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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 624 - (hide annotations) (download)
Tue Jul 19 10:43:28 2011 UTC (3 years, 3 months ago) by ph10
File MIME type: text/plain
File size: 240983 byte(s)
Fix unset variable bug introduced during recursion refactoring.

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 604 compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int *,
549     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 ph10 602 that do not influence this. For some calls, it makes sense to skip negative
1401     forward and all backward assertions, and also the \b assertion; for others it
1402     does not.
1403 nigel 77
1404     Arguments:
1405     code pointer to the start of the group
1406     skipassert TRUE if certain assertions are to be skipped
1407    
1408     Returns: pointer to the first significant opcode
1409     */
1410    
1411     static const uschar*
1412 ph10 604 first_significant_code(const uschar *code, BOOL skipassert)
1413 nigel 77 {
1414     for (;;)
1415     {
1416     switch ((int)*code)
1417     {
1418     case OP_ASSERT_NOT:
1419     case OP_ASSERTBACK:
1420     case OP_ASSERTBACK_NOT:
1421     if (!skipassert) return code;
1422     do code += GET(code, 1); while (*code == OP_ALT);
1423     code += _pcre_OP_lengths[*code];
1424     break;
1425    
1426     case OP_WORD_BOUNDARY:
1427     case OP_NOT_WORD_BOUNDARY:
1428     if (!skipassert) return code;
1429     /* Fall through */
1430    
1431     case OP_CALLOUT:
1432     case OP_CREF:
1433 ph10 459 case OP_NCREF:
1434 nigel 93 case OP_RREF:
1435 ph10 459 case OP_NRREF:
1436 nigel 93 case OP_DEF:
1437 nigel 77 code += _pcre_OP_lengths[*code];
1438     break;
1439    
1440     default:
1441     return code;
1442     }
1443     }
1444     /* Control never reaches here */
1445     }
1446    
1447    
1448    
1449    
1450     /*************************************************
1451 ph10 454 * Find the fixed length of a branch *
1452 nigel 77 *************************************************/
1453    
1454 ph10 454 /* Scan a branch and compute the fixed length of subject that will match it,
1455 nigel 77 if the length is fixed. This is needed for dealing with backward assertions.
1456 ph10 461 In UTF8 mode, the result is in characters rather than bytes. The branch is
1457 ph10 454 temporarily terminated with OP_END when this function is called.
1458 nigel 77
1459 ph10 461 This function is called when a backward assertion is encountered, so that if it
1460     fails, the error message can point to the correct place in the pattern.
1461 ph10 454 However, we cannot do this when the assertion contains subroutine calls,
1462 ph10 461 because they can be forward references. We solve this by remembering this case
1463 ph10 454 and doing the check at the end; a flag specifies which mode we are running in.
1464    
1465 nigel 77 Arguments:
1466     code points to the start of the pattern (the bracket)
1467 ph10 604 utf8 TRUE in UTF-8 mode
1468 ph10 461 atend TRUE if called when the pattern is complete
1469     cd the "compile data" structure
1470 nigel 77
1471 ph10 461 Returns: the fixed length,
1472 ph10 454 or -1 if there is no fixed length,
1473 nigel 77 or -2 if \C was encountered
1474 ph10 454 or -3 if an OP_RECURSE item was encountered and atend is FALSE
1475 nigel 77 */
1476    
1477     static int
1478 ph10 604 find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd)
1479 nigel 77 {
1480     int length = -1;
1481    
1482     register int branchlength = 0;
1483     register uschar *cc = code + 1 + LINK_SIZE;
1484    
1485     /* Scan along the opcodes for this branch. If we get to the end of the
1486     branch, check the length against that of the other branches. */
1487    
1488     for (;;)
1489     {
1490     int d;
1491 ph10 454 uschar *ce, *cs;
1492 nigel 77 register int op = *cc;
1493     switch (op)
1494     {
1495 ph10 604 /* We only need to continue for OP_CBRA (normal capturing bracket) and
1496     OP_BRA (normal non-capturing bracket) because the other variants of these
1497     opcodes are all concerned with unlimited repeated groups, which of course
1498     are not of fixed length. They will cause a -1 response from the default
1499     case of this switch. */
1500    
1501 nigel 93 case OP_CBRA:
1502 nigel 77 case OP_BRA:
1503     case OP_ONCE:
1504     case OP_COND:
1505 ph10 604 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd);
1506 nigel 77 if (d < 0) return d;
1507     branchlength += d;
1508     do cc += GET(cc, 1); while (*cc == OP_ALT);
1509     cc += 1 + LINK_SIZE;
1510     break;
1511    
1512     /* Reached end of a branch; if it's a ket it is the end of a nested
1513     call. If it's ALT it is an alternation in a nested call. If it is
1514 ph10 604 END it's the end of the outer call. All can be handled by the same code.
1515     Note that we must not include the OP_KETRxxx opcodes here, because they
1516     all imply an unlimited repeat. */
1517 nigel 77
1518     case OP_ALT:
1519     case OP_KET:
1520     case OP_END:
1521     if (length < 0) length = branchlength;
1522     else if (length != branchlength) return -1;
1523     if (*cc != OP_ALT) return length;
1524     cc += 1 + LINK_SIZE;
1525     branchlength = 0;
1526     break;
1527 ph10 461
1528 ph10 454 /* A true recursion implies not fixed length, but a subroutine call may
1529     be OK. If the subroutine is a forward reference, we can't deal with
1530     it until the end of the pattern, so return -3. */
1531 ph10 461
1532 ph10 454 case OP_RECURSE:
1533     if (!atend) return -3;
1534     cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
1535     do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
1536     if (cc > cs && cc < ce) return -1; /* Recursion */
1537 ph10 604 d = find_fixedlength(cs + 2, utf8, atend, cd);
1538 ph10 461 if (d < 0) return d;
1539 ph10 454 branchlength += d;
1540     cc += 1 + LINK_SIZE;
1541 ph10 461 break;
1542 nigel 77
1543     /* Skip over assertive subpatterns */
1544    
1545     case OP_ASSERT:
1546     case OP_ASSERT_NOT:
1547     case OP_ASSERTBACK:
1548     case OP_ASSERTBACK_NOT:
1549     do cc += GET(cc, 1); while (*cc == OP_ALT);
1550     /* Fall through */
1551    
1552     /* Skip over things that don't match chars */
1553    
1554     case OP_REVERSE:
1555     case OP_CREF:
1556 ph10 459 case OP_NCREF:
1557 nigel 93 case OP_RREF:
1558 ph10 459 case OP_NRREF:
1559 nigel 93 case OP_DEF:
1560 nigel 77 case OP_CALLOUT:
1561     case OP_SOD:
1562     case OP_SOM:
1563 ph10 500 case OP_SET_SOM:
1564 nigel 77 case OP_EOD:
1565     case OP_EODN:
1566     case OP_CIRC:
1567 ph10 602 case OP_CIRCM:
1568 nigel 77 case OP_DOLL:
1569 ph10 602 case OP_DOLLM:
1570 nigel 77 case OP_NOT_WORD_BOUNDARY:
1571     case OP_WORD_BOUNDARY:
1572     cc += _pcre_OP_lengths[*cc];
1573     break;
1574    
1575     /* Handle literal characters */
1576    
1577     case OP_CHAR:
1578 ph10 602 case OP_CHARI:
1579 nigel 91 case OP_NOT:
1580 ph10 604 case OP_NOTI:
1581 nigel 77 branchlength++;
1582     cc += 2;
1583     #ifdef SUPPORT_UTF8
1584 ph10 604 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1585 nigel 77 #endif
1586     break;
1587    
1588     /* Handle exact repetitions. The count is already in characters, but we
1589     need to skip over a multibyte character in UTF8 mode. */
1590    
1591     case OP_EXACT:
1592     branchlength += GET2(cc,1);
1593     cc += 4;
1594     #ifdef SUPPORT_UTF8
1595 ph10 604 if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1596 nigel 77 #endif
1597     break;
1598    
1599     case OP_TYPEEXACT:
1600     branchlength += GET2(cc,1);
1601 ph10 220 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1602 nigel 77 cc += 4;
1603     break;
1604    
1605     /* Handle single-char matchers */
1606    
1607     case OP_PROP:
1608     case OP_NOTPROP:
1609 nigel 87 cc += 2;
1610 nigel 77 /* Fall through */
1611    
1612     case OP_NOT_DIGIT:
1613     case OP_DIGIT:
1614     case OP_NOT_WHITESPACE:
1615     case OP_WHITESPACE:
1616     case OP_NOT_WORDCHAR:
1617     case OP_WORDCHAR:
1618     case OP_ANY:
1619 ph10 342 case OP_ALLANY:
1620 nigel 77 branchlength++;
1621     cc++;
1622     break;
1623    
1624     /* The single-byte matcher isn't allowed */
1625    
1626     case OP_ANYBYTE:
1627     return -2;
1628    
1629     /* Check a class for variable quantification */
1630    
1631     #ifdef SUPPORT_UTF8
1632     case OP_XCLASS:
1633     cc += GET(cc, 1) - 33;
1634     /* Fall through */
1635     #endif
1636    
1637     case OP_CLASS:
1638     case OP_NCLASS:
1639     cc += 33;
1640    
1641     switch (*cc)
1642     {
1643     case OP_CRSTAR:
1644     case OP_CRMINSTAR:
1645     case OP_CRQUERY:
1646     case OP_CRMINQUERY:
1647     return -1;
1648    
1649     case OP_CRRANGE:
1650     case OP_CRMINRANGE:
1651     if (GET2(cc,1) != GET2(cc,3)) return -1;
1652     branchlength += GET2(cc,1);
1653     cc += 5;
1654     break;
1655    
1656     default:
1657     branchlength++;
1658     }
1659     break;
1660    
1661     /* Anything else is variable length */
1662    
1663     default:
1664     return -1;
1665     }
1666     }
1667     /* Control never gets here */
1668     }
1669    
1670    
1671    
1672    
1673     /*************************************************
1674 ph10 454 * Scan compiled regex for specific bracket *
1675 nigel 77 *************************************************/
1676    
1677     /* This little function scans through a compiled pattern until it finds a
1678 ph10 454 capturing bracket with the given number, or, if the number is negative, an
1679 ph10 461 instance of OP_REVERSE for a lookbehind. The function is global in the C sense
1680     so that it can be called from pcre_study() when finding the minimum matching
1681 ph10 455 length.
1682 nigel 77
1683     Arguments:
1684     code points to start of expression
1685     utf8 TRUE in UTF-8 mode
1686 ph10 454 number the required bracket number or negative to find a lookbehind
1687 nigel 77
1688     Returns: pointer to the opcode for the bracket, or NULL if not found
1689     */
1690    
1691 ph10 455 const uschar *
1692     _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1693 nigel 77 {
1694     for (;;)
1695     {
1696     register int c = *code;
1697 ph10 618
1698 nigel 77 if (c == OP_END) return NULL;
1699 nigel 91
1700     /* XCLASS is used for classes that cannot be represented just by a bit
1701     map. This includes negated single high-valued characters. The length in
1702     the table is zero; the actual length is stored in the compiled code. */
1703    
1704     if (c == OP_XCLASS) code += GET(code, 1);
1705 ph10 461
1706 ph10 454 /* Handle recursion */
1707 ph10 461
1708 ph10 454 else if (c == OP_REVERSE)
1709     {
1710 ph10 461 if (number < 0) return (uschar *)code;
1711 ph10 454 code += _pcre_OP_lengths[c];
1712     }
1713 nigel 91
1714 nigel 93 /* Handle capturing bracket */
1715 nigel 91
1716 ph10 604 else if (c == OP_CBRA || c == OP_SCBRA ||
1717     c == OP_CBRAPOS || c == OP_SCBRAPOS)
1718 nigel 77 {
1719 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1720 nigel 77 if (n == number) return (uschar *)code;
1721 nigel 93 code += _pcre_OP_lengths[c];
1722 nigel 77 }
1723 nigel 91
1724 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1725     repeated character types, we have to test for \p and \P, which have an extra
1726 ph10 512 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1727 ph10 510 must add in its length. */
1728 nigel 91
1729 nigel 77 else
1730     {
1731 ph10 218 switch(c)
1732     {
1733     case OP_TYPESTAR:
1734     case OP_TYPEMINSTAR:
1735     case OP_TYPEPLUS:
1736     case OP_TYPEMINPLUS:
1737     case OP_TYPEQUERY:
1738     case OP_TYPEMINQUERY:
1739     case OP_TYPEPOSSTAR:
1740     case OP_TYPEPOSPLUS:
1741     case OP_TYPEPOSQUERY:
1742     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1743 ph10 220 break;
1744 ph10 221
1745     case OP_TYPEUPTO:
1746     case OP_TYPEMINUPTO:
1747     case OP_TYPEEXACT:
1748     case OP_TYPEPOSUPTO:
1749     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1750     break;
1751 ph10 512
1752 ph10 510 case OP_MARK:
1753     case OP_PRUNE_ARG:
1754     case OP_SKIP_ARG:
1755     code += code[1];
1756 ph10 512 break;
1757 ph10 550
1758     case OP_THEN_ARG:
1759     code += code[1+LINK_SIZE];
1760     break;
1761 ph10 220 }
1762    
1763 ph10 218 /* Add in the fixed length from the table */
1764 ph10 220
1765 nigel 77 code += _pcre_OP_lengths[c];
1766 ph10 220
1767 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1768     a multi-byte character. The length in the table is a minimum, so we have to
1769     arrange to skip the extra bytes. */
1770 ph10 220
1771 ph10 107 #ifdef SUPPORT_UTF8
1772 nigel 77 if (utf8) switch(c)
1773     {
1774     case OP_CHAR:
1775 ph10 602 case OP_CHARI:
1776 nigel 77 case OP_EXACT:
1777 ph10 602 case OP_EXACTI:
1778 nigel 77 case OP_UPTO:
1779 ph10 602 case OP_UPTOI:
1780 nigel 77 case OP_MINUPTO:
1781 ph10 602 case OP_MINUPTOI:
1782 nigel 93 case OP_POSUPTO:
1783 ph10 602 case OP_POSUPTOI:
1784 nigel 77 case OP_STAR:
1785 ph10 602 case OP_STARI:
1786 nigel 77 case OP_MINSTAR:
1787 ph10 602 case OP_MINSTARI:
1788 nigel 93 case OP_POSSTAR:
1789 ph10 602 case OP_POSSTARI:
1790 nigel 77 case OP_PLUS:
1791 ph10 602 case OP_PLUSI:
1792 nigel 77 case OP_MINPLUS:
1793 ph10 602 case OP_MINPLUSI:
1794 nigel 93 case OP_POSPLUS:
1795 ph10 602 case OP_POSPLUSI:
1796 nigel 77 case OP_QUERY:
1797 ph10 602 case OP_QUERYI:
1798 nigel 77 case OP_MINQUERY:
1799 ph10 602 case OP_MINQUERYI:
1800 nigel 93 case OP_POSQUERY:
1801 ph10 602 case OP_POSQUERYI:
1802 nigel 93 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1803 nigel 77 break;
1804     }
1805 ph10 369 #else
1806     (void)(utf8); /* Keep compiler happy by referencing function argument */
1807 ph10 111 #endif
1808 nigel 77 }
1809     }
1810     }
1811    
1812    
1813    
1814     /*************************************************
1815     * Scan compiled regex for recursion reference *
1816     *************************************************/
1817    
1818     /* This little function scans through a compiled pattern until it finds an
1819     instance of OP_RECURSE.
1820    
1821     Arguments:
1822     code points to start of expression
1823     utf8 TRUE in UTF-8 mode
1824    
1825     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1826     */
1827    
1828     static const uschar *
1829     find_recurse(const uschar *code, BOOL utf8)
1830     {
1831     for (;;)
1832     {
1833     register int c = *code;
1834     if (c == OP_END) return NULL;
1835 nigel 91 if (c == OP_RECURSE) return code;
1836 ph10 220
1837 nigel 91 /* XCLASS is used for classes that cannot be represented just by a bit
1838     map. This includes negated single high-valued characters. The length in
1839     the table is zero; the actual length is stored in the compiled code. */
1840    
1841     if (c == OP_XCLASS) code += GET(code, 1);
1842    
1843 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1844     repeated character types, we have to test for \p and \P, which have an extra
1845 ph10 512 two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
1846 ph10 510 must add in its length. */
1847 nigel 91
1848 nigel 77 else
1849     {
1850 ph10 218 switch(c)
1851     {
1852     case OP_TYPESTAR:
1853     case OP_TYPEMINSTAR:
1854     case OP_TYPEPLUS:
1855     case OP_TYPEMINPLUS:
1856     case OP_TYPEQUERY:
1857     case OP_TYPEMINQUERY:
1858     case OP_TYPEPOSSTAR:
1859     case OP_TYPEPOSPLUS:
1860     case OP_TYPEPOSQUERY:
1861     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1862 ph10 220 break;
1863 ph10 221
1864     case OP_TYPEPOSUPTO:
1865     case OP_TYPEUPTO:
1866     case OP_TYPEMINUPTO:
1867     case OP_TYPEEXACT:
1868     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1869     break;
1870 ph10 512
1871 ph10 510 case OP_MARK:
1872     case OP_PRUNE_ARG:
1873     case OP_SKIP_ARG:
1874     code += code[1];
1875 ph10 512 break;
1876 ph10 550
1877     case OP_THEN_ARG:
1878     code += code[1+LINK_SIZE];
1879     break;
1880 ph10 220 }
1881    
1882 ph10 218 /* Add in the fixed length from the table */
1883    
1884 nigel 77 code += _pcre_OP_lengths[c];
1885 ph10 220
1886 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1887     by a multi-byte character. The length in the table is a minimum, so we have
1888     to arrange to skip the extra bytes. */
1889 ph10 220
1890 ph10 107 #ifdef SUPPORT_UTF8
1891 nigel 77 if (utf8) switch(c)
1892     {
1893     case OP_CHAR:
1894 ph10 602 case OP_CHARI:
1895 nigel 77 case OP_EXACT:
1896 ph10 602 case OP_EXACTI:
1897 nigel 77 case OP_UPTO:
1898 ph10 602 case OP_UPTOI:
1899 nigel 77 case OP_MINUPTO:
1900 ph10 602 case OP_MINUPTOI:
1901 nigel 93 case OP_POSUPTO:
1902 ph10 602 case OP_POSUPTOI:
1903 nigel 77 case OP_STAR:
1904 ph10 602 case OP_STARI:
1905 nigel 77 case OP_MINSTAR:
1906 ph10 602 case OP_MINSTARI:
1907 nigel 93 case OP_POSSTAR:
1908 ph10 602 case OP_POSSTARI:
1909 nigel 77 case OP_PLUS:
1910 ph10 602 case OP_PLUSI:
1911 nigel 77 case OP_MINPLUS:
1912 ph10 602 case OP_MINPLUSI:
1913 nigel 93 case OP_POSPLUS:
1914 ph10 602 case OP_POSPLUSI:
1915 nigel 77 case OP_QUERY:
1916 ph10 602 case OP_QUERYI:
1917 nigel 77 case OP_MINQUERY:
1918 ph10 602 case OP_MINQUERYI:
1919 nigel 93 case OP_POSQUERY:
1920 ph10 602 case OP_POSQUERYI:
1921 nigel 93 if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1922 nigel 77 break;
1923     }
1924 ph10 369 #else
1925     (void)(utf8); /* Keep compiler happy by referencing function argument */
1926 ph10 111 #endif
1927 nigel 77 }
1928     }
1929     }
1930    
1931    
1932    
1933     /*************************************************
1934     * Scan compiled branch for non-emptiness *
1935     *************************************************/
1936    
1937     /* This function scans through a branch of a compiled pattern to see whether it
1938 nigel 93 can match the empty string or not. It is called from could_be_empty()
1939     below and from compile_branch() when checking for an unlimited repeat of a
1940     group that can match nothing. Note that first_significant_code() skips over
1941 ph10 282 backward and negative forward assertions when its final argument is TRUE. If we
1942     hit an unclosed bracket, we return "empty" - this means we've struck an inner
1943     bracket whose current branch will already have been scanned.
1944 nigel 77
1945     Arguments:
1946     code points to start of search
1947     endcode points to where to stop
1948     utf8 TRUE if in UTF8 mode
1949 ph10 503 cd contains pointers to tables etc.
1950 nigel 77
1951     Returns: TRUE if what is matched could be empty
1952     */
1953    
1954     static BOOL
1955 ph10 503 could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
1956     compile_data *cd)
1957 nigel 77 {
1958     register int c;
1959 ph10 604 for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE);
1960 nigel 77 code < endcode;
1961 ph10 604 code = first_significant_code(code + _pcre_OP_lengths[c], TRUE))
1962 nigel 77 {
1963     const uschar *ccode;
1964    
1965     c = *code;
1966 ph10 507
1967 ph10 286 /* Skip over forward assertions; the other assertions are skipped by
1968 ph10 282 first_significant_code() with a TRUE final argument. */
1969 ph10 286
1970 ph10 282 if (c == OP_ASSERT)
1971 ph10 286 {
1972 ph10 282 do code += GET(code, 1); while (*code == OP_ALT);
1973     c = *code;
1974     continue;
1975 ph10 286 }
1976 ph10 172
1977 ph10 503 /* For a recursion/subroutine call, if its end has been reached, which
1978 ph10 624 implies a backward reference subroutine call, we can scan it. If it's a
1979     forward reference subroutine call, we can't. To detect forward reference
1980     we have to scan up the list that is kept in the workspace. This function is
1981     called only when doing the real compile, not during the pre-compile that
1982     measures the size of the compiled pattern. */
1983 ph10 507
1984 ph10 503 if (c == OP_RECURSE)
1985     {
1986 ph10 624 const uschar *scode;
1987     BOOL empty_branch;
1988    
1989     /* Test for forward reference */
1990    
1991     for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE)
1992     if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE;
1993    
1994     /* Not a forward reference, test for completed backward reference */
1995    
1996     empty_branch = FALSE;
1997     scode = cd->start_code + GET(code, 1);
1998 ph10 503 if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
1999 ph10 624
2000     /* Completed backwards reference */
2001    
2002 ph10 503 do
2003     {
2004 ph10 504 if (could_be_empty_branch(scode, endcode, utf8, cd))
2005     {
2006     empty_branch = TRUE;
2007 ph10 507 break;
2008     }
2009 ph10 503 scode += GET(scode, 1);
2010     }
2011     while (*scode == OP_ALT);
2012 ph10 624
2013 ph10 504 if (!empty_branch) return FALSE; /* All branches are non-empty */
2014 ph10 503 continue;
2015 ph10 507 }
2016 ph10 170
2017 ph10 604 /* Groups with zero repeats can of course be empty; skip them. */
2018    
2019     if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO ||
2020     c == OP_BRAPOSZERO)
2021     {
2022     code += _pcre_OP_lengths[c];
2023     do code += GET(code, 1); while (*code == OP_ALT);
2024     c = *code;
2025     continue;
2026     }
2027    
2028     /* A nested group that is already marked as "could be empty" can just be
2029     skipped. */
2030    
2031     if (c == OP_SBRA || c == OP_SBRAPOS ||
2032     c == OP_SCBRA || c == OP_SCBRAPOS)
2033     {
2034     do code += GET(code, 1); while (*code == OP_ALT);
2035     c = *code;
2036     continue;
2037     }
2038    
2039 ph10 170 /* For other groups, scan the branches. */
2040 ph10 172
2041 ph10 604 if (c == OP_BRA || c == OP_BRAPOS ||
2042     c == OP_CBRA || c == OP_CBRAPOS ||
2043     c == OP_ONCE || c == OP_COND)
2044 nigel 77 {
2045     BOOL empty_branch;
2046     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
2047 ph10 406
2048     /* If a conditional group has only one branch, there is a second, implied,
2049 ph10 395 empty branch, so just skip over the conditional, because it could be empty.
2050     Otherwise, scan the individual branches of the group. */
2051 ph10 406
2052 ph10 395 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
2053 nigel 77 code += GET(code, 1);
2054 ph10 395 else
2055 ph10 406 {
2056 ph10 395 empty_branch = FALSE;
2057     do
2058     {
2059 ph10 503 if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
2060 ph10 395 empty_branch = TRUE;
2061     code += GET(code, 1);
2062     }
2063     while (*code == OP_ALT);
2064     if (!empty_branch) return FALSE; /* All branches are non-empty */
2065 nigel 77 }
2066 ph10 406
2067 ph10 172 c = *code;
2068 nigel 93 continue;
2069 nigel 77 }
2070    
2071 nigel 93 /* Handle the other opcodes */
2072    
2073     switch (c)
2074 nigel 77 {
2075 ph10 216 /* Check for quantifiers after a class. XCLASS is used for classes that
2076     cannot be represented just by a bit map. This includes negated single
2077     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
2078 ph10 220 actual length is stored in the compiled code, so we must update "code"
2079 ph10 216 here. */
2080 nigel 77
2081     #ifdef SUPPORT_UTF8
2082     case OP_XCLASS:
2083 ph10 216 ccode = code += GET(code, 1);
2084 nigel 77 goto CHECK_CLASS_REPEAT;
2085     #endif
2086    
2087     case OP_CLASS:
2088     case OP_NCLASS:
2089     ccode = code + 33;
2090    
2091     #ifdef SUPPORT_UTF8
2092     CHECK_CLASS_REPEAT:
2093     #endif
2094    
2095     switch (*ccode)
2096     {
2097     case OP_CRSTAR: /* These could be empty; continue */
2098     case OP_CRMINSTAR:
2099     case OP_CRQUERY:
2100     case OP_CRMINQUERY:
2101     break;
2102    
2103     default: /* Non-repeat => class must match */
2104     case OP_CRPLUS: /* These repeats aren't empty */
2105     case OP_CRMINPLUS:
2106     return FALSE;
2107    
2108     case OP_CRRANGE:
2109     case OP_CRMINRANGE:
2110     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2111     break;
2112     }
2113     break;
2114    
2115     /* Opcodes that must match a character */
2116    
2117     case OP_PROP:
2118     case OP_NOTPROP:
2119     case OP_EXTUNI:
2120     case OP_NOT_DIGIT:
2121     case OP_DIGIT:
2122     case OP_NOT_WHITESPACE:
2123     case OP_WHITESPACE:
2124     case OP_NOT_WORDCHAR:
2125     case OP_WORDCHAR:
2126     case OP_ANY:
2127 ph10 345 case OP_ALLANY:
2128 nigel 77 case OP_ANYBYTE:
2129     case OP_CHAR:
2130 ph10 602 case OP_CHARI:
2131 nigel 77 case OP_NOT:
2132 ph10 602 case OP_NOTI:
2133 nigel 77 case OP_PLUS:
2134     case OP_MINPLUS:
2135 nigel 93 case OP_POSPLUS:
2136 nigel 77 case OP_EXACT:
2137     case OP_NOTPLUS:
2138     case OP_NOTMINPLUS:
2139 nigel 93 case OP_NOTPOSPLUS:
2140 nigel 77 case OP_NOTEXACT:
2141     case OP_TYPEPLUS:
2142     case OP_TYPEMINPLUS:
2143 nigel 93 case OP_TYPEPOSPLUS:
2144 nigel 77 case OP_TYPEEXACT:
2145     return FALSE;
2146 ph10 227
2147     /* These are going to continue, as they may be empty, but we have to
2148     fudge the length for the \p and \P cases. */
2149    
2150 ph10 224 case OP_TYPESTAR:
2151     case OP_TYPEMINSTAR:
2152     case OP_TYPEPOSSTAR:
2153     case OP_TYPEQUERY:
2154     case OP_TYPEMINQUERY:
2155     case OP_TYPEPOSQUERY:
2156     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2157 ph10 227 break;
2158    
2159 ph10 224 /* Same for these */
2160 ph10 227
2161 ph10 224 case OP_TYPEUPTO:
2162     case OP_TYPEMINUPTO:
2163     case OP_TYPEPOSUPTO:
2164     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2165     break;
2166 nigel 77
2167     /* End of branch */
2168    
2169     case OP_KET:
2170     case OP_KETRMAX:
2171     case OP_KETRMIN:
2172 ph10 604 case OP_KETRPOS:
2173 nigel 77 case OP_ALT:
2174     return TRUE;
2175    
2176 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2177     MINUPTO, and POSUPTO may be followed by a multibyte character */
2178 nigel 77
2179     #ifdef SUPPORT_UTF8
2180     case OP_STAR:
2181 ph10 602 case OP_STARI:
2182 nigel 77 case OP_MINSTAR:
2183 ph10 602 case OP_MINSTARI:
2184 nigel 93 case OP_POSSTAR:
2185 ph10 602 case OP_POSSTARI:
2186 nigel 77 case OP_QUERY:
2187 ph10 602 case OP_QUERYI:
2188 nigel 77 case OP_MINQUERY:
2189 ph10 602 case OP_MINQUERYI:
2190 nigel 93 case OP_POSQUERY:
2191 ph10 602 case OP_POSQUERYI:
2192 ph10 426 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
2193     break;
2194 ph10 461
2195 nigel 77 case OP_UPTO:
2196 ph10 602 case OP_UPTOI:
2197 nigel 77 case OP_MINUPTO:
2198 ph10 602 case OP_MINUPTOI:
2199 nigel 93 case OP_POSUPTO:
2200 ph10 602 case OP_POSUPTOI:
2201 ph10 426 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
2202 nigel 77 break;
2203     #endif
2204 ph10 503
2205 ph10 510 /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
2206     string. */
2207    
2208     case OP_MARK:
2209     case OP_PRUNE_ARG:
2210     case OP_SKIP_ARG:
2211     code += code[1];
2212 ph10 512 break;
2213 ph10 510
2214 ph10 550 case OP_THEN_ARG:
2215     code += code[1+LINK_SIZE];
2216     break;
2217    
2218 ph10 503 /* None of the remaining opcodes are required to match a character. */
2219 ph10 507
2220 ph10 503 default:
2221 ph10 507 break;
2222 nigel 77 }
2223     }
2224    
2225     return TRUE;
2226     }
2227    
2228    
2229    
2230     /*************************************************
2231     * Scan compiled regex for non-emptiness *
2232     *************************************************/
2233    
2234     /* This function is called to check for left recursive calls. We want to check
2235     the current branch of the current pattern to see if it could match the empty
2236     string. If it could, we must look outwards for branches at other levels,
2237     stopping when we pass beyond the bracket which is the subject of the recursion.
2238 ph10 624 This function is called only during the real compile, not during the
2239     pre-compile.
2240 nigel 77
2241     Arguments:
2242     code points to start of the recursion
2243     endcode points to where to stop (current RECURSE item)
2244     bcptr points to the chain of current (unclosed) branch starts
2245     utf8 TRUE if in UTF-8 mode
2246 ph10 507 cd pointers to tables etc
2247 nigel 77
2248     Returns: TRUE if what is matched could be empty
2249     */
2250    
2251     static BOOL
2252     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
2253 ph10 503 BOOL utf8, compile_data *cd)
2254 nigel 77 {
2255 ph10 475 while (bcptr != NULL && bcptr->current_branch >= code)
2256 nigel 77 {
2257 ph10 503 if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
2258 ph10 475 return FALSE;
2259 nigel 77 bcptr = bcptr->outer;
2260     }
2261     return TRUE;
2262     }
2263    
2264    
2265    
2266     /*************************************************
2267     * Check for POSIX class syntax *
2268     *************************************************/
2269    
2270     /* This function is called when the sequence "[:" or "[." or "[=" is
2271 ph10 295 encountered in a character class. It checks whether this is followed by a
2272 ph10 298 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2273 ph10 295 reach an unescaped ']' without the special preceding character, return FALSE.
2274 nigel 77
2275 ph10 298 Originally, this function only recognized a sequence of letters between the
2276     terminators, but it seems that Perl recognizes any sequence of characters,
2277     though of course unknown POSIX names are subsequently rejected. Perl gives an
2278     "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2279     didn't consider this to be a POSIX class. Likewise for [:1234:].
2280 ph10 295
2281 ph10 298 The problem in trying to be exactly like Perl is in the handling of escapes. We
2282     have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2283     class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2284     below handles the special case of \], but does not try to do any other escape
2285     processing. This makes it different from Perl for cases such as [:l\ower:]
2286 ph10 295 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2287 ph10 298 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2288 ph10 295 I think.
2289    
2290     Arguments:
2291 nigel 77 ptr pointer to the initial [
2292     endptr where to return the end pointer
2293    
2294     Returns: TRUE or FALSE
2295     */
2296    
2297     static BOOL
2298 ph10 295 check_posix_syntax(const uschar *ptr, const uschar **endptr)
2299 nigel 77 {
2300     int terminator; /* Don't combine these lines; the Solaris cc */
2301     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2302 ph10 295 for (++ptr; *ptr != 0; ptr++)
2303 nigel 77 {
2304 ph10 391 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2305 ph10 298 {
2306 ph10 391 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2307     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2308 ph10 295 {
2309     *endptr = ptr;
2310     return TRUE;
2311 ph10 298 }
2312     }
2313     }
2314 nigel 77 return FALSE;
2315     }
2316    
2317    
2318    
2319    
2320     /*************************************************
2321     * Check POSIX class name *
2322     *************************************************/
2323    
2324     /* This function is called to check the name given in a POSIX-style class entry
2325     such as [:alnum:].
2326    
2327     Arguments:
2328     ptr points to the first letter
2329     len the length of the name
2330    
2331     Returns: a value representing the name, or -1 if unknown
2332     */
2333    
2334     static int
2335     check_posix_name(const uschar *ptr, int len)
2336     {
2337 ph10 240 const char *pn = posix_names;
2338 nigel 77 register int yield = 0;
2339     while (posix_name_lengths[yield] != 0)
2340     {
2341     if (len == posix_name_lengths[yield] &&
2342 ph10 240 strncmp((const char *)ptr, pn, len) == 0) return yield;
2343 ph10 243 pn += posix_name_lengths[yield] + 1;
2344 nigel 77 yield++;
2345     }
2346     return -1;
2347     }
2348    
2349    
2350     /*************************************************
2351     * Adjust OP_RECURSE items in repeated group *
2352     *************************************************/
2353    
2354     /* OP_RECURSE items contain an offset from the start of the regex to the group
2355     that is referenced. This means that groups can be replicated for fixed
2356     repetition simply by copying (because the recursion is allowed to refer to
2357     earlier groups that are outside the current group). However, when a group is
2358 ph10 335 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2359     inserted before it, after it has been compiled. This means that any OP_RECURSE
2360     items within it that refer to the group itself or any contained groups have to
2361     have their offsets adjusted. That one of the jobs of this function. Before it
2362     is called, the partially compiled regex must be temporarily terminated with
2363     OP_END.
2364 nigel 77
2365 nigel 93 This function has been extended with the possibility of forward references for
2366     recursions and subroutine calls. It must also check the list of such references
2367     for the group we are dealing with. If it finds that one of the recursions in
2368     the current group is on this list, it adjusts the offset in the list, not the
2369     value in the reference (which is a group number).
2370    
2371 nigel 77 Arguments:
2372     group points to the start of the group
2373     adjust the amount by which the group is to be moved
2374     utf8 TRUE in UTF-8 mode
2375     cd contains pointers to tables etc.
2376 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
2377 nigel 77
2378     Returns: nothing
2379     */
2380    
2381     static void
2382 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2383     uschar *save_hwm)
2384 nigel 77 {
2385     uschar *ptr = group;
2386 ph10 224
2387 nigel 77 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2388     {
2389 nigel 93 int offset;
2390     uschar *hc;
2391    
2392     /* See if this recursion is on the forward reference list. If so, adjust the
2393     reference. */
2394 ph10 345
2395 nigel 93 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2396     {
2397     offset = GET(hc, 0);
2398     if (cd->start_code + offset == ptr + 1)
2399     {
2400     PUT(hc, 0, offset + adjust);
2401     break;
2402     }
2403     }
2404    
2405     /* Otherwise, adjust the recursion offset if it's after the start of this
2406     group. */
2407    
2408     if (hc >= cd->hwm)
2409     {
2410     offset = GET(ptr, 1);
2411     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2412     }
2413    
2414 nigel 77 ptr += 1 + LINK_SIZE;
2415     }
2416     }
2417    
2418    
2419    
2420     /*************************************************
2421     * Insert an automatic callout point *
2422     *************************************************/
2423    
2424     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2425     callout points before each pattern item.
2426    
2427     Arguments:
2428     code current code pointer
2429     ptr current pattern pointer
2430     cd pointers to tables etc
2431    
2432     Returns: new code pointer
2433     */
2434    
2435     static uschar *
2436     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2437     {
2438     *code++ = OP_CALLOUT;
2439     *code++ = 255;
2440 ph10 530 PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
2441     PUT(code, LINK_SIZE, 0); /* Default length */
2442 nigel 77 return code + 2*LINK_SIZE;
2443     }
2444    
2445    
2446    
2447     /*************************************************
2448     * Complete a callout item *
2449     *************************************************/
2450    
2451     /* A callout item contains the length of the next item in the pattern, which
2452     we can't fill in till after we have reached the relevant point. This is used
2453     for both automatic and manual callouts.
2454    
2455     Arguments:
2456     previous_callout points to previous callout item
2457     ptr current pattern pointer
2458     cd pointers to tables etc
2459    
2460     Returns: nothing
2461     */
2462    
2463     static void
2464     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2465     {
2466 ph10 530 int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
2467 nigel 77 PUT(previous_callout, 2 + LINK_SIZE, length);
2468     }
2469    
2470    
2471    
2472     #ifdef SUPPORT_UCP
2473     /*************************************************
2474     * Get othercase range *
2475     *************************************************/
2476    
2477     /* This function is passed the start and end of a class range, in UTF-8 mode
2478     with UCP support. It searches up the characters, looking for internal ranges of
2479     characters in the "other" case. Each call returns the next one, updating the
2480     start address.
2481    
2482     Arguments:
2483     cptr points to starting character value; updated
2484     d end value
2485     ocptr where to put start of othercase range
2486     odptr where to put end of othercase range
2487    
2488     Yield: TRUE when range returned; FALSE when no more
2489     */
2490    
2491     static BOOL
2492 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2493     unsigned int *odptr)
2494 nigel 77 {
2495 nigel 93 unsigned int c, othercase, next;
2496 nigel 77
2497     for (c = *cptr; c <= d; c++)
2498 ph10 349 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2499 nigel 77
2500     if (c > d) return FALSE;
2501    
2502     *ocptr = othercase;
2503     next = othercase + 1;
2504    
2505     for (++c; c <= d; c++)
2506     {
2507 ph10 349 if (UCD_OTHERCASE(c) != next) break;
2508 nigel 77 next++;
2509     }
2510    
2511     *odptr = next - 1;
2512     *cptr = c;
2513    
2514     return TRUE;
2515     }
2516 ph10 532
2517    
2518    
2519     /*************************************************
2520     * Check a character and a property *
2521     *************************************************/
2522    
2523     /* This function is called by check_auto_possessive() when a property item
2524     is adjacent to a fixed character.
2525    
2526     Arguments:
2527     c the character
2528     ptype the property type
2529     pdata the data for the type
2530     negated TRUE if it's a negated property (\P or \p{^)
2531 ph10 535
2532 ph10 532 Returns: TRUE if auto-possessifying is OK
2533 ph10 535 */
2534 ph10 532
2535     static BOOL
2536     check_char_prop(int c, int ptype, int pdata, BOOL negated)
2537     {
2538     const ucd_record *prop = GET_UCD(c);
2539     switch(ptype)
2540     {
2541     case PT_LAMP:
2542     return (prop->chartype == ucp_Lu ||
2543     prop->chartype == ucp_Ll ||
2544     prop->chartype == ucp_Lt) == negated;
2545    
2546     case PT_GC:
2547     return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
2548    
2549     case PT_PC:
2550     return (pdata == prop->chartype) == negated;
2551    
2552     case PT_SC:
2553     return (pdata == prop->script) == negated;
2554    
2555     /* These are specials */
2556    
2557     case PT_ALNUM:
2558     return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2559     _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
2560    
2561     case PT_SPACE: /* Perl space */
2562     return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2563     c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
2564     == negated;
2565    
2566     case PT_PXSPACE: /* POSIX space */
2567     return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
2568     c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
2569     c == CHAR_FF || c == CHAR_CR)
2570     == negated;
2571    
2572     case PT_WORD:
2573     return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
2574     _pcre_ucp_gentype[prop->chartype] == ucp_N ||
2575     c == CHAR_UNDERSCORE) == negated;
2576     }
2577 ph10 535 return FALSE;
2578 ph10 532 }
2579 nigel 77 #endif /* SUPPORT_UCP */
2580    
2581    
2582 nigel 93
2583 nigel 77 /*************************************************
2584 nigel 93 * Check if auto-possessifying is possible *
2585     *************************************************/
2586    
2587     /* This function is called for unlimited repeats of certain items, to see
2588     whether the next thing could possibly match the repeated item. If not, it makes
2589     sense to automatically possessify the repeated item.
2590    
2591     Arguments:
2592 ph10 532 previous pointer to the repeated opcode
2593 nigel 93 utf8 TRUE in UTF-8 mode
2594     ptr next character in pattern
2595     options options bits
2596     cd contains pointers to tables etc.
2597    
2598     Returns: TRUE if possessifying is wanted
2599     */
2600    
2601     static BOOL
2602 ph10 535 check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
2603 ph10 532 int options, compile_data *cd)
2604 nigel 93 {
2605 ph10 532 int c, next;
2606     int op_code = *previous++;
2607 nigel 93
2608     /* Skip whitespace and comments in extended mode */
2609    
2610     if ((options & PCRE_EXTENDED) != 0)
2611     {
2612     for (;;)
2613     {
2614     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2615 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2616 nigel 93 {
2617 ph10 579 ptr++;
2618 ph10 556 while (*ptr != 0)
2619     {
2620 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2621 ph10 556 ptr++;
2622 ph10 579 #ifdef SUPPORT_UTF8
2623 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2624     #endif
2625     }
2626 nigel 93 }
2627     else break;
2628     }
2629     }
2630    
2631     /* If the next item is one that we can handle, get its value. A non-negative
2632     value is a character, a negative value is an escape value. */
2633    
2634 ph10 391 if (*ptr == CHAR_BACKSLASH)
2635 nigel 93 {
2636     int temperrorcode = 0;
2637     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2638     if (temperrorcode != 0) return FALSE;
2639     ptr++; /* Point after the escape sequence */
2640     }
2641    
2642     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2643     {
2644     #ifdef SUPPORT_UTF8
2645     if (utf8) { GETCHARINC(next, ptr); } else
2646     #endif
2647     next = *ptr++;
2648     }
2649    
2650     else return FALSE;
2651    
2652     /* Skip whitespace and comments in extended mode */
2653    
2654     if ((options & PCRE_EXTENDED) != 0)
2655     {
2656     for (;;)
2657     {
2658     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2659 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2660 nigel 93 {
2661 ph10 579 ptr++;
2662 ph10 556 while (*ptr != 0)
2663     {
2664 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2665 ph10 556 ptr++;
2666 ph10 579 #ifdef SUPPORT_UTF8
2667 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
2668     #endif
2669     }
2670 nigel 93 }
2671     else break;
2672     }
2673     }
2674    
2675     /* If the next thing is itself optional, we have to give up. */
2676    
2677 ph10 392 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2678 ph10 391 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2679     return FALSE;
2680 nigel 93
2681 ph10 532 /* Now compare the next item with the previous opcode. First, handle cases when
2682     the next item is a character. */
2683 nigel 93
2684     if (next >= 0) switch(op_code)
2685     {
2686     case OP_CHAR:
2687 ph10 535 #ifdef SUPPORT_UTF8
2688 ph10 532 GETCHARTEST(c, previous);
2689 ph10 369 #else
2690 ph10 532 c = *previous;
2691 ph10 535 #endif
2692     return c != next;
2693 nigel 93
2694 ph10 602 /* For CHARI (caseless character) we must check the other case. If we have
2695 nigel 93 Unicode property support, we can use it to test the other case of
2696     high-valued characters. */
2697    
2698 ph10 602 case OP_CHARI:
2699 ph10 535 #ifdef SUPPORT_UTF8
2700 ph10 532 GETCHARTEST(c, previous);
2701     #else
2702     c = *previous;
2703 ph10 535 #endif
2704 ph10 532 if (c == next) return FALSE;
2705 nigel 93 #ifdef SUPPORT_UTF8
2706     if (utf8)
2707     {
2708     unsigned int othercase;
2709     if (next < 128) othercase = cd->fcc[next]; else
2710     #ifdef SUPPORT_UCP
2711 ph10 349 othercase = UCD_OTHERCASE((unsigned int)next);
2712 nigel 93 #else
2713     othercase = NOTACHAR;
2714     #endif
2715 ph10 532 return (unsigned int)c != othercase;
2716 nigel 93 }
2717     else
2718     #endif /* SUPPORT_UTF8 */
2719 ph10 532 return (c != cd->fcc[next]); /* Non-UTF-8 mode */
2720 nigel 93
2721 ph10 602 /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These
2722 ph10 604 opcodes are not used for multi-byte characters, because they are coded using
2723 ph10 602 an XCLASS instead. */
2724 nigel 93
2725     case OP_NOT:
2726 ph10 602 return (c = *previous) == next;
2727 ph10 604
2728     case OP_NOTI:
2729 ph10 532 if ((c = *previous) == next) return TRUE;
2730 nigel 93 #ifdef SUPPORT_UTF8
2731     if (utf8)
2732     {
2733     unsigned int othercase;
2734     if (next < 128) othercase = cd->fcc[next]; else
2735     #ifdef SUPPORT_UCP
2736 ph10 349 othercase = UCD_OTHERCASE(next);
2737 nigel 93 #else
2738     othercase = NOTACHAR;
2739     #endif
2740 ph10 532 return (unsigned int)c == othercase;
2741 nigel 93 }
2742     else
2743     #endif /* SUPPORT_UTF8 */
2744 ph10 532 return (c == cd->fcc[next]); /* Non-UTF-8 mode */
2745 nigel 93
2746 ph10 535 /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
2747     When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
2748    
2749 nigel 93 case OP_DIGIT:
2750     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2751    
2752     case OP_NOT_DIGIT:
2753     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2754    
2755     case OP_WHITESPACE:
2756     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2757    
2758     case OP_NOT_WHITESPACE:
2759     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2760    
2761     case OP_WORDCHAR:
2762     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2763    
2764     case OP_NOT_WORDCHAR:
2765     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2766    
2767 ph10 180 case OP_HSPACE:
2768     case OP_NOT_HSPACE:
2769     switch(next)
2770     {
2771     case 0x09:
2772     case 0x20:
2773     case 0xa0:
2774     case 0x1680:
2775     case 0x180e:
2776     case 0x2000:
2777     case 0x2001:
2778     case 0x2002:
2779     case 0x2003:
2780     case 0x2004:
2781     case 0x2005:
2782     case 0x2006:
2783     case 0x2007:
2784     case 0x2008:
2785     case 0x2009:
2786     case 0x200A:
2787     case 0x202f:
2788     case 0x205f:
2789     case 0x3000:
2790 ph10 528 return op_code == OP_NOT_HSPACE;
2791 ph10 180 default:
2792 ph10 528 return op_code != OP_NOT_HSPACE;
2793 ph10 180 }
2794    
2795 ph10 528 case OP_ANYNL:
2796 ph10 180 case OP_VSPACE:
2797     case OP_NOT_VSPACE:
2798     switch(next)
2799     {
2800     case 0x0a:
2801     case 0x0b:
2802     case 0x0c:
2803     case 0x0d:
2804     case 0x85:
2805     case 0x2028:
2806     case 0x2029:
2807 ph10 528 return op_code == OP_NOT_VSPACE;
2808 ph10 180 default:
2809 ph10 528 return op_code != OP_NOT_VSPACE;
2810 ph10 180 }
2811    
2812 ph10 532 #ifdef SUPPORT_UCP
2813     case OP_PROP:
2814     return check_char_prop(next, previous[0], previous[1], FALSE);
2815 ph10 535
2816 ph10 532 case OP_NOTPROP:
2817     return check_char_prop(next, previous[0], previous[1], TRUE);
2818     #endif
2819    
2820 nigel 93 default:
2821     return FALSE;
2822     }
2823    
2824    
2825 ph10 535 /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
2826     is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
2827     generated only when PCRE_UCP is *not* set, that is, when only ASCII
2828     characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
2829 ph10 532 replaced by OP_PROP codes when PCRE_UCP is set. */
2830 nigel 93
2831     switch(op_code)
2832     {
2833     case OP_CHAR:
2834 ph10 602 case OP_CHARI:
2835 ph10 535 #ifdef SUPPORT_UTF8
2836 ph10 532 GETCHARTEST(c, previous);
2837     #else
2838     c = *previous;
2839 ph10 535 #endif
2840 nigel 93 switch(-next)
2841     {
2842     case ESC_d:
2843 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
2844 nigel 93
2845     case ESC_D:
2846 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
2847 nigel 93
2848     case ESC_s:
2849 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
2850 nigel 93
2851     case ESC_S:
2852 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
2853 nigel 93
2854     case ESC_w:
2855 ph10 532 return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
2856 nigel 93
2857     case ESC_W:
2858 ph10 532 return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
2859 ph10 182
2860 ph10 180 case ESC_h:
2861     case ESC_H:
2862 ph10 532 switch(c)
2863 ph10 180 {
2864     case 0x09:
2865     case 0x20:
2866     case 0xa0:
2867     case 0x1680:
2868     case 0x180e:
2869     case 0x2000:
2870     case 0x2001:
2871     case 0x2002:
2872     case 0x2003:
2873     case 0x2004:
2874     case 0x2005:
2875     case 0x2006:
2876     case 0x2007:
2877     case 0x2008:
2878     case 0x2009:
2879     case 0x200A:
2880     case 0x202f:
2881     case 0x205f:
2882     case 0x3000:
2883     return -next != ESC_h;
2884     default:
2885     return -next == ESC_h;
2886 ph10 182 }
2887    
2888 ph10 180 case ESC_v:
2889     case ESC_V:
2890 ph10 532 switch(c)
2891 ph10 180 {
2892     case 0x0a:
2893     case 0x0b:
2894     case 0x0c:
2895     case 0x0d:
2896     case 0x85:
2897     case 0x2028:
2898     case 0x2029:
2899     return -next != ESC_v;
2900     default:
2901     return -next == ESC_v;
2902 ph10 182 }
2903 ph10 535
2904     /* When PCRE_UCP is set, these values get generated for \d etc. Find
2905     their substitutions and process them. The result will always be either
2906 ph10 532 -ESC_p or -ESC_P. Then fall through to process those values. */
2907 ph10 535
2908 ph10 532 #ifdef SUPPORT_UCP
2909     case ESC_du:
2910     case ESC_DU:
2911     case ESC_wu:
2912     case ESC_WU:
2913     case ESC_su:
2914     case ESC_SU:
2915     {
2916     int temperrorcode = 0;
2917     ptr = substitutes[-next - ESC_DU];
2918     next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
2919     if (temperrorcode != 0) return FALSE;
2920     ptr++; /* For compatibility */
2921     }
2922 ph10 535 /* Fall through */
2923 nigel 93
2924 ph10 532 case ESC_p:
2925     case ESC_P:
2926     {
2927     int ptype, pdata, errorcodeptr;
2928 ph10 535 BOOL negated;
2929    
2930 ph10 532 ptr--; /* Make ptr point at the p or P */
2931     ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
2932     if (ptype < 0) return FALSE;
2933     ptr++; /* Point past the final curly ket */
2934 ph10 535
2935 ph10 532 /* If the property item is optional, we have to give up. (When generated
2936     from \d etc by PCRE_UCP, this test will have been applied much earlier,
2937     to the original \d etc. At this point, ptr will point to a zero byte. */
2938 ph10 535
2939 ph10 532 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2940     strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2941     return FALSE;
2942 ph10 535
2943 ph10 532 /* Do the property check. */
2944 ph10 535
2945 ph10 532 return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
2946 ph10 535 }
2947 ph10 532 #endif
2948    
2949 nigel 93 default:
2950     return FALSE;
2951     }
2952    
2953 ph10 535 /* In principle, support for Unicode properties should be integrated here as
2954     well. It means re-organizing the above code so as to get hold of the property
2955     values before switching on the op-code. However, I wonder how many patterns
2956     combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
2957     these op-codes are never generated.) */
2958    
2959 nigel 93 case OP_DIGIT:
2960 ph10 180 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2961 ph10 528 next == -ESC_h || next == -ESC_v || next == -ESC_R;
2962 nigel 93
2963     case OP_NOT_DIGIT:
2964     return next == -ESC_d;
2965    
2966     case OP_WHITESPACE:
2967 ph10 528 return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
2968 nigel 93
2969     case OP_NOT_WHITESPACE:
2970 ph10 180 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2971 nigel 93
2972 ph10 180 case OP_HSPACE:
2973 ph10 535 return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
2974 ph10 528 next == -ESC_w || next == -ESC_v || next == -ESC_R;
2975 ph10 180
2976     case OP_NOT_HSPACE:
2977     return next == -ESC_h;
2978 ph10 182
2979 ph10 180 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2980 ph10 535 case OP_ANYNL:
2981 ph10 182 case OP_VSPACE:
2982 ph10 180 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2983    
2984     case OP_NOT_VSPACE:
2985 ph10 528 return next == -ESC_v || next == -ESC_R;
2986 ph10 180
2987 nigel 93 case OP_WORDCHAR:
2988 ph10 535 return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
2989 ph10 528 next == -ESC_v || next == -ESC_R;
2990 nigel 93
2991     case OP_NOT_WORDCHAR:
2992     return next == -ESC_w || next == -ESC_d;
2993 ph10 182
2994 nigel 93 default:
2995     return FALSE;
2996     }
2997    
2998     /* Control does not reach here */
2999     }
3000    
3001    
3002    
3003     /*************************************************
3004 nigel 77 * Compile one branch *
3005     *************************************************/
3006    
3007 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
3008 nigel 77 changed during the branch, the pointer is used to change the external options
3009 nigel 93 bits. This function is used during the pre-compile phase when we are trying
3010     to find out the amount of memory needed, as well as during the real compile
3011     phase. The value of lengthptr distinguishes the two phases.
3012 nigel 77
3013     Arguments:
3014     optionsptr pointer to the option bits
3015     codeptr points to the pointer to the current code point
3016     ptrptr points to the current pattern pointer
3017     errorcodeptr points to error code variable
3018     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
3019     reqbyteptr set to the last literal character required, else < 0
3020     bcptr points to current branch chain
3021     cd contains pointers to tables etc.
3022 nigel 93 lengthptr NULL during the real compile phase
3023     points to length accumulator during pre-compile phase
3024 nigel 77
3025     Returns: TRUE on success
3026     FALSE, with *errorcodeptr set non-zero on error
3027     */
3028    
3029     static BOOL
3030 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
3031     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
3032     compile_data *cd, int *lengthptr)
3033 nigel 77 {
3034     int repeat_type, op_type;
3035     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
3036     int bravalue = 0;
3037     int greedy_default, greedy_non_default;
3038     int firstbyte, reqbyte;
3039     int zeroreqbyte, zerofirstbyte;
3040     int req_caseopt, reqvary, tempreqvary;
3041     int options = *optionsptr;
3042     int after_manual_callout = 0;
3043 nigel 93 int length_prevgroup = 0;
3044 nigel 77 register int c;
3045     register uschar *code = *codeptr;
3046 nigel 93 uschar *last_code = code;
3047     uschar *orig_code = code;
3048 nigel 77 uschar *tempcode;
3049     BOOL inescq = FALSE;
3050     BOOL groupsetfirstbyte = FALSE;
3051     const uschar *ptr = *ptrptr;
3052     const uschar *tempptr;
3053 ph10 518 const uschar *nestptr = NULL;
3054 nigel 77 uschar *previous = NULL;
3055     uschar *previous_callout = NULL;
3056 nigel 93 uschar *save_hwm = NULL;
3057 nigel 77 uschar classbits[32];
3058    
3059     #ifdef SUPPORT_UTF8
3060     BOOL class_utf8;
3061     BOOL utf8 = (options & PCRE_UTF8) != 0;
3062     uschar *class_utf8data;
3063 ph10 300 uschar *class_utf8data_base;
3064 nigel 77 uschar utf8_char[6];
3065     #else
3066     BOOL utf8 = FALSE;
3067 nigel 93 uschar *utf8_char = NULL;
3068 nigel 77 #endif
3069    
3070 ph10 475 #ifdef PCRE_DEBUG
3071 nigel 93 if (lengthptr != NULL) DPRINTF((">> start branch\n"));
3072     #endif
3073    
3074 nigel 77 /* Set up the default and non-default settings for greediness */
3075    
3076     greedy_default = ((options & PCRE_UNGREEDY) != 0);
3077     greedy_non_default = greedy_default ^ 1;
3078    
3079     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
3080     matching encountered yet". It gets changed to REQ_NONE if we hit something that
3081     matches a non-fixed char first char; reqbyte just remains unset if we never
3082     find one.
3083    
3084     When we hit a repeat whose minimum is zero, we may have to adjust these values
3085     to take the zero repeat into account. This is implemented by setting them to
3086     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
3087     item types that can be repeated set these backoff variables appropriately. */
3088    
3089     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
3090    
3091     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
3092     according to the current setting of the caseless flag. REQ_CASELESS is a bit
3093     value > 255. It is added into the firstbyte or reqbyte variables to record the
3094     case status of the value. This is used only for ASCII characters. */
3095    
3096     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
3097    
3098     /* Switch on next character until the end of the branch */
3099    
3100     for (;; ptr++)
3101     {
3102     BOOL negate_class;
3103 ph10 286 BOOL should_flip_negation;
3104 nigel 77 BOOL possessive_quantifier;
3105     BOOL is_quantifier;
3106 nigel 93 BOOL is_recurse;
3107 ph10 180 BOOL reset_bracount;
3108 nigel 77 int class_charcount;
3109     int class_lastchar;
3110     int newoptions;
3111     int recno;
3112 ph10 172 int refsign;
3113 nigel 77 int skipbytes;
3114     int subreqbyte;
3115     int subfirstbyte;
3116 nigel 93 int terminator;
3117 nigel 77 int mclength;
3118     uschar mcbuffer[8];
3119    
3120 nigel 93 /* Get next byte in the pattern */
3121 nigel 77
3122     c = *ptr;
3123 ph10 345
3124 ph10 535 /* If we are at the end of a nested substitution, revert to the outer level
3125 ph10 518 string. Nesting only happens one level deep. */
3126    
3127     if (c == 0 && nestptr != NULL)
3128     {
3129     ptr = nestptr;
3130     nestptr = NULL;
3131     c = *ptr;
3132     }
3133    
3134 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
3135     previous cycle of this loop. */
3136    
3137     if (lengthptr != NULL)
3138     {
3139 ph10 475 #ifdef PCRE_DEBUG
3140 nigel 93 if (code > cd->hwm) cd->hwm = code; /* High water info */
3141     #endif
3142 ph10 505 if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
3143 nigel 93 {
3144     *errorcodeptr = ERR52;
3145     goto FAILED;
3146     }
3147    
3148     /* There is at least one situation where code goes backwards: this is the
3149     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3150     the class is simply eliminated. However, it is created first, so we have to
3151     allow memory for it. Therefore, don't ever reduce the length at this point.
3152     */
3153    
3154     if (code < last_code) code = last_code;
3155 ph10 202
3156     /* Paranoid check for integer overflow */
3157    
3158     if (OFLOW_MAX - *lengthptr < code - last_code)
3159     {
3160     *errorcodeptr = ERR20;
3161     goto FAILED;
3162     }
3163    
3164 ph10 530 *lengthptr += (int)(code - last_code);
3165 nigel 93 DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3166    
3167     /* If "previous" is set and it is not at the start of the work space, move
3168     it back to there, in order to avoid filling up the work space. Otherwise,
3169     if "previous" is NULL, reset the current code pointer to the start. */
3170    
3171     if (previous != NULL)
3172     {
3173     if (previous > orig_code)
3174     {
3175     memmove(orig_code, previous, code - previous);
3176     code -= previous - orig_code;
3177     previous = orig_code;
3178     }
3179     }
3180     else code = orig_code;
3181    
3182     /* Remember where this code item starts so we can pick up the length
3183     next time round. */
3184    
3185     last_code = code;
3186     }
3187    
3188     /* In the real compile phase, just check the workspace used by the forward
3189     reference list. */
3190    
3191 ph10 505 else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
3192 nigel 93 {
3193     *errorcodeptr = ERR52;
3194     goto FAILED;
3195     }
3196    
3197 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
3198    
3199     if (inescq && c != 0)
3200     {
3201 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3202 nigel 77 {
3203     inescq = FALSE;
3204     ptr++;
3205     continue;
3206     }
3207     else
3208     {
3209     if (previous_callout != NULL)
3210     {
3211 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3212     complete_callout(previous_callout, ptr, cd);
3213 nigel 77 previous_callout = NULL;
3214     }
3215     if ((options & PCRE_AUTO_CALLOUT) != 0)
3216     {
3217     previous_callout = code;
3218     code = auto_callout(code, ptr, cd);
3219     }
3220     goto NORMAL_CHAR;
3221     }
3222     }
3223    
3224     /* Fill in length of a previous callout, except when the next thing is
3225     a quantifier. */
3226    
3227 ph10 392 is_quantifier =
3228 ph10 391 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3229     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
3230 nigel 77
3231     if (!is_quantifier && previous_callout != NULL &&
3232     after_manual_callout-- <= 0)
3233     {
3234 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3235     complete_callout(previous_callout, ptr, cd);
3236 nigel 77 previous_callout = NULL;
3237     }
3238    
3239     /* In extended mode, skip white space and comments */
3240    
3241     if ((options & PCRE_EXTENDED) != 0)
3242     {
3243     if ((cd->ctypes[c] & ctype_space) != 0) continue;
3244 ph10 391 if (c == CHAR_NUMBER_SIGN)
3245 nigel 77 {
3246 ph10 579 ptr++;
3247 ph10 556 while (*ptr != 0)
3248 nigel 91 {
3249 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3250 ph10 556 ptr++;
3251 ph10 579 #ifdef SUPPORT_UTF8
3252 ph10 556 if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++;
3253     #endif
3254 nigel 91 }
3255 nigel 93 if (*ptr != 0) continue;
3256    
3257 nigel 91 /* Else fall through to handle end of string */
3258     c = 0;
3259 nigel 77 }
3260     }
3261    
3262     /* No auto callout for quantifiers. */
3263    
3264     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3265     {
3266     previous_callout = code;
3267     code = auto_callout(code, ptr, cd);
3268     }
3269    
3270     switch(c)
3271     {
3272 nigel 93 /* ===================================================================*/
3273     case 0: /* The branch terminates at string end */
3274 ph10 391 case CHAR_VERTICAL_LINE: /* or | or ) */
3275     case CHAR_RIGHT_PARENTHESIS:
3276 nigel 77 *firstbyteptr = firstbyte;
3277     *reqbyteptr = reqbyte;
3278     *codeptr = code;
3279     *ptrptr = ptr;
3280 nigel 93 if (lengthptr != NULL)
3281     {
3282 ph10 202 if (OFLOW_MAX - *lengthptr < code - last_code)
3283     {
3284     *errorcodeptr = ERR20;
3285     goto FAILED;
3286     }
3287 ph10 530 *lengthptr += (int)(code - last_code); /* To include callout length */
3288 nigel 93 DPRINTF((">> end branch\n"));
3289     }
3290 nigel 77 return TRUE;
3291    
3292 nigel 93
3293     /* ===================================================================*/
3294 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
3295     the setting of any following char as a first character. */
3296    
3297 ph10 391 case CHAR_CIRCUMFLEX_ACCENT:
3298 ph10 602 previous = NULL;
3299 nigel 77 if ((options & PCRE_MULTILINE) != 0)
3300     {
3301     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3302 ph10 602 *code++ = OP_CIRCM;
3303 nigel 77 }
3304 ph10 602 else *code++ = OP_CIRC;
3305 nigel 77 break;
3306    
3307 ph10 391 case CHAR_DOLLAR_SIGN:
3308 nigel 77 previous = NULL;
3309 ph10 602 *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL;
3310 nigel 77 break;
3311    
3312     /* There can never be a first char if '.' is first, whatever happens about
3313     repeats. The value of reqbyte doesn't change either. */
3314    
3315 ph10 391 case CHAR_DOT:
3316 nigel 77 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3317     zerofirstbyte = firstbyte;
3318     zeroreqbyte = reqbyte;
3319     previous = code;
3320 ph10 342 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3321 nigel 77 break;
3322    
3323 nigel 93
3324     /* ===================================================================*/
3325 nigel 87 /* Character classes. If the included characters are all < 256, we build a
3326     32-byte bitmap of the permitted characters, except in the special case
3327     where there is only one such character. For negated classes, we build the
3328     map as usual, then invert it at the end. However, we use a different opcode
3329     so that data characters > 255 can be handled correctly.
3330 nigel 77
3331     If the class contains characters outside the 0-255 range, a different
3332     opcode is compiled. It may optionally have a bit map for characters < 256,
3333     but those above are are explicitly listed afterwards. A flag byte tells
3334     whether the bitmap is present, and whether this is a negated class or not.
3335 ph10 345
3336 ph10 336 In JavaScript compatibility mode, an isolated ']' causes an error. In
3337     default (Perl) mode, it is treated as a data character. */
3338 ph10 345
3339 ph10 391 case CHAR_RIGHT_SQUARE_BRACKET:
3340 ph10 336 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3341     {
3342     *errorcodeptr = ERR64;
3343 ph10 345 goto FAILED;
3344 ph10 336 }
3345 ph10 345 goto NORMAL_CHAR;
3346 nigel 77
3347 ph10 391 case CHAR_LEFT_SQUARE_BRACKET:
3348 nigel 77 previous = code;
3349    
3350     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3351     they are encountered at the top level, so we'll do that too. */
3352    
3353 ph10 392 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3354 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) &&
3355 ph10 295 check_posix_syntax(ptr, &tempptr))
3356 nigel 77 {
3357 ph10 391 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
3358 nigel 77 goto FAILED;
3359     }
3360    
3361 ph10 205 /* If the first character is '^', set the negation flag and skip it. Also,
3362 ph10 208 if the first few characters (either before or after ^) are \Q\E or \E we
3363 ph10 205 skip them too. This makes for compatibility with Perl. */
3364 ph10 208
3365 ph10 205 negate_class = FALSE;
3366     for (;;)
3367 nigel 77 {
3368     c = *(++ptr);
3369 ph10 391 if (c == CHAR_BACKSLASH)
3370 ph10 205 {
3371 ph10 392 if (ptr[1] == CHAR_E)
3372 ph10 391 ptr++;
3373 ph10 392 else if (strncmp((const char *)ptr+1,
3374     STR_Q STR_BACKSLASH STR_E, 3) == 0)
3375 ph10 391 ptr += 3;
3376 ph10 392 else
3377 ph10 391 break;
3378 ph10 205 }
3379 ph10 391 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
3380 ph10 205 negate_class = TRUE;
3381     else break;
3382 ph10 208 }
3383 ph10 345
3384     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3385     an initial ']' is taken as a data character -- the code below handles
3386 ph10 341 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3387     [^] must match any character, so generate OP_ALLANY. */
3388 ph10 345
3389 ph10 392 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3390 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3391 ph10 341 {
3392     *code++ = negate_class? OP_ALLANY : OP_FAIL;
3393     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3394     zerofirstbyte = firstbyte;
3395     break;
3396 ph10 345 }
3397 nigel 77
3398 ph10 286 /* If a class contains a negative special such as \S, we need to flip the
3399     negation flag at the end, so that support for characters > 255 works
3400 ph10 264 correctly (they are all included in the class). */
3401    
3402     should_flip_negation = FALSE;
3403    
3404 nigel 77 /* Keep a count of chars with values < 256 so that we can optimize the case
3405 nigel 93 of just a single character (as long as it's < 256). However, For higher
3406     valued UTF-8 characters, we don't yet do any optimization. */
3407 nigel 77
3408     class_charcount = 0;
3409     class_lastchar = -1;
3410    
3411 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
3412     temporary bit of memory, in case the class contains only 1 character (less
3413     than 256), because in that case the compiled code doesn't use the bit map.
3414     */
3415    
3416     memset(classbits, 0, 32 * sizeof(uschar));
3417    
3418 nigel 77 #ifdef SUPPORT_UTF8
3419     class_utf8 = FALSE; /* No chars >= 256 */
3420 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3421 ph10 309 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3422 nigel 77 #endif
3423    
3424     /* Process characters until ] is reached. By writing this as a "do" it
3425 nigel 93 means that an initial ] is taken as a data character. At the start of the
3426     loop, c contains the first byte of the character. */
3427 nigel 77
3428 nigel 93 if (c != 0) do
3429 nigel 77 {
3430 nigel 93 const uschar *oldptr;
3431    
3432 nigel 77 #ifdef SUPPORT_UTF8
3433     if (utf8 && c > 127)
3434     { /* Braces are required because the */
3435     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3436     }
3437 ph10 535
3438 ph10 300 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3439 ph10 309 data and reset the pointer. This is so that very large classes that
3440 ph10 300 contain a zillion UTF-8 characters no longer overwrite the work space
3441 ph10 309 (which is on the stack). */
3442    
3443 ph10 300 if (lengthptr != NULL)
3444     {
3445     *lengthptr += class_utf8data - class_utf8data_base;
3446 ph10 309 class_utf8data = class_utf8data_base;
3447     }
3448    
3449 nigel 77 #endif
3450    
3451     /* Inside \Q...\E everything is literal except \E */
3452    
3453     if (inescq)
3454     {
3455 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
3456 nigel 77 {
3457 nigel 93 inescq = FALSE; /* Reset literal state */
3458     ptr++; /* Skip the 'E' */
3459     continue; /* Carry on with next */
3460 nigel 77 }
3461 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
3462 nigel 77 }
3463    
3464     /* Handle POSIX class names. Perl allows a negation extension of the
3465     form [:^name:]. A square bracket that doesn't match the syntax is
3466     treated as a literal. We also recognize the POSIX constructions
3467     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3468     5.6 and 5.8 do. */
3469    
3470 ph10 391 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3471 ph10 392 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3472 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3473 nigel 77 {
3474     BOOL local_negate = FALSE;
3475 nigel 87 int posix_class, taboffset, tabopt;
3476 nigel 77 register const uschar *cbits = cd->cbits;
3477 nigel 87 uschar pbits[32];
3478 nigel 77
3479 ph10 391 if (ptr[1] != CHAR_COLON)
3480 nigel 77 {
3481     *errorcodeptr = ERR31;
3482     goto FAILED;
3483     }
3484    
3485     ptr += 2;
3486 ph10 391 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3487 nigel 77 {
3488     local_negate = TRUE;
3489 ph10 286 should_flip_negation = TRUE; /* Note negative special */
3490 nigel 77 ptr++;
3491     }
3492    
3493 ph10 530 posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
3494 nigel 77 if (posix_class < 0)
3495     {
3496     *errorcodeptr = ERR30;
3497     goto FAILED;
3498     }
3499    
3500     /* If matching is caseless, upper and lower are converted to
3501     alpha. This relies on the fact that the class table starts with
3502     alpha, lower, upper as the first 3 entries. */
3503    
3504     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3505     posix_class = 0;
3506 ph10 535
3507     /* When PCRE_UCP is set, some of the POSIX classes are converted to
3508 ph10 518 different escape sequences that use Unicode properties. */
3509 ph10 535
3510 ph10 518 #ifdef SUPPORT_UCP
3511     if ((options & PCRE_UCP) != 0)
3512     {
3513     int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
3514     if (posix_substitutes[pc] != NULL)
3515     {
3516 ph10 535 nestptr = tempptr + 1;
3517 ph10 518 ptr = posix_substitutes[pc] - 1;
3518 ph10 535 continue;
3519     }
3520     }
3521     #endif
3522 ph10 518 /* In the non-UCP case, we build the bit map for the POSIX class in a
3523     chunk of local store because we may be adding and subtracting from it,
3524     and we don't want to subtract bits that may be in the main map already.
3525     At the end we or the result into the bit map that is being built. */
3526 nigel 77
3527     posix_class *= 3;
3528 nigel 87
3529     /* Copy in the first table (always present) */
3530    
3531     memcpy(pbits, cbits + posix_class_maps[posix_class],
3532     32 * sizeof(uschar));
3533    
3534     /* If there is a second table, add or remove it as required. */
3535    
3536     taboffset = posix_class_maps[posix_class + 1];
3537     tabopt = posix_class_maps[posix_class + 2];
3538    
3539     if (taboffset >= 0)
3540 nigel 77 {
3541 nigel 87 if (tabopt >= 0)
3542     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3543 nigel 77 else
3544 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3545 nigel 77 }
3546    
3547 nigel 87 /* Not see if we need to remove any special characters. An option
3548     value of 1 removes vertical space and 2 removes underscore. */
3549    
3550     if (tabopt < 0) tabopt = -tabopt;
3551     if (tabopt == 1) pbits[1] &= ~0x3c;
3552     else if (tabopt == 2) pbits[11] &= 0x7f;
3553    
3554     /* Add the POSIX table or its complement into the main table that is
3555     being built and we are done. */
3556    
3557     if (local_negate)
3558     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3559     else
3560     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3561    
3562 nigel 77 ptr = tempptr + 1;
3563     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3564     continue; /* End of POSIX syntax handling */
3565     }
3566    
3567     /* Backslash may introduce a single character, or it may introduce one
3568 nigel 93 of the specials, which just set a flag. The sequence \b is a special
3569 ph10 513 case. Inside a class (and only there) it is treated as backspace. We
3570     assume that other escapes have more than one character in them, so set
3571     class_charcount bigger than one. Unrecognized escapes fall through and
3572     are either treated as literal characters (by default), or are faulted if
3573     PCRE_EXTRA is set. */
3574 nigel 77
3575 ph10 391 if (c == CHAR_BACKSLASH)
3576 nigel 77 {
3577 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3578     if (*errorcodeptr != 0) goto FAILED;
3579 nigel 77
3580 ph10 513 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3581 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
3582     {
3583 ph10 391 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3584 nigel 77 {
3585     ptr += 2; /* avoid empty string */
3586     }
3587     else inescq = TRUE;
3588     continue;
3589     }
3590 ph10 220 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3591 nigel 77
3592     if (c < 0)
3593     {
3594     register const uschar *cbits = cd->cbits;
3595     class_charcount += 2; /* Greater than 1 is what matters */
3596 nigel 93
3597 ph10 518 switch (-c)
3598 nigel 77 {
3599 ph10 518 #ifdef SUPPORT_UCP
3600     case ESC_du: /* These are the values given for \d etc */
3601     case ESC_DU: /* when PCRE_UCP is set. We replace the */
3602     case ESC_wu: /* escape sequence with an appropriate \p */
3603     case ESC_WU: /* or \P to test Unicode properties instead */
3604     case ESC_su: /* of the default ASCII testing. */
3605     case ESC_SU:
3606     nestptr = ptr;
3607     ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
3608 ph10 535 class_charcount -= 2; /* Undo! */
3609 ph10 518 continue;
3610     #endif
3611 nigel 77 case ESC_d:
3612     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3613     continue;
3614    
3615     case ESC_D:
3616 ph10 286 should_flip_negation = TRUE;
3617 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3618     continue;
3619    
3620     case ESC_w:
3621     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3622     continue;
3623    
3624     case ESC_W:
3625 ph10 286 should_flip_negation = TRUE;
3626 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3627     continue;
3628    
3629 ph10 552 /* Perl 5.004 onwards omits VT from \s, but we must preserve it
3630 ph10 579 if it was previously set by something earlier in the character
3631     class. */
3632 ph10 552
3633 nigel 77 case ESC_s:
3634 ph10 552 classbits[0] |= cbits[cbit_space];
3635 ph10 579 classbits[1] |= cbits[cbit_space+1] & ~0x08;
3636 ph10 552 for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3637 nigel 77 continue;
3638    
3639     case ESC_S:
3640 ph10 286 should_flip_negation = TRUE;
3641 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3642     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3643     continue;
3644    
3645 ph10 518 case ESC_h:
3646 ph10 178 SETBIT(classbits, 0x09); /* VT */
3647     SETBIT(classbits, 0x20); /* SPACE */
3648 ph10 180 SETBIT(classbits, 0xa0); /* NSBP */
3649 ph10 178 #ifdef SUPPORT_UTF8
3650     if (utf8)
3651 ph10 180 {
3652 ph10 178 class_utf8 = TRUE;
3653     *class_utf8data++ = XCL_SINGLE;
3654 ph10 180 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3655 ph10 178 *class_utf8data++ = XCL_SINGLE;
3656 ph10 180 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3657     *class_utf8data++ = XCL_RANGE;
3658     class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3659     class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3660 ph10 178 *class_utf8data++ = XCL_SINGLE;
3661 ph10 180 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3662 ph10 178 *class_utf8data++ = XCL_SINGLE;
3663 ph10 180 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3664 ph10 178 *class_utf8data++ = XCL_SINGLE;
3665 ph10 180 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3666     }
3667     #endif
3668     continue;
3669 nigel 93
3670 ph10 518 case ESC_H:
3671 ph10 178 for (c = 0; c < 32; c++)
3672     {
3673     int x = 0xff;
3674     switch (c)
3675 ph10 180 {
3676 ph10 178 case 0x09/8: x ^= 1 << (0x09%8); break;
3677     case 0x20/8: x ^= 1 << (0x20%8); break;
3678     case 0xa0/8: x ^= 1 << (0xa0%8); break;
3679     default: break;
3680     }
3681     classbits[c] |= x;
3682 ph10 180 }
3683    
3684 ph10 178 #ifdef SUPPORT_UTF8
3685     if (utf8)
3686 ph10 180 {
3687 ph10 178 class_utf8 = TRUE;
3688 ph10 180 *class_utf8data++ = XCL_RANGE;
3689     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3690     class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3691     *class_utf8data++ = XCL_RANGE;
3692     class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3693     class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3694     *class_utf8data++ = XCL_RANGE;
3695     class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3696     class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3697     *class_utf8data++ = XCL_RANGE;
3698     class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3699     class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3700     *class_utf8data++ = XCL_RANGE;
3701     class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3702     class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3703     *class_utf8data++ = XCL_RANGE;
3704     class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3705     class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3706     *class_utf8data++ = XCL_RANGE;
3707     class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3708     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3709     }
3710     #endif
3711     continue;
3712 ph10 178
3713 ph10 518 case ESC_v:
3714 ph10 178 SETBIT(classbits, 0x0a); /* LF */
3715     SETBIT(classbits, 0x0b); /* VT */
3716 ph10 180 SETBIT(classbits, 0x0c); /* FF */
3717     SETBIT(classbits, 0x0d); /* CR */
3718     SETBIT(classbits, 0x85); /* NEL */
3719 ph10 178 #ifdef SUPPORT_UTF8
3720     if (utf8)
3721 ph10 180 {
3722 ph10 178 class_utf8 = TRUE;
3723 ph10 180 *class_utf8data++ = XCL_RANGE;
3724     class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3725     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3726     }
3727     #endif
3728     continue;
3729 ph10 178
3730 ph10 518 case ESC_V:
3731 ph10 178 for (c = 0; c < 32; c++)
3732     {
3733     int x = 0xff;
3734     switch (c)
3735 ph10 180 {
3736 ph10 178 case 0x0a/8: x ^= 1 << (0x0a%8);
3737     x ^= 1 << (0x0b%8);
3738     x ^= 1 << (0x0c%8);
3739 ph10 180 x ^= 1 << (0x0d%8);
3740 ph10 178 break;
3741     case 0x85/8: x ^= 1 << (0x85%8); break;
3742     default: break;
3743     }
3744     classbits[c] |= x;
3745 ph10 180 }
3746    
3747 ph10 178 #ifdef SUPPORT_UTF8
3748     if (utf8)
3749 ph10 180 {
3750 ph10 178 class_utf8 = TRUE;
3751 ph10 180 *class_utf8data++ = XCL_RANGE;
3752     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3753     class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3754     *class_utf8data++ = XCL_RANGE;
3755     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3756     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3757     }
3758     #endif
3759     continue;
3760 ph10 178
3761 nigel 77 #ifdef SUPPORT_UCP
3762 ph10 518 case ESC_p:
3763     case ESC_P:
3764     {
3765     BOOL negated;
3766     int pdata;
3767     int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3768     if (ptype < 0) goto FAILED;
3769     class_utf8 = TRUE;
3770     *class_utf8data++ = ((-c == ESC_p) != negated)?
3771     XCL_PROP : XCL_NOTPROP;
3772     *class_utf8data++ = ptype;
3773     *class_utf8data++ = pdata;
3774     class_charcount -= 2; /* Not a < 256 character */
3775     continue;
3776     }
3777 nigel 77 #endif
3778 ph10 518 /* Unrecognized escapes are faulted if PCRE is running in its
3779     strict mode. By default, for compatibility with Perl, they are
3780     treated as literals. */
3781 nigel 77
3782 ph10 518 default:
3783     if ((options & PCRE_EXTRA) != 0)
3784     {
3785     *errorcodeptr = ERR7;
3786     goto FAILED;
3787     }
3788     class_charcount -= 2; /* Undo the default count from above */
3789     c = *ptr; /* Get the final character and fall through */
3790     break;
3791 nigel 93 }
3792 nigel 77 }
3793    
3794     /* Fall through if we have a single character (c >= 0). This may be
3795 nigel 93 greater than 256 in UTF-8 mode. */
3796 nigel 77
3797     } /* End of backslash handling */
3798    
3799     /* A single character may be followed by '-' to form a range. However,
3800     Perl does not permit ']' to be the end of the range. A '-' character
3801 nigel 93 at the end is treated as a literal. Perl ignores orphaned \E sequences
3802     entirely. The code for handling \Q and \E is messy. */
3803 nigel 77
3804 nigel 93 CHECK_RANGE:
3805 ph10 391 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3806 nigel 77 {
3807 nigel 93 inescq = FALSE;
3808     ptr += 2;
3809     }
3810    
3811     oldptr = ptr;
3812 ph10 231
3813 ph10 230 /* Remember \r or \n */
3814 ph10 231
3815 ph10 391 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3816 ph10 231
3817 ph10 230 /* Check for range */
3818 nigel 93
3819 ph10 391 if (!inescq && ptr[1] == CHAR_MINUS)
3820 nigel 93 {
3821 nigel 77 int d;
3822     ptr += 2;
3823 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3824 nigel 77
3825 nigel 93 /* If we hit \Q (not followed by \E) at this point, go into escaped
3826     mode. */
3827    
3828 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3829 nigel 93 {
3830     ptr += 2;
3831 ph10 392 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3832 ph10 391 { ptr += 2; continue; }
3833 nigel 93 inescq = TRUE;
3834     break;
3835     }
3836    
3837 ph10 391 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3838 nigel 93 {
3839     ptr = oldptr;
3840     goto LONE_SINGLE_CHARACTER;
3841     }
3842    
3843 nigel 77 #ifdef SUPPORT_UTF8
3844     if (utf8)
3845     { /* Braces are required because the */
3846     GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3847     }
3848     else
3849     #endif
3850