == -ESC_d;
3018 ph10 182
3019 nigel 93 default:
3020     return FALSE;
3021     }
3022    
3023     /* Control does not reach here */
3024     }
3025    
3026    
3027    
3028     /*************************************************
3029 nigel 77 * Compile one branch *
3030     *************************************************/
3031    
3032 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
3033 nigel 77 changed during the branch, the pointer is used to change the external options
3034 nigel 93 bits. This function is used during the pre-compile phase when we are trying
3035     to find out the amount of memory needed, as well as during the real compile
3036     phase. The value of lengthptr distinguishes the two phases.
3037 nigel 77
3038     Arguments:
3039     optionsptr pointer to the option bits
3040     codeptr points to the pointer to the current code point
3041     ptrptr points to the current pattern pointer
3042     errorcodeptr points to error code variable
3043     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3044     reqbyteptr set to the last literal character required, else < 0
3045     bcptr points to current branch chain
3046 ph10 654 cond_depth conditional nesting depth
3047 nigel 77 cd contains pointers to tables etc.
3048 nigel 93 lengthptr NULL during the real compile phase
3049     points to length accumulator during pre-compile phase
3050 nigel 77
3051     Returns: TRUE on success
3052     FALSE, with *errorcodeptr set non-zero on error
3053     */
3054    
3055     static BOOL
3056 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3057     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3058 ph10 642 int cond_depth, compile_data *cd, int *lengthptr)
3059 nigel 77 {
3060     int repeat_type, op_type;
3061     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3062     int bravalue = 0;
3063     int greedy_default, greedy_non_default;
3064     int firstbyte, reqbyte;
3065     int zeroreqbyte, zerofirstbyte;
3066     int req_caseopt, reqvary, tempreqvary;
3067 ph10 635 int options = *optionsptr; /* May change dynamically */
3068 nigel 77 int after_manual_callout = 0;
3069 nigel 93 int length_prevgroup = 0;
3070 nigel 77 register int c;
3071     register uschar *code = *codeptr;
3072 nigel 93 uschar *last_code = code;
3073     uschar *orig_code = code;
3074 nigel 77 uschar *tempcode;
3075     BOOL inescq = FALSE;
3076     BOOL groupsetfirstbyte = FALSE;
3077     const uschar *ptr = *ptrptr;
3078     const uschar *tempptr;
3079 ph10 518 const uschar *nestptr = NULL;
3080 nigel 77 uschar *previous = NULL;
3081     uschar *previous_callout = NULL;
3082 nigel 93 uschar *save_hwm = NULL;
3083 nigel 77 uschar classbits[32];
3084    
3085 ph10 635 /* We can fish out the UTF-8 setting once and for all into a BOOL, but we
3086 ph10 654 must not do this for other options (e.g. PCRE_EXTENDED) because they may change
3087 ph10 635 dynamically as we process the pattern. */
3088    
3089 nigel 77 #ifdef SUPPORT_UTF8
3090     BOOL class_utf8;
3091     BOOL utf8 = (options & PCRE_UTF8) != 0;
3092     uschar *class_utf8data;
3093 ph10 300 uschar *class_utf8data_base;
3094 nigel 77 uschar utf8_char[6];
3095     #else
3096     BOOL utf8 = FALSE;
3097     #endif
3098    
3099 ph10 475 #ifdef PCRE_DEBUG
3100 nigel 93 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3101     #endif
3102    
3103 nigel 77 /* Set up the default and non-default settings for greediness */
3104    
3105     greedy_default = ((options & PCRE_UNGREEDY) != 0);
3106     greedy_non_default = greedy_default ^ 1;
3107    
3108     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3109     matching encountered yet". It gets changed to REQ_NONE if we hit something that
3110     matches a non-fixed char first char; reqbyte just remains unset if we never
3111     find one.
3112    
3113     When we hit a repeat whose minimum is zero, we may have to adjust these values
3114     to take the zero repeat into account. This is implemented by setting them to
3115     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3116     item types that can be repeated set these backoff variables appropriately. */
3117    
3118     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3119    
3120     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3121     according to the current setting of the caseless flag. REQ_CASELESS is a bit
3122     value > 255. It is added into the firstbyte or reqbyte variables to record the
3123     case status of the value. This is used only for ASCII characters. */
3124    
3125     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3126    
3127     /* Switch on next character until the end of the branch */
3128    
3129     for (;; ptr++)
3130     {
3131     BOOL negate_class;
3132 ph10 286 BOOL should_flip_negation;
3133 nigel 77 BOOL possessive_quantifier;
3134     BOOL is_quantifier;
3135 nigel 93 BOOL is_recurse;
3136 ph10 180 BOOL reset_bracount;
3137 nigel 77 int class_charcount;
3138     int class_lastchar;
3139     int newoptions;
3140     int recno;
3141 ph10 172 int refsign;
3142 nigel 77 int skipbytes;
3143     int subreqbyte;
3144     int subfirstbyte;
3145 nigel 93 int terminator;
3146 nigel 77 int mclength;
3147 ph10 723 int tempbracount;
3148 nigel 77 uschar mcbuffer[8];
3149    
3150 nigel 93 /* Get next byte in the pattern */
3151 nigel 77
3152     c = *ptr;
3153 ph10 345
3154 ph10 535 /* If we are at the end of a nested substitution, revert to the outer level
3155 ph10 518 string. Nesting only happens one level deep. */
3156    
3157     if (c == 0 && nestptr != NULL)
3158     {
3159     ptr = nestptr;
3160     nestptr = NULL;
3161     c = *ptr;
3162     }
3163    
3164 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
3165     previous cycle of this loop. */
3166    
3167     if (lengthptr != NULL)
3168     {
3169 ph10 475 #ifdef PCRE_DEBUG
3170 nigel 93 if (code > cd->hwm) cd->hwm = code; /* High water info */
3171     #endif
3172 ph10 505 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3173 nigel 93 {
3174     *errorcodeptr = ERR52;
3175     goto FAILED;
3176     }
3177    
3178     /* There is at least one situation where code goes backwards: this is the
3179     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3180     the class is simply eliminated. However, it is created first, so we have to
3181     allow memory for it. Therefore, don't ever reduce the length at this point.
3182     */
3183    
3184     if (code < last_code) code = last_code;
3185 ph10 202
3186     /* Paranoid check for integer overflow */
3187    
3188     if (OFLOW_MAX - *lengthptr < code - last_code)
3189     {
3190     *errorcodeptr = ERR20;
3191     goto FAILED;
3192     }
3193    
3194 ph10 530 *lengthptr += (int)(code - last_code);
3195 nigel 93 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3196    
3197     /* If "previous" is set and it is not at the start of the work space, move
3198     it back to there, in order to avoid filling up the work space. Otherwise,
3199     if "previous" is NULL, reset the current code pointer to the start. */
3200    
3201     if (previous != NULL)
3202     {
3203     if (previous > orig_code)
3204     {
3205     memmove(orig_code, previous, code - previous);
3206     code -= previous - orig_code;
3207     previous = orig_code;
3208     }
3209     }
3210     else code = orig_code;
3211    
3212     /* Remember where this code item starts so we can pick up the length
3213     next time round. */
3214    
3215     last_code = code;
3216     }
3217    
3218     /* In the real compile phase, just check the workspace used by the forward
3219     reference list. */
3220    
3221 ph10 505 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3222 nigel 93 {
3223     *errorcodeptr = ERR52;
3224     goto FAILED;
3225     }
3226    
3227 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
3228    
3229     if (inescq && c != 0)
3230     {
3231 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3232 nigel 77 {
3233     inescq = FALSE;
3234     ptr++;
3235     continue;
3236     }
3237     else
3238     {
3239     if (previous_callout != NULL)
3240     {
3241 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3242     complete_callout(previous_callout, ptr, cd);
3243 nigel 77 previous_callout = NULL;
3244     }
3245     if ((options & PCRE_AUTO_CALLOUT) != 0)
3246     {
3247     previous_callout = code;
3248     code = auto_callout(code, ptr, cd);
3249     }
3250     goto NORMAL_CHAR;
3251     }
3252     }
3253    
3254     /* Fill in length of a previous callout, except when the next thing is
3255     a quantifier. */
3256    
3257 ph10 392 is_quantifier =
3258 ph10 391 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3259     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3260 nigel 77
3261     if (!is_quantifier && previous_callout != NULL &&
3262     after_manual_callout-- <= 0)
3263     {
3264 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3265     complete_callout(previous_callout, ptr, cd);
3266 nigel 77 previous_callout = NULL;
3267     }
3268    
3269 ph10 635 /* In extended mode, skip white space and comments. */
3270 nigel 77
3271     if ((options & PCRE_EXTENDED) != 0)
3272     {
3273     if ((cd->ctypes[c] & ctype_space) != 0) continue;
3274 ph10 391 if (c == CHAR_NUMBER_SIGN)
3275 nigel 77 {
3276 ph10 579 ptr++;
3277 ph10 556 while (*ptr != 0)
3278 nigel 91 {
3279 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3280 ph10 556 ptr++;
3281 ph10 579 #ifdef SUPPORT_UTF8
3282 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3283     #endif
3284 nigel 91 }
3285 nigel 93 if (*ptr != 0) continue;
3286    
3287 nigel 91 /* Else fall through to handle end of string */
3288     c = 0;
3289 nigel 77 }
3290     }
3291    
3292     /* No auto callout for quantifiers. */
3293    
3294     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3295     {
3296     previous_callout = code;
3297     code = auto_callout(code, ptr, cd);
3298     }
3299    
3300     switch(c)
3301     {
3302 nigel 93 /* ===================================================================*/
3303     case 0: /* The branch terminates at string end */
3304 ph10 391 case CHAR_VERTICAL_LINE: /* or | or ) */
3305     case CHAR_RIGHT_PARENTHESIS:
3306 nigel 77 *firstbyteptr = firstbyte;
3307     *reqbyteptr = reqbyte;
3308     *codeptr = code;
3309     *ptrptr = ptr;
3310 nigel 93 if (lengthptr != NULL)
3311     {
3312 ph10 202 if (OFLOW_MAX - *lengthptr < code - last_code)
3313     {
3314     *errorcodeptr = ERR20;
3315     goto FAILED;
3316     }
3317 ph10 530 *lengthptr += (int)(code - last_code); /* To include callout length */
3318 nigel 93 DPRINTF((">> end branch\n"));
3319     }
3320 nigel 77 return TRUE;
3321    
3322 nigel 93
3323     /* ===================================================================*/
3324 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
3325     the setting of any following char as a first character. */
3326    
3327 ph10 391 case CHAR_CIRCUMFLEX_ACCENT:
3328 ph10 602 previous = NULL;
3329 nigel 77 if ((options & PCRE_MULTILINE) != 0)
3330     {
3331     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3332 ph10 602 *code++ = OP_CIRCM;
3333 nigel 77 }
3334 ph10 602 else *code++ = OP_CIRC;
3335 nigel 77 break;
3336    
3337 ph10 391 case CHAR_DOLLAR_SIGN:
3338 nigel 77 previous = NULL;
3339 ph10 602 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3340 nigel 77 break;
3341    
3342     /* There can never be a first char if '.' is first, whatever happens about
3343     repeats. The value of reqbyte doesn't change either. */
3344    
3345 ph10 391 case CHAR_DOT:
3346 nigel 77 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3347     zerofirstbyte = firstbyte;
3348     zeroreqbyte = reqbyte;
3349     previous = code;
3350 ph10 342 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3351 nigel 77 break;
3352    
3353 nigel 93
3354     /* ===================================================================*/
3355 nigel 87 /* Character classes. If the included characters are all < 256, we build a
3356     32-byte bitmap of the permitted characters, except in the special case
3357     where there is only one such character. For negated classes, we build the
3358     map as usual, then invert it at the end. However, we use a different opcode
3359     so that data characters > 255 can be handled correctly.
3360 nigel 77
3361     If the class contains characters outside the 0-255 range, a different
3362     opcode is compiled. It may optionally have a bit map for characters < 256,
3363     but those above are are explicitly listed afterwards. A flag byte tells
3364     whether the bitmap is present, and whether this is a negated class or not.
3365 ph10 345
3366 ph10 336 In JavaScript compatibility mode, an isolated ']' causes an error. In
3367     default (Perl) mode, it is treated as a data character. */
3368 ph10 345
3369 ph10 391 case CHAR_RIGHT_SQUARE_BRACKET:
3370 ph10 336 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3371     {
3372     *errorcodeptr = ERR64;
3373 ph10 345 goto FAILED;
3374 ph10 336 }
3375 ph10 345 goto NORMAL_CHAR;
3376 nigel 77
3377 ph10 391 case CHAR_LEFT_SQUARE_BRACKET:
3378 nigel 77 previous = code;
3379    
3380     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3381     they are encountered at the top level, so we'll do that too. */
3382    
3383 ph10 392 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3384 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) &&
3385 ph10 295 check_posix_syntax(ptr, &tempptr))
3386 nigel 77 {
3387 ph10 391 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3388 nigel 77 goto FAILED;
3389     }
3390    
3391 ph10 205 /* If the first character is '^', set the negation flag and skip it. Also,
3392 ph10 208 if the first few characters (either before or after ^) are \Q\E or \E we
3393 ph10 205 skip them too. This makes for compatibility with Perl. */
3394 ph10 208
3395 ph10 205 negate_class = FALSE;
3396     for (;;)
3397 nigel 77 {
3398     c = *(++ptr);
3399 ph10 391 if (c == CHAR_BACKSLASH)
3400 ph10 205 {
3401 ph10 392 if (ptr[1] == CHAR_E)
3402 ph10 391 ptr++;
3403 ph10 392 else if (strncmp((const char *)ptr+1,
3404     STR_Q STR_BACKSLASH STR_E, 3) == 0)
3405 ph10 391 ptr += 3;
3406 ph10 392 else
3407 ph10 391 break;
3408 ph10 205 }
3409 ph10 391 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3410 ph10 205 negate_class = TRUE;
3411     else break;
3412 ph10 208 }
3413 ph10 345
3414     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3415     an initial ']' is taken as a data character -- the code below handles
3416 ph10 341 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3417     [^] must match any character, so generate OP_ALLANY. */
3418 ph10 345
3419 ph10 392 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3420 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3421 ph10 341 {
3422     *code++ = negate_class? OP_ALLANY : OP_FAIL;
3423     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3424     zerofirstbyte = firstbyte;
3425     break;
3426 ph10 345 }
3427 nigel 77
3428 ph10 286 /* If a class contains a negative special such as \S, we need to flip the
3429     negation flag at the end, so that support for characters > 255 works
3430 ph10 264 correctly (they are all included in the class). */
3431    
3432     should_flip_negation = FALSE;
3433    
3434 nigel 77 /* Keep a count of chars with values < 256 so that we can optimize the case
3435 nigel 93 of just a single character (as long as it's < 256). However, For higher
3436     valued UTF-8 characters, we don't yet do any optimization. */
3437 nigel 77
3438     class_charcount = 0;
3439     class_lastchar = -1;
3440    
3441 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
3442     temporary bit of memory, in case the class contains only 1 character (less
3443     than 256), because in that case the compiled code doesn't use the bit map.
3444     */
3445    
3446     memset(classbits, 0, 32 * sizeof(uschar));
3447    
3448 nigel 77 #ifdef SUPPORT_UTF8
3449     class_utf8 = FALSE; /* No chars >= 256 */
3450 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3451 ph10 309 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3452 nigel 77 #endif
3453    
3454     /* Process characters until ] is reached. By writing this as a "do" it
3455 nigel 93 means that an initial ] is taken as a data character. At the start of the
3456     loop, c contains the first byte of the character. */
3457 nigel 77
3458 nigel 93 if (c != 0) do
3459 nigel 77 {
3460 nigel 93 const uschar *oldptr;
3461    
3462 nigel 77 #ifdef SUPPORT_UTF8
3463     if (utf8 && c > 127)
3464     { /* Braces are required because the */
3465     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3466     }
3467 ph10 535
3468 ph10 300 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3469 ph10 309 data and reset the pointer. This is so that very large classes that
3470 ph10 300 contain a zillion UTF-8 characters no longer overwrite the work space
3471 ph10 309 (which is on the stack). */
3472    
3473 ph10 300 if (lengthptr != NULL)
3474     {
3475     *lengthptr += class_utf8data - class_utf8data_base;
3476 ph10 309 class_utf8data = class_utf8data_base;
3477     }
3478    
3479 nigel 77 #endif
3480    
3481     /* Inside \Q...\E everything is literal except \E */
3482    
3483     if (inescq)
3484     {
3485 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3486 nigel 77 {
3487 nigel 93 inescq = FALSE; /* Reset literal state */
3488     ptr++; /* Skip the 'E' */
3489     continue; /* Carry on with next */
3490 nigel 77 }
3491 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
3492 nigel 77 }
3493    
3494     /* Handle POSIX class names. Perl allows a negation extension of the
3495     form [:^name:]. A square bracket that doesn't match the syntax is
3496     treated as a literal. We also recognize the POSIX constructions
3497     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3498     5.6 and 5.8 do. */
3499    
3500 ph10 391 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3501 ph10 392 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3502 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3503 nigel 77 {
3504     BOOL local_negate = FALSE;
3505 nigel 87 int posix_class, taboffset, tabopt;
3506 nigel 77 register const uschar *cbits = cd->cbits;
3507 nigel 87 uschar pbits[32];
3508 nigel 77
3509 ph10 391 if (ptr[1] != CHAR_COLON)
3510 nigel 77 {
3511     *errorcodeptr = ERR31;
3512     goto FAILED;
3513     }
3514    
3515     ptr += 2;
3516 ph10 391 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3517 nigel 77 {
3518     local_negate = TRUE;
3519 ph10 286 should_flip_negation = TRUE; /* Note negative special */
3520 nigel 77 ptr++;
3521     }
3522    
3523 ph10 530 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3524 nigel 77 if (posix_class < 0)
3525     {
3526     *errorcodeptr = ERR30;
3527     goto FAILED;
3528     }
3529    
3530     /* If matching is caseless, upper and lower are converted to
3531     alpha. This relies on the fact that the class table starts with
3532     alpha, lower, upper as the first 3 entries. */
3533    
3534     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3535     posix_class = 0;
3536 ph10 535
3537     /* When PCRE_UCP is set, some of the POSIX classes are converted to
3538 ph10 518 different escape sequences that use Unicode properties. */
3539 ph10 535
3540 ph10 518 #ifdef SUPPORT_UCP
3541     if ((options & PCRE_UCP) != 0)
3542     {
3543     int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3544     if (posix_substitutes[pc] != NULL)
3545     {
3546 ph10 535 nestptr = tempptr + 1;
3547 ph10 518 ptr = posix_substitutes[pc] - 1;
3548 ph10 535 continue;
3549     }
3550     }
3551     #endif
3552 ph10 518 /* In the non-UCP case, we build the bit map for the POSIX class in a
3553     chunk of local store because we may be adding and subtracting from it,
3554     and we don't want to subtract bits that may be in the main map already.
3555     At the end we or the result into the bit map that is being built. */
3556 nigel 77
3557     posix_class *= 3;
3558 nigel 87
3559     /* Copy in the first table (always present) */
3560    
3561     memcpy(pbits, cbits + posix_class_maps[posix_class],
3562     32 * sizeof(uschar));
3563    
3564     /* If there is a second table, add or remove it as required. */
3565    
3566     taboffset = posix_class_maps[posix_class + 1];
3567     tabopt = posix_class_maps[posix_class + 2];
3568    
3569     if (taboffset >= 0)
3570 nigel 77 {
3571 nigel 87 if (tabopt >= 0)
3572     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3573 nigel 77 else
3574 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3575 nigel 77 }
3576    
3577 nigel 87 /* Not see if we need to remove any special characters. An option
3578     value of 1 removes vertical space and 2 removes underscore. */
3579    
3580     if (tabopt < 0) tabopt = -tabopt;
3581     if (tabopt == 1) pbits[1] &= ~0x3c;
3582     else if (tabopt == 2) pbits[11] &= 0x7f;
3583    
3584     /* Add the POSIX table or its complement into the main table that is
3585     being built and we are done. */
3586    
3587     if (local_negate)
3588     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3589     else
3590     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3591    
3592 nigel 77 ptr = tempptr + 1;
3593     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3594     continue; /* End of POSIX syntax handling */
3595     }
3596    
3597     /* Backslash may introduce a single character, or it may introduce one
3598 nigel 93 of the specials, which just set a flag. The sequence \b is a special
3599 ph10 513 case. Inside a class (and only there) it is treated as backspace. We
3600     assume that other escapes have more than one character in them, so set
3601     class_charcount bigger than one. Unrecognized escapes fall through and
3602     are either treated as literal characters (by default), or are faulted if
3603     PCRE_EXTRA is set. */
3604 nigel 77
3605 ph10 391 if (c == CHAR_BACKSLASH)
3606 nigel 77 {
3607 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3608     if (*errorcodeptr != 0) goto FAILED;
3609 nigel 77
3610 ph10 513 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3611 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
3612     {
3613 ph10 391 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3614 nigel 77 {
3615     ptr += 2; /* avoid empty string */
3616     }
3617     else inescq = TRUE;
3618     continue;
3619     }
3620 ph10 220 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3621 nigel 77
3622     if (c < 0)
3623     {
3624     register const uschar *cbits = cd->cbits;
3625     class_charcount += 2; /* Greater than 1 is what matters */
3626 nigel 93
3627 ph10 518 switch (-c)
3628 nigel 77 {
3629 ph10 518 #ifdef SUPPORT_UCP
3630     case ESC_du: /* These are the values given for \d etc */
3631     case ESC_DU: /* when PCRE_UCP is set. We replace the */
3632     case ESC_wu: /* escape sequence with an appropriate \p */
3633     case ESC_WU: /* or \P to test Unicode properties instead */
3634     case ESC_su: /* of the default ASCII testing. */
3635     case ESC_SU:
3636     nestptr = ptr;
3637     ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3638 ph10 535 class_charcount -= 2; /* Undo! */
3639 ph10 518 continue;
3640     #endif
3641 nigel 77 case ESC_d:
3642     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3643     continue;
3644    
3645     case ESC_D:
3646 ph10 286 should_flip_negation = TRUE;
3647 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3648     continue;
3649    
3650     case ESC_w:
3651     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3652     continue;
3653    
3654     case ESC_W:
3655 ph10 286 should_flip_negation = TRUE;
3656 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3657     continue;
3658    
3659 ph10 552 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3660 ph10 579 if it was previously set by something earlier in the character
3661     class. */
3662 ph10 552
3663 nigel 77 case ESC_s:
3664 ph10 552 classbits[0] |= cbits[cbit_space];
3665 ph10 579 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3666 ph10 552 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3667 nigel 77 continue;
3668    
3669     case ESC_S:
3670 ph10 286 should_flip_negation = TRUE;
3671 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3672     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3673     continue;
3674    
3675 ph10 518 case ESC_h:
3676 ph10 178 SETBIT(classbits, 0x09); /* VT */
3677     SETBIT(classbits, 0x20); /* SPACE */
3678 ph10 180 SETBIT(classbits, 0xa0); /* NSBP */
3679 ph10 178 #ifdef SUPPORT_UTF8
3680     if (utf8)
3681 ph10 180 {
3682 ph10 178 class_utf8 = TRUE;
3683     *class_utf8data++ = XCL_SINGLE;
3684 ph10 180 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3685 ph10 178 *class_utf8data++ = XCL_SINGLE;
3686 ph10 180 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3687     *class_utf8data++ = XCL_RANGE;
3688     class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3689     class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3690 ph10 178 *class_utf8data++ = XCL_SINGLE;
3691 ph10 180 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3692 ph10 178 *class_utf8data++ = XCL_SINGLE;
3693 ph10 180 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3694 ph10 178 *class_utf8data++ = XCL_SINGLE;
3695 ph10 180 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3696     }
3697     #endif
3698     continue;
3699 nigel 93
3700 ph10 518 case ESC_H:
3701 ph10 178 for (c = 0; c < 32; c++)
3702     {
3703     int x = 0xff;
3704     switch (c)
3705 ph10 180 {
3706 ph10 178 case 0x09/8: x ^= 1 << (0x09%8); break;
3707     case 0x20/8: x ^= 1 << (0x20%8); break;
3708     case 0xa0/8: x ^= 1 << (0xa0%8); break;
3709     default: break;
3710     }
3711     classbits[c] |= x;
3712 ph10 180 }
3713    
3714 ph10 178 #ifdef SUPPORT_UTF8
3715     if (utf8)
3716 ph10 180 {
3717 ph10 178 class_utf8 = TRUE;
3718 ph10 180 *class_utf8data++ = XCL_RANGE;
3719     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3720     class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3721     *class_utf8data++ = XCL_RANGE;
3722     class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3723     class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3724     *class_utf8data++ = XCL_RANGE;
3725     class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3726     class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3727     *class_utf8data++ = XCL_RANGE;
3728     class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3729     class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3730     *class_utf8data++ = XCL_RANGE;
3731     class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3732     class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3733     *class_utf8data++ = XCL_RANGE;
3734     class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3735     class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3736     *class_utf8data++ = XCL_RANGE;
3737     class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3738     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3739     }
3740     #endif
3741     continue;
3742 ph10 178
3743 ph10 518 case ESC_v:
3744 ph10 178 SETBIT(classbits, 0x0a); /* LF */
3745     SETBIT(classbits, 0x0b); /* VT */
3746 ph10 180 SETBIT(classbits, 0x0c); /* FF */
3747     SETBIT(classbits, 0x0d); /* CR */
3748     SETBIT(classbits, 0x85); /* NEL */
3749 ph10 178 #ifdef SUPPORT_UTF8
3750     if (utf8)
3751 ph10 180 {
3752 ph10 178 class_utf8 = TRUE;
3753 ph10 180 *class_utf8data++ = XCL_RANGE;
3754     class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3755     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3756     }
3757     #endif
3758     continue;
3759 ph10 178
3760 ph10 518 case ESC_V:
3761 ph10 178 for (c = 0; c < 32; c++)
3762     {
3763     int x = 0xff;
3764     switch (c)
3765 ph10 180 {
3766 ph10