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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 598 - (hide annotations) (download)
Sat May 7 15:37:31 2011 UTC (3 years, 5 months ago) by ph10
File MIME type: text/plain
File size: 237170 byte(s)
Pass back detailed info when UTF-8 check fails at runtime.

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 240 "repeating a DEFINE group is not allowed\0"
397     "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 510 ;
413 nigel 77
414     /* Table to identify digits and hex digits. This is used when compiling
415     patterns. Note that the tables in chartables are dependent on the locale, and
416     may mark arbitrary characters as digits - but the PCRE compiling code expects
417     to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
418     a private table here. It costs 256 bytes, but it is a lot faster than doing
419     character value tests (at least in some simple cases I timed), and in some
420     applications one wants PCRE to compile efficiently as well as match
421     efficiently.
422    
423     For convenience, we use the same bit definitions as in chartables:
424    
425     0x04 decimal digit
426     0x08 hexadecimal digit
427    
428     Then we can use ctype_digit and ctype_xdigit in the code. */
429    
430 ph10 392 #ifndef EBCDIC
431 ph10 391
432 ph10 392 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
433 ph10 391 UTF-8 mode. */
434    
435 nigel 77 static const unsigned char digitab[] =
436     {
437     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
438     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
439     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
440     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
441     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
442     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
443     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
444     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
445     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
446     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
447     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
448     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
449     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
450     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
451     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
452     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
453     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
454     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
455     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
456     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
457     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
458     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
459     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
460     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
461     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
462     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
463     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
464     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
465     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
466     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
467     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
468     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
469    
470 ph10 392 #else
471 ph10 391
472     /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
473    
474 nigel 77 static const unsigned char digitab[] =
475     {
476     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
477     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
478     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
479     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
480     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
481     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
482     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
483     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
484     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
485     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
486     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
487 ph10 97 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
488 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
489     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
490     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
491     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
492     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
493     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
494     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
495     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
496     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
497     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
498     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
499     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
500     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
501     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
502     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
503     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
504     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
505     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
506     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
507     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
508    
509     static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
510     0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
511     0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
512     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
513     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
514     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
515     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
516     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
517     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
518     0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
519     0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
520     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
521 ph10 97 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
522 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
523     0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
524     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
525     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
526     0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
527     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
528     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
529     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
530     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
531     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
532     0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
533     0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
534     0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
535     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
536     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
537     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
538     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
539     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
540     0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
541     0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
542     #endif
543    
544    
545     /* Definition to allow mutual recursion */
546    
547     static BOOL
548 ph10 180 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
549 ph10 175 int *, int *, branch_chain *, compile_data *, int *);
550 nigel 77
551    
552    
553     /*************************************************
554 ph10 240 * Find an error text *
555     *************************************************/
556    
557 ph10 243 /* The error texts are now all in one long string, to save on relocations. As
558     some of the text is of unknown length, we can't use a table of offsets.
559     Instead, just count through the strings. This is not a performance issue
560 ph10 240 because it happens only when there has been a compilation error.
561    
562     Argument: the error number
563     Returns: pointer to the error string
564     */
565    
566     static const char *
567     find_error_text(int n)
568     {
569     const char *s = error_texts;
570 ph10 507 for (; n > 0; n--)
571 ph10 499 {
572     while (*s++ != 0) {};
573     if (*s == 0) return "Error text not found (please report)";
574 ph10 507 }
575 ph10 240 return s;
576     }
577    
578    
579     /*************************************************
580 nigel 77 * Handle escapes *
581     *************************************************/
582    
583     /* This function is called when a \ has been encountered. It either returns a
584     positive value for a simple escape such as \n, or a negative value which
585 nigel 93 encodes one of the more complicated things such as \d. A backreference to group
586     n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
587     UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
588     ptr is pointing at the \. On exit, it is on the final character of the escape
589     sequence.
590 nigel 77
591     Arguments:
592     ptrptr points to the pattern position pointer
593     errorcodeptr points to the errorcode variable
594     bracount number of previous extracting brackets
595     options the options bits
596     isclass TRUE if inside a character class
597    
598     Returns: zero or positive => a data character
599     negative => a special escape sequence
600 ph10 213 on error, errorcodeptr is set
601 nigel 77 */
602    
603     static int
604     check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
605     int options, BOOL isclass)
606     {
607 nigel 87 BOOL utf8 = (options & PCRE_UTF8) != 0;
608     const uschar *ptr = *ptrptr + 1;
609 nigel 77 int c, i;
610    
611 nigel 87 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
612     ptr--; /* Set pointer back to the last byte */
613    
614 nigel 77 /* If backslash is at the end of the pattern, it's an error. */
615    
616     if (c == 0) *errorcodeptr = ERR1;
617    
618 ph10 274 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
619     in a table. A non-zero result is something that can be returned immediately.
620 nigel 77 Otherwise further processing may be required. */
621    
622 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
623     else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
624     else if ((i = escapes[c - CHAR_0]) != 0) c = i;
625 nigel 77
626 ph10 97 #else /* EBCDIC coding */
627 ph10 274 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
628 nigel 77 else if ((i = escapes[c - 0x48]) != 0) c = i;
629     #endif
630    
631     /* Escapes that need further processing, or are illegal. */
632    
633     else
634     {
635     const uschar *oldptr;
636 nigel 93 BOOL braced, negated;
637    
638 nigel 77 switch (c)
639     {
640     /* A number of Perl escapes are not handled by PCRE. We give an explicit
641     error. */
642    
643 ph10 391 case CHAR_l:
644     case CHAR_L:
645     case CHAR_u:
646     case CHAR_U:
647 nigel 77 *errorcodeptr = ERR37;
648     break;
649    
650 ph10 333 /* \g must be followed by one of a number of specific things:
651 ph10 345
652 ph10 333 (1) A number, either plain or braced. If positive, it is an absolute
653     backreference. If negative, it is a relative backreference. This is a Perl
654     5.10 feature.
655 ph10 345
656 ph10 333 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
657     is part of Perl's movement towards a unified syntax for back references. As
658     this is synonymous with \k{name}, we fudge it up by pretending it really
659     was \k.
660 ph10 345
661     (3) For Oniguruma compatibility we also support \g followed by a name or a
662     number either in angle brackets or in single quotes. However, these are
663     (possibly recursive) subroutine calls, _not_ backreferences. Just return
664 ph10 333 the -ESC_g code (cf \k). */
665 nigel 93
666 ph10 391 case CHAR_g:
667     if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
668 ph10 333 {
669     c = -ESC_g;
670 ph10 345 break;
671     }
672 ph10 333
673     /* Handle the Perl-compatible cases */
674 ph10 345
675 ph10 391 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
676 nigel 93 {
677 ph10 171 const uschar *p;
678 ph10 391 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
679     if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
680     if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
681 ph10 171 {
682     c = -ESC_k;
683     break;
684 ph10 172 }
685 nigel 93 braced = TRUE;
686     ptr++;
687     }
688     else braced = FALSE;
689    
690 ph10 391 if (ptr[1] == CHAR_MINUS)
691 nigel 93 {
692     negated = TRUE;
693     ptr++;
694     }
695     else negated = FALSE;
696    
697     c = 0;
698     while ((digitab[ptr[1]] & ctype_digit) != 0)
699 ph10 391 c = c * 10 + *(++ptr) - CHAR_0;
700 ph10 220
701 ph10 333 if (c < 0) /* Integer overflow */
702 ph10 213 {
703     *errorcodeptr = ERR61;
704     break;
705 ph10 220 }
706 ph10 345
707 ph10 391 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
708 nigel 93 {
709     *errorcodeptr = ERR57;
710 ph10 213 break;
711 nigel 93 }
712 ph10 345
713 ph10 333 if (c == 0)
714     {
715     *errorcodeptr = ERR58;
716     break;
717 ph10 345 }
718 nigel 93
719     if (negated)
720     {
721     if (c > bracount)
722     {
723     *errorcodeptr = ERR15;
724 ph10 213 break;
725 nigel 93 }
726     c = bracount - (c - 1);
727     }
728    
729     c = -(ESC_REF + c);
730     break;
731    
732 nigel 77 /* The handling of escape sequences consisting of a string of digits
733     starting with one that is not zero is not straightforward. By experiment,
734     the way Perl works seems to be as follows:
735    
736     Outside a character class, the digits are read as a decimal number. If the
737     number is less than 10, or if there are that many previous extracting
738     left brackets, then it is a back reference. Otherwise, up to three octal
739     digits are read to form an escaped byte. Thus \123 is likely to be octal
740     123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
741     value is greater than 377, the least significant 8 bits are taken. Inside a
742     character class, \ followed by a digit is always an octal number. */
743    
744 ph10 391 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
745     case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
746 nigel 77
747     if (!isclass)
748     {
749     oldptr = ptr;
750 ph10 391 c -= CHAR_0;
751 nigel 77 while ((digitab[ptr[1]] & ctype_digit) != 0)
752 ph10 391 c = c * 10 + *(++ptr) - CHAR_0;
753 ph10 333 if (c < 0) /* Integer overflow */
754 ph10 213 {
755     *errorcodeptr = ERR61;
756 ph10 220 break;
757     }
758 nigel 77 if (c < 10 || c <= bracount)
759     {
760     c = -(ESC_REF + c);
761     break;
762     }
763     ptr = oldptr; /* Put the pointer back and fall through */
764     }
765    
766     /* Handle an octal number following \. If the first digit is 8 or 9, Perl
767     generates a binary zero byte and treats the digit as a following literal.
768     Thus we have to pull back the pointer by one. */
769    
770 ph10 391 if ((c = *ptr) >= CHAR_8)
771 nigel 77 {
772     ptr--;
773     c = 0;
774     break;
775     }
776    
777     /* \0 always starts an octal number, but we may drop through to here with a
778 nigel 91 larger first octal digit. The original code used just to take the least
779     significant 8 bits of octal numbers (I think this is what early Perls used
780     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
781     than 3 octal digits. */
782 nigel 77
783 ph10 391 case CHAR_0:
784     c -= CHAR_0;
785     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
786     c = c * 8 + *(++ptr) - CHAR_0;
787 nigel 91 if (!utf8 && c > 255) *errorcodeptr = ERR51;
788 nigel 77 break;
789    
790 nigel 87 /* \x is complicated. \x{ddd} is a character number which can be greater
791     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
792     treated as a data character. */
793 nigel 77
794 ph10 391 case CHAR_x:
795     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
796 nigel 77 {
797     const uschar *pt = ptr + 2;
798 nigel 87 int count = 0;
799    
800 nigel 77 c = 0;
801     while ((digitab[*pt] & ctype_xdigit) != 0)
802     {
803 nigel 87 register int cc = *pt++;
804 ph10 391 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
805 nigel 77 count++;
806 nigel 87
807 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
808     if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
809     c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
810 ph10 97 #else /* EBCDIC coding */
811 ph10 391 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
812     c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
813 nigel 77 #endif
814     }
815 nigel 87
816 ph10 391 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
817 nigel 77 {
818 nigel 87 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
819 nigel 77 ptr = pt;
820     break;
821     }
822 nigel 87
823 nigel 77 /* If the sequence of hex digits does not end with '}', then we don't
824     recognize this construct; fall through to the normal \x handling. */
825     }
826    
827 nigel 87 /* Read just a single-byte hex-defined char */
828 nigel 77
829     c = 0;
830     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
831     {
832 ph10 391 int cc; /* Some compilers don't like */
833     cc = *(++ptr); /* ++ in initializers */
834     #ifndef EBCDIC /* ASCII/UTF-8 coding */
835     if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
836     c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
837 ph10 97 #else /* EBCDIC coding */
838 ph10 391 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
839     c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
840 nigel 77 #endif
841     }
842     break;
843    
844 nigel 93 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
845 ph10 574 An error is given if the byte following \c is not an ASCII character. This
846     coding is ASCII-specific, but then the whole concept of \cx is
847 nigel 93 ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
848 nigel 77
849 ph10 391 case CHAR_c:
850 nigel 77 c = *(++ptr);
851     if (c == 0)
852     {
853     *errorcodeptr = ERR2;
854 ph10 213 break;
855 nigel 77 }
856 ph10 574 #ifndef EBCDIC /* ASCII/UTF-8 coding */
857     if (c > 127) /* Excludes all non-ASCII in either mode */
858     {
859     *errorcodeptr = ERR68;
860 ph10 579 break;
861     }
862 ph10 391 if (c >= CHAR_a && c <= CHAR_z) c -= 32;
863 nigel 77 c ^= 0x40;
864 ph10 574 #else /* EBCDIC coding */
865 ph10 391 if (c >= CHAR_a && c <= CHAR_z) c += 64;
866 nigel 77 c ^= 0xC0;
867     #endif
868     break;
869    
870     /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
871 ph10 274 other alphanumeric following \ is an error if PCRE_EXTRA was set;
872     otherwise, for Perl compatibility, it is a literal. This code looks a bit
873     odd, but there used to be some cases other than the default, and there may
874     be again in future, so I haven't "optimized" it. */
875 nigel 77
876     default:
877     if ((options & PCRE_EXTRA) != 0) switch(c)
878     {
879     default:
880     *errorcodeptr = ERR3;
881     break;
882     }
883     break;
884     }
885     }
886 ph10 518
887     /* Perl supports \N{name} for character names, as well as plain \N for "not
888 ph10 514 newline". PCRE does not support \N{name}. */
889 nigel 77
890 ph10 514 if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
891 ph10 518 *errorcodeptr = ERR37;
892 ph10 514
893 ph10 518 /* If PCRE_UCP is set, we change the values for \d etc. */
894    
895     if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
896     c -= (ESC_DU - ESC_D);
897    
898     /* Set the pointer to the final character before returning. */
899    
900 nigel 77 *ptrptr = ptr;
901     return c;
902     }
903    
904    
905    
906     #ifdef SUPPORT_UCP
907     /*************************************************
908     * Handle \P and \p *
909     *************************************************/
910    
911     /* This function is called after \P or \p has been encountered, provided that
912     PCRE is compiled with support for Unicode properties. On entry, ptrptr is
913     pointing at the P or p. On exit, it is pointing at the final character of the
914     escape sequence.
915    
916     Argument:
917     ptrptr points to the pattern position pointer
918     negptr points to a boolean that is set TRUE for negation else FALSE
919 nigel 87 dptr points to an int that is set to the detailed property value
920 nigel 77 errorcodeptr points to the error code variable
921    
922 nigel 87 Returns: type value from ucp_type_table, or -1 for an invalid type
923 nigel 77 */
924    
925     static int
926 nigel 87 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
927 nigel 77 {
928     int c, i, bot, top;
929     const uschar *ptr = *ptrptr;
930 nigel 87 char name[32];
931 nigel 77
932     c = *(++ptr);
933     if (c == 0) goto ERROR_RETURN;
934    
935     *negptr = FALSE;
936    
937 nigel 87 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
938     negation. */
939 nigel 77
940 ph10 391 if (c == CHAR_LEFT_CURLY_BRACKET)
941 nigel 77 {
942 ph10 391 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
943 nigel 77 {
944     *negptr = TRUE;
945     ptr++;
946     }
947 ph10 199 for (i = 0; i < (int)sizeof(name) - 1; i++)
948 nigel 77 {
949     c = *(++ptr);
950     if (c == 0) goto ERROR_RETURN;
951 ph10 391 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
952 nigel 77 name[i] = c;
953     }
954 ph10 391 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
955 nigel 77 name[i] = 0;
956     }
957    
958     /* Otherwise there is just one following character */
959    
960     else
961     {
962     name[0] = c;
963     name[1] = 0;
964     }
965    
966     *ptrptr = ptr;
967    
968     /* Search for a recognized property name using binary chop */
969    
970     bot = 0;
971     top = _pcre_utt_size;
972    
973     while (bot < top)
974     {
975 nigel 87 i = (bot + top) >> 1;
976 ph10 240 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
977 nigel 87 if (c == 0)
978     {
979     *dptr = _pcre_utt[i].value;
980     return _pcre_utt[i].type;
981     }
982 nigel 77 if (c > 0) bot = i + 1; else top = i;
983     }
984    
985     *errorcodeptr = ERR47;
986     *ptrptr = ptr;
987     return -1;
988    
989     ERROR_RETURN:
990     *errorcodeptr = ERR46;
991     *ptrptr = ptr;
992     return -1;
993     }
994     #endif
995    
996    
997    
998    
999     /*************************************************
1000     * Check for counted repeat *
1001     *************************************************/
1002    
1003     /* This function is called when a '{' is encountered in a place where it might
1004     start a quantifier. It looks ahead to see if it really is a quantifier or not.
1005     It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1006     where the ddds are digits.
1007    
1008     Arguments:
1009     p pointer to the first char after '{'
1010    
1011     Returns: TRUE or FALSE
1012     */
1013    
1014     static BOOL
1015     is_counted_repeat(const uschar *p)
1016     {
1017     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1018     while ((digitab[*p] & ctype_digit) != 0) p++;
1019 ph10 391 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1020 nigel 77
1021 ph10 391 if (*p++ != CHAR_COMMA) return FALSE;
1022     if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1023 nigel 77
1024     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1025     while ((digitab[*p] & ctype_digit) != 0) p++;
1026    
1027 ph10 391 return (*p == CHAR_RIGHT_CURLY_BRACKET);
1028 nigel 77 }
1029    
1030    
1031    
1032     /*************************************************
1033     * Read repeat counts *
1034     *************************************************/
1035    
1036     /* Read an item of the form {n,m} and return the values. This is called only
1037     after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1038     so the syntax is guaranteed to be correct, but we need to check the values.
1039    
1040     Arguments:
1041     p pointer to first char after '{'
1042     minp pointer to int for min
1043     maxp pointer to int for max
1044     returned as -1 if no max
1045     errorcodeptr points to error code variable
1046    
1047     Returns: pointer to '}' on success;
1048     current ptr on error, with errorcodeptr set non-zero
1049     */
1050    
1051     static const uschar *
1052     read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1053     {
1054     int min = 0;
1055     int max = -1;
1056    
1057 nigel 81 /* Read the minimum value and do a paranoid check: a negative value indicates
1058     an integer overflow. */
1059    
1060 ph10 391 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1061 nigel 81 if (min < 0 || min > 65535)
1062     {
1063     *errorcodeptr = ERR5;
1064     return p;
1065     }
1066 nigel 77
1067 nigel 81 /* Read the maximum value if there is one, and again do a paranoid on its size.
1068     Also, max must not be less than min. */
1069    
1070 ph10 391 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1071 nigel 77 {
1072 ph10 391 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1073 nigel 77 {
1074     max = 0;
1075 ph10 391 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1076 nigel 81 if (max < 0 || max > 65535)
1077     {
1078     *errorcodeptr = ERR5;
1079     return p;
1080     }
1081 nigel 77 if (max < min)
1082     {
1083     *errorcodeptr = ERR4;
1084     return p;
1085     }
1086     }
1087     }
1088    
1089 nigel 81 /* Fill in the required variables, and pass back the pointer to the terminating
1090     '}'. */
1091 nigel 77
1092 nigel 81 *minp = min;
1093     *maxp = max;
1094 nigel 77 return p;
1095     }
1096    
1097    
1098    
1099     /*************************************************
1100 ph10 408 * Subroutine for finding forward reference *
1101 nigel 91 *************************************************/
1102    
1103 ph10 408 /* This recursive function is called only from find_parens() below. The
1104     top-level call starts at the beginning of the pattern. All other calls must
1105     start at a parenthesis. It scans along a pattern's text looking for capturing
1106 nigel 93 subpatterns, and counting them. If it finds a named pattern that matches the
1107     name it is given, it returns its number. Alternatively, if the name is NULL, it
1108 ph10 578 returns when it reaches a given numbered subpattern. Recursion is used to keep
1109     track of subpatterns that reset the capturing group numbers - the (?| feature.
1110 nigel 91
1111 ph10 578 This function was originally called only from the second pass, in which we know
1112     that if (?< or (?' or (?P< is encountered, the name will be correctly
1113     terminated because that is checked in the first pass. There is now one call to
1114     this function in the first pass, to check for a recursive back reference by
1115     name (so that we can make the whole group atomic). In this case, we need check
1116 ph10 579 only up to the current position in the pattern, and that is still OK because
1117     and previous occurrences will have been checked. To make this work, the test
1118     for "end of pattern" is a check against cd->end_pattern in the main loop,
1119 ph10 578 instead of looking for a binary zero. This means that the special first-pass
1120 ph10 579 call can adjust cd->end_pattern temporarily. (Checks for binary zero while
1121     processing items within the loop are OK, because afterwards the main loop will
1122 ph10 578 terminate.)
1123    
1124 nigel 91 Arguments:
1125 ph10 408 ptrptr address of the current character pointer (updated)
1126 ph10 345 cd compile background data
1127 nigel 93 name name to seek, or NULL if seeking a numbered subpattern
1128     lorn name length, or subpattern number if name is NULL
1129     xmode TRUE if we are in /x mode
1130 ph10 579 utf8 TRUE if we are in UTF-8 mode
1131 ph10 411 count pointer to the current capturing subpattern number (updated)
1132 nigel 91
1133     Returns: the number of the named subpattern, or -1 if not found
1134     */
1135    
1136     static int
1137 ph10 408 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1138 ph10 556 BOOL xmode, BOOL utf8, int *count)
1139 nigel 91 {
1140 ph10 408 uschar *ptr = *ptrptr;
1141     int start_count = *count;
1142     int hwm_count = start_count;
1143     BOOL dup_parens = FALSE;
1144 nigel 93
1145 ph10 411 /* If the first character is a parenthesis, check on the type of group we are
1146 ph10 408 dealing with. The very first call may not start with a parenthesis. */
1147    
1148     if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1149     {
1150 ph10 544 /* Handle specials such as (*SKIP) or (*UTF8) etc. */
1151 ph10 545
1152 ph10 544 if (ptr[1] == CHAR_ASTERISK) ptr += 2;
1153 ph10 545
1154 ph10 544 /* Handle a normal, unnamed capturing parenthesis. */
1155 ph10 408
1156 ph10 544 else if (ptr[1] != CHAR_QUESTION_MARK)
1157 ph10 408 {
1158     *count += 1;
1159     if (name == NULL && *count == lorn) return *count;
1160 ph10 411 ptr++;
1161 ph10 408 }
1162    
1163 ph10 544 /* All cases now have (? at the start. Remember when we are in a group
1164     where the parenthesis numbers are duplicated. */
1165    
1166     else if (ptr[2] == CHAR_VERTICAL_LINE)
1167     {
1168     ptr += 3;
1169     dup_parens = TRUE;
1170     }
1171 ph10 545
1172 ph10 544 /* Handle comments; all characters are allowed until a ket is reached. */
1173    
1174     else if (ptr[2] == CHAR_NUMBER_SIGN)
1175     {
1176     for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
1177     goto FAIL_EXIT;
1178 ph10 545 }
1179 ph10 544
1180 ph10 408 /* Handle a condition. If it is an assertion, just carry on so that it
1181     is processed as normal. If not, skip to the closing parenthesis of the
1182 ph10 544 condition (there can't be any nested parens). */
1183 ph10 411
1184 ph10 408 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1185     {
1186 ph10 411 ptr += 2;
1187 ph10 408 if (ptr[1] != CHAR_QUESTION_MARK)
1188     {
1189     while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1190 ph10 411 if (*ptr != 0) ptr++;
1191 ph10 408 }
1192 ph10 411 }
1193    
1194 ph10 544 /* Start with (? but not a condition. */
1195 ph10 408
1196     else
1197 ph10 411 {
1198 ph10 408 ptr += 2;
1199     if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1200    
1201     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1202 ph10 411
1203 ph10 408 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1204     ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1205     {
1206     int term;
1207     const uschar *thisname;
1208     *count += 1;
1209     if (name == NULL && *count == lorn) return *count;
1210     term = *ptr++;
1211     if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1212     thisname = ptr;
1213     while (*ptr != term) ptr++;
1214     if (name != NULL && lorn == ptr - thisname &&
1215     strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1216     return *count;
1217 ph10 461 term++;
1218 ph10 411 }
1219 ph10 408 }
1220 ph10 411 }
1221 ph10 408
1222 ph10 411 /* Past any initial parenthesis handling, scan for parentheses or vertical
1223 ph10 579 bars. Stop if we get to cd->end_pattern. Note that this is important for the
1224     first-pass call when this value is temporarily adjusted to stop at the current
1225 ph10 578 position. So DO NOT change this to a test for binary zero. */
1226 ph10 408
1227 ph10 578 for (; ptr < cd->end_pattern; ptr++)
1228 nigel 91 {
1229 nigel 93 /* Skip over backslashed characters and also entire \Q...\E */
1230    
1231 ph10 391 if (*ptr == CHAR_BACKSLASH)
1232 nigel 93 {
1233 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1234 ph10 391 if (*ptr == CHAR_Q) for (;;)
1235 nigel 93 {
1236 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1237 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1238 ph10 391 if (*(++ptr) == CHAR_E) break;
1239 nigel 93 }
1240     continue;
1241     }
1242    
1243 ph10 340 /* Skip over character classes; this logic must be similar to the way they
1244     are handled for real. If the first character is '^', skip it. Also, if the
1245     first few characters (either before or after ^) are \Q\E or \E we skip them
1246 ph10 392 too. This makes for compatibility with Perl. Note the use of STR macros to
1247 ph10 391 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1248 nigel 93
1249 ph10 391 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1250 nigel 93 {
1251 ph10 340 BOOL negate_class = FALSE;
1252     for (;;)
1253     {
1254 ph10 438 if (ptr[1] == CHAR_BACKSLASH)
1255 ph10 340 {
1256 ph10 438 if (ptr[2] == CHAR_E)
1257     ptr+= 2;
1258     else if (strncmp((const char *)ptr+2,
1259 ph10 392 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1260 ph10 438 ptr += 4;
1261 ph10 392 else
1262 ph10 391 break;
1263 ph10 340 }
1264 ph10 438 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1265 ph10 461 {
1266 ph10 340 negate_class = TRUE;
1267 ph10 438 ptr++;
1268 ph10 461 }
1269 ph10 340 else break;
1270     }
1271    
1272     /* If the next character is ']', it is a data character that must be
1273 ph10 341 skipped, except in JavaScript compatibility mode. */
1274 ph10 345
1275 ph10 392 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1276 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1277 ph10 345 ptr++;
1278    
1279 ph10 391 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1280 nigel 93 {
1281 ph10 220 if (*ptr == 0) return -1;
1282 ph10 391 if (*ptr == CHAR_BACKSLASH)
1283 nigel 93 {
1284 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1285 ph10 391 if (*ptr == CHAR_Q) for (;;)
1286 nigel 93 {
1287 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1288 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1289 ph10 391 if (*(++ptr) == CHAR_E) break;
1290 nigel 93 }
1291     continue;
1292     }
1293     }
1294     continue;
1295     }
1296    
1297     /* Skip comments in /x mode */
1298    
1299 ph10 391 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1300 nigel 93 {
1301 ph10 579 ptr++;
1302 ph10 556 while (*ptr != 0)
1303     {
1304     if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
1305     ptr++;
1306 ph10 579 #ifdef SUPPORT_UTF8
1307 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
1308     #endif
1309     }
1310 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1311 nigel 93 continue;
1312     }
1313    
1314 ph10 408 /* Check for the special metacharacters */
1315 ph10 411
1316 ph10 408 if (*ptr == CHAR_LEFT_PARENTHESIS)
1317 nigel 93 {
1318 ph10 556 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count);
1319 ph10 408 if (rc > 0) return rc;
1320     if (*ptr == 0) goto FAIL_EXIT;
1321 nigel 93 }
1322 ph10 411
1323 ph10 408 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1324     {
1325     if (dup_parens && *count < hwm_count) *count = hwm_count;
1326 ph10 545 goto FAIL_EXIT;
1327 ph10 408 }
1328 ph10 411
1329     else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1330 ph10 408 {
1331     if (*count > hwm_count) hwm_count = *count;
1332     *count = start_count;
1333 ph10 411 }
1334 ph10 408 }
1335 nigel 93
1336 ph10 408 FAIL_EXIT:
1337     *ptrptr = ptr;
1338     return -1;
1339     }
1340 nigel 93
1341    
1342    
1343    
1344 ph10 408 /*************************************************
1345     * Find forward referenced subpattern *
1346     *************************************************/
1347 nigel 93
1348 ph10 408 /* This function scans along a pattern's text looking for capturing
1349     subpatterns, and counting them. If it finds a named pattern that matches the
1350     name it is given, it returns its number. Alternatively, if the name is NULL, it
1351     returns when it reaches a given numbered subpattern. This is used for forward
1352     references to subpatterns. We used to be able to start this scan from the
1353     current compiling point, using the current count value from cd->bracount, and
1354     do it all in a single loop, but the addition of the possibility of duplicate
1355     subpattern numbers means that we have to scan from the very start, in order to
1356     take account of such duplicates, and to use a recursive function to keep track
1357     of the different types of group.
1358    
1359     Arguments:
1360     cd compile background data
1361     name name to seek, or NULL if seeking a numbered subpattern
1362     lorn name length, or subpattern number if name is NULL
1363     xmode TRUE if we are in /x mode
1364 ph10 579 utf8 TRUE if we are in UTF-8 mode
1365 ph10 408
1366     Returns: the number of the found subpattern, or -1 if not found
1367     */
1368    
1369     static int
1370 ph10 556 find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode,
1371     BOOL utf8)
1372 ph10 408 {
1373     uschar *ptr = (uschar *)cd->start_pattern;
1374     int count = 0;
1375     int rc;
1376    
1377     /* If the pattern does not start with an opening parenthesis, the first call
1378     to find_parens_sub() will scan right to the end (if necessary). However, if it
1379     does start with a parenthesis, find_parens_sub() will return when it hits the
1380     matching closing parens. That is why we have to have a loop. */
1381    
1382 ph10 411 for (;;)
1383     {
1384 ph10 556 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count);
1385 ph10 411 if (rc > 0 || *ptr++ == 0) break;
1386     }
1387    
1388 ph10 408 return rc;
1389 nigel 91 }
1390    
1391    
1392    
1393 ph10 408
1394 nigel 91 /*************************************************
1395 nigel 77 * Find first significant op code *
1396     *************************************************/
1397    
1398     /* This is called by several functions that scan a compiled expression looking
1399     for a fixed first character, or an anchoring op code etc. It skips over things
1400     that do not influence this. For some calls, a change of option is important.
1401     For some calls, it makes sense to skip negative forward and all backward
1402     assertions, and also the \b assertion; for others it does not.
1403    
1404     Arguments:
1405     code pointer to the start of the group
1406     options pointer to external options
1407     optbit the option bit whose changing is significant, or
1408     zero if none are
1409     skipassert TRUE if certain assertions are to be skipped
1410    
1411     Returns: pointer to the first significant opcode
1412     */
1413    
1414     static const uschar*
1415     first_significant_code(const uschar *code, int *options, int optbit,
1416     BOOL skipassert)
1417     {
1418     for (;;)
1419     {
1420     switch ((int)*code)
1421     {
1422     case OP_OPT:
1423     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1424     *options = (int)code[1];
1425     code += 2;
1426     break;
1427    
1428     case OP_ASSERT_NOT:
1429     case OP_ASSERTBACK:
1430     case OP_ASSERTBACK_NOT:
1431     if (!skipassert) return code;
1432     do code += GET(code, 1); while (*code == OP_ALT);
1433     code += _pcre_OP_lengths[*code];
1434     break;
1435    
1436     case OP_WORD_BOUNDARY:
1437     case OP_NOT_WORD_BOUNDARY:
1438     if (!skipassert) return code;
1439     /* Fall through */
1440    
1441     case OP_CALLOUT:
1442     case OP_CREF:
1443 ph10 459 case OP_NCREF:
1444 nigel 93 case OP_RREF:
1445 ph10 459 case OP_NRREF:
1446 nigel 93 case OP_DEF:
1447 nigel 77 code += _pcre_OP_lengths[*code];
1448     break;
1449    
1450     default:
1451     return code;
1452     }
1453     }
1454     /* Control never reaches here */
1455     }
1456    
1457    
1458    
1459    
1460     /*************************************************
1461 ph10 454 * Find the fixed length of a branch *
1462 nigel 77 *************************************************/
1463    
1464 ph10 454 /* Scan a branch and compute the fixed length of subject that will match it,
1465 nigel 77 if the length is fixed. This is needed for dealing with backward assertions.
1466 ph10 461 In UTF8 mode, the result is in characters rather than bytes. The branch is
1467 ph10 454 temporarily terminated with OP_END when this function is called.
1468 nigel 77
1469 ph10 461 This function is called when a backward assertion is encountered, so that if it
1470     fails, the error message can point to the correct place in the pattern.
1471 ph10 454 However, we cannot do this when the assertion contains subroutine calls,
1472 ph10 461 because they can be forward references. We solve this by remembering this case
1473 ph10 454 and doing the check at the end; a flag specifies which mode we are running in.
1474    
1475 nigel 77 Arguments:
1476     code points to the start of the pattern (the bracket)
1477     options the compiling options
1478 ph10 461 atend TRUE if called when the pattern is complete
1479     cd the "compile data" structure
1480 nigel 77
1481 ph10 461 Returns: the fixed length,
1482 ph10 454 or -1 if there is no fixed length,
1483 nigel 77 or -2 if \C was encountered
1484 ph10 454 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1485 nigel 77 */
1486    
1487     static int
1488 ph10 454 find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1489 nigel 77 {
1490     int length = -1;
1491    
1492     register int branchlength = 0;
1493     register uschar *cc = code + 1 + LINK_SIZE;
1494    
1495     /* Scan along the opcodes for this branch. If we get to the end of the
1496     branch, check the length against that of the other branches. */
1497    
1498     for (;;)
1499     {
1500     int d;
1501 ph10 454 uschar *ce, *cs;
1502 nigel 77 register int op = *cc;
1503     switch (op)
1504     {
1505 nigel 93 case OP_CBRA:
1506 nigel 77 case OP_BRA:
1507     case OP_ONCE:
1508     case OP_COND:
1509 ph10 454 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1510 nigel 77 if (d < 0) return d;
1511     branchlength += d;
1512     do cc += GET(cc, 1); while (*cc == OP_ALT);
1513     cc += 1 + LINK_SIZE;
1514     break;
1515    
1516     /* Reached end of a branch; if it's a ket it is the end of a nested
1517     call. If it's ALT it is an alternation in a nested call. If it is
1518     END it's the end of the outer call. All can be handled by the same code. */
1519    
1520     case OP_ALT:
1521     case OP_KET:
1522     case OP_KETRMAX:
1523     case OP_KETRMIN:
1524     case OP_END:
1525     if (length < 0) length = branchlength;
1526     else if (length != branchlength) return -1;
1527     if (*cc != OP_ALT) return length;
1528     cc += 1 + LINK_SIZE;
1529     branchlength = 0;
1530     break;
1531 ph10 461
1532 ph10 454 /* A true recursion implies not fixed length, but a subroutine call may
1533     be OK. If the subroutine is a forward reference, we can't deal with
1534     it until the end of the pattern, so return -3. */
1535 ph10 461
1536 ph10 454 case OP_RECURSE:
1537     if (!atend) return -3;
1538     cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1539     do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1540     if (cc > cs && cc < ce) return -1; /* Recursion */
1541     d = find_fixedlength(cs + 2, options, atend, cd);
1542 ph10 461 if (d < 0) return d;
1543 ph10 454 branchlength += d;
1544     cc += 1 + LINK_SIZE;
1545 ph10 461 break;
1546 nigel 77
1547     /* Skip over assertive subpatterns */
1548    
1549     case OP_ASSERT:
1550     case OP_ASSERT_NOT:
1551     case OP_ASSERTBACK:
1552     case OP_ASSERTBACK_NOT:
1553     do cc += GET(cc, 1); while (*cc == OP_ALT);
1554     /* Fall through */
1555    
1556     /* Skip over things that don't match chars */
1557    
1558     case OP_REVERSE:
1559     case OP_CREF:
1560 ph10 459 case OP_NCREF:
1561 nigel 93 case OP_RREF:
1562 ph10 459 case OP_NRREF:
1563 nigel 93 case OP_DEF:
1564 nigel 77 case OP_OPT:
1565     case OP_CALLOUT:
1566     case OP_SOD:
1567     case OP_SOM:
1568 ph10 500 case OP_SET_SOM:
1569 nigel 77 case OP_EOD:
1570     case OP_EODN:
1571     case OP_CIRC:
1572     case OP_DOLL:
1573     case OP_NOT_WORD_BOUNDARY:
1574     case OP_WORD_BOUNDARY:
1575     cc += _pcre_OP_lengths[*cc];
1576     break;
1577    
1578     /* Handle literal characters */
1579    
1580     case OP_CHAR:
1581     case OP_CHARNC:
1582 nigel 91 case OP_NOT:
1583 nigel 77 branchlength++;
1584     cc += 2;
1585     #ifdef SUPPORT_UTF8
1586 ph10 461 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1587 ph10 426 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1588 nigel 77 #endif
1589     break;
1590    
1591     /* Handle exact repetitions. The count is already in characters, but we
1592     need to skip over a multibyte character in UTF8 mode. */
1593    
1594     case OP_EXACT:
1595     branchlength += GET2(cc,1);
1596     cc += 4;
1597     #ifdef SUPPORT_UTF8
1598 ph10 461 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1599 ph10 426 cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1600 nigel 77 #endif
1601     break;
1602    
1603     case OP_TYPEEXACT:
1604     branchlength += GET2(cc,1);
1605 ph10 220 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1606 nigel 77 cc += 4;
1607     break;
1608    
1609     /* Handle single-char matchers */
1610    
1611     case OP_PROP:
1612     case OP_NOTPROP:
1613 nigel 87 cc += 2;
1614 nigel 77 /* Fall through */
1615    
1616     case OP_NOT_DIGIT:
1617     case OP_DIGIT:
1618     case OP_NOT_WHITESPACE:
1619     case OP_WHITESPACE:
1620     case OP_NOT_WORDCHAR:
1621     case OP_WORDCHAR:
1622     case OP_ANY:
1623 ph10 342 case OP_ALLANY:
1624 nigel 77 branchlength++;
1625     cc++;
1626     break;
1627    
1628     /* The single-byte matcher isn't allowed */
1629    
1630     case OP_ANYBYTE:
1631     return -2;
1632    
1633     /* Check a class for variable quantification */
1634    
1635     #ifdef SUPPORT_UTF8
1636     case OP_XCLASS:
1637     cc += GET(cc, 1) - 33;
1638     /* Fall through */
1639     #endif
1640    
1641     case OP_CLASS:
1642     case OP_NCLASS:
1643     cc += 33;
1644    
1645     switch (*cc)
1646     {
1647     case OP_CRSTAR:
1648     case OP_CRMINSTAR:
1649     case OP_CRQUERY:
1650     case OP_CRMINQUERY:
1651     return -1;
1652    
1653     case OP_CRRANGE:
1654     case OP_CRMINRANGE:
1655     if (GET2(cc,1) != GET2(cc,3)) return -1;
1656     branchlength += GET2(cc,1);
1657     cc += 5;
1658     break;
1659    
1660     default:
1661     branchlength++;
1662     }
1663     break;
1664    
1665     /* Anything else is variable length */
1666    
1667     default:
1668     return -1;
1669     }
1670     }
1671     /* Control never gets here */
1672     }
1673    
1674    
1675    
1676    
1677     /*************************************************
1678 ph10 454 * Scan compiled regex for specific bracket *
1679 nigel 77 *************************************************/
1680    
1681     /* This little function scans through a compiled pattern until it finds a
1682 ph10 454 capturing bracket with the given number, or, if the number is negative, an
1683 ph10 461 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1684     so that it can be called from pcre_study() when finding the minimum matching
1685 ph10 455 length.
1686 nigel 77
1687     Arguments:
1688     code points to start of expression
1689     utf8 TRUE in UTF-8 mode
1690 ph10 454 number the required bracket number or negative to find a lookbehind
1691 nigel 77
1692     Returns: pointer to the opcode for the bracket, or NULL if not found
1693     */
1694    
1695 ph10 455 const uschar *
1696     _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1697 nigel 77 {
1698     for (;;)
1699     {
1700     register int c = *code;
1701     if (c == OP_END) return NULL;
1702 nigel 91
1703     /* XCLASS is used for classes that cannot be represented just by a bit
1704     map. This includes negated single high-valued characters. The length in
1705     the table is zero; the actual length is stored in the compiled code. */
1706    
1707     if (c == OP_XCLASS) code += GET(code, 1);
1708 ph10 461
1709 ph10 454 /* Handle recursion */
1710 ph10 461
1711 ph10 454 else if (c == OP_REVERSE)
1712     {
1713 ph10 461 if (number < 0) return (uschar *)code;
1714 ph10 454 code += _pcre_OP_lengths[c];
1715     }
1716 nigel 91
1717 nigel 93 /* Handle capturing bracket */
1718 nigel 91
1719 nigel 93 else if (c == OP_CBRA)
1720 nigel 77 {
1721 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1722 nigel 77 if (n == number) return (uschar *)code;
1723 nigel 93 code += _pcre_OP_lengths[c];
1724 nigel 77 }
1725 nigel 91
1726 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1727     repeated character types, we have to test for \p and \P, which have an extra
1728 ph10 512 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1729 ph10 510 must add in its length. */
1730 nigel 91
1731 nigel 77 else
1732     {
1733 ph10 218 switch(c)
1734     {
1735     case OP_TYPESTAR:
1736     case OP_TYPEMINSTAR:
1737     case OP_TYPEPLUS:
1738     case OP_TYPEMINPLUS:
1739     case OP_TYPEQUERY:
1740     case OP_TYPEMINQUERY:
1741     case OP_TYPEPOSSTAR:
1742     case OP_TYPEPOSPLUS:
1743     case OP_TYPEPOSQUERY:
1744     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1745 ph10 220 break;
1746 ph10 221
1747     case OP_TYPEUPTO:
1748     case OP_TYPEMINUPTO:
1749     case OP_TYPEEXACT:
1750     case OP_TYPEPOSUPTO:
1751     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1752     break;
1753 ph10 512
1754 ph10 510 case OP_MARK:
1755     case OP_PRUNE_ARG:
1756     case OP_SKIP_ARG:
1757     code += code[1];
1758 ph10 512 break;
1759 ph10 550
1760     case OP_THEN_ARG:
1761     code += code[1+LINK_SIZE];
1762     break;
1763 ph10 220 }
1764    
1765 ph10 218 /* Add in the fixed length from the table */
1766 ph10 220
1767 nigel 77 code += _pcre_OP_lengths[c];
1768 ph10 220
1769 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1770     a multi-byte character. The length in the table is a minimum, so we have to
1771     arrange to skip the extra bytes. */
1772 ph10 220
1773 ph10 107 #ifdef SUPPORT_UTF8
1774 nigel 77 if (utf8) switch(c)
1775     {
1776     case OP_CHAR:
1777     case OP_CHARNC:
1778     case OP_EXACT:
1779     case OP_UPTO:
1780     case OP_MINUPTO:
1781 nigel 93 case OP_POSUPTO:
1782 nigel 77 case OP_STAR:
1783     case OP_MINSTAR:
1784 nigel 93 case OP_POSSTAR:
1785 nigel 77 case OP_PLUS:
1786     case OP_MINPLUS:
1787 nigel 93 case OP_POSPLUS:
1788 nigel 77 case OP_QUERY:
1789     case OP_MINQUERY:
1790 nigel 93 case OP_POSQUERY:
1791     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1792 nigel 77 break;
1793     }
1794 ph10 369 #else
1795     (void)(utf8); /* Keep compiler happy by referencing function argument */
1796 ph10 111 #endif
1797 nigel 77 }
1798     }
1799     }
1800    
1801    
1802    
1803     /*************************************************
1804     * Scan compiled regex for recursion reference *
1805     *************************************************/
1806    
1807     /* This little function scans through a compiled pattern until it finds an
1808     instance of OP_RECURSE.
1809    
1810     Arguments:
1811     code points to start of expression
1812     utf8 TRUE in UTF-8 mode
1813    
1814     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1815     */
1816    
1817     static const uschar *
1818     find_recurse(const uschar *code, BOOL utf8)
1819     {
1820     for (;;)
1821     {
1822     register int c = *code;
1823     if (c == OP_END) return NULL;
1824 nigel 91 if (c == OP_RECURSE) return code;
1825 ph10 220
1826 nigel 91 /* XCLASS is used for classes that cannot be represented just by a bit
1827     map. This includes negated single high-valued characters. The length in
1828     the table is zero; the actual length is stored in the compiled code. */
1829    
1830     if (c == OP_XCLASS) code += GET(code, 1);
1831    
1832 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1833     repeated character types, we have to test for \p and \P, which have an extra
1834 ph10 512 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1835 ph10 510 must add in its length. */
1836 nigel 91
1837 nigel 77 else
1838     {
1839 ph10 218 switch(c)
1840     {
1841     case OP_TYPESTAR:
1842     case OP_TYPEMINSTAR:
1843     case OP_TYPEPLUS:
1844     case OP_TYPEMINPLUS:
1845     case OP_TYPEQUERY:
1846     case OP_TYPEMINQUERY:
1847     case OP_TYPEPOSSTAR:
1848     case OP_TYPEPOSPLUS:
1849     case OP_TYPEPOSQUERY:
1850     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1851 ph10 220 break;
1852 ph10 221
1853     case OP_TYPEPOSUPTO:
1854     case OP_TYPEUPTO:
1855     case OP_TYPEMINUPTO:
1856     case OP_TYPEEXACT:
1857     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1858     break;
1859 ph10 512
1860 ph10 510 case OP_MARK:
1861     case OP_PRUNE_ARG:
1862     case OP_SKIP_ARG:
1863     code += code[1];
1864 ph10 512 break;
1865 ph10 550
1866     case OP_THEN_ARG:
1867     code += code[1+LINK_SIZE];
1868     break;
1869 ph10 220 }
1870    
1871 ph10 218 /* Add in the fixed length from the table */
1872    
1873 nigel 77 code += _pcre_OP_lengths[c];
1874 ph10 220
1875 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1876     by a multi-byte character. The length in the table is a minimum, so we have
1877     to arrange to skip the extra bytes. */
1878 ph10 220
1879 ph10 107 #ifdef SUPPORT_UTF8
1880 nigel 77 if (utf8) switch(c)
1881     {
1882     case OP_CHAR:
1883     case OP_CHARNC:
1884     case OP_EXACT:
1885     case OP_UPTO:
1886     case OP_MINUPTO:
1887 nigel 93 case OP_POSUPTO:
1888 nigel 77 case OP_STAR:
1889     case OP_MINSTAR:
1890 nigel 93 case OP_POSSTAR:
1891 nigel 77 case OP_PLUS:
1892     case OP_MINPLUS:
1893 nigel 93 case OP_POSPLUS:
1894 nigel 77 case OP_QUERY:
1895     case OP_MINQUERY:
1896 nigel 93 case OP_POSQUERY:
1897     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1898 nigel 77 break;
1899     }
1900 ph10 369 #else
1901     (void)(utf8); /* Keep compiler happy by referencing function argument */
1902 ph10 111 #endif
1903 nigel 77 }
1904     }
1905     }
1906    
1907    
1908    
1909     /*************************************************
1910     * Scan compiled branch for non-emptiness *
1911     *************************************************/
1912    
1913     /* This function scans through a branch of a compiled pattern to see whether it
1914 nigel 93 can match the empty string or not. It is called from could_be_empty()
1915     below and from compile_branch() when checking for an unlimited repeat of a
1916     group that can match nothing. Note that first_significant_code() skips over
1917 ph10 282 backward and negative forward assertions when its final argument is TRUE. If we
1918     hit an unclosed bracket, we return "empty" - this means we've struck an inner
1919     bracket whose current branch will already have been scanned.
1920 nigel 77
1921     Arguments:
1922     code points to start of search
1923     endcode points to where to stop
1924     utf8 TRUE if in UTF8 mode
1925 ph10 503 cd contains pointers to tables etc.
1926 nigel 77
1927     Returns: TRUE if what is matched could be empty
1928     */
1929    
1930     static BOOL
1931 ph10 503 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1932     compile_data *cd)
1933 nigel 77 {
1934     register int c;
1935 nigel 93 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1936 nigel 77 code < endcode;
1937     code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1938     {
1939     const uschar *ccode;
1940    
1941     c = *code;
1942 ph10 507
1943 ph10 286 /* Skip over forward assertions; the other assertions are skipped by
1944 ph10 282 first_significant_code() with a TRUE final argument. */
1945 ph10 286
1946 ph10 282 if (c == OP_ASSERT)
1947 ph10 286 {
1948 ph10 282 do code += GET(code, 1); while (*code == OP_ALT);
1949     c = *code;
1950     continue;
1951 ph10 286 }
1952 ph10 172
1953 ph10 170 /* Groups with zero repeats can of course be empty; skip them. */
1954 nigel 77
1955 ph10 335 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1956 ph10 170 {
1957 ph10 172 code += _pcre_OP_lengths[c];
1958 ph10 170 do code += GET(code, 1); while (*code == OP_ALT);
1959     c = *code;
1960     continue;
1961     }
1962 ph10 507
1963 ph10 503 /* For a recursion/subroutine call, if its end has been reached, which
1964     implies a subroutine call, we can scan it. */
1965 ph10 507
1966 ph10 503 if (c == OP_RECURSE)
1967     {
1968 ph10 507 BOOL empty_branch = FALSE;
1969 ph10 503 const uschar *scode = cd->start_code + GET(code, 1);
1970     if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1971     do
1972     {
1973 ph10 504 if (could_be_empty_branch(scode, endcode, utf8, cd))
1974     {
1975     empty_branch = TRUE;
1976 ph10 507 break;
1977     }
1978 ph10 503 scode += GET(scode, 1);
1979     }
1980     while (*scode == OP_ALT);
1981 ph10 504 if (!empty_branch) return FALSE; /* All branches are non-empty */
1982 ph10 503 continue;
1983 ph10 507 }
1984 ph10 170
1985     /* For other groups, scan the branches. */
1986 ph10 172
1987 ph10 206 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1988 nigel 77 {
1989     BOOL empty_branch;
1990     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1991 ph10 406
1992     /* If a conditional group has only one branch, there is a second, implied,
1993 ph10 395 empty branch, so just skip over the conditional, because it could be empty.
1994     Otherwise, scan the individual branches of the group. */
1995 ph10 406
1996 ph10 395 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1997 nigel 77 code += GET(code, 1);
1998 ph10 395 else
1999 ph10 406 {
2000 ph10 395 empty_branch = FALSE;
2001     do
2002     {
2003 ph10 503 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2004 ph10 395 empty_branch = TRUE;
2005     code += GET(code, 1);
2006     }
2007     while (*code == OP_ALT);
2008     if (!empty_branch) return FALSE; /* All branches are non-empty */
2009 nigel 77 }
2010 ph10 406
2011 ph10 172 c = *code;
2012 nigel 93 continue;
2013 nigel 77 }
2014    
2015 nigel 93 /* Handle the other opcodes */
2016    
2017     switch (c)
2018 nigel 77 {
2019 ph10 216 /* Check for quantifiers after a class. XCLASS is used for classes that
2020     cannot be represented just by a bit map. This includes negated single
2021     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2022 ph10 220 actual length is stored in the compiled code, so we must update "code"
2023 ph10 216 here. */
2024 nigel 77
2025     #ifdef SUPPORT_UTF8
2026     case OP_XCLASS:
2027 ph10 216 ccode = code += GET(code, 1);
2028 nigel 77 goto CHECK_CLASS_REPEAT;
2029     #endif
2030    
2031     case OP_CLASS:
2032     case OP_NCLASS:
2033     ccode = code + 33;
2034    
2035     #ifdef SUPPORT_UTF8
2036     CHECK_CLASS_REPEAT:
2037     #endif
2038    
2039     switch (*ccode)
2040     {
2041     case OP_CRSTAR: /* These could be empty; continue */
2042     case OP_CRMINSTAR:
2043     case OP_CRQUERY:
2044     case OP_CRMINQUERY:
2045     break;
2046    
2047     default: /* Non-repeat => class must match */
2048     case OP_CRPLUS: /* These repeats aren't empty */
2049     case OP_CRMINPLUS:
2050     return FALSE;
2051    
2052     case OP_CRRANGE:
2053     case OP_CRMINRANGE:
2054     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2055     break;
2056     }
2057     break;
2058    
2059     /* Opcodes that must match a character */
2060    
2061     case OP_PROP:
2062     case OP_NOTPROP:
2063     case OP_EXTUNI:
2064     case OP_NOT_DIGIT:
2065     case OP_DIGIT:
2066     case OP_NOT_WHITESPACE:
2067     case OP_WHITESPACE:
2068     case OP_NOT_WORDCHAR:
2069     case OP_WORDCHAR:
2070     case OP_ANY:
2071 ph10 345 case OP_ALLANY:
2072 nigel 77 case OP_ANYBYTE:
2073     case OP_CHAR:
2074     case OP_CHARNC:
2075     case OP_NOT:
2076     case OP_PLUS:
2077     case OP_MINPLUS:
2078 nigel 93 case OP_POSPLUS:
2079 nigel 77 case OP_EXACT:
2080     case OP_NOTPLUS:
2081     case OP_NOTMINPLUS:
2082 nigel 93 case OP_NOTPOSPLUS:
2083 nigel 77 case OP_NOTEXACT:
2084     case OP_TYPEPLUS:
2085     case OP_TYPEMINPLUS:
2086 nigel 93 case OP_TYPEPOSPLUS:
2087 nigel 77 case OP_TYPEEXACT:
2088     return FALSE;
2089 ph10 227
2090     /* These are going to continue, as they may be empty, but we have to
2091     fudge the length for the \p and \P cases. */
2092    
2093 ph10 224 case OP_TYPESTAR:
2094     case OP_TYPEMINSTAR:
2095     case OP_TYPEPOSSTAR:
2096     case OP_TYPEQUERY:
2097     case OP_TYPEMINQUERY:
2098     case OP_TYPEPOSQUERY:
2099     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2100 ph10 227 break;
2101    
2102 ph10 224 /* Same for these */
2103 ph10 227
2104 ph10 224 case OP_TYPEUPTO:
2105     case OP_TYPEMINUPTO:
2106     case OP_TYPEPOSUPTO:
2107     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2108     break;
2109 nigel 77
2110     /* End of branch */
2111    
2112     case OP_KET:
2113     case OP_KETRMAX:
2114     case OP_KETRMIN:
2115     case OP_ALT:
2116     return TRUE;
2117    
2118 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2119     MINUPTO, and POSUPTO may be followed by a multibyte character */
2120 nigel 77
2121     #ifdef SUPPORT_UTF8
2122     case OP_STAR:
2123     case OP_MINSTAR:
2124 nigel 93 case OP_POSSTAR:
2125 nigel 77 case OP_QUERY:
2126     case OP_MINQUERY:
2127 nigel 93 case OP_POSQUERY:
2128 ph10 426 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2129     break;
2130 ph10 461
2131 nigel 77 case OP_UPTO:
2132     case OP_MINUPTO:
2133 nigel 93 case OP_POSUPTO:
2134 ph10 426 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2135 nigel 77 break;
2136     #endif
2137 ph10 503
2138 ph10 510 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2139     string. */
2140    
2141     case OP_MARK:
2142     case OP_PRUNE_ARG:
2143     case OP_SKIP_ARG:
2144     code += code[1];
2145 ph10 512 break;
2146 ph10 510
2147 ph10 550 case OP_THEN_ARG:
2148     code += code[1+LINK_SIZE];
2149     break;
2150    
2151 ph10 503 /* None of the remaining opcodes are required to match a character. */
2152 ph10 507
2153 ph10 503 default:
2154 ph10 507 break;
2155 nigel 77 }
2156     }
2157    
2158     return TRUE;
2159     }
2160    
2161    
2162    
2163     /*************************************************
2164     * Scan compiled regex for non-emptiness *
2165     *************************************************/
2166    
2167     /* This function is called to check for left recursive calls. We want to check
2168     the current branch of the current pattern to see if it could match the empty
2169     string. If it could, we must look outwards for branches at other levels,
2170     stopping when we pass beyond the bracket which is the subject of the recursion.
2171    
2172     Arguments:
2173     code points to start of the recursion
2174     endcode points to where to stop (current RECURSE item)
2175     bcptr points to the chain of current (unclosed) branch starts
2176     utf8 TRUE if in UTF-8 mode
2177 ph10 507 cd pointers to tables etc
2178 nigel 77
2179     Returns: TRUE if what is matched could be empty
2180     */
2181    
2182     static BOOL
2183     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2184 ph10 503 BOOL utf8, compile_data *cd)
2185 nigel 77 {
2186 ph10 475 while (bcptr != NULL && bcptr->current_branch >= code)
2187 nigel 77 {
2188 ph10 503 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2189 ph10 475 return FALSE;
2190 nigel 77 bcptr = bcptr->outer;
2191     }
2192     return TRUE;
2193     }
2194    
2195    
2196    
2197     /*************************************************
2198     * Check for POSIX class syntax *
2199     *************************************************/
2200    
2201     /* This function is called when the sequence "[:" or "[." or "[=" is
2202 ph10 295 encountered in a character class. It checks whether this is followed by a
2203 ph10 298 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2204 ph10 295 reach an unescaped ']' without the special preceding character, return FALSE.
2205 nigel 77
2206 ph10 298 Originally, this function only recognized a sequence of letters between the
2207     terminators, but it seems that Perl recognizes any sequence of characters,
2208     though of course unknown POSIX names are subsequently rejected. Perl gives an
2209     "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2210     didn't consider this to be a POSIX class. Likewise for [:1234:].
2211 ph10 295
2212 ph10 298 The problem in trying to be exactly like Perl is in the handling of escapes. We
2213     have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2214     class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2215     below handles the special case of \], but does not try to do any other escape
2216     processing. This makes it different from Perl for cases such as [:l\ower:]
2217 ph10 295 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2218 ph10 298 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2219 ph10 295 I think.
2220    
2221     Arguments:
2222 nigel 77 ptr pointer to the initial [
2223     endptr where to return the end pointer
2224    
2225     Returns: TRUE or FALSE
2226     */
2227    
2228     static BOOL
2229 ph10 295 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2230 nigel 77 {
2231     int terminator; /* Don't combine these lines; the Solaris cc */
2232     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2233 ph10 295 for (++ptr; *ptr != 0; ptr++)
2234 nigel 77 {
2235 ph10 391 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2236 ph10 298 {
2237 ph10 391 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2238     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2239 ph10 295 {
2240     *endptr = ptr;
2241     return TRUE;
2242 ph10 298 }
2243     }
2244     }
2245 nigel 77 return FALSE;
2246     }
2247    
2248    
2249    
2250    
2251     /*************************************************
2252     * Check POSIX class name *
2253     *************************************************/
2254    
2255     /* This function is called to check the name given in a POSIX-style class entry
2256     such as [:alnum:].
2257    
2258     Arguments:
2259     ptr points to the first letter
2260     len the length of the name
2261    
2262     Returns: a value representing the name, or -1 if unknown
2263     */
2264    
2265     static int
2266     check_posix_name(const uschar *ptr, int len)
2267     {
2268 ph10 240 const char *pn = posix_names;
2269 nigel 77 register int yield = 0;
2270     while (posix_name_lengths[yield] != 0)
2271     {
2272     if (len == posix_name_lengths[yield] &&
2273 ph10 240 strncmp((const char *)ptr, pn, len) == 0) return yield;
2274 ph10 243 pn += posix_name_lengths[yield] + 1;
2275 nigel 77 yield++;
2276     }
2277     return -1;
2278     }
2279    
2280    
2281     /*************************************************
2282     * Adjust OP_RECURSE items in repeated group *
2283     *************************************************/
2284    
2285     /* OP_RECURSE items contain an offset from the start of the regex to the group
2286     that is referenced. This means that groups can be replicated for fixed
2287     repetition simply by copying (because the recursion is allowed to refer to
2288     earlier groups that are outside the current group). However, when a group is
2289 ph10 335 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2290     inserted before it, after it has been compiled. This means that any OP_RECURSE
2291     items within it that refer to the group itself or any contained groups have to
2292     have their offsets adjusted. That one of the jobs of this function. Before it
2293     is called, the partially compiled regex must be temporarily terminated with
2294     OP_END.
2295 nigel 77
2296 nigel 93 This function has been extended with the possibility of forward references for
2297     recursions and subroutine calls. It must also check the list of such references
2298     for the group we are dealing with. If it finds that one of the recursions in
2299     the current group is on this list, it adjusts the offset in the list, not the
2300     value in the reference (which is a group number).
2301    
2302 nigel 77 Arguments:
2303     group points to the start of the group
2304     adjust the amount by which the group is to be moved
2305     utf8 TRUE in UTF-8 mode
2306     cd contains pointers to tables etc.
2307 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
2308 nigel 77
2309     Returns: nothing
2310     */
2311    
2312     static void
2313 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2314     uschar *save_hwm)
2315 nigel 77 {
2316     uschar *ptr = group;
2317 ph10 224
2318 nigel 77 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2319     {
2320 nigel 93 int offset;
2321     uschar *hc;
2322    
2323     /* See if this recursion is on the forward reference list. If so, adjust the
2324     reference. */
2325 ph10 345
2326 nigel 93 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2327     {
2328     offset = GET(hc, 0);
2329     if (cd->start_code + offset == ptr + 1)
2330     {
2331     PUT(hc, 0, offset + adjust);
2332     break;
2333     }
2334     }
2335    
2336     /* Otherwise, adjust the recursion offset if it's after the start of this
2337     group. */
2338    
2339     if (hc >= cd->hwm)
2340     {
2341     offset = GET(ptr, 1);
2342     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2343     }
2344    
2345 nigel 77 ptr += 1 + LINK_SIZE;
2346     }
2347     }
2348    
2349    
2350    
2351     /*************************************************
2352     * Insert an automatic callout point *
2353     *************************************************/
2354    
2355     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2356     callout points before each pattern item.
2357    
2358     Arguments:
2359     code current code pointer
2360     ptr current pattern pointer
2361     cd pointers to tables etc
2362    
2363     Returns: new code pointer
2364     */
2365    
2366     static uschar *
2367     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2368     {
2369     *code++ = OP_CALLOUT;
2370     *code++ = 255;
2371 ph10 530 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2372     PUT(code, LINK_SIZE, 0); /* Default length */
2373 nigel 77 return code + 2*LINK_SIZE;
2374     }
2375    
2376    
2377    
2378     /*************************************************
2379     * Complete a callout item *
2380     *************************************************/
2381    
2382     /* A callout item contains the length of the next item in the pattern, which
2383     we can't fill in till after we have reached the relevant point. This is used
2384     for both automatic and manual callouts.
2385    
2386     Arguments:
2387     previous_callout points to previous callout item
2388     ptr current pattern pointer
2389     cd pointers to tables etc
2390    
2391     Returns: nothing
2392     */
2393    
2394     static void
2395     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2396     {
2397 ph10 530 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2398 nigel 77 PUT(previous_callout, 2 + LINK_SIZE, length);
2399     }
2400    
2401    
2402    
2403     #ifdef SUPPORT_UCP
2404     /*************************************************
2405     * Get othercase range *
2406     *************************************************/
2407    
2408     /* This function is passed the start and end of a class range, in UTF-8 mode
2409     with UCP support. It searches up the characters, looking for internal ranges of
2410     characters in the "other" case. Each call returns the next one, updating the
2411     start address.
2412    
2413     Arguments:
2414     cptr points to starting character value; updated
2415     d end value
2416     ocptr where to put start of othercase range
2417     odptr where to put end of othercase range
2418    
2419     Yield: TRUE when range returned; FALSE when no more
2420     */
2421    
2422     static BOOL
2423 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2424     unsigned int *odptr)
2425 nigel 77 {
2426 nigel 93 unsigned int c, othercase, next;
2427 nigel 77
2428     for (c = *cptr; c <= d; c++)
2429 ph10 349 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2430 nigel 77
2431     if (c > d) return FALSE;
2432    
2433     *ocptr = othercase;
2434     next = othercase + 1;
2435    
2436     for (++c; c <= d; c++)
2437     {
2438 ph10 349 if (UCD_OTHERCASE(c) != next) break;
2439 nigel 77 next++;
2440     }
2441    
2442     *odptr = next - 1;
2443     *cptr = c;
2444    
2445     return TRUE;
2446     }
2447 ph10 532
2448    
2449    
2450     /*************************************************
2451     * Check a character and a property *
2452     *************************************************/
2453    
2454     /* This function is called by check_auto_possessive() when a property item
2455     is adjacent to a fixed character.
2456    
2457     Arguments:
2458     c the character
2459     ptype the property type
2460     pdata the data for the type
2461     negated TRUE if it's a negated property (\P or \p{^)
2462 ph10 535
2463 ph10 532 Returns: TRUE if auto-possessifying is OK
2464 ph10 535 */
2465 ph10 532
2466     static BOOL
2467     check_char_prop(int c, int ptype, int pdata, BOOL negated)
2468     {
2469     const ucd_record *prop = GET_UCD(c);
2470     switch(ptype)
2471     {
2472     case PT_LAMP:
2473     return (prop->chartype == ucp_Lu ||
2474     prop->chartype == ucp_Ll ||
2475     prop->chartype == ucp_Lt) == negated;
2476    
2477     case PT_GC:
2478     return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2479    
2480     case PT_PC:
2481     return (pdata == prop->chartype) == negated;
2482    
2483     case PT_SC:
2484     return (pdata == prop->script) == negated;
2485    
2486     /* These are specials */
2487    
2488     case PT_ALNUM:
2489     return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2490     _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2491    
2492     case PT_SPACE: /* Perl space */
2493     return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2494     c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2495     == negated;
2496    
2497     case PT_PXSPACE: /* POSIX space */
2498     return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2499     c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2500     c == CHAR_FF || c == CHAR_CR)
2501     == negated;
2502    
2503     case PT_WORD:
2504     return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2505     _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2506     c == CHAR_UNDERSCORE) == negated;
2507     }
2508 ph10 535 return FALSE;
2509 ph10 532 }
2510 nigel 77 #endif /* SUPPORT_UCP */
2511    
2512    
2513 nigel 93
2514 nigel 77 /*************************************************
2515 nigel 93 * Check if auto-possessifying is possible *
2516     *************************************************/
2517    
2518     /* This function is called for unlimited repeats of certain items, to see
2519     whether the next thing could possibly match the repeated item. If not, it makes
2520     sense to automatically possessify the repeated item.
2521    
2522     Arguments:
2523 ph10 532 previous pointer to the repeated opcode
2524 nigel 93 utf8 TRUE in UTF-8 mode
2525     ptr next character in pattern
2526     options options bits
2527     cd contains pointers to tables etc.
2528    
2529     Returns: TRUE if possessifying is wanted
2530     */
2531    
2532     static BOOL
2533 ph10 535 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2534 ph10 532 int options, compile_data *cd)
2535 nigel 93 {
2536 ph10 532 int c, next;
2537     int op_code = *previous++;
2538 nigel 93
2539     /* Skip whitespace and comments in extended mode */
2540    
2541     if ((options & PCRE_EXTENDED) != 0)
2542     {
2543     for (;;)
2544     {
2545     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2546 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2547 nigel 93 {
2548 ph10 579 ptr++;
2549 ph10 556 while (*ptr != 0)
2550     {
2551 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2552 ph10 556 ptr++;
2553 ph10 579 #ifdef SUPPORT_UTF8
2554 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2555     #endif
2556     }
2557 nigel 93 }
2558     else break;
2559     }
2560     }
2561    
2562     /* If the next item is one that we can handle, get its value. A non-negative
2563     value is a character, a negative value is an escape value. */
2564    
2565 ph10 391 if (*ptr == CHAR_BACKSLASH)
2566 nigel 93 {
2567     int temperrorcode = 0;
2568     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2569     if (temperrorcode != 0) return FALSE;
2570     ptr++; /* Point after the escape sequence */
2571     }
2572    
2573     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2574     {
2575     #ifdef SUPPORT_UTF8
2576     if (utf8) { GETCHARINC(next, ptr); } else
2577     #endif
2578     next = *ptr++;
2579     }
2580    
2581     else return FALSE;
2582    
2583     /* Skip whitespace and comments in extended mode */
2584    
2585     if ((options & PCRE_EXTENDED) != 0)
2586     {
2587     for (;;)
2588     {
2589     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2590 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2591 nigel 93 {
2592 ph10 579 ptr++;
2593 ph10 556 while (*ptr != 0)
2594     {
2595 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2596 ph10 556 ptr++;
2597 ph10 579 #ifdef SUPPORT_UTF8
2598 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2599     #endif
2600     }
2601 nigel 93 }
2602     else break;
2603     }
2604     }
2605    
2606     /* If the next thing is itself optional, we have to give up. */
2607    
2608 ph10 392 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2609 ph10 391 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2610     return FALSE;
2611 nigel 93
2612 ph10 532 /* Now compare the next item with the previous opcode. First, handle cases when
2613     the next item is a character. */
2614 nigel 93
2615     if (next >= 0) switch(op_code)
2616     {
2617     case OP_CHAR:
2618 ph10 535 #ifdef SUPPORT_UTF8
2619 ph10 532 GETCHARTEST(c, previous);
2620 ph10 369 #else
2621 ph10 532 c = *previous;
2622 ph10 535 #endif
2623     return c != next;
2624 nigel 93
2625     /* For CHARNC (caseless character) we must check the other case. If we have
2626     Unicode property support, we can use it to test the other case of
2627     high-valued characters. */
2628    
2629     case OP_CHARNC:
2630 ph10 535 #ifdef SUPPORT_UTF8
2631 ph10 532 GETCHARTEST(c, previous);
2632     #else
2633     c = *previous;
2634 ph10 535 #endif
2635 ph10 532 if (c == next) return FALSE;
2636 nigel 93 #ifdef SUPPORT_UTF8
2637     if (utf8)
2638     {
2639     unsigned int othercase;
2640     if (next < 128) othercase = cd->fcc[next]; else
2641     #ifdef SUPPORT_UCP
2642 ph10 349 othercase = UCD_OTHERCASE((unsigned int)next);
2643 nigel 93 #else
2644     othercase = NOTACHAR;
2645     #endif
2646 ph10 532 return (unsigned int)c != othercase;
2647 nigel 93 }
2648     else
2649     #endif /* SUPPORT_UTF8 */
2650 ph10 532 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2651 nigel 93
2652 ph10 532 /* For OP_NOT, its data is always a single-byte character. */
2653 nigel 93
2654     case OP_NOT:
2655 ph10 532 if ((c = *previous) == next) return TRUE;
2656 nigel 93 if ((options & PCRE_CASELESS) == 0) return FALSE;
2657     #ifdef SUPPORT_UTF8
2658     if (utf8)
2659     {
2660     unsigned int othercase;
2661     if (next < 128) othercase = cd->fcc[next]; else
2662     #ifdef SUPPORT_UCP
2663 ph10 349 othercase = UCD_OTHERCASE(next);
2664 nigel 93 #else
2665     othercase = NOTACHAR;
2666     #endif
2667 ph10 532 return (unsigned int)c == othercase;
2668 nigel 93 }
2669     else
2670     #endif /* SUPPORT_UTF8 */
2671 ph10 532 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2672 nigel 93
2673 ph10 535 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2674     When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2675    
2676 nigel 93 case OP_DIGIT:
2677     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2678    
2679     case OP_NOT_DIGIT:
2680     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2681    
2682     case OP_WHITESPACE:
2683     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2684    
2685     case OP_NOT_WHITESPACE:
2686     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2687    
2688     case OP_WORDCHAR:
2689     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2690    
2691     case OP_NOT_WORDCHAR:
2692     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2693    
2694 ph10 180 case OP_HSPACE:
2695     case OP_NOT_HSPACE:
2696     switch(next)
2697     {
2698     case 0x09:
2699     case 0x20:
2700     case 0xa0:
2701     case 0x1680:
2702     case 0x180e:
2703     case 0x2000:
2704     case 0x2001:
2705     case 0x2002:
2706     case 0x2003:
2707     case 0x2004:
2708     case 0x2005:
2709     case 0x2006:
2710     case 0x2007:
2711     case 0x2008:
2712     case 0x2009:
2713     case 0x200A:
2714     case 0x202f:
2715     case 0x205f:
2716     case 0x3000:
2717 ph10 528 return op_code == OP_NOT_HSPACE;
2718 ph10 180 default:
2719 ph10 528 return op_code != OP_NOT_HSPACE;
2720 ph10 180 }
2721    
2722 ph10 528 case OP_ANYNL:
2723 ph10 180 case OP_VSPACE:
2724     case OP_NOT_VSPACE:
2725     switch(next)
2726     {
2727     case 0x0a:
2728     case 0x0b:
2729     case 0x0c:
2730     case 0x0d:
2731     case 0x85:
2732     case 0x2028:
2733     case 0x2029:
2734 ph10 528 return op_code == OP_NOT_VSPACE;
2735 ph10 180 default:
2736 ph10 528 return op_code != OP_NOT_VSPACE;
2737 ph10 180 }
2738    
2739 ph10 532 #ifdef SUPPORT_UCP
2740     case OP_PROP:
2741     return check_char_prop(next, previous[0], previous[1], FALSE);
2742 ph10 535
2743 ph10 532 case OP_NOTPROP:
2744     return check_char_prop(next, previous[0], previous[1], TRUE);
2745     #endif
2746    
2747 nigel 93 default:
2748     return FALSE;
2749     }
2750    
2751    
2752 ph10 535 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2753     is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2754     generated only when PCRE_UCP is *not* set, that is, when only ASCII
2755     characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2756 ph10 532 replaced by OP_PROP codes when PCRE_UCP is set. */
2757 nigel 93
2758     switch(op_code)
2759     {
2760     case OP_CHAR:
2761     case OP_CHARNC:
2762 ph10 535 #ifdef SUPPORT_UTF8
2763 ph10 532 GETCHARTEST(c, previous);
2764     #else
2765     c = *previous;
2766 ph10 535 #endif
2767 nigel 93 switch(-next)
2768     {
2769     case ESC_d:
2770 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2771 nigel 93
2772     case ESC_D:
2773 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2774 nigel 93
2775     case ESC_s:
2776 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2777 nigel 93
2778     case ESC_S:
2779 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2780 nigel 93
2781     case ESC_w:
2782 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2783 nigel 93
2784     case ESC_W:
2785 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2786 ph10 182
2787 ph10 180 case ESC_h:
2788     case ESC_H:
2789 ph10 532 switch(c)
2790 ph10 180 {
2791     case 0x09:
2792     case 0x20:
2793     case 0xa0:
2794     case 0x1680:
2795     case 0x180e:
2796     case 0x2000:
2797     case 0x2001:
2798     case 0x2002:
2799     case 0x2003:
2800     case 0x2004:
2801     case 0x2005:
2802     case 0x2006:
2803     case 0x2007:
2804     case 0x2008:
2805     case 0x2009:
2806     case 0x200A:
2807     case 0x202f:
2808     case 0x205f:
2809     case 0x3000:
2810     return -next != ESC_h;
2811     default:
2812     return -next == ESC_h;
2813 ph10 182 }
2814    
2815 ph10 180 case ESC_v:
2816     case ESC_V:
2817 ph10 532 switch(c)
2818 ph10 180 {
2819     case 0x0a:
2820     case 0x0b:
2821     case 0x0c:
2822     case 0x0d:
2823     case 0x85:
2824     case 0x2028:
2825     case 0x2029:
2826     return -next != ESC_v;
2827     default:
2828     return -next == ESC_v;
2829 ph10 182 }
2830 ph10 535
2831     /* When PCRE_UCP is set, these values get generated for \d etc. Find
2832     their substitutions and process them. The result will always be either
2833 ph10 532 -ESC_p or -ESC_P. Then fall through to process those values. */
2834 ph10 535
2835 ph10 532 #ifdef SUPPORT_UCP
2836     case ESC_du:
2837     case ESC_DU:
2838     case ESC_wu:
2839     case ESC_WU:
2840     case ESC_su:
2841     case ESC_SU:
2842     {
2843     int temperrorcode = 0;
2844     ptr = substitutes[-next - ESC_DU];
2845     next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2846     if (temperrorcode != 0) return FALSE;
2847     ptr++; /* For compatibility */
2848     }
2849 ph10 535 /* Fall through */
2850 nigel 93
2851 ph10 532 case ESC_p:
2852     case ESC_P:
2853     {
2854     int ptype, pdata, errorcodeptr;
2855 ph10 535 BOOL negated;
2856    
2857 ph10 532 ptr--; /* Make ptr point at the p or P */
2858     ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2859     if (ptype < 0) return FALSE;
2860     ptr++; /* Point past the final curly ket */
2861 ph10 535
2862 ph10 532 /* If the property item is optional, we have to give up. (When generated
2863     from \d etc by PCRE_UCP, this test will have been applied much earlier,
2864     to the original \d etc. At this point, ptr will point to a zero byte. */
2865 ph10 535
2866 ph10 532 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2867     strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2868     return FALSE;
2869 ph10 535
2870 ph10 532 /* Do the property check. */
2871 ph10 535
2872 ph10 532 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2873 ph10 535 }
2874 ph10 532 #endif
2875    
2876 nigel 93 default:
2877     return FALSE;
2878     }
2879    
2880 ph10 535 /* In principle, support for Unicode properties should be integrated here as
2881     well. It means re-organizing the above code so as to get hold of the property
2882     values before switching on the op-code. However, I wonder how many patterns
2883     combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2884     these op-codes are never generated.) */
2885    
2886 nigel 93 case OP_DIGIT:
2887 ph10 180 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2888 ph10 528 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2889 nigel 93
2890     case OP_NOT_DIGIT:
2891     return next == -ESC_d;
2892    
2893     case OP_WHITESPACE:
2894 ph10 528 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2895 nigel 93
2896     case OP_NOT_WHITESPACE:
2897 ph10 180 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2898 nigel 93
2899 ph10 180 case OP_HSPACE:
2900 ph10 535 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2901 ph10 528 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2902 ph10 180
2903     case OP_NOT_HSPACE:
2904     return next == -ESC_h;
2905 ph10 182
2906 ph10 180 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2907 ph10 535 case OP_ANYNL:
2908 ph10 182 case OP_VSPACE:
2909 ph10 180 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2910    
2911     case OP_NOT_VSPACE:
2912 ph10 528 return next == -ESC_v || next == -ESC_R;
2913 ph10 180
2914 nigel 93 case OP_WORDCHAR:
2915 ph10 535 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2916 ph10 528 next == -ESC_v || next == -ESC_R;
2917 nigel 93
2918     case OP_NOT_WORDCHAR:
2919     return next == -ESC_w || next == -ESC_d;
2920 ph10 182
2921 nigel 93 default:
2922     return FALSE;
2923     }
2924    
2925     /* Control does not reach here */
2926     }
2927    
2928    
2929    
2930     /*************************************************
2931 nigel 77 * Compile one branch *
2932     *************************************************/
2933    
2934 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
2935 nigel 77 changed during the branch, the pointer is used to change the external options
2936 nigel 93 bits. This function is used during the pre-compile phase when we are trying
2937     to find out the amount of memory needed, as well as during the real compile
2938     phase. The value of lengthptr distinguishes the two phases.
2939 nigel 77
2940     Arguments:
2941     optionsptr pointer to the option bits
2942     codeptr points to the pointer to the current code point
2943     ptrptr points to the current pattern pointer
2944     errorcodeptr points to error code variable
2945     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2946     reqbyteptr set to the last literal character required, else < 0
2947     bcptr points to current branch chain
2948     cd contains pointers to tables etc.
2949 nigel 93 lengthptr NULL during the real compile phase
2950     points to length accumulator during pre-compile phase
2951 nigel 77
2952     Returns: TRUE on success
2953     FALSE, with *errorcodeptr set non-zero on error
2954     */
2955    
2956     static BOOL
2957 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2958     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2959     compile_data *cd, int *lengthptr)
2960 nigel 77 {
2961     int repeat_type, op_type;
2962     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2963     int bravalue = 0;
2964     int greedy_default, greedy_non_default;
2965     int firstbyte, reqbyte;
2966     int zeroreqbyte, zerofirstbyte;
2967     int req_caseopt, reqvary, tempreqvary;
2968     int options = *optionsptr;
2969     int after_manual_callout = 0;
2970 nigel 93 int length_prevgroup = 0;
2971 nigel 77 register int c;
2972     register uschar *code = *codeptr;
2973 nigel 93 uschar *last_code = code;
2974     uschar *orig_code = code;
2975 nigel 77 uschar *tempcode;
2976     BOOL inescq = FALSE;
2977     BOOL groupsetfirstbyte = FALSE;
2978     const uschar *ptr = *ptrptr;
2979     const uschar *tempptr;
2980 ph10 518 const uschar *nestptr = NULL;
2981 nigel 77 uschar *previous = NULL;
2982     uschar *previous_callout = NULL;
2983 nigel 93 uschar *save_hwm = NULL;
2984 nigel 77 uschar classbits[32];
2985    
2986     #ifdef SUPPORT_UTF8
2987     BOOL class_utf8;
2988     BOOL utf8 = (options & PCRE_UTF8) != 0;
2989     uschar *class_utf8data;
2990 ph10 300 uschar *class_utf8data_base;
2991 nigel 77 uschar utf8_char[6];
2992     #else
2993     BOOL utf8 = FALSE;
2994 nigel 93 uschar *utf8_char = NULL;
2995 nigel 77 #endif
2996    
2997 ph10 475 #ifdef PCRE_DEBUG
2998 nigel 93 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2999     #endif
3000    
3001 nigel 77 /* Set up the default and non-default settings for greediness */
3002    
3003     greedy_default = ((options & PCRE_UNGREEDY) != 0);
3004     greedy_non_default = greedy_default ^ 1;
3005    
3006     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3007     matching encountered yet". It gets changed to REQ_NONE if we hit something that
3008     matches a non-fixed char first char; reqbyte just remains unset if we never
3009     find one.
3010    
3011     When we hit a repeat whose minimum is zero, we may have to adjust these values
3012     to take the zero repeat into account. This is implemented by setting them to
3013     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3014     item types that can be repeated set these backoff variables appropriately. */
3015    
3016     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3017    
3018     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3019     according to the current setting of the caseless flag. REQ_CASELESS is a bit
3020     value > 255. It is added into the firstbyte or reqbyte variables to record the
3021     case status of the value. This is used only for ASCII characters. */
3022    
3023     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3024    
3025     /* Switch on next character until the end of the branch */
3026    
3027     for (;; ptr++)
3028     {
3029     BOOL negate_class;
3030 ph10 286 BOOL should_flip_negation;
3031 nigel 77 BOOL possessive_quantifier;
3032     BOOL is_quantifier;
3033 nigel 93 BOOL is_recurse;
3034 ph10 180 BOOL reset_bracount;
3035 nigel 77 int class_charcount;
3036     int class_lastchar;
3037     int newoptions;
3038     int recno;
3039 ph10 172 int refsign;
3040 nigel 77 int skipbytes;
3041     int subreqbyte;
3042     int subfirstbyte;
3043 nigel 93 int terminator;
3044 nigel 77 int mclength;
3045     uschar mcbuffer[8];
3046    
3047 nigel 93 /* Get next byte in the pattern */
3048 nigel 77
3049     c = *ptr;
3050 ph10 345
3051 ph10 535 /* If we are at the end of a nested substitution, revert to the outer level
3052 ph10 518 string. Nesting only happens one level deep. */
3053    
3054     if (c == 0 && nestptr != NULL)
3055     {
3056     ptr = nestptr;
3057     nestptr = NULL;
3058     c = *ptr;
3059     }
3060    
3061 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
3062     previous cycle of this loop. */
3063    
3064     if (lengthptr != NULL)
3065     {
3066 ph10 475 #ifdef PCRE_DEBUG
3067 nigel 93 if (code > cd->hwm) cd->hwm = code; /* High water info */
3068     #endif
3069 ph10 505 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3070 nigel 93 {
3071     *errorcodeptr = ERR52;
3072     goto FAILED;
3073     }
3074    
3075     /* There is at least one situation where code goes backwards: this is the
3076     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3077     the class is simply eliminated. However, it is created first, so we have to
3078     allow memory for it. Therefore, don't ever reduce the length at this point.
3079     */
3080    
3081     if (code < last_code) code = last_code;
3082 ph10 202
3083     /* Paranoid check for integer overflow */
3084    
3085     if (OFLOW_MAX - *lengthptr < code - last_code)
3086     {
3087     *errorcodeptr = ERR20;
3088     goto FAILED;
3089     }
3090    
3091 ph10 530 *lengthptr += (int)(code - last_code);
3092 nigel 93 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3093    
3094     /* If "previous" is set and it is not at the start of the work space, move
3095     it back to there, in order to avoid filling up the work space. Otherwise,
3096     if "previous" is NULL, reset the current code pointer to the start. */
3097    
3098     if (previous != NULL)
3099     {
3100     if (previous > orig_code)
3101     {
3102     memmove(orig_code, previous, code - previous);
3103     code -= previous - orig_code;
3104     previous = orig_code;
3105     }
3106     }
3107     else code = orig_code;
3108    
3109     /* Remember where this code item starts so we can pick up the length
3110     next time round. */
3111    
3112     last_code = code;
3113     }
3114    
3115     /* In the real compile phase, just check the workspace used by the forward
3116     reference list. */
3117    
3118 ph10 505 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3119 nigel 93 {
3120     *errorcodeptr = ERR52;
3121     goto FAILED;
3122     }
3123    
3124 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
3125    
3126     if (inescq && c != 0)
3127     {
3128 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3129 nigel 77 {
3130     inescq = FALSE;
3131     ptr++;
3132     continue;
3133     }
3134     else
3135     {
3136     if (previous_callout != NULL)
3137     {
3138 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3139     complete_callout(previous_callout, ptr, cd);
3140 nigel 77 previous_callout = NULL;
3141     }
3142     if ((options & PCRE_AUTO_CALLOUT) != 0)
3143     {
3144     previous_callout = code;
3145     code = auto_callout(code, ptr, cd);
3146     }
3147     goto NORMAL_CHAR;
3148     }
3149     }
3150    
3151     /* Fill in length of a previous callout, except when the next thing is
3152     a quantifier. */
3153    
3154 ph10 392 is_quantifier =
3155 ph10 391 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3156     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3157 nigel 77
3158     if (!is_quantifier && previous_callout != NULL &&
3159     after_manual_callout-- <= 0)
3160     {
3161 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3162     complete_callout(previous_callout, ptr, cd);
3163 nigel 77 previous_callout = NULL;
3164     }
3165    
3166     /* In extended mode, skip white space and comments */
3167    
3168     if ((options & PCRE_EXTENDED) != 0)
3169     {
3170     if ((cd->ctypes[c] & ctype_space) != 0) continue;
3171 ph10 391 if (c == CHAR_NUMBER_SIGN)
3172 nigel 77 {
3173 ph10 579 ptr++;
3174 ph10 556 while (*ptr != 0)
3175 nigel 91 {
3176 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3177 ph10 556 ptr++;
3178 ph10 579 #ifdef SUPPORT_UTF8
3179 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3180     #endif
3181 nigel 91 }
3182 nigel 93 if (*ptr != 0) continue;
3183    
3184 nigel 91 /* Else fall through to handle end of string */
3185     c = 0;
3186 nigel 77 }
3187     }
3188    
3189     /* No auto callout for quantifiers. */
3190    
3191     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3192     {
3193     previous_callout = code;
3194     code = auto_callout(code, ptr, cd);
3195     }
3196    
3197     switch(c)
3198     {
3199 nigel 93 /* ===================================================================*/
3200     case 0: /* The branch terminates at string end */
3201 ph10 391 case CHAR_VERTICAL_LINE: /* or | or ) */
3202     case CHAR_RIGHT_PARENTHESIS:
3203 nigel 77 *firstbyteptr = firstbyte;
3204     *reqbyteptr = reqbyte;
3205     *codeptr = code;
3206     *ptrptr = ptr;
3207 nigel 93 if (lengthptr != NULL)
3208     {
3209 ph10 202 if (OFLOW_MAX - *lengthptr < code - last_code)
3210     {
3211     *errorcodeptr = ERR20;
3212     goto FAILED;
3213     }
3214 ph10 530 *lengthptr += (int)(code - last_code); /* To include callout length */
3215 nigel 93 DPRINTF((">> end branch\n"));
3216     }
3217 nigel 77 return TRUE;
3218    
3219 nigel 93
3220     /* ===================================================================*/
3221 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
3222     the setting of any following char as a first character. */
3223    
3224 ph10 391 case CHAR_CIRCUMFLEX_ACCENT:
3225 nigel 77 if ((options & PCRE_MULTILINE) != 0)
3226     {
3227     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3228     }
3229     previous = NULL;
3230     *code++ = OP_CIRC;
3231     break;
3232    
3233 ph10 391 case CHAR_DOLLAR_SIGN:
3234 nigel 77 previous = NULL;
3235     *code++ = OP_DOLL;
3236     break;
3237    
3238     /* There can never be a first char if '.' is first, whatever happens about
3239     repeats. The value of reqbyte doesn't change either. */
3240    
3241 ph10 391 case CHAR_DOT:
3242 nigel 77 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3243     zerofirstbyte = firstbyte;
3244     zeroreqbyte = reqbyte;
3245     previous = code;
3246 ph10 342 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3247 nigel 77 break;
3248    
3249 nigel 93
3250     /* ===================================================================*/
3251 nigel 87 /* Character classes. If the included characters are all < 256, we build a
3252     32-byte bitmap of the permitted characters, except in the special case
3253     where there is only one such character. For negated classes, we build the
3254     map as usual, then invert it at the end. However, we use a different opcode
3255     so that data characters > 255 can be handled correctly.
3256 nigel 77
3257     If the class contains characters outside the 0-255 range, a different
3258     opcode is compiled. It may optionally have a bit map for characters < 256,
3259     but those above are are explicitly listed afterwards. A flag byte tells
3260     whether the bitmap is present, and whether this is a negated class or not.
3261 ph10 345
3262 ph10 336 In JavaScript compatibility mode, an isolated ']' causes an error. In
3263     default (Perl) mode, it is treated as a data character. */
3264 ph10 345
3265 ph10 391 case CHAR_RIGHT_SQUARE_BRACKET:
3266 ph10 336 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3267     {
3268     *errorcodeptr = ERR64;
3269 ph10 345 goto FAILED;
3270 ph10 336 }
3271 ph10 345 goto NORMAL_CHAR;
3272 nigel 77
3273 ph10 391 case CHAR_LEFT_SQUARE_BRACKET:
3274 nigel 77 previous = code;
3275    
3276     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3277     they are encountered at the top level, so we'll do that too. */
3278    
3279 ph10 392 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3280 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) &&
3281 ph10 295 check_posix_syntax(ptr, &tempptr))
3282 nigel 77 {
3283 ph10 391 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3284 nigel 77 goto FAILED;
3285     }
3286    
3287 ph10 205 /* If the first character is '^', set the negation flag and skip it. Also,
3288 ph10 208 if the first few characters (either before or after ^) are \Q\E or \E we
3289 ph10 205 skip them too. This makes for compatibility with Perl. */
3290 ph10 208
3291 ph10 205 negate_class = FALSE;
3292     for (;;)
3293 nigel 77 {
3294     c = *(++ptr);
3295 ph10 391 if (c == CHAR_BACKSLASH)
3296 ph10 205 {
3297 ph10 392 if (ptr[1] == CHAR_E)
3298 ph10 391 ptr++;
3299 ph10 392 else if (strncmp((const char *)ptr+1,
3300     STR_Q STR_BACKSLASH STR_E, 3) == 0)
3301 ph10 391 ptr += 3;
3302 ph10 392 else
3303 ph10 391 break;
3304 ph10 205 }
3305 ph10 391 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3306 ph10 205 negate_class = TRUE;
3307     else break;
3308 ph10 208 }
3309 ph10 345
3310     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3311     an initial ']' is taken as a data character -- the code below handles
3312 ph10 341 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3313     [^] must match any character, so generate OP_ALLANY. */
3314 ph10 345
3315 ph10 392 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3316 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3317 ph10 341 {
3318     *code++ = negate_class? OP_ALLANY : OP_FAIL;
3319     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3320     zerofirstbyte = firstbyte;
3321     break;
3322 ph10 345 }
3323 nigel 77
3324 ph10 286 /* If a class contains a negative special such as \S, we need to flip the
3325     negation flag at the end, so that support for characters > 255 works
3326 ph10 264 correctly (they are all included in the class). */
3327    
3328     should_flip_negation = FALSE;
3329    
3330 nigel 77 /* Keep a count of chars with values < 256 so that we can optimize the case
3331 nigel 93 of just a single character (as long as it's < 256). However, For higher
3332     valued UTF-8 characters, we don't yet do any optimization. */
3333 nigel 77
3334     class_charcount = 0;
3335     class_lastchar = -1;
3336    
3337 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
3338     temporary bit of memory, in case the class contains only 1 character (less
3339     than 256), because in that case the compiled code doesn't use the bit map.
3340     */
3341    
3342     memset(classbits, 0, 32 * sizeof(uschar));
3343    
3344 nigel 77 #ifdef SUPPORT_UTF8
3345     class_utf8 = FALSE; /* No chars >= 256 */
3346 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3347 ph10 309 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3348 nigel 77 #endif
3349    
3350     /* Process characters until ] is reached. By writing this as a "do" it
3351 nigel 93 means that an initial ] is taken as a data character. At the start of the
3352     loop, c contains the first byte of the character. */
3353 nigel 77
3354 nigel 93 if (c != 0) do
3355 nigel 77 {
3356 nigel 93 const uschar *oldptr;
3357    
3358 nigel 77 #ifdef SUPPORT_UTF8
3359     if (utf8 && c > 127)
3360     { /* Braces are required because the */
3361     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3362     }
3363 ph10 535
3364 ph10 300 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3365 ph10 309 data and reset the pointer. This is so that very large classes that
3366 ph10 300 contain a zillion UTF-8 characters no longer overwrite the work space
3367 ph10 309 (which is on the stack). */
3368    
3369 ph10 300 if (lengthptr != NULL)
3370     {
3371     *lengthptr += class_utf8data - class_utf8data_base;
3372 ph10 309 class_utf8data = class_utf8data_base;
3373     }
3374    
3375 nigel 77 #endif
3376    
3377     /* Inside \Q...\E everything is literal except \E */
3378    
3379     if (inescq)
3380     {
3381 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3382 nigel 77 {
3383 nigel 93 inescq = FALSE; /* Reset literal state */
3384     ptr++; /* Skip the 'E' */
3385     continue; /* Carry on with next */
3386 nigel 77 }
3387 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
3388 nigel 77 }
3389    
3390     /* Handle POSIX class names. Perl allows a negation extension of the
3391     form [:^name:]. A square bracket that doesn't match the syntax is
3392     treated as a literal. We also recognize the POSIX constructions
3393     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3394     5.6 and 5.8 do. */
3395    
3396 ph10 391 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3397 ph10 392 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3398 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3399 nigel 77 {
3400     BOOL local_negate = FALSE;
3401 nigel 87 int posix_class, taboffset, tabopt;
3402 nigel 77 register const uschar *cbits = cd->cbits;
3403 nigel 87 uschar pbits[32];
3404 nigel 77
3405 ph10 391 if (ptr[1] != CHAR_COLON)
3406 nigel 77 {
3407     *errorcodeptr = ERR31;
3408     goto FAILED;
3409     }
3410    
3411     ptr += 2;
3412 ph10 391 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3413 nigel 77 {
3414     local_negate = TRUE;
3415 ph10 286 should_flip_negation = TRUE; /* Note negative special */
3416 nigel 77 ptr++;
3417     }
3418    
3419 ph10 530 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3420 nigel 77 if (posix_class < 0)
3421     {
3422     *errorcodeptr = ERR30;
3423     goto FAILED;
3424     }
3425    
3426     /* If matching is caseless, upper and lower are converted to
3427     alpha. This relies on the fact that the class table starts with
3428     alpha, lower, upper as the first 3 entries. */
3429    
3430     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3431     posix_class = 0;
3432 ph10 535
3433     /* When PCRE_UCP is set, some of the POSIX classes are converted to
3434 ph10 518 different escape sequences that use Unicode properties. */
3435 ph10 535
3436 ph10 518 #ifdef SUPPORT_UCP
3437     if ((options & PCRE_UCP) != 0)
3438     {
3439     int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3440     if (posix_substitutes[pc] != NULL)
3441     {
3442 ph10 535 nestptr = tempptr + 1;
3443 ph10 518 ptr = posix_substitutes[pc] - 1;
3444 ph10 535 continue;
3445     }
3446     }
3447     #endif
3448 ph10 518 /* In the non-UCP case, we build the bit map for the POSIX class in a
3449     chunk of local store because we may be adding and subtracting from it,
3450     and we don't want to subtract bits that may be in the main map already.
3451     At the end we or the result into the bit map that is being built. */
3452 nigel 77
3453     posix_class *= 3;
3454 nigel 87
3455     /* Copy in the first table (always present) */
3456    
3457     memcpy(pbits, cbits + posix_class_maps[posix_class],
3458     32 * sizeof(uschar));
3459    
3460     /* If there is a second table, add or remove it as required. */
3461    
3462     taboffset = posix_class_maps[posix_class + 1];
3463     tabopt = posix_class_maps[posix_class + 2];
3464    
3465     if (taboffset >= 0)
3466 nigel 77 {
3467 nigel 87 if (tabopt >= 0)
3468     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3469 nigel 77 else
3470 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3471 nigel 77 }
3472    
3473 nigel 87 /* Not see if we need to remove any special characters. An option
3474     value of 1 removes vertical space and 2 removes underscore. */
3475    
3476     if (tabopt < 0) tabopt = -tabopt;
3477     if (tabopt == 1) pbits[1] &= ~0x3c;
3478     else if (tabopt == 2) pbits[11] &= 0x7f;
3479    
3480     /* Add the POSIX table or its complement into the main table that is
3481     being built and we are done. */
3482    
3483     if (local_negate)
3484     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3485     else
3486     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3487    
3488 nigel 77 ptr = tempptr + 1;
3489     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3490     continue; /* End of POSIX syntax handling */
3491     }
3492    
3493     /* Backslash may introduce a single character, or it may introduce one
3494 nigel 93 of the specials, which just set a flag. The sequence \b is a special
3495 ph10 513 case. Inside a class (and only there) it is treated as backspace. We
3496     assume that other escapes have more than one character in them, so set
3497     class_charcount bigger than one. Unrecognized escapes fall through and
3498     are either treated as literal characters (by default), or are faulted if
3499     PCRE_EXTRA is set. */
3500 nigel 77
3501 ph10 391 if (c == CHAR_BACKSLASH)
3502 nigel 77 {
3503 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3504     if (*errorcodeptr != 0) goto FAILED;
3505 nigel 77
3506 ph10 513 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3507 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
3508     {
3509 ph10 391 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3510 nigel 77 {
3511     ptr += 2; /* avoid empty string */
3512     }
3513     else inescq = TRUE;
3514     continue;
3515     }
3516 ph10 220 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3517 nigel 77
3518     if (c < 0)
3519     {
3520     register const uschar *cbits = cd->cbits;
3521     class_charcount += 2; /* Greater than 1 is what matters */
3522 nigel 93
3523 ph10 518 switch (-c)
3524 nigel 77 {
3525 ph10 518 #ifdef SUPPORT_UCP
3526     case ESC_du: /* These are the values given for \d etc */
3527     case ESC_DU: /* when PCRE_UCP is set. We replace the */
3528     case ESC_wu: /* escape sequence with an appropriate \p */
3529     case ESC_WU: /* or \P to test Unicode properties instead */
3530     case ESC_su: /* of the default ASCII testing. */
3531     case ESC_SU:
3532     nestptr = ptr;
3533     ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3534 ph10 535 class_charcount -= 2; /* Undo! */
3535 ph10 518 continue;
3536     #endif
3537 nigel 77 case ESC_d:
3538     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3539     continue;
3540    
3541     case ESC_D:
3542 ph10 286 should_flip_negation = TRUE;
3543 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3544     continue;
3545    
3546     case ESC_w:
3547     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3548     continue;
3549    
3550     case ESC_W:
3551 ph10 286 should_flip_negation = TRUE;
3552 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3553     continue;
3554    
3555 ph10 552 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3556 ph10 579 if it was previously set by something earlier in the character
3557     class. */
3558 ph10 552
3559 nigel 77 case ESC_s:
3560 ph10 552 classbits[0] |= cbits[cbit_space];
3561 ph10 579 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3562 ph10 552 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3563 nigel 77 continue;
3564    
3565     case ESC_S:
3566 ph10 286 should_flip_negation = TRUE;
3567 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3568     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3569     continue;
3570    
3571 ph10 518 case ESC_h:
3572 ph10 178 SETBIT(classbits, 0x09); /* VT */
3573     SETBIT(classbits, 0x20); /* SPACE */
3574 ph10 180 SETBIT(classbits, 0xa0); /* NSBP */
3575 ph10 178 #ifdef SUPPORT_UTF8
3576     if (utf8)
3577 ph10 180 {
3578 ph10 178 class_utf8 = TRUE;
3579     *class_utf8data++ = XCL_SINGLE;
3580 ph10 180 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3581 ph10 178 *class_utf8data++ = XCL_SINGLE;
3582 ph10 180 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3583     *class_utf8data++ = XCL_RANGE;
3584     class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3585     class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3586 ph10 178 *class_utf8data++ = XCL_SINGLE;
3587 ph10 180 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3588 ph10 178 *class_utf8data++ = XCL_SINGLE;
3589 ph10 180 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3590 ph10 178 *class_utf8data++ = XCL_SINGLE;
3591 ph10 180 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3592     }
3593     #endif
3594     continue;
3595 nigel 93
3596 ph10 518 case ESC_H:
3597 ph10 178 for (c = 0; c < 32; c++)
3598     {
3599     int x = 0xff;
3600     switch (c)
3601 ph10 180 {
3602 ph10 178 case 0x09/8: x ^= 1 << (0x09%8); break;
3603     case 0x20/8: x ^= 1 << (0x20%8); break;
3604     case 0xa0/8: x ^= 1 << (0xa0%8); break;
3605     default: break;
3606     }
3607     classbits[c] |= x;
3608 ph10 180 }
3609    
3610 ph10 178 #ifdef SUPPORT_UTF8
3611     if (utf8)
3612 ph10 180 {
3613 ph10 178 class_utf8 = TRUE;
3614 ph10 180 *class_utf8data++ = XCL_RANGE;
3615     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3616     class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3617     *class_utf8data++ = XCL_RANGE;
3618     class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3619     class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3620     *class_utf8data++ = XCL_RANGE;
3621     class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3622     class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3623     *class_utf8data++ = XCL_RANGE;
3624     class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3625     class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3626     *class_utf8data++ = XCL_RANGE;
3627     class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3628     class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3629     *class_utf8data++ = XCL_RANGE;
3630     class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3631     class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3632     *class_utf8data++ = XCL_RANGE;
3633     class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3634     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3635     }
3636     #endif
3637     continue;
3638 ph10 178
3639 ph10 518 case ESC_v:
3640 ph10 178 SETBIT(classbits, 0x0a); /* LF */
3641     SETBIT(classbits, 0x0b); /* VT */
3642 ph10 180 SETBIT(classbits, 0x0c); /* FF */
3643     SETBIT(classbits, 0x0d); /* CR */
3644     SETBIT(classbits, 0x85); /* NEL */
3645 ph10 178 #ifdef SUPPORT_UTF8
3646     if (utf8)
3647 ph10 180 {
3648 ph10 178 class_utf8 = TRUE;
3649 ph10 180 *class_utf8data++ = XCL_RANGE;
3650     class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3651     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3652     }
3653     #endif
3654     continue;
3655 ph10 178
3656 ph10 518 case ESC_V:
3657 ph10 178 for (c = 0; c < 32; c++)
3658     {
3659     int x = 0xff;
3660     switch (c)
3661 ph10 180 {
3662 ph10 178 case 0x0a/8: x ^= 1 << (0x0a%8);
3663     x ^= 1 << (0x0b%8);
3664     x ^= 1 << (0x0c%8);
3665 ph10 180 x ^= 1 << (0x0d%8);
3666 ph10 178 break;
3667     case 0x85/8: x ^= 1 << (0x85%8); break;
3668     default: break;
3669     }
3670     classbits[c] |= x;
3671 ph10 180 }
3672    
3673 ph10 178 #ifdef SUPPORT_UTF8
3674     if (utf8)
3675 ph10 180 {
3676 ph10 178 class_utf8 = TRUE;
3677 ph10 180 *class_utf8data++ = XCL_RANGE;
3678     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3679     class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3680     *class_utf8data++ = XCL_RANGE;
3681     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3682     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3683     }
3684     #endif
3685     continue;
3686 ph10 178
3687 nigel 77 #ifdef SUPPORT_UCP
3688 ph10 518 case ESC_p:
3689     case ESC_P:
3690     {
3691     BOOL negated;
3692     int pdata;
3693     int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3694     if (ptype < 0) goto FAILED;
3695     class_utf8 = TRUE;
3696     *class_utf8data++ = ((-c == ESC_p) != negated)?
3697     XCL_PROP : XCL_NOTPROP;
3698     *class_utf8data++ = ptype;
3699     *class_utf8data++ = pdata;
3700     class_charcount -= 2; /* Not a < 256 character */
3701     continue;
3702     }
3703 nigel 77 #endif
3704 ph10 518 /* Unrecognized escapes are faulted if PCRE is running in its
3705     strict mode. By default, for compatibility with Perl, they are
3706     treated as literals. */
3707 nigel 77
3708 ph10 518 default:
3709     if ((options & PCRE_EXTRA) != 0)
3710     {
3711     *errorcodeptr = ERR7;
3712     goto FAILED;
3713     }
3714     class_charcount -= 2; /* Undo the default count from above */
3715     c = *ptr; /* Get the final character and fall through */
3716     break;
3717 nigel 93 }
3718 nigel 77 }
3719    
3720     /* Fall through if we have a single character (c >= 0). This may be
3721 nigel 93 greater than 256 in UTF-8 mode. */
3722 nigel 77
3723     } /* End of backslash handling */
3724    
3725     /* A single character may be followed by '-' to form a range. However,
3726     Perl does not permit ']' to be the end of the range. A '-' character
3727 nigel 93 at the end is treated as a literal. Perl ignores orphaned \E sequences
3728     entirely. The code for handling \Q and \E is messy. */
3729 nigel 77
3730 nigel 93 CHECK_RANGE:
3731 ph10 391 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3732 nigel 77 {
3733 nigel 93 inescq = FALSE;
3734     ptr += 2;
3735     }
3736    
3737     oldptr = ptr;
3738 ph10 231
3739 ph10 230 /* Remember \r or \n */
3740 ph10 231
3741 ph10 391 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3742 ph10 231
3743 ph10 230 /* Check for range */
3744 nigel 93
3745 ph10 391 if (!inescq && ptr[1] == CHAR_MINUS)
3746 nigel 93 {
3747 nigel 77 int d;
3748     ptr += 2;
3749 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3750 nigel 77
3751 nigel 93 /* If we hit \Q (not followed by \E) at this point, go into escaped
3752     mode. */
3753    
3754 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3755 nigel 93 {
3756     ptr += 2;
3757 ph10 392 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3758 ph10 391 { ptr += 2; continue; }
3759 nigel 93 inescq = TRUE;
3760     break;
3761     }
3762    
3763 ph10 391 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3764 nigel 93 {
3765     ptr = oldptr;
3766     goto LONE_SINGLE_CHARACTER;
3767     }
3768    
3769 nigel 77 #ifdef SUPPORT_UTF8
3770     if (utf8)
3771     { /* Braces are required because the */
3772     GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3773     }
3774     else
3775     #endif
3776     d = *ptr; /* Not UTF-8 mode */
3777    
3778     /* The second part of a range can be a single-charact