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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 438 - (hide annotations) (download)
Sun Sep 6 20:00:47 2009 UTC (5 years, 1 month ago) by ph10
File MIME type: text/plain
File size: 212803 byte(s)
Fix internal error for forward reference with [^m] interposing.

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 381 Copyright (c) 1997-2009 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 nigel 85 /* When DEBUG is defined, we need the pcre_printint() function, which is also
57     used by pcretest. DEBUG is not defined when building a production library. */
58    
59     #ifdef DEBUG
60     #include "pcre_printint.src"
61     #endif
62    
63    
64 ph10 178 /* Macro for setting individual bits in class bitmaps. */
65    
66     #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
67    
68 ph10 202 /* Maximum length value to check against when making sure that the integer that
69     holds the compiled pattern length does not overflow. We make it a bit less than
70     INT_MAX to allow for adding in group terminating bytes, so that we don't have
71     to check them every time. */
72 ph10 178
73 ph10 202 #define OFLOW_MAX (INT_MAX - 20)
74    
75    
76 nigel 77 /*************************************************
77     * Code parameters and static tables *
78     *************************************************/
79    
80 nigel 93 /* This value specifies the size of stack workspace that is used during the
81     first pre-compile phase that determines how much memory is required. The regex
82     is partly compiled into this space, but the compiled parts are discarded as
83     soon as they can be, so that hopefully there will never be an overrun. The code
84     does, however, check for an overrun. The largest amount I've seen used is 218,
85     so this number is very generous.
86 nigel 77
87 nigel 93 The same workspace is used during the second, actual compile phase for
88     remembering forward references to groups so that they can be filled in at the
89     end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
90     is 4 there is plenty of room. */
91 nigel 77
92 nigel 93 #define COMPILE_WORK_SIZE (4096)
93 nigel 77
94 nigel 93
95 nigel 77 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
96     are simple data values; negative values are for special things like \d and so
97     on. Zero means further processing is needed (for things like \x), or the escape
98     is invalid. */
99    
100 ph10 391 #ifndef EBCDIC
101    
102     /* This is the "normal" table for ASCII systems or for EBCDIC systems running
103 ph10 392 in UTF-8 mode. */
104 ph10 391
105 ph10 392 static const short int escapes[] = {
106 ph10 391 0, 0,
107     0, 0,
108 ph10 392 0, 0,
109     0, 0,
110     0, 0,
111 ph10 391 CHAR_COLON, CHAR_SEMICOLON,
112 ph10 392 CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN,
113 ph10 391 CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK,
114 ph10 392 CHAR_COMMERCIAL_AT, -ESC_A,
115     -ESC_B, -ESC_C,
116     -ESC_D, -ESC_E,
117     0, -ESC_G,
118     -ESC_H, 0,
119     0, -ESC_K,
120 ph10 391 0, 0,
121 ph10 392 0, 0,
122 ph10 391 -ESC_P, -ESC_Q,
123     -ESC_R, -ESC_S,
124 ph10 392 0, 0,
125     -ESC_V, -ESC_W,
126     -ESC_X, 0,
127     -ESC_Z, CHAR_LEFT_SQUARE_BRACKET,
128 ph10 391 CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET,
129 ph10 392 CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE,
130 ph10 391 CHAR_GRAVE_ACCENT, 7,
131 ph10 392 -ESC_b, 0,
132     -ESC_d, ESC_e,
133 ph10 391 ESC_f, 0,
134     -ESC_h, 0,
135 ph10 392 0, -ESC_k,
136 ph10 391 0, 0,
137     ESC_n, 0,
138 ph10 392 -ESC_p, 0,
139     ESC_r, -ESC_s,
140 ph10 391 ESC_tee, 0,
141 ph10 392 -ESC_v, -ESC_w,
142     0, 0,
143 ph10 391 -ESC_z
144 nigel 77 };
145    
146 ph10 392 #else
147 ph10 391
148     /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
149    
150 nigel 77 static const short int escapes[] = {
151     /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
152     /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
153     /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
154     /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
155     /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
156     /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
157     /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
158     /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
159 ph10 178 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
160 nigel 93 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
161 nigel 77 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
162 ph10 178 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
163 nigel 77 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
164     /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
165     /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
166     /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
167 ph10 178 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
168 ph10 195 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
169 nigel 93 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
170 ph10 178 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
171 nigel 77 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
172     /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
173     /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
174     };
175     #endif
176    
177    
178 ph10 243 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
179     searched linearly. Put all the names into a single string, in order to reduce
180 ph10 392 the number of relocations when a shared library is dynamically linked. The
181     string is built from string macros so that it works in UTF-8 mode on EBCDIC
182 ph10 391 platforms. */
183 ph10 210
184     typedef struct verbitem {
185     int len;
186     int op;
187 ph10 211 } verbitem;
188 ph10 210
189 ph10 240 static const char verbnames[] =
190 ph10 391 STRING_ACCEPT0
191     STRING_COMMIT0
192     STRING_F0
193     STRING_FAIL0
194     STRING_PRUNE0
195     STRING_SKIP0
196     STRING_THEN;
197 ph10 240
198 ph10 327 static const verbitem verbs[] = {
199 ph10 240 { 6, OP_ACCEPT },
200     { 6, OP_COMMIT },
201     { 1, OP_FAIL },
202     { 4, OP_FAIL },
203     { 5, OP_PRUNE },
204     { 4, OP_SKIP },
205     { 4, OP_THEN }
206 ph10 210 };
207    
208 ph10 327 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
209 ph10 210
210    
211 ph10 243 /* Tables of names of POSIX character classes and their lengths. The names are
212     now all in a single string, to reduce the number of relocations when a shared
213 ph10 240 library is dynamically loaded. The list of lengths is terminated by a zero
214     length entry. The first three must be alpha, lower, upper, as this is assumed
215     for handling case independence. */
216 nigel 77
217 ph10 240 static const char posix_names[] =
218 ph10 392 STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0
219     STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0
220 ph10 391 STRING_graph0 STRING_print0 STRING_punct0 STRING_space0
221     STRING_word0 STRING_xdigit;
222 nigel 77
223     static const uschar posix_name_lengths[] = {
224     5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
225    
226 nigel 87 /* Table of class bit maps for each POSIX class. Each class is formed from a
227     base map, with an optional addition or removal of another map. Then, for some
228     classes, there is some additional tweaking: for [:blank:] the vertical space
229     characters are removed, and for [:alpha:] and [:alnum:] the underscore
230     character is removed. The triples in the table consist of the base map offset,
231     second map offset or -1 if no second map, and a non-negative value for map
232     addition or a negative value for map subtraction (if there are two maps). The
233     absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
234     remove vertical space characters, 2 => remove underscore. */
235 nigel 77
236     static const int posix_class_maps[] = {
237 nigel 87 cbit_word, cbit_digit, -2, /* alpha */
238     cbit_lower, -1, 0, /* lower */
239     cbit_upper, -1, 0, /* upper */
240     cbit_word, -1, 2, /* alnum - word without underscore */
241     cbit_print, cbit_cntrl, 0, /* ascii */
242     cbit_space, -1, 1, /* blank - a GNU extension */
243     cbit_cntrl, -1, 0, /* cntrl */
244     cbit_digit, -1, 0, /* digit */
245     cbit_graph, -1, 0, /* graph */
246     cbit_print, -1, 0, /* print */
247     cbit_punct, -1, 0, /* punct */
248     cbit_space, -1, 0, /* space */
249     cbit_word, -1, 0, /* word - a Perl extension */
250     cbit_xdigit,-1, 0 /* xdigit */
251 nigel 77 };
252    
253    
254 nigel 93 #define STRING(a) # a
255     #define XSTRING(s) STRING(s)
256    
257 nigel 77 /* The texts of compile-time error messages. These are "char *" because they
258 nigel 93 are passed to the outside world. Do not ever re-use any error number, because
259     they are documented. Always add a new error instead. Messages marked DEAD below
260 ph10 243 are no longer used. This used to be a table of strings, but in order to reduce
261     the number of relocations needed when a shared library is loaded dynamically,
262     it is now one long string. We cannot use a table of offsets, because the
263     lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
264     simply count through to the one we want - this isn't a performance issue
265 ph10 240 because these strings are used only when there is a compilation error. */
266 nigel 77
267 ph10 240 static const char error_texts[] =
268     "no error\0"
269     "\\ at end of pattern\0"
270     "\\c at end of pattern\0"
271     "unrecognized character follows \\\0"
272     "numbers out of order in {} quantifier\0"
273 nigel 77 /* 5 */
274 ph10 240 "number too big in {} quantifier\0"
275     "missing terminating ] for character class\0"
276     "invalid escape sequence in character class\0"
277     "range out of order in character class\0"
278     "nothing to repeat\0"
279 nigel 77 /* 10 */
280 ph10 240 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
281     "internal error: unexpected repeat\0"
282 ph10 269 "unrecognized character after (? or (?-\0"
283 ph10 240 "POSIX named classes are supported only within a class\0"
284     "missing )\0"
285 nigel 77 /* 15 */
286 ph10 240 "reference to non-existent subpattern\0"
287     "erroffset passed as NULL\0"
288     "unknown option bit(s) set\0"
289     "missing ) after comment\0"
290     "parentheses nested too deeply\0" /** DEAD **/
291 nigel 77 /* 20 */
292 ph10 240 "regular expression is too large\0"
293     "failed to get memory\0"
294     "unmatched parentheses\0"
295     "internal error: code overflow\0"
296     "unrecognized character after (?<\0"
297 nigel 77 /* 25 */
298 ph10 240 "lookbehind assertion is not fixed length\0"
299     "malformed number or name after (?(\0"
300     "conditional group contains more than two branches\0"
301     "assertion expected after (?(\0"
302     "(?R or (?[+-]digits must be followed by )\0"
303 nigel 77 /* 30 */
304 ph10 240 "unknown POSIX class name\0"
305     "POSIX collating elements are not supported\0"
306     "this version of PCRE is not compiled with PCRE_UTF8 support\0"
307     "spare error\0" /** DEAD **/
308     "character value in \\x{...} sequence is too large\0"
309 nigel 77 /* 35 */
310 ph10 240 "invalid condition (?(0)\0"
311     "\\C not allowed in lookbehind assertion\0"
312     "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
313     "number after (?C is > 255\0"
314     "closing ) for (?C expected\0"
315 nigel 77 /* 40 */
316 ph10 240 "recursive call could loop indefinitely\0"
317     "unrecognized character after (?P\0"
318     "syntax error in subpattern name (missing terminator)\0"
319     "two named subpatterns have the same name\0"
320     "invalid UTF-8 string\0"
321 nigel 77 /* 45 */
322 ph10 240 "support for \\P, \\p, and \\X has not been compiled\0"
323     "malformed \\P or \\p sequence\0"
324     "unknown property name after \\P or \\p\0"
325     "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
326     "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
327 nigel 91 /* 50 */
328 ph10 240 "repeated subpattern is too long\0" /** DEAD **/
329     "octal value is greater than \\377 (not in UTF-8 mode)\0"
330     "internal error: overran compiling workspace\0"
331     "internal error: previously-checked referenced subpattern not found\0"
332     "DEFINE group contains more than one branch\0"
333 nigel 93 /* 55 */
334 ph10 240 "repeating a DEFINE group is not allowed\0"
335     "inconsistent NEWLINE options\0"
336 ph10 333 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
337     "a numbered reference must not be zero\0"
338 ph10 240 "(*VERB) with an argument is not supported\0"
339 ph10 211 /* 60 */
340 ph10 240 "(*VERB) not recognized\0"
341 ph10 268 "number is too big\0"
342 ph10 272 "subpattern name expected\0"
343 ph10 336 "digit expected after (?+\0"
344 ph10 345 "] is an invalid data character in JavaScript compatibility mode";
345 nigel 77
346    
347     /* Table to identify digits and hex digits. This is used when compiling
348     patterns. Note that the tables in chartables are dependent on the locale, and
349     may mark arbitrary characters as digits - but the PCRE compiling code expects
350     to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
351     a private table here. It costs 256 bytes, but it is a lot faster than doing
352     character value tests (at least in some simple cases I timed), and in some
353     applications one wants PCRE to compile efficiently as well as match
354     efficiently.
355    
356     For convenience, we use the same bit definitions as in chartables:
357    
358     0x04 decimal digit
359     0x08 hexadecimal digit
360    
361     Then we can use ctype_digit and ctype_xdigit in the code. */
362    
363 ph10 392 #ifndef EBCDIC
364 ph10 391
365 ph10 392 /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
366 ph10 391 UTF-8 mode. */
367    
368 nigel 77 static const unsigned char digitab[] =
369     {
370     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
371     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
372     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
373     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
374     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
375     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
376     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
377     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
378     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
379     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
380     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
381     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
382     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
383     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
384     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
385     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
386     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
387     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
388     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
389     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
390     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
391     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
392     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
393     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
394     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
395     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
396     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
397     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
398     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
399     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
400     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
401     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
402    
403 ph10 392 #else
404 ph10 391
405     /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
406    
407 nigel 77 static const unsigned char digitab[] =
408     {
409     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
410     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
411     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
412     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
413     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
414     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
415     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
416     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
417     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
418     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
419     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
420 ph10 97 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
421 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
422     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
423     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
424     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
425     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
426     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
427     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
428     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
429     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
430     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
431     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
432     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
433     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
434     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
435     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
436     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
437     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
438     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
439     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
440     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
441    
442     static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
443     0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
444     0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
445     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
446     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
447     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
448     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
449     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
450     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
451     0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
452     0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
453     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
454 ph10 97 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
455 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
456     0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
457     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
458     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
459     0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
460     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
461     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
462     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
463     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
464     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
465     0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
466     0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
467     0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
468     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
469     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
470     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
471     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
472     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
473     0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
474     0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
475     #endif
476    
477    
478     /* Definition to allow mutual recursion */
479    
480     static BOOL
481 ph10 180 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
482 ph10 175 int *, int *, branch_chain *, compile_data *, int *);
483 nigel 77
484    
485    
486     /*************************************************
487 ph10 240 * Find an error text *
488     *************************************************/
489    
490 ph10 243 /* The error texts are now all in one long string, to save on relocations. As
491     some of the text is of unknown length, we can't use a table of offsets.
492     Instead, just count through the strings. This is not a performance issue
493 ph10 240 because it happens only when there has been a compilation error.
494    
495     Argument: the error number
496     Returns: pointer to the error string
497     */
498    
499     static const char *
500     find_error_text(int n)
501     {
502     const char *s = error_texts;
503 ph10 369 for (; n > 0; n--) while (*s++ != 0) {};
504 ph10 240 return s;
505     }
506    
507    
508     /*************************************************
509 nigel 77 * Handle escapes *
510     *************************************************/
511    
512     /* This function is called when a \ has been encountered. It either returns a
513     positive value for a simple escape such as \n, or a negative value which
514 nigel 93 encodes one of the more complicated things such as \d. A backreference to group
515     n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
516     UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
517     ptr is pointing at the \. On exit, it is on the final character of the escape
518     sequence.
519 nigel 77
520     Arguments:
521     ptrptr points to the pattern position pointer
522     errorcodeptr points to the errorcode variable
523     bracount number of previous extracting brackets
524     options the options bits
525     isclass TRUE if inside a character class
526    
527     Returns: zero or positive => a data character
528     negative => a special escape sequence
529 ph10 213 on error, errorcodeptr is set
530 nigel 77 */
531    
532     static int
533     check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
534     int options, BOOL isclass)
535     {
536 nigel 87 BOOL utf8 = (options & PCRE_UTF8) != 0;
537     const uschar *ptr = *ptrptr + 1;
538 nigel 77 int c, i;
539    
540 nigel 87 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
541     ptr--; /* Set pointer back to the last byte */
542    
543 nigel 77 /* If backslash is at the end of the pattern, it's an error. */
544    
545     if (c == 0) *errorcodeptr = ERR1;
546    
547 ph10 274 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
548     in a table. A non-zero result is something that can be returned immediately.
549 nigel 77 Otherwise further processing may be required. */
550    
551 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
552     else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
553     else if ((i = escapes[c - CHAR_0]) != 0) c = i;
554 nigel 77
555 ph10 97 #else /* EBCDIC coding */
556 ph10 274 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
557 nigel 77 else if ((i = escapes[c - 0x48]) != 0) c = i;
558     #endif
559    
560     /* Escapes that need further processing, or are illegal. */
561    
562     else
563     {
564     const uschar *oldptr;
565 nigel 93 BOOL braced, negated;
566    
567 nigel 77 switch (c)
568     {
569     /* A number of Perl escapes are not handled by PCRE. We give an explicit
570     error. */
571    
572 ph10 391 case CHAR_l:
573     case CHAR_L:
574     case CHAR_N:
575     case CHAR_u:
576     case CHAR_U:
577 nigel 77 *errorcodeptr = ERR37;
578     break;
579    
580 ph10 333 /* \g must be followed by one of a number of specific things:
581 ph10 345
582 ph10 333 (1) A number, either plain or braced. If positive, it is an absolute
583     backreference. If negative, it is a relative backreference. This is a Perl
584     5.10 feature.
585 ph10 345
586 ph10 333 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
587     is part of Perl's movement towards a unified syntax for back references. As
588     this is synonymous with \k{name}, we fudge it up by pretending it really
589     was \k.
590 ph10 345
591     (3) For Oniguruma compatibility we also support \g followed by a name or a
592     number either in angle brackets or in single quotes. However, these are
593     (possibly recursive) subroutine calls, _not_ backreferences. Just return
594 ph10 333 the -ESC_g code (cf \k). */
595 nigel 93
596 ph10 391 case CHAR_g:
597     if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
598 ph10 333 {
599     c = -ESC_g;
600 ph10 345 break;
601     }
602 ph10 333
603     /* Handle the Perl-compatible cases */
604 ph10 345
605 ph10 391 if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
606 nigel 93 {
607 ph10 171 const uschar *p;
608 ph10 391 for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
609     if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
610     if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
611 ph10 171 {
612     c = -ESC_k;
613     break;
614 ph10 172 }
615 nigel 93 braced = TRUE;
616     ptr++;
617     }
618     else braced = FALSE;
619    
620 ph10 391 if (ptr[1] == CHAR_MINUS)
621 nigel 93 {
622     negated = TRUE;
623     ptr++;
624     }
625     else negated = FALSE;
626    
627     c = 0;
628     while ((digitab[ptr[1]] & ctype_digit) != 0)
629 ph10 391 c = c * 10 + *(++ptr) - CHAR_0;
630 ph10 220
631 ph10 333 if (c < 0) /* Integer overflow */
632 ph10 213 {
633     *errorcodeptr = ERR61;
634     break;
635 ph10 220 }
636 ph10 345
637 ph10 391 if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
638 nigel 93 {
639     *errorcodeptr = ERR57;
640 ph10 213 break;
641 nigel 93 }
642 ph10 345
643 ph10 333 if (c == 0)
644     {
645     *errorcodeptr = ERR58;
646     break;
647 ph10 345 }
648 nigel 93
649     if (negated)
650     {
651     if (c > bracount)
652     {
653     *errorcodeptr = ERR15;
654 ph10 213 break;
655 nigel 93 }
656     c = bracount - (c - 1);
657     }
658    
659     c = -(ESC_REF + c);
660     break;
661    
662 nigel 77 /* The handling of escape sequences consisting of a string of digits
663     starting with one that is not zero is not straightforward. By experiment,
664     the way Perl works seems to be as follows:
665    
666     Outside a character class, the digits are read as a decimal number. If the
667     number is less than 10, or if there are that many previous extracting
668     left brackets, then it is a back reference. Otherwise, up to three octal
669     digits are read to form an escaped byte. Thus \123 is likely to be octal
670     123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
671     value is greater than 377, the least significant 8 bits are taken. Inside a
672     character class, \ followed by a digit is always an octal number. */
673    
674 ph10 391 case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
675     case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
676 nigel 77
677     if (!isclass)
678     {
679     oldptr = ptr;
680 ph10 391 c -= CHAR_0;
681 nigel 77 while ((digitab[ptr[1]] & ctype_digit) != 0)
682 ph10 391 c = c * 10 + *(++ptr) - CHAR_0;
683 ph10 333 if (c < 0) /* Integer overflow */
684 ph10 213 {
685     *errorcodeptr = ERR61;
686 ph10 220 break;
687     }
688 nigel 77 if (c < 10 || c <= bracount)
689     {
690     c = -(ESC_REF + c);
691     break;
692     }
693     ptr = oldptr; /* Put the pointer back and fall through */
694     }
695    
696     /* Handle an octal number following \. If the first digit is 8 or 9, Perl
697     generates a binary zero byte and treats the digit as a following literal.
698     Thus we have to pull back the pointer by one. */
699    
700 ph10 391 if ((c = *ptr) >= CHAR_8)
701 nigel 77 {
702     ptr--;
703     c = 0;
704     break;
705     }
706    
707     /* \0 always starts an octal number, but we may drop through to here with a
708 nigel 91 larger first octal digit. The original code used just to take the least
709     significant 8 bits of octal numbers (I think this is what early Perls used
710     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
711     than 3 octal digits. */
712 nigel 77
713 ph10 391 case CHAR_0:
714     c -= CHAR_0;
715     while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
716     c = c * 8 + *(++ptr) - CHAR_0;
717 nigel 91 if (!utf8 && c > 255) *errorcodeptr = ERR51;
718 nigel 77 break;
719    
720 nigel 87 /* \x is complicated. \x{ddd} is a character number which can be greater
721     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
722     treated as a data character. */
723 nigel 77
724 ph10 391 case CHAR_x:
725     if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
726 nigel 77 {
727     const uschar *pt = ptr + 2;
728 nigel 87 int count = 0;
729    
730 nigel 77 c = 0;
731     while ((digitab[*pt] & ctype_xdigit) != 0)
732     {
733 nigel 87 register int cc = *pt++;
734 ph10 391 if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
735 nigel 77 count++;
736 nigel 87
737 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
738     if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
739     c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
740 ph10 97 #else /* EBCDIC coding */
741 ph10 391 if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
742     c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
743 nigel 77 #endif
744     }
745 nigel 87
746 ph10 391 if (*pt == CHAR_RIGHT_CURLY_BRACKET)
747 nigel 77 {
748 nigel 87 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
749 nigel 77 ptr = pt;
750     break;
751     }
752 nigel 87
753 nigel 77 /* If the sequence of hex digits does not end with '}', then we don't
754     recognize this construct; fall through to the normal \x handling. */
755     }
756    
757 nigel 87 /* Read just a single-byte hex-defined char */
758 nigel 77
759     c = 0;
760     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
761     {
762 ph10 391 int cc; /* Some compilers don't like */
763     cc = *(++ptr); /* ++ in initializers */
764     #ifndef EBCDIC /* ASCII/UTF-8 coding */
765     if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
766     c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
767 ph10 97 #else /* EBCDIC coding */
768 ph10 391 if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
769     c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
770 nigel 77 #endif
771     }
772     break;
773    
774 nigel 93 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
775     This coding is ASCII-specific, but then the whole concept of \cx is
776     ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
777 nigel 77
778 ph10 391 case CHAR_c:
779 nigel 77 c = *(++ptr);
780     if (c == 0)
781     {
782     *errorcodeptr = ERR2;
783 ph10 213 break;
784 nigel 77 }
785    
786 ph10 391 #ifndef EBCDIC /* ASCII/UTF-8 coding */
787     if (c >= CHAR_a && c <= CHAR_z) c -= 32;
788 nigel 77 c ^= 0x40;
789 ph10 97 #else /* EBCDIC coding */
790 ph10 391 if (c >= CHAR_a && c <= CHAR_z) c += 64;
791 nigel 77 c ^= 0xC0;
792     #endif
793     break;
794    
795     /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
796 ph10 274 other alphanumeric following \ is an error if PCRE_EXTRA was set;
797     otherwise, for Perl compatibility, it is a literal. This code looks a bit
798     odd, but there used to be some cases other than the default, and there may
799     be again in future, so I haven't "optimized" it. */
800 nigel 77
801     default:
802     if ((options & PCRE_EXTRA) != 0) switch(c)
803     {
804     default:
805     *errorcodeptr = ERR3;
806     break;
807     }
808     break;
809     }
810     }
811    
812     *ptrptr = ptr;
813     return c;
814     }
815    
816    
817    
818     #ifdef SUPPORT_UCP
819     /*************************************************
820     * Handle \P and \p *
821     *************************************************/
822    
823     /* This function is called after \P or \p has been encountered, provided that
824     PCRE is compiled with support for Unicode properties. On entry, ptrptr is
825     pointing at the P or p. On exit, it is pointing at the final character of the
826     escape sequence.
827    
828     Argument:
829     ptrptr points to the pattern position pointer
830     negptr points to a boolean that is set TRUE for negation else FALSE
831 nigel 87 dptr points to an int that is set to the detailed property value
832 nigel 77 errorcodeptr points to the error code variable
833    
834 nigel 87 Returns: type value from ucp_type_table, or -1 for an invalid type
835 nigel 77 */
836    
837     static int
838 nigel 87 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
839 nigel 77 {
840     int c, i, bot, top;
841     const uschar *ptr = *ptrptr;
842 nigel 87 char name[32];
843 nigel 77
844     c = *(++ptr);
845     if (c == 0) goto ERROR_RETURN;
846    
847     *negptr = FALSE;
848    
849 nigel 87 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
850     negation. */
851 nigel 77
852 ph10 391 if (c == CHAR_LEFT_CURLY_BRACKET)
853 nigel 77 {
854 ph10 391 if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
855 nigel 77 {
856     *negptr = TRUE;
857     ptr++;
858     }
859 ph10 199 for (i = 0; i < (int)sizeof(name) - 1; i++)
860 nigel 77 {
861     c = *(++ptr);
862     if (c == 0) goto ERROR_RETURN;
863 ph10 391 if (c == CHAR_RIGHT_CURLY_BRACKET) break;
864 nigel 77 name[i] = c;
865     }
866 ph10 391 if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
867 nigel 77 name[i] = 0;
868     }
869    
870     /* Otherwise there is just one following character */
871    
872     else
873     {
874     name[0] = c;
875     name[1] = 0;
876     }
877    
878     *ptrptr = ptr;
879    
880     /* Search for a recognized property name using binary chop */
881    
882     bot = 0;
883     top = _pcre_utt_size;
884    
885     while (bot < top)
886     {
887 nigel 87 i = (bot + top) >> 1;
888 ph10 240 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
889 nigel 87 if (c == 0)
890     {
891     *dptr = _pcre_utt[i].value;
892     return _pcre_utt[i].type;
893     }
894 nigel 77 if (c > 0) bot = i + 1; else top = i;
895     }
896    
897     *errorcodeptr = ERR47;
898     *ptrptr = ptr;
899     return -1;
900    
901     ERROR_RETURN:
902     *errorcodeptr = ERR46;
903     *ptrptr = ptr;
904     return -1;
905     }
906     #endif
907    
908    
909    
910    
911     /*************************************************
912     * Check for counted repeat *
913     *************************************************/
914    
915     /* This function is called when a '{' is encountered in a place where it might
916     start a quantifier. It looks ahead to see if it really is a quantifier or not.
917     It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
918     where the ddds are digits.
919    
920     Arguments:
921     p pointer to the first char after '{'
922    
923     Returns: TRUE or FALSE
924     */
925    
926     static BOOL
927     is_counted_repeat(const uschar *p)
928     {
929     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
930     while ((digitab[*p] & ctype_digit) != 0) p++;
931 ph10 391 if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
932 nigel 77
933 ph10 391 if (*p++ != CHAR_COMMA) return FALSE;
934     if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
935 nigel 77
936     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
937     while ((digitab[*p] & ctype_digit) != 0) p++;
938    
939 ph10 391 return (*p == CHAR_RIGHT_CURLY_BRACKET);
940 nigel 77 }
941    
942    
943    
944     /*************************************************
945     * Read repeat counts *
946     *************************************************/
947    
948     /* Read an item of the form {n,m} and return the values. This is called only
949     after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
950     so the syntax is guaranteed to be correct, but we need to check the values.
951    
952     Arguments:
953     p pointer to first char after '{'
954     minp pointer to int for min
955     maxp pointer to int for max
956     returned as -1 if no max
957     errorcodeptr points to error code variable
958    
959     Returns: pointer to '}' on success;
960     current ptr on error, with errorcodeptr set non-zero
961     */
962    
963     static const uschar *
964     read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
965     {
966     int min = 0;
967     int max = -1;
968    
969 nigel 81 /* Read the minimum value and do a paranoid check: a negative value indicates
970     an integer overflow. */
971    
972 ph10 391 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
973 nigel 81 if (min < 0 || min > 65535)
974     {
975     *errorcodeptr = ERR5;
976     return p;
977     }
978 nigel 77
979 nigel 81 /* Read the maximum value if there is one, and again do a paranoid on its size.
980     Also, max must not be less than min. */
981    
982 ph10 391 if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
983 nigel 77 {
984 ph10 391 if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
985 nigel 77 {
986     max = 0;
987 ph10 391 while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
988 nigel 81 if (max < 0 || max > 65535)
989     {
990     *errorcodeptr = ERR5;
991     return p;
992     }
993 nigel 77 if (max < min)
994     {
995     *errorcodeptr = ERR4;
996     return p;
997     }
998     }
999     }
1000    
1001 nigel 81 /* Fill in the required variables, and pass back the pointer to the terminating
1002     '}'. */
1003 nigel 77
1004 nigel 81 *minp = min;
1005     *maxp = max;
1006 nigel 77 return p;
1007     }
1008    
1009    
1010    
1011     /*************************************************
1012 ph10 408 * Subroutine for finding forward reference *
1013 nigel 91 *************************************************/
1014    
1015 ph10 408 /* This recursive function is called only from find_parens() below. The
1016     top-level call starts at the beginning of the pattern. All other calls must
1017     start at a parenthesis. It scans along a pattern's text looking for capturing
1018 nigel 93 subpatterns, and counting them. If it finds a named pattern that matches the
1019     name it is given, it returns its number. Alternatively, if the name is NULL, it
1020 ph10 408 returns when it reaches a given numbered subpattern. We know that if (?P< is
1021     encountered, the name will be terminated by '>' because that is checked in the
1022 ph10 411 first pass. Recursion is used to keep track of subpatterns that reset the
1023 ph10 408 capturing group numbers - the (?| feature.
1024 nigel 91
1025     Arguments:
1026 ph10 408 ptrptr address of the current character pointer (updated)
1027 ph10 345 cd compile background data
1028 nigel 93 name name to seek, or NULL if seeking a numbered subpattern
1029     lorn name length, or subpattern number if name is NULL
1030     xmode TRUE if we are in /x mode
1031 ph10 411 count pointer to the current capturing subpattern number (updated)
1032 nigel 91
1033     Returns: the number of the named subpattern, or -1 if not found
1034     */
1035    
1036     static int
1037 ph10 408 find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1038     BOOL xmode, int *count)
1039 nigel 91 {
1040 ph10 408 uschar *ptr = *ptrptr;
1041     int start_count = *count;
1042     int hwm_count = start_count;
1043     BOOL dup_parens = FALSE;
1044 nigel 93
1045 ph10 411 /* If the first character is a parenthesis, check on the type of group we are
1046 ph10 408 dealing with. The very first call may not start with a parenthesis. */
1047    
1048     if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1049     {
1050     if (ptr[1] == CHAR_QUESTION_MARK &&
1051 ph10 411 ptr[2] == CHAR_VERTICAL_LINE)
1052 ph10 408 {
1053     ptr += 3;
1054 ph10 411 dup_parens = TRUE;
1055     }
1056 ph10 408
1057     /* Handle a normal, unnamed capturing parenthesis */
1058 ph10 411
1059 ph10 408 else if (ptr[1] != CHAR_QUESTION_MARK && ptr[1] != CHAR_ASTERISK)
1060     {
1061     *count += 1;
1062     if (name == NULL && *count == lorn) return *count;
1063 ph10 411 ptr++;
1064 ph10 408 }
1065    
1066     /* Handle a condition. If it is an assertion, just carry on so that it
1067     is processed as normal. If not, skip to the closing parenthesis of the
1068 ph10 411 condition (there can't be any nested parens. */
1069    
1070 ph10 408 else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1071     {
1072 ph10 411 ptr += 2;
1073 ph10 408 if (ptr[1] != CHAR_QUESTION_MARK)
1074     {
1075     while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1076 ph10 411 if (*ptr != 0) ptr++;
1077 ph10 408 }
1078 ph10 411 }
1079    
1080 ph10 408 /* We have either (? or (* and not a condition */
1081    
1082     else
1083 ph10 411 {
1084 ph10 408 ptr += 2;
1085     if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1086    
1087     /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1088 ph10 411
1089 ph10 408 if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1090     ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1091     {
1092     int term;
1093     const uschar *thisname;
1094     *count += 1;
1095     if (name == NULL && *count == lorn) return *count;
1096     term = *ptr++;
1097     if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1098     thisname = ptr;
1099     while (*ptr != term) ptr++;
1100     if (name != NULL && lorn == ptr - thisname &&
1101     strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1102     return *count;
1103 ph10 438 term++;
1104 ph10 411 }
1105 ph10 408 }
1106 ph10 411 }
1107 ph10 408
1108 ph10 411 /* Past any initial parenthesis handling, scan for parentheses or vertical
1109 ph10 408 bars. */
1110    
1111 nigel 91 for (; *ptr != 0; ptr++)
1112     {
1113 nigel 93 /* Skip over backslashed characters and also entire \Q...\E */
1114    
1115 ph10 391 if (*ptr == CHAR_BACKSLASH)
1116 nigel 93 {
1117 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1118 ph10 391 if (*ptr == CHAR_Q) for (;;)
1119 nigel 93 {
1120 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1121 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1122 ph10 391 if (*(++ptr) == CHAR_E) break;
1123 nigel 93 }
1124     continue;
1125     }
1126    
1127 ph10 340 /* Skip over character classes; this logic must be similar to the way they
1128     are handled for real. If the first character is '^', skip it. Also, if the
1129     first few characters (either before or after ^) are \Q\E or \E we skip them
1130 ph10 392 too. This makes for compatibility with Perl. Note the use of STR macros to
1131 ph10 391 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1132 nigel 93
1133 ph10 391 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1134 nigel 93 {
1135 ph10 340 BOOL negate_class = FALSE;
1136     for (;;)
1137     {
1138 ph10 438 if (ptr[1] == CHAR_BACKSLASH)
1139 ph10 340 {
1140 ph10 438 if (ptr[2] == CHAR_E)
1141     ptr+= 2;
1142     else if (strncmp((const char *)ptr+2,
1143 ph10 392 STR_Q STR_BACKSLASH STR_E, 3) == 0)
1144 ph10 438 ptr += 4;
1145 ph10 392 else
1146 ph10 391 break;
1147 ph10 340 }
1148 ph10 438 else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1149     {
1150 ph10 340 negate_class = TRUE;
1151 ph10 438 ptr++;
1152     }
1153 ph10 340 else break;
1154     }
1155    
1156     /* If the next character is ']', it is a data character that must be
1157 ph10 341 skipped, except in JavaScript compatibility mode. */
1158 ph10 345
1159 ph10 392 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1160 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1161 ph10 345 ptr++;
1162    
1163 ph10 391 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1164 nigel 93 {
1165 ph10 220 if (*ptr == 0) return -1;
1166 ph10 391 if (*ptr == CHAR_BACKSLASH)
1167 nigel 93 {
1168 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1169 ph10 391 if (*ptr == CHAR_Q) for (;;)
1170 nigel 93 {
1171 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1172 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1173 ph10 391 if (*(++ptr) == CHAR_E) break;
1174 nigel 93 }
1175     continue;
1176     }
1177     }
1178     continue;
1179     }
1180    
1181     /* Skip comments in /x mode */
1182    
1183 ph10 391 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1184 nigel 93 {
1185 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1186 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1187 nigel 93 continue;
1188     }
1189    
1190 ph10 408 /* Check for the special metacharacters */
1191 ph10 411
1192 ph10 408 if (*ptr == CHAR_LEFT_PARENTHESIS)
1193 nigel 93 {
1194 ph10 408 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1195     if (rc > 0) return rc;
1196     if (*ptr == 0) goto FAIL_EXIT;
1197 nigel 93 }
1198 ph10 411
1199 ph10 408 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1200     {
1201     if (dup_parens && *count < hwm_count) *count = hwm_count;
1202     *ptrptr = ptr;
1203     return -1;
1204     }
1205 ph10 411
1206     else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1207 ph10 408 {
1208     if (*count > hwm_count) hwm_count = *count;
1209     *count = start_count;
1210 ph10 411 }
1211 ph10 408 }
1212 nigel 93
1213 ph10 408 FAIL_EXIT:
1214     *ptrptr = ptr;
1215     return -1;
1216     }
1217 nigel 93
1218    
1219    
1220    
1221 ph10 408 /*************************************************
1222     * Find forward referenced subpattern *
1223     *************************************************/
1224 nigel 93
1225 ph10 408 /* This function scans along a pattern's text looking for capturing
1226     subpatterns, and counting them. If it finds a named pattern that matches the
1227     name it is given, it returns its number. Alternatively, if the name is NULL, it
1228     returns when it reaches a given numbered subpattern. This is used for forward
1229     references to subpatterns. We used to be able to start this scan from the
1230     current compiling point, using the current count value from cd->bracount, and
1231     do it all in a single loop, but the addition of the possibility of duplicate
1232     subpattern numbers means that we have to scan from the very start, in order to
1233     take account of such duplicates, and to use a recursive function to keep track
1234     of the different types of group.
1235    
1236     Arguments:
1237     cd compile background data
1238     name name to seek, or NULL if seeking a numbered subpattern
1239     lorn name length, or subpattern number if name is NULL
1240     xmode TRUE if we are in /x mode
1241    
1242     Returns: the number of the found subpattern, or -1 if not found
1243     */
1244    
1245     static int
1246     find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1247     {
1248     uschar *ptr = (uschar *)cd->start_pattern;
1249     int count = 0;
1250     int rc;
1251    
1252     /* If the pattern does not start with an opening parenthesis, the first call
1253     to find_parens_sub() will scan right to the end (if necessary). However, if it
1254     does start with a parenthesis, find_parens_sub() will return when it hits the
1255     matching closing parens. That is why we have to have a loop. */
1256    
1257 ph10 411 for (;;)
1258     {
1259 ph10 408 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1260 ph10 411 if (rc > 0 || *ptr++ == 0) break;
1261     }
1262    
1263 ph10 408 return rc;
1264 nigel 91 }
1265    
1266    
1267    
1268 ph10 408
1269 nigel 91 /*************************************************
1270 nigel 77 * Find first significant op code *
1271     *************************************************/
1272    
1273     /* This is called by several functions that scan a compiled expression looking
1274     for a fixed first character, or an anchoring op code etc. It skips over things
1275     that do not influence this. For some calls, a change of option is important.
1276     For some calls, it makes sense to skip negative forward and all backward
1277     assertions, and also the \b assertion; for others it does not.
1278    
1279     Arguments:
1280     code pointer to the start of the group
1281     options pointer to external options
1282     optbit the option bit whose changing is significant, or
1283     zero if none are
1284     skipassert TRUE if certain assertions are to be skipped
1285    
1286     Returns: pointer to the first significant opcode
1287     */
1288    
1289     static const uschar*
1290     first_significant_code(const uschar *code, int *options, int optbit,
1291     BOOL skipassert)
1292     {
1293     for (;;)
1294     {
1295     switch ((int)*code)
1296     {
1297     case OP_OPT:
1298     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1299     *options = (int)code[1];
1300     code += 2;
1301     break;
1302    
1303     case OP_ASSERT_NOT:
1304     case OP_ASSERTBACK:
1305     case OP_ASSERTBACK_NOT:
1306     if (!skipassert) return code;
1307     do code += GET(code, 1); while (*code == OP_ALT);
1308     code += _pcre_OP_lengths[*code];
1309     break;
1310    
1311     case OP_WORD_BOUNDARY:
1312     case OP_NOT_WORD_BOUNDARY:
1313     if (!skipassert) return code;
1314     /* Fall through */
1315    
1316     case OP_CALLOUT:
1317     case OP_CREF:
1318 nigel 93 case OP_RREF:
1319     case OP_DEF:
1320 nigel 77 code += _pcre_OP_lengths[*code];
1321     break;
1322    
1323     default:
1324     return code;
1325     }
1326     }
1327     /* Control never reaches here */
1328     }
1329    
1330    
1331    
1332    
1333     /*************************************************
1334     * Find the fixed length of a pattern *
1335     *************************************************/
1336    
1337     /* Scan a pattern and compute the fixed length of subject that will match it,
1338     if the length is fixed. This is needed for dealing with backward assertions.
1339     In UTF8 mode, the result is in characters rather than bytes.
1340    
1341     Arguments:
1342     code points to the start of the pattern (the bracket)
1343     options the compiling options
1344    
1345     Returns: the fixed length, or -1 if there is no fixed length,
1346     or -2 if \C was encountered
1347     */
1348    
1349     static int
1350     find_fixedlength(uschar *code, int options)
1351     {
1352     int length = -1;
1353    
1354     register int branchlength = 0;
1355     register uschar *cc = code + 1 + LINK_SIZE;
1356    
1357     /* Scan along the opcodes for this branch. If we get to the end of the
1358     branch, check the length against that of the other branches. */
1359    
1360     for (;;)
1361     {
1362     int d;
1363     register int op = *cc;
1364     switch (op)
1365     {
1366 nigel 93 case OP_CBRA:
1367 nigel 77 case OP_BRA:
1368     case OP_ONCE:
1369     case OP_COND:
1370 nigel 93 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1371 nigel 77 if (d < 0) return d;
1372     branchlength += d;
1373     do cc += GET(cc, 1); while (*cc == OP_ALT);
1374     cc += 1 + LINK_SIZE;
1375     break;
1376    
1377     /* Reached end of a branch; if it's a ket it is the end of a nested
1378     call. If it's ALT it is an alternation in a nested call. If it is
1379     END it's the end of the outer call. All can be handled by the same code. */
1380    
1381     case OP_ALT:
1382     case OP_KET:
1383     case OP_KETRMAX:
1384     case OP_KETRMIN:
1385     case OP_END:
1386     if (length < 0) length = branchlength;
1387     else if (length != branchlength) return -1;
1388     if (*cc != OP_ALT) return length;
1389     cc += 1 + LINK_SIZE;
1390     branchlength = 0;
1391     break;
1392    
1393     /* Skip over assertive subpatterns */
1394    
1395     case OP_ASSERT:
1396     case OP_ASSERT_NOT:
1397     case OP_ASSERTBACK:
1398     case OP_ASSERTBACK_NOT:
1399     do cc += GET(cc, 1); while (*cc == OP_ALT);
1400     /* Fall through */
1401    
1402     /* Skip over things that don't match chars */
1403    
1404     case OP_REVERSE:
1405     case OP_CREF:
1406 nigel 93 case OP_RREF:
1407     case OP_DEF:
1408 nigel 77 case OP_OPT:
1409     case OP_CALLOUT:
1410     case OP_SOD:
1411     case OP_SOM:
1412     case OP_EOD:
1413     case OP_EODN:
1414     case OP_CIRC:
1415     case OP_DOLL:
1416     case OP_NOT_WORD_BOUNDARY:
1417     case OP_WORD_BOUNDARY:
1418     cc += _pcre_OP_lengths[*cc];
1419     break;
1420    
1421     /* Handle literal characters */
1422    
1423     case OP_CHAR:
1424     case OP_CHARNC:
1425 nigel 91 case OP_NOT:
1426 nigel 77 branchlength++;
1427     cc += 2;
1428     #ifdef SUPPORT_UTF8
1429 ph10 426 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1430     cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1431 nigel 77 #endif
1432     break;
1433    
1434     /* Handle exact repetitions. The count is already in characters, but we
1435     need to skip over a multibyte character in UTF8 mode. */
1436    
1437     case OP_EXACT:
1438     branchlength += GET2(cc,1);
1439     cc += 4;
1440     #ifdef SUPPORT_UTF8
1441 ph10 426 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1442     cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1443 nigel 77 #endif
1444     break;
1445    
1446     case OP_TYPEEXACT:
1447     branchlength += GET2(cc,1);
1448 ph10 220 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1449 nigel 77 cc += 4;
1450     break;
1451    
1452     /* Handle single-char matchers */
1453    
1454     case OP_PROP:
1455     case OP_NOTPROP:
1456 nigel 87 cc += 2;
1457 nigel 77 /* Fall through */
1458    
1459     case OP_NOT_DIGIT:
1460     case OP_DIGIT:
1461     case OP_NOT_WHITESPACE:
1462     case OP_WHITESPACE:
1463     case OP_NOT_WORDCHAR:
1464     case OP_WORDCHAR:
1465     case OP_ANY:
1466 ph10 342 case OP_ALLANY:
1467 nigel 77 branchlength++;
1468     cc++;
1469     break;
1470    
1471     /* The single-byte matcher isn't allowed */
1472    
1473     case OP_ANYBYTE:
1474     return -2;
1475    
1476     /* Check a class for variable quantification */
1477    
1478     #ifdef SUPPORT_UTF8
1479     case OP_XCLASS:
1480     cc += GET(cc, 1) - 33;
1481     /* Fall through */
1482     #endif
1483    
1484     case OP_CLASS:
1485     case OP_NCLASS:
1486     cc += 33;
1487    
1488     switch (*cc)
1489     {
1490     case OP_CRSTAR:
1491     case OP_CRMINSTAR:
1492     case OP_CRQUERY:
1493     case OP_CRMINQUERY:
1494     return -1;
1495    
1496     case OP_CRRANGE:
1497     case OP_CRMINRANGE:
1498     if (GET2(cc,1) != GET2(cc,3)) return -1;
1499     branchlength += GET2(cc,1);
1500     cc += 5;
1501     break;
1502    
1503     default:
1504     branchlength++;
1505     }
1506     break;
1507    
1508     /* Anything else is variable length */
1509    
1510     default:
1511     return -1;
1512     }
1513     }
1514     /* Control never gets here */
1515     }
1516    
1517    
1518    
1519    
1520     /*************************************************
1521     * Scan compiled regex for numbered bracket *
1522     *************************************************/
1523    
1524     /* This little function scans through a compiled pattern until it finds a
1525     capturing bracket with the given number.
1526    
1527     Arguments:
1528     code points to start of expression
1529     utf8 TRUE in UTF-8 mode
1530     number the required bracket number
1531    
1532     Returns: pointer to the opcode for the bracket, or NULL if not found
1533     */
1534    
1535     static const uschar *
1536     find_bracket(const uschar *code, BOOL utf8, int number)
1537     {
1538     for (;;)
1539     {
1540     register int c = *code;
1541     if (c == OP_END) return NULL;
1542 nigel 91
1543     /* XCLASS is used for classes that cannot be represented just by a bit
1544     map. This includes negated single high-valued characters. The length in
1545     the table is zero; the actual length is stored in the compiled code. */
1546    
1547     if (c == OP_XCLASS) code += GET(code, 1);
1548    
1549 nigel 93 /* Handle capturing bracket */
1550 nigel 91
1551 nigel 93 else if (c == OP_CBRA)
1552 nigel 77 {
1553 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1554 nigel 77 if (n == number) return (uschar *)code;
1555 nigel 93 code += _pcre_OP_lengths[c];
1556 nigel 77 }
1557 nigel 91
1558 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1559     repeated character types, we have to test for \p and \P, which have an extra
1560 ph10 218 two bytes of parameters. */
1561 nigel 91
1562 nigel 77 else
1563     {
1564 ph10 218 switch(c)
1565     {
1566     case OP_TYPESTAR:
1567     case OP_TYPEMINSTAR:
1568     case OP_TYPEPLUS:
1569     case OP_TYPEMINPLUS:
1570     case OP_TYPEQUERY:
1571     case OP_TYPEMINQUERY:
1572     case OP_TYPEPOSSTAR:
1573     case OP_TYPEPOSPLUS:
1574     case OP_TYPEPOSQUERY:
1575     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1576 ph10 220 break;
1577 ph10 221
1578     case OP_TYPEUPTO:
1579     case OP_TYPEMINUPTO:
1580     case OP_TYPEEXACT:
1581     case OP_TYPEPOSUPTO:
1582     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1583     break;
1584 ph10 220 }
1585    
1586 ph10 218 /* Add in the fixed length from the table */
1587 ph10 220
1588 nigel 77 code += _pcre_OP_lengths[c];
1589 ph10 220
1590 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1591     a multi-byte character. The length in the table is a minimum, so we have to
1592     arrange to skip the extra bytes. */
1593 ph10 220
1594 ph10 107 #ifdef SUPPORT_UTF8
1595 nigel 77 if (utf8) switch(c)
1596     {
1597     case OP_CHAR:
1598     case OP_CHARNC:
1599     case OP_EXACT:
1600     case OP_UPTO:
1601     case OP_MINUPTO:
1602 nigel 93 case OP_POSUPTO:
1603 nigel 77 case OP_STAR:
1604     case OP_MINSTAR:
1605 nigel 93 case OP_POSSTAR:
1606 nigel 77 case OP_PLUS:
1607     case OP_MINPLUS:
1608 nigel 93 case OP_POSPLUS:
1609 nigel 77 case OP_QUERY:
1610     case OP_MINQUERY:
1611 nigel 93 case OP_POSQUERY:
1612     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1613 nigel 77 break;
1614     }
1615 ph10 369 #else
1616     (void)(utf8); /* Keep compiler happy by referencing function argument */
1617 ph10 111 #endif
1618 nigel 77 }
1619     }
1620     }
1621    
1622    
1623    
1624     /*************************************************
1625     * Scan compiled regex for recursion reference *
1626     *************************************************/
1627    
1628     /* This little function scans through a compiled pattern until it finds an
1629     instance of OP_RECURSE.
1630    
1631     Arguments:
1632     code points to start of expression
1633     utf8 TRUE in UTF-8 mode
1634    
1635     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1636     */
1637    
1638     static const uschar *
1639     find_recurse(const uschar *code, BOOL utf8)
1640     {
1641     for (;;)
1642     {
1643     register int c = *code;
1644     if (c == OP_END) return NULL;
1645 nigel 91 if (c == OP_RECURSE) return code;
1646 ph10 220
1647 nigel 91 /* XCLASS is used for classes that cannot be represented just by a bit
1648     map. This includes negated single high-valued characters. The length in
1649     the table is zero; the actual length is stored in the compiled code. */
1650    
1651     if (c == OP_XCLASS) code += GET(code, 1);
1652    
1653 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1654     repeated character types, we have to test for \p and \P, which have an extra
1655 ph10 218 two bytes of parameters. */
1656 nigel 91
1657 nigel 77 else
1658     {
1659 ph10 218 switch(c)
1660     {
1661     case OP_TYPESTAR:
1662     case OP_TYPEMINSTAR:
1663     case OP_TYPEPLUS:
1664     case OP_TYPEMINPLUS:
1665     case OP_TYPEQUERY:
1666     case OP_TYPEMINQUERY:
1667     case OP_TYPEPOSSTAR:
1668     case OP_TYPEPOSPLUS:
1669     case OP_TYPEPOSQUERY:
1670     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1671 ph10 220 break;
1672 ph10 221
1673     case OP_TYPEPOSUPTO:
1674     case OP_TYPEUPTO:
1675     case OP_TYPEMINUPTO:
1676     case OP_TYPEEXACT:
1677     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1678     break;
1679 ph10 220 }
1680    
1681 ph10 218 /* Add in the fixed length from the table */
1682    
1683 nigel 77 code += _pcre_OP_lengths[c];
1684 ph10 220
1685 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1686     by a multi-byte character. The length in the table is a minimum, so we have
1687     to arrange to skip the extra bytes. */
1688 ph10 220
1689 ph10 107 #ifdef SUPPORT_UTF8
1690 nigel 77 if (utf8) switch(c)
1691     {
1692     case OP_CHAR:
1693     case OP_CHARNC:
1694     case OP_EXACT:
1695     case OP_UPTO:
1696     case OP_MINUPTO:
1697 nigel 93 case OP_POSUPTO:
1698 nigel 77 case OP_STAR:
1699     case OP_MINSTAR:
1700 nigel 93 case OP_POSSTAR:
1701 nigel 77 case OP_PLUS:
1702     case OP_MINPLUS:
1703 nigel 93 case OP_POSPLUS:
1704 nigel 77 case OP_QUERY:
1705     case OP_MINQUERY:
1706 nigel 93 case OP_POSQUERY:
1707     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1708 nigel 77 break;
1709     }
1710 ph10 369 #else
1711     (void)(utf8); /* Keep compiler happy by referencing function argument */
1712 ph10 111 #endif
1713 nigel 77 }
1714     }
1715     }
1716    
1717    
1718    
1719     /*************************************************
1720     * Scan compiled branch for non-emptiness *
1721     *************************************************/
1722    
1723     /* This function scans through a branch of a compiled pattern to see whether it
1724 nigel 93 can match the empty string or not. It is called from could_be_empty()
1725     below and from compile_branch() when checking for an unlimited repeat of a
1726     group that can match nothing. Note that first_significant_code() skips over
1727 ph10 282 backward and negative forward assertions when its final argument is TRUE. If we
1728     hit an unclosed bracket, we return "empty" - this means we've struck an inner
1729     bracket whose current branch will already have been scanned.
1730 nigel 77
1731     Arguments:
1732     code points to start of search
1733     endcode points to where to stop
1734     utf8 TRUE if in UTF8 mode
1735    
1736     Returns: TRUE if what is matched could be empty
1737     */
1738    
1739     static BOOL
1740     could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1741     {
1742     register int c;
1743 nigel 93 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1744 nigel 77 code < endcode;
1745     code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1746     {
1747     const uschar *ccode;
1748    
1749     c = *code;
1750 ph10 286
1751     /* Skip over forward assertions; the other assertions are skipped by
1752 ph10 282 first_significant_code() with a TRUE final argument. */
1753 ph10 286
1754 ph10 282 if (c == OP_ASSERT)
1755 ph10 286 {
1756 ph10 282 do code += GET(code, 1); while (*code == OP_ALT);
1757     c = *code;
1758     continue;
1759 ph10 286 }
1760 ph10 172
1761 ph10 170 /* Groups with zero repeats can of course be empty; skip them. */
1762 nigel 77
1763 ph10 335 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1764 ph10 170 {
1765 ph10 172 code += _pcre_OP_lengths[c];
1766 ph10 170 do code += GET(code, 1); while (*code == OP_ALT);
1767     c = *code;
1768     continue;
1769     }
1770    
1771     /* For other groups, scan the branches. */
1772 ph10 172
1773 ph10 206 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1774 nigel 77 {
1775     BOOL empty_branch;
1776     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1777 ph10 406
1778     /* If a conditional group has only one branch, there is a second, implied,
1779 ph10 395 empty branch, so just skip over the conditional, because it could be empty.
1780     Otherwise, scan the individual branches of the group. */
1781 ph10 406
1782 ph10 395 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1783 nigel 77 code += GET(code, 1);
1784 ph10 395 else
1785 ph10 406 {
1786 ph10 395 empty_branch = FALSE;
1787     do
1788     {
1789     if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1790     empty_branch = TRUE;
1791     code += GET(code, 1);
1792     }
1793     while (*code == OP_ALT);
1794     if (!empty_branch) return FALSE; /* All branches are non-empty */
1795 nigel 77 }
1796 ph10 406
1797 ph10 172 c = *code;
1798 nigel 93 continue;
1799 nigel 77 }
1800    
1801 nigel 93 /* Handle the other opcodes */
1802    
1803     switch (c)
1804 nigel 77 {
1805 ph10 216 /* Check for quantifiers after a class. XCLASS is used for classes that
1806     cannot be represented just by a bit map. This includes negated single
1807     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1808 ph10 220 actual length is stored in the compiled code, so we must update "code"
1809 ph10 216 here. */
1810 nigel 77
1811     #ifdef SUPPORT_UTF8
1812     case OP_XCLASS:
1813 ph10 216 ccode = code += GET(code, 1);
1814 nigel 77 goto CHECK_CLASS_REPEAT;
1815     #endif
1816    
1817     case OP_CLASS:
1818     case OP_NCLASS:
1819     ccode = code + 33;
1820    
1821     #ifdef SUPPORT_UTF8
1822     CHECK_CLASS_REPEAT:
1823     #endif
1824    
1825     switch (*ccode)
1826     {
1827     case OP_CRSTAR: /* These could be empty; continue */
1828     case OP_CRMINSTAR:
1829     case OP_CRQUERY:
1830     case OP_CRMINQUERY:
1831     break;
1832    
1833     default: /* Non-repeat => class must match */
1834     case OP_CRPLUS: /* These repeats aren't empty */
1835     case OP_CRMINPLUS:
1836     return FALSE;
1837    
1838     case OP_CRRANGE:
1839     case OP_CRMINRANGE:
1840     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1841     break;
1842     }
1843     break;
1844    
1845     /* Opcodes that must match a character */
1846    
1847     case OP_PROP:
1848     case OP_NOTPROP:
1849     case OP_EXTUNI:
1850     case OP_NOT_DIGIT:
1851     case OP_DIGIT:
1852     case OP_NOT_WHITESPACE:
1853     case OP_WHITESPACE:
1854     case OP_NOT_WORDCHAR:
1855     case OP_WORDCHAR:
1856     case OP_ANY:
1857 ph10 345 case OP_ALLANY:
1858 nigel 77 case OP_ANYBYTE:
1859     case OP_CHAR:
1860     case OP_CHARNC:
1861     case OP_NOT:
1862     case OP_PLUS:
1863     case OP_MINPLUS:
1864 nigel 93 case OP_POSPLUS:
1865 nigel 77 case OP_EXACT:
1866     case OP_NOTPLUS:
1867     case OP_NOTMINPLUS:
1868 nigel 93 case OP_NOTPOSPLUS:
1869 nigel 77 case OP_NOTEXACT:
1870     case OP_TYPEPLUS:
1871     case OP_TYPEMINPLUS:
1872 nigel 93 case OP_TYPEPOSPLUS:
1873 nigel 77 case OP_TYPEEXACT:
1874     return FALSE;
1875 ph10 227
1876     /* These are going to continue, as they may be empty, but we have to
1877     fudge the length for the \p and \P cases. */
1878    
1879 ph10 224 case OP_TYPESTAR:
1880     case OP_TYPEMINSTAR:
1881     case OP_TYPEPOSSTAR:
1882     case OP_TYPEQUERY:
1883     case OP_TYPEMINQUERY:
1884     case OP_TYPEPOSQUERY:
1885     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1886 ph10 227 break;
1887    
1888 ph10 224 /* Same for these */
1889 ph10 227
1890 ph10 224 case OP_TYPEUPTO:
1891     case OP_TYPEMINUPTO:
1892     case OP_TYPEPOSUPTO:
1893     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1894     break;
1895 nigel 77
1896     /* End of branch */
1897    
1898     case OP_KET:
1899     case OP_KETRMAX:
1900     case OP_KETRMIN:
1901     case OP_ALT:
1902     return TRUE;
1903    
1904 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1905     MINUPTO, and POSUPTO may be followed by a multibyte character */
1906 nigel 77
1907     #ifdef SUPPORT_UTF8
1908     case OP_STAR:
1909     case OP_MINSTAR:
1910 nigel 93 case OP_POSSTAR:
1911 nigel 77 case OP_QUERY:
1912     case OP_MINQUERY:
1913 nigel 93 case OP_POSQUERY:
1914 ph10 426 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1915     break;
1916    
1917 nigel 77 case OP_UPTO:
1918     case OP_MINUPTO:
1919 nigel 93 case OP_POSUPTO:
1920 ph10 426 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1921 nigel 77 break;
1922     #endif
1923     }
1924     }
1925    
1926     return TRUE;
1927     }
1928    
1929    
1930    
1931     /*************************************************
1932     * Scan compiled regex for non-emptiness *
1933     *************************************************/
1934    
1935     /* This function is called to check for left recursive calls. We want to check
1936     the current branch of the current pattern to see if it could match the empty
1937     string. If it could, we must look outwards for branches at other levels,
1938     stopping when we pass beyond the bracket which is the subject of the recursion.
1939    
1940     Arguments:
1941     code points to start of the recursion
1942     endcode points to where to stop (current RECURSE item)
1943     bcptr points to the chain of current (unclosed) branch starts
1944     utf8 TRUE if in UTF-8 mode
1945    
1946     Returns: TRUE if what is matched could be empty
1947     */
1948    
1949     static BOOL
1950     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1951     BOOL utf8)
1952     {
1953     while (bcptr != NULL && bcptr->current >= code)
1954     {
1955     if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1956     bcptr = bcptr->outer;
1957     }
1958     return TRUE;
1959     }
1960    
1961    
1962    
1963     /*************************************************
1964     * Check for POSIX class syntax *
1965     *************************************************/
1966    
1967     /* This function is called when the sequence "[:" or "[." or "[=" is
1968 ph10 295 encountered in a character class. It checks whether this is followed by a
1969 ph10 298 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1970 ph10 295 reach an unescaped ']' without the special preceding character, return FALSE.
1971 nigel 77
1972 ph10 298 Originally, this function only recognized a sequence of letters between the
1973     terminators, but it seems that Perl recognizes any sequence of characters,
1974     though of course unknown POSIX names are subsequently rejected. Perl gives an
1975     "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1976     didn't consider this to be a POSIX class. Likewise for [:1234:].
1977 ph10 295
1978 ph10 298 The problem in trying to be exactly like Perl is in the handling of escapes. We
1979     have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1980     class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1981     below handles the special case of \], but does not try to do any other escape
1982     processing. This makes it different from Perl for cases such as [:l\ower:]
1983 ph10 295 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1984 ph10 298 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1985 ph10 295 I think.
1986    
1987     Arguments:
1988 nigel 77 ptr pointer to the initial [
1989     endptr where to return the end pointer
1990    
1991     Returns: TRUE or FALSE
1992     */
1993    
1994     static BOOL
1995 ph10 295 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1996 nigel 77 {
1997     int terminator; /* Don't combine these lines; the Solaris cc */
1998     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1999 ph10 295 for (++ptr; *ptr != 0; ptr++)
2000 nigel 77 {
2001 ph10 391 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
2002 ph10 298 {
2003 ph10 391 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2004     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2005 ph10 295 {
2006     *endptr = ptr;
2007     return TRUE;
2008 ph10 298 }
2009     }
2010     }
2011 nigel 77 return FALSE;
2012     }
2013    
2014    
2015    
2016    
2017     /*************************************************
2018     * Check POSIX class name *
2019     *************************************************/
2020    
2021     /* This function is called to check the name given in a POSIX-style class entry
2022     such as [:alnum:].
2023    
2024     Arguments:
2025     ptr points to the first letter
2026     len the length of the name
2027    
2028     Returns: a value representing the name, or -1 if unknown
2029     */
2030    
2031     static int
2032     check_posix_name(const uschar *ptr, int len)
2033     {
2034 ph10 240 const char *pn = posix_names;
2035 nigel 77 register int yield = 0;
2036     while (posix_name_lengths[yield] != 0)
2037     {
2038     if (len == posix_name_lengths[yield] &&
2039 ph10 240 strncmp((const char *)ptr, pn, len) == 0) return yield;
2040 ph10 243 pn += posix_name_lengths[yield] + 1;
2041 nigel 77 yield++;
2042     }
2043     return -1;
2044     }
2045    
2046    
2047     /*************************************************
2048     * Adjust OP_RECURSE items in repeated group *
2049     *************************************************/
2050    
2051     /* OP_RECURSE items contain an offset from the start of the regex to the group
2052     that is referenced. This means that groups can be replicated for fixed
2053     repetition simply by copying (because the recursion is allowed to refer to
2054     earlier groups that are outside the current group). However, when a group is
2055 ph10 335 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2056     inserted before it, after it has been compiled. This means that any OP_RECURSE
2057     items within it that refer to the group itself or any contained groups have to
2058     have their offsets adjusted. That one of the jobs of this function. Before it
2059     is called, the partially compiled regex must be temporarily terminated with
2060     OP_END.
2061 nigel 77
2062 nigel 93 This function has been extended with the possibility of forward references for
2063     recursions and subroutine calls. It must also check the list of such references
2064     for the group we are dealing with. If it finds that one of the recursions in
2065     the current group is on this list, it adjusts the offset in the list, not the
2066     value in the reference (which is a group number).
2067    
2068 nigel 77 Arguments:
2069     group points to the start of the group
2070     adjust the amount by which the group is to be moved
2071     utf8 TRUE in UTF-8 mode
2072     cd contains pointers to tables etc.
2073 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
2074 nigel 77
2075     Returns: nothing
2076     */
2077    
2078     static void
2079 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2080     uschar *save_hwm)
2081 nigel 77 {
2082     uschar *ptr = group;
2083 ph10 224
2084 nigel 77 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2085     {
2086 nigel 93 int offset;
2087     uschar *hc;
2088    
2089     /* See if this recursion is on the forward reference list. If so, adjust the
2090     reference. */
2091 ph10 345
2092 nigel 93 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2093     {
2094     offset = GET(hc, 0);
2095     if (cd->start_code + offset == ptr + 1)
2096     {
2097     PUT(hc, 0, offset + adjust);
2098     break;
2099     }
2100     }
2101    
2102     /* Otherwise, adjust the recursion offset if it's after the start of this
2103     group. */
2104    
2105     if (hc >= cd->hwm)
2106     {
2107     offset = GET(ptr, 1);
2108     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2109     }
2110    
2111 nigel 77 ptr += 1 + LINK_SIZE;
2112     }
2113     }
2114    
2115    
2116    
2117     /*************************************************
2118     * Insert an automatic callout point *
2119     *************************************************/
2120    
2121     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2122     callout points before each pattern item.
2123    
2124     Arguments:
2125     code current code pointer
2126     ptr current pattern pointer
2127     cd pointers to tables etc
2128    
2129     Returns: new code pointer
2130     */
2131    
2132     static uschar *
2133     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2134     {
2135     *code++ = OP_CALLOUT;
2136     *code++ = 255;
2137     PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2138     PUT(code, LINK_SIZE, 0); /* Default length */
2139     return code + 2*LINK_SIZE;
2140     }
2141    
2142    
2143    
2144     /*************************************************
2145     * Complete a callout item *
2146     *************************************************/
2147    
2148     /* A callout item contains the length of the next item in the pattern, which
2149     we can't fill in till after we have reached the relevant point. This is used
2150     for both automatic and manual callouts.
2151    
2152     Arguments:
2153     previous_callout points to previous callout item
2154     ptr current pattern pointer
2155     cd pointers to tables etc
2156    
2157     Returns: nothing
2158     */
2159    
2160     static void
2161     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2162     {
2163     int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2164     PUT(previous_callout, 2 + LINK_SIZE, length);
2165     }
2166    
2167    
2168    
2169     #ifdef SUPPORT_UCP
2170     /*************************************************
2171     * Get othercase range *
2172     *************************************************/
2173    
2174     /* This function is passed the start and end of a class range, in UTF-8 mode
2175     with UCP support. It searches up the characters, looking for internal ranges of
2176     characters in the "other" case. Each call returns the next one, updating the
2177     start address.
2178    
2179     Arguments:
2180     cptr points to starting character value; updated
2181     d end value
2182     ocptr where to put start of othercase range
2183     odptr where to put end of othercase range
2184    
2185     Yield: TRUE when range returned; FALSE when no more
2186     */
2187    
2188     static BOOL
2189 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2190     unsigned int *odptr)
2191 nigel 77 {
2192 nigel 93 unsigned int c, othercase, next;
2193 nigel 77
2194     for (c = *cptr; c <= d; c++)
2195 ph10 349 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2196 nigel 77
2197     if (c > d) return FALSE;
2198    
2199     *ocptr = othercase;
2200     next = othercase + 1;
2201    
2202     for (++c; c <= d; c++)
2203     {
2204 ph10 349 if (UCD_OTHERCASE(c) != next) break;
2205 nigel 77 next++;
2206     }
2207    
2208     *odptr = next - 1;
2209     *cptr = c;
2210    
2211     return TRUE;
2212     }
2213     #endif /* SUPPORT_UCP */
2214    
2215    
2216 nigel 93
2217 nigel 77 /*************************************************
2218 nigel 93 * Check if auto-possessifying is possible *
2219     *************************************************/
2220    
2221     /* This function is called for unlimited repeats of certain items, to see
2222     whether the next thing could possibly match the repeated item. If not, it makes
2223     sense to automatically possessify the repeated item.
2224    
2225     Arguments:
2226     op_code the repeated op code
2227     this data for this item, depends on the opcode
2228     utf8 TRUE in UTF-8 mode
2229     utf8_char used for utf8 character bytes, NULL if not relevant
2230     ptr next character in pattern
2231     options options bits
2232     cd contains pointers to tables etc.
2233    
2234     Returns: TRUE if possessifying is wanted
2235     */
2236    
2237     static BOOL
2238     check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2239     const uschar *ptr, int options, compile_data *cd)
2240     {
2241     int next;
2242    
2243     /* Skip whitespace and comments in extended mode */
2244    
2245     if ((options & PCRE_EXTENDED) != 0)
2246     {
2247     for (;;)
2248     {
2249     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2250 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2251 nigel 93 {
2252     while (*(++ptr) != 0)
2253     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2254     }
2255     else break;
2256     }
2257     }
2258    
2259     /* If the next item is one that we can handle, get its value. A non-negative
2260     value is a character, a negative value is an escape value. */
2261    
2262 ph10 391 if (*ptr == CHAR_BACKSLASH)
2263 nigel 93 {
2264     int temperrorcode = 0;
2265     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2266     if (temperrorcode != 0) return FALSE;
2267     ptr++; /* Point after the escape sequence */
2268     }
2269    
2270     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2271     {
2272     #ifdef SUPPORT_UTF8
2273     if (utf8) { GETCHARINC(next, ptr); } else
2274     #endif
2275     next = *ptr++;
2276     }
2277    
2278     else return FALSE;
2279    
2280     /* Skip whitespace and comments in extended mode */
2281    
2282     if ((options & PCRE_EXTENDED) != 0)
2283     {
2284     for (;;)
2285     {
2286     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2287 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2288 nigel 93 {
2289     while (*(++ptr) != 0)
2290     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2291     }
2292     else break;
2293     }
2294     }
2295    
2296     /* If the next thing is itself optional, we have to give up. */
2297    
2298 ph10 392 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2299 ph10 391 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2300     return FALSE;
2301 nigel 93
2302     /* Now compare the next item with the previous opcode. If the previous is a
2303     positive single character match, "item" either contains the character or, if
2304     "item" is greater than 127 in utf8 mode, the character's bytes are in
2305     utf8_char. */
2306    
2307    
2308     /* Handle cases when the next item is a character. */
2309    
2310     if (next >= 0) switch(op_code)
2311     {
2312     case OP_CHAR:
2313     #ifdef SUPPORT_UTF8
2314     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2315 ph10 369 #else
2316     (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2317 nigel 93 #endif
2318     return item != next;
2319    
2320     /* For CHARNC (caseless character) we must check the other case. If we have
2321     Unicode property support, we can use it to test the other case of
2322     high-valued characters. */
2323    
2324     case OP_CHARNC:
2325     #ifdef SUPPORT_UTF8
2326     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2327     #endif
2328     if (item == next) return FALSE;
2329     #ifdef SUPPORT_UTF8
2330     if (utf8)
2331     {
2332     unsigned int othercase;
2333     if (next < 128) othercase = cd->fcc[next]; else
2334     #ifdef SUPPORT_UCP
2335 ph10 349 othercase = UCD_OTHERCASE((unsigned int)next);
2336 nigel 93 #else
2337     othercase = NOTACHAR;
2338     #endif
2339     return (unsigned int)item != othercase;
2340     }
2341     else
2342     #endif /* SUPPORT_UTF8 */
2343     return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2344    
2345     /* For OP_NOT, "item" must be a single-byte character. */
2346    
2347     case OP_NOT:
2348     if (item == next) return TRUE;
2349     if ((options & PCRE_CASELESS) == 0) return FALSE;
2350     #ifdef SUPPORT_UTF8
2351     if (utf8)
2352     {
2353     unsigned int othercase;
2354     if (next < 128) othercase = cd->fcc[next]; else
2355     #ifdef SUPPORT_UCP
2356 ph10 349 othercase = UCD_OTHERCASE(next);
2357 nigel 93 #else
2358     othercase = NOTACHAR;
2359     #endif
2360     return (unsigned int)item == othercase;
2361     }
2362     else
2363     #endif /* SUPPORT_UTF8 */
2364     return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2365    
2366     case OP_DIGIT:
2367     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2368    
2369     case OP_NOT_DIGIT:
2370     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2371    
2372     case OP_WHITESPACE:
2373     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2374    
2375     case OP_NOT_WHITESPACE:
2376     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2377    
2378     case OP_WORDCHAR:
2379     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2380    
2381     case OP_NOT_WORDCHAR:
2382     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2383    
2384 ph10 180 case OP_HSPACE:
2385     case OP_NOT_HSPACE:
2386     switch(next)
2387     {
2388     case 0x09:
2389     case 0x20:
2390     case 0xa0:
2391     case 0x1680:
2392     case 0x180e:
2393     case 0x2000:
2394     case 0x2001:
2395     case 0x2002:
2396     case 0x2003:
2397     case 0x2004:
2398     case 0x2005:
2399     case 0x2006:
2400     case 0x2007:
2401     case 0x2008:
2402     case 0x2009:
2403     case 0x200A:
2404     case 0x202f:
2405     case 0x205f:
2406     case 0x3000:
2407     return op_code != OP_HSPACE;
2408     default:
2409     return op_code == OP_HSPACE;
2410     }
2411    
2412     case OP_VSPACE:
2413     case OP_NOT_VSPACE:
2414     switch(next)
2415     {
2416     case 0x0a:
2417     case 0x0b:
2418     case 0x0c:
2419     case 0x0d:
2420     case 0x85:
2421     case 0x2028:
2422     case 0x2029:
2423     return op_code != OP_VSPACE;
2424     default:
2425     return op_code == OP_VSPACE;
2426     }
2427    
2428 nigel 93 default:
2429     return FALSE;
2430     }
2431    
2432    
2433     /* Handle the case when the next item is \d, \s, etc. */
2434    
2435     switch(op_code)
2436     {
2437     case OP_CHAR:
2438     case OP_CHARNC:
2439     #ifdef SUPPORT_UTF8
2440     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2441     #endif
2442     switch(-next)
2443     {
2444     case ESC_d:
2445     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2446    
2447     case ESC_D:
2448     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2449    
2450     case ESC_s:
2451     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2452    
2453     case ESC_S:
2454     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2455    
2456     case ESC_w:
2457     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2458    
2459     case ESC_W:
2460     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2461 ph10 182
2462 ph10 180 case ESC_h:
2463     case ESC_H:
2464     switch(item)
2465     {
2466     case 0x09:
2467     case 0x20:
2468     case 0xa0:
2469     case 0x1680:
2470     case 0x180e:
2471     case 0x2000:
2472     case 0x2001:
2473     case 0x2002:
2474     case 0x2003:
2475     case 0x2004:
2476     case 0x2005:
2477     case 0x2006:
2478     case 0x2007:
2479     case 0x2008:
2480     case 0x2009:
2481     case 0x200A:
2482     case 0x202f:
2483     case 0x205f:
2484     case 0x3000:
2485     return -next != ESC_h;
2486     default:
2487     return -next == ESC_h;
2488 ph10 182 }
2489    
2490 ph10 180 case ESC_v:
2491     case ESC_V:
2492     switch(item)
2493     {
2494     case 0x0a:
2495     case 0x0b:
2496     case 0x0c:
2497     case 0x0d:
2498     case 0x85:
2499     case 0x2028:
2500     case 0x2029:
2501     return -next != ESC_v;
2502     default:
2503     return -next == ESC_v;
2504 ph10 182 }
2505 nigel 93
2506     default:
2507     return FALSE;
2508     }
2509    
2510     case OP_DIGIT:
2511 ph10 180 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2512     next == -ESC_h || next == -ESC_v;
2513 nigel 93
2514     case OP_NOT_DIGIT:
2515     return next == -ESC_d;
2516    
2517     case OP_WHITESPACE:
2518     return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2519    
2520     case OP_NOT_WHITESPACE:
2521 ph10 180 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2522 nigel 93
2523 ph10 180 case OP_HSPACE:
2524     return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2525    
2526     case OP_NOT_HSPACE:
2527     return next == -ESC_h;
2528 ph10 182
2529 ph10 180 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2530 ph10 182 case OP_VSPACE:
2531 ph10 180 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2532    
2533     case OP_NOT_VSPACE:
2534 ph10 182 return next == -ESC_v;
2535 ph10 180
2536 nigel 93 case OP_WORDCHAR:
2537 ph10 180 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2538 nigel 93
2539     case OP_NOT_WORDCHAR:
2540     return next == -ESC_w || next == -ESC_d;
2541 ph10 182
2542 nigel 93 default:
2543     return FALSE;
2544     }
2545    
2546     /* Control does not reach here */
2547     }
2548    
2549    
2550    
2551     /*************************************************
2552 nigel 77 * Compile one branch *
2553     *************************************************/
2554    
2555 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
2556 nigel 77 changed during the branch, the pointer is used to change the external options
2557 nigel 93 bits. This function is used during the pre-compile phase when we are trying
2558     to find out the amount of memory needed, as well as during the real compile
2559     phase. The value of lengthptr distinguishes the two phases.
2560 nigel 77
2561     Arguments:
2562     optionsptr pointer to the option bits
2563     codeptr points to the pointer to the current code point
2564     ptrptr points to the current pattern pointer
2565     errorcodeptr points to error code variable
2566     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2567     reqbyteptr set to the last literal character required, else < 0
2568     bcptr points to current branch chain
2569     cd contains pointers to tables etc.
2570 nigel 93 lengthptr NULL during the real compile phase
2571     points to length accumulator during pre-compile phase
2572 nigel 77
2573     Returns: TRUE on success
2574     FALSE, with *errorcodeptr set non-zero on error
2575     */
2576    
2577     static BOOL
2578 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2579     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2580     compile_data *cd, int *lengthptr)
2581 nigel 77 {
2582     int repeat_type, op_type;
2583     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2584     int bravalue = 0;
2585     int greedy_default, greedy_non_default;
2586     int firstbyte, reqbyte;
2587     int zeroreqbyte, zerofirstbyte;
2588     int req_caseopt, reqvary, tempreqvary;
2589     int options = *optionsptr;
2590     int after_manual_callout = 0;
2591 nigel 93 int length_prevgroup = 0;
2592 nigel 77 register int c;
2593     register uschar *code = *codeptr;
2594 nigel 93 uschar *last_code = code;
2595     uschar *orig_code = code;
2596 nigel 77 uschar *tempcode;
2597     BOOL inescq = FALSE;
2598     BOOL groupsetfirstbyte = FALSE;
2599     const uschar *ptr = *ptrptr;
2600     const uschar *tempptr;
2601     uschar *previous = NULL;
2602     uschar *previous_callout = NULL;
2603 nigel 93 uschar *save_hwm = NULL;
2604 nigel 77 uschar classbits[32];
2605    
2606     #ifdef SUPPORT_UTF8
2607     BOOL class_utf8;
2608     BOOL utf8 = (options & PCRE_UTF8) != 0;
2609     uschar *class_utf8data;
2610 ph10 300 uschar *class_utf8data_base;
2611 nigel 77 uschar utf8_char[6];
2612     #else
2613     BOOL utf8 = FALSE;
2614 nigel 93 uschar *utf8_char = NULL;
2615 nigel 77 #endif
2616    
2617 nigel 93 #ifdef DEBUG
2618     if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2619     #endif
2620    
2621 nigel 77 /* Set up the default and non-default settings for greediness */
2622    
2623     greedy_default = ((options & PCRE_UNGREEDY) != 0);
2624     greedy_non_default = greedy_default ^ 1;
2625    
2626     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2627     matching encountered yet". It gets changed to REQ_NONE if we hit something that
2628     matches a non-fixed char first char; reqbyte just remains unset if we never
2629     find one.
2630    
2631     When we hit a repeat whose minimum is zero, we may have to adjust these values
2632     to take the zero repeat into account. This is implemented by setting them to
2633     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2634     item types that can be repeated set these backoff variables appropriately. */
2635    
2636     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2637    
2638     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2639     according to the current setting of the caseless flag. REQ_CASELESS is a bit
2640     value > 255. It is added into the firstbyte or reqbyte variables to record the
2641     case status of the value. This is used only for ASCII characters. */
2642    
2643     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2644    
2645     /* Switch on next character until the end of the branch */
2646    
2647     for (;; ptr++)
2648     {
2649     BOOL negate_class;
2650 ph10 286 BOOL should_flip_negation;
2651 nigel 77 BOOL possessive_quantifier;
2652     BOOL is_quantifier;
2653 nigel 93 BOOL is_recurse;
2654 ph10 180 BOOL reset_bracount;
2655 nigel 77 int class_charcount;
2656     int class_lastchar;
2657     int newoptions;
2658     int recno;
2659 ph10 172 int refsign;
2660 nigel 77 int skipbytes;
2661     int subreqbyte;
2662     int subfirstbyte;
2663 nigel 93 int terminator;
2664 nigel 77 int mclength;
2665     uschar mcbuffer[8];
2666    
2667 nigel 93 /* Get next byte in the pattern */
2668 nigel 77
2669     c = *ptr;
2670 ph10 345
2671 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
2672     previous cycle of this loop. */
2673    
2674     if (lengthptr != NULL)
2675     {
2676     #ifdef DEBUG
2677     if (code > cd->hwm) cd->hwm = code; /* High water info */
2678     #endif
2679     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2680     {
2681     *errorcodeptr = ERR52;
2682     goto FAILED;
2683     }
2684    
2685     /* There is at least one situation where code goes backwards: this is the
2686     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2687     the class is simply eliminated. However, it is created first, so we have to
2688     allow memory for it. Therefore, don't ever reduce the length at this point.
2689     */
2690    
2691     if (code < last_code) code = last_code;
2692 ph10 202
2693     /* Paranoid check for integer overflow */
2694    
2695     if (OFLOW_MAX - *lengthptr < code - last_code)
2696     {
2697     *errorcodeptr = ERR20;
2698     goto FAILED;
2699     }
2700    
2701 nigel 93 *lengthptr += code - last_code;
2702     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2703    
2704     /* If "previous" is set and it is not at the start of the work space, move
2705     it back to there, in order to avoid filling up the work space. Otherwise,
2706     if "previous" is NULL, reset the current code pointer to the start. */
2707    
2708     if (previous != NULL)
2709     {
2710     if (previous > orig_code)
2711     {
2712     memmove(orig_code, previous, code - previous);
2713     code -= previous - orig_code;
2714     previous = orig_code;
2715     }
2716     }
2717     else code = orig_code;
2718    
2719     /* Remember where this code item starts so we can pick up the length
2720     next time round. */
2721    
2722     last_code = code;
2723     }
2724    
2725     /* In the real compile phase, just check the workspace used by the forward
2726     reference list. */
2727    
2728     else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2729     {
2730     *errorcodeptr = ERR52;
2731     goto FAILED;
2732     }
2733    
2734 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
2735    
2736     if (inescq && c != 0)
2737     {
2738 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2739 nigel 77 {
2740     inescq = FALSE;
2741     ptr++;
2742     continue;
2743     }
2744     else
2745     {
2746     if (previous_callout != NULL)
2747     {
2748 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2749     complete_callout(previous_callout, ptr, cd);
2750 nigel 77 previous_callout = NULL;
2751     }
2752     if ((options & PCRE_AUTO_CALLOUT) != 0)
2753     {
2754     previous_callout = code;
2755     code = auto_callout(code, ptr, cd);
2756     }
2757     goto NORMAL_CHAR;
2758     }
2759     }
2760    
2761     /* Fill in length of a previous callout, except when the next thing is
2762     a quantifier. */
2763    
2764 ph10 392 is_quantifier =
2765 ph10 391 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2766     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2767 nigel 77
2768     if (!is_quantifier && previous_callout != NULL &&
2769     after_manual_callout-- <= 0)
2770     {
2771 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2772     complete_callout(previous_callout, ptr, cd);
2773 nigel 77 previous_callout = NULL;
2774     }
2775    
2776     /* In extended mode, skip white space and comments */
2777    
2778     if ((options & PCRE_EXTENDED) != 0)
2779     {
2780     if ((cd->ctypes[c] & ctype_space) != 0) continue;
2781 ph10 391 if (c == CHAR_NUMBER_SIGN)
2782 nigel 77 {
2783 nigel 93 while (*(++ptr) != 0)
2784 nigel 91 {
2785 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2786 nigel 91 }
2787 nigel 93 if (*ptr != 0) continue;
2788    
2789 nigel 91 /* Else fall through to handle end of string */
2790     c = 0;
2791 nigel 77 }
2792     }
2793    
2794     /* No auto callout for quantifiers. */
2795    
2796     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2797     {
2798     previous_callout = code;
2799     code = auto_callout(code, ptr, cd);
2800     }
2801    
2802     switch(c)
2803     {
2804 nigel 93 /* ===================================================================*/
2805     case 0: /* The branch terminates at string end */
2806 ph10 391 case CHAR_VERTICAL_LINE: /* or | or ) */
2807     case CHAR_RIGHT_PARENTHESIS:
2808 nigel 77 *firstbyteptr = firstbyte;
2809     *reqbyteptr = reqbyte;
2810     *codeptr = code;
2811     *ptrptr = ptr;
2812 nigel 93 if (lengthptr != NULL)
2813     {
2814 ph10 202 if (OFLOW_MAX - *lengthptr < code - last_code)
2815     {
2816     *errorcodeptr = ERR20;
2817     goto FAILED;
2818     }
2819 nigel 93 *lengthptr += code - last_code; /* To include callout length */
2820     DPRINTF((">> end branch\n"));
2821     }
2822 nigel 77 return TRUE;
2823    
2824 nigel 93
2825     /* ===================================================================*/
2826 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
2827     the setting of any following char as a first character. */
2828    
2829 ph10 391 case CHAR_CIRCUMFLEX_ACCENT:
2830 nigel 77 if ((options & PCRE_MULTILINE) != 0)
2831     {
2832     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2833     }
2834     previous = NULL;
2835     *code++ = OP_CIRC;
2836     break;
2837    
2838 ph10 391 case CHAR_DOLLAR_SIGN:
2839 nigel 77 previous = NULL;
2840     *code++ = OP_DOLL;
2841     break;
2842    
2843     /* There can never be a first char if '.' is first, whatever happens about
2844     repeats. The value of reqbyte doesn't change either. */
2845    
2846 ph10 391 case CHAR_DOT:
2847 nigel 77 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2848     zerofirstbyte = firstbyte;
2849     zeroreqbyte = reqbyte;
2850     previous = code;
2851 ph10 342 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2852 nigel 77 break;
2853    
2854 nigel 93
2855     /* ===================================================================*/
2856 nigel 87 /* Character classes. If the included characters are all < 256, we build a
2857     32-byte bitmap of the permitted characters, except in the special case
2858     where there is only one such character. For negated classes, we build the
2859     map as usual, then invert it at the end. However, we use a different opcode
2860     so that data characters > 255 can be handled correctly.
2861 nigel 77
2862     If the class contains characters outside the 0-255 range, a different
2863     opcode is compiled. It may optionally have a bit map for characters < 256,
2864     but those above are are explicitly listed afterwards. A flag byte tells
2865     whether the bitmap is present, and whether this is a negated class or not.
2866 ph10 345
2867 ph10 336 In JavaScript compatibility mode, an isolated ']' causes an error. In
2868     default (Perl) mode, it is treated as a data character. */
2869 ph10 345
2870 ph10 391 case CHAR_RIGHT_SQUARE_BRACKET:
2871 ph10 336 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2872     {
2873     *errorcodeptr = ERR64;
2874 ph10 345 goto FAILED;
2875 ph10 336 }
2876 ph10 345 goto NORMAL_CHAR;
2877 nigel 77
2878 ph10 391 case CHAR_LEFT_SQUARE_BRACKET:
2879 nigel 77 previous = code;
2880    
2881     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2882     they are encountered at the top level, so we'll do that too. */
2883    
2884 ph10 392 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2885 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) &&
2886 ph10 295 check_posix_syntax(ptr, &tempptr))
2887 nigel 77 {
2888 ph10 391 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2889 nigel 77 goto FAILED;
2890     }
2891    
2892 ph10 205 /* If the first character is '^', set the negation flag and skip it. Also,
2893 ph10 208 if the first few characters (either before or after ^) are \Q\E or \E we
2894 ph10 205 skip them too. This makes for compatibility with Perl. */
2895 ph10 208
2896 ph10 205 negate_class = FALSE;
2897     for (;;)
2898 nigel 77 {
2899     c = *(++ptr);
2900 ph10 391 if (c == CHAR_BACKSLASH)
2901 ph10 205 {
2902 ph10 392 if (ptr[1] == CHAR_E)
2903 ph10 391 ptr++;
2904 ph10 392 else if (strncmp((const char *)ptr+1,
2905     STR_Q STR_BACKSLASH STR_E, 3) == 0)
2906 ph10 391 ptr += 3;
2907 ph10 392 else
2908 ph10 391 break;
2909 ph10 205 }
2910 ph10 391 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2911 ph10 205 negate_class = TRUE;
2912     else break;
2913 ph10 208 }
2914 ph10 345
2915     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2916     an initial ']' is taken as a data character -- the code below handles
2917 ph10 341 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2918     [^] must match any character, so generate OP_ALLANY. */
2919 ph10 345
2920 ph10 392 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2921 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2922 ph10 341 {
2923     *code++ = negate_class? OP_ALLANY : OP_FAIL;
2924     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2925     zerofirstbyte = firstbyte;
2926     break;
2927 ph10 345 }
2928 nigel 77
2929 ph10 286 /* If a class contains a negative special such as \S, we need to flip the
2930     negation flag at the end, so that support for characters > 255 works
2931 ph10 264 correctly (they are all included in the class). */
2932    
2933     should_flip_negation = FALSE;
2934    
2935 nigel 77 /* Keep a count of chars with values < 256 so that we can optimize the case
2936 nigel 93 of just a single character (as long as it's < 256). However, For higher
2937     valued UTF-8 characters, we don't yet do any optimization. */
2938 nigel 77
2939     class_charcount = 0;
2940     class_lastchar = -1;
2941    
2942 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
2943     temporary bit of memory, in case the class contains only 1 character (less
2944     than 256), because in that case the compiled code doesn't use the bit map.
2945     */
2946    
2947     memset(classbits, 0, 32 * sizeof(uschar));
2948    
2949 nigel 77 #ifdef SUPPORT_UTF8
2950     class_utf8 = FALSE; /* No chars >= 256 */
2951 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2952 ph10 309 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2953 nigel 77 #endif
2954    
2955     /* Process characters until ] is reached. By writing this as a "do" it
2956 nigel 93 means that an initial ] is taken as a data character. At the start of the
2957     loop, c contains the first byte of the character. */
2958 nigel 77
2959 nigel 93 if (c != 0) do
2960 nigel 77 {
2961 nigel 93 const uschar *oldptr;
2962    
2963 nigel 77 #ifdef SUPPORT_UTF8
2964     if (utf8 && c > 127)
2965     { /* Braces are required because the */
2966     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2967     }
2968 ph10 309
2969 ph10 300 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2970 ph10 309 data and reset the pointer. This is so that very large classes that
2971 ph10 300 contain a zillion UTF-8 characters no longer overwrite the work space
2972 ph10 309 (which is on the stack). */
2973    
2974 ph10 300 if (lengthptr != NULL)
2975     {
2976     *lengthptr += class_utf8data - class_utf8data_base;
2977 ph10 309 class_utf8data = class_utf8data_base;
2978     }
2979    
2980 nigel 77 #endif
2981    
2982     /* Inside \Q...\E everything is literal except \E */
2983    
2984     if (inescq)
2985     {
2986 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
2987 nigel 77 {
2988 nigel 93 inescq = FALSE; /* Reset literal state */
2989     ptr++; /* Skip the 'E' */
2990     continue; /* Carry on with next */
2991 nigel 77 }
2992 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
2993 nigel 77 }
2994    
2995     /* Handle POSIX class names. Perl allows a negation extension of the
2996     form [:^name:]. A square bracket that doesn't match the syntax is
2997     treated as a literal. We also recognize the POSIX constructions
2998     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2999     5.6 and 5.8 do. */
3000    
3001 ph10 391 if (c == CHAR_LEFT_SQUARE_BRACKET &&
3002 ph10 392 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3003 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3004 nigel 77 {
3005     BOOL local_negate = FALSE;
3006 nigel 87 int posix_class, taboffset, tabopt;
3007 nigel 77 register const uschar *cbits = cd->cbits;
3008 nigel 87 uschar pbits[32];
3009 nigel 77
3010 ph10 391 if (ptr[1] != CHAR_COLON)
3011 nigel 77 {
3012     *errorcodeptr = ERR31;
3013     goto FAILED;
3014     }
3015    
3016     ptr += 2;
3017 ph10 391 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3018 nigel 77 {
3019     local_negate = TRUE;
3020 ph10 286 should_flip_negation = TRUE; /* Note negative special */
3021 nigel 77 ptr++;
3022     }
3023    
3024     posix_class = check_posix_name(ptr, tempptr - ptr);
3025     if (posix_class < 0)
3026     {
3027     *errorcodeptr = ERR30;
3028     goto FAILED;
3029     }
3030    
3031     /* If matching is caseless, upper and lower are converted to
3032     alpha. This relies on the fact that the class table starts with
3033     alpha, lower, upper as the first 3 entries. */
3034    
3035     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3036     posix_class = 0;
3037    
3038 nigel 87 /* We build the bit map for the POSIX class in a chunk of local store
3039     because we may be adding and subtracting from it, and we don't want to
3040     subtract bits that may be in the main map already. At the end we or the
3041     result into the bit map that is being built. */
3042 nigel 77
3043     posix_class *= 3;
3044 nigel 87
3045     /* Copy in the first table (always present) */
3046    
3047     memcpy(pbits, cbits + posix_class_maps[posix_class],
3048     32 * sizeof(uschar));
3049    
3050     /* If there is a second table, add or remove it as required. */
3051    
3052     taboffset = posix_class_maps[posix_class + 1];
3053     tabopt = posix_class_maps[posix_class + 2];
3054    
3055     if (taboffset >= 0)
3056 nigel 77 {
3057 nigel 87 if (tabopt >= 0)
3058     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3059 nigel 77 else
3060 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3061 nigel 77 }
3062    
3063 nigel 87 /* Not see if we need to remove any special characters. An option
3064     value of 1 removes vertical space and 2 removes underscore. */
3065    
3066     if (tabopt < 0) tabopt = -tabopt;
3067     if (tabopt == 1) pbits[1] &= ~0x3c;
3068     else if (tabopt == 2) pbits[11] &= 0x7f;
3069    
3070     /* Add the POSIX table or its complement into the main table that is
3071     being built and we are done. */
3072    
3073     if (local_negate)
3074     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3075     else
3076     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3077    
3078 nigel 77 ptr = tempptr + 1;
3079     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3080     continue; /* End of POSIX syntax handling */
3081     }
3082    
3083     /* Backslash may introduce a single character, or it may introduce one
3084 nigel 93 of the specials, which just set a flag. The sequence \b is a special
3085     case. Inside a class (and only there) it is treated as backspace.
3086     Elsewhere it marks a word boundary. Other escapes have preset maps ready
3087 ph10 205 to 'or' into the one we are building. We assume they have more than one
3088 nigel 77 character in them, so set class_charcount bigger than one. */
3089    
3090 ph10 391 if (c == CHAR_BACKSLASH)
3091 nigel 77 {
3092 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3093     if (*errorcodeptr != 0) goto FAILED;
3094 nigel 77
3095 ph10 391 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3096     else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3097     else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3098 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
3099     {
3100 ph10 391 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3101 nigel 77 {
3102     ptr += 2; /* avoid empty string */
3103     }
3104     else inescq = TRUE;
3105     continue;
3106     }
3107 ph10 220 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3108 nigel 77
3109     if (c < 0)
3110     {
3111     register const uschar *cbits = cd->cbits;
3112     class_charcount += 2; /* Greater than 1 is what matters */
3113 nigel 93
3114     /* Save time by not doing this in the pre-compile phase. */
3115    
3116     if (lengthptr == NULL) switch (-c)
3117 nigel 77 {
3118     case ESC_d:
3119     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3120     continue;
3121    
3122     case ESC_D:
3123 ph10 286 should_flip_negation = TRUE;
3124 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3125     continue;
3126    
3127     case ESC_w:
3128     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3129     continue;
3130    
3131     case ESC_W:
3132 ph10 286 should_flip_negation = TRUE;
3133 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3134     continue;
3135    
3136     case ESC_s:
3137     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3138     classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3139     continue;
3140    
3141     case ESC_S:
3142 ph10 286 should_flip_negation = TRUE;
3143 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3144     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3145     continue;
3146    
3147 nigel 93 default: /* Not recognized; fall through */
3148     break; /* Need "default" setting to stop compiler warning. */
3149     }
3150    
3151     /* In the pre-compile phase, just do the recognition. */
3152    
3153     else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3154     c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3155 ph10 180
3156 ph10 178 /* We need to deal with \H, \h, \V, and \v in both phases because
3157     they use extra memory. */
3158 ph10 180
3159 ph10 178 if (-c == ESC_h)
3160     {
3161     SETBIT(classbits, 0x09); /* VT */
3162     SETBIT(classbits, 0x20); /* SPACE */
3163 ph10 180 SETBIT(classbits, 0xa0); /* NSBP */
3164 ph10 178 #ifdef SUPPORT_UTF8
3165     if (utf8)
3166 ph10 180 {
3167 ph10 178 class_utf8 = TRUE;
3168     *class_utf8data++ = XCL_SINGLE;
3169 ph10 180 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3170 ph10 178 *class_utf8data++ = XCL_SINGLE;
3171 ph10 180 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3172     *class_utf8data++ = XCL_RANGE;
3173     class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3174     class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3175 ph10 178 *class_utf8data++ = XCL_SINGLE;
3176 ph10 180 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3177 ph10 178 *class_utf8data++ = XCL_SINGLE;
3178 ph10 180 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3179 ph10 178 *class_utf8data++ = XCL_SINGLE;
3180 ph10 180 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3181     }
3182     #endif
3183     continue;
3184     }
3185 nigel 93
3186 ph10 178 if (-c == ESC_H)
3187     {
3188     for (c = 0; c < 32; c++)
3189     {
3190     int x = 0xff;
3191     switch (c)
3192 ph10 180 {
3193 ph10 178 case 0x09/8: x ^= 1 << (0x09%8); break;
3194     case 0x20/8: x ^= 1 << (0x20%8); break;
3195     case 0xa0/8: x ^= 1 << (0xa0%8); break;
3196     default: break;
3197     }
3198     classbits[c] |= x;
3199 ph10 180 }
3200    
3201 ph10 178 #ifdef SUPPORT_UTF8
3202     if (utf8)
3203 ph10 180 {
3204 ph10 178 class_utf8 = TRUE;
3205 ph10 180 *class_utf8data++ = XCL_RANGE;
3206     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3207     class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3208     *class_utf8data++ = XCL_RANGE;
3209     class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3210     class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3211     *class_utf8data++ = XCL_RANGE;
3212     class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3213     class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3214     *class_utf8data++ = XCL_RANGE;
3215     class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3216     class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3217     *class_utf8data++ = XCL_RANGE;
3218     class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3219     class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3220     *class_utf8data++ = XCL_RANGE;
3221     class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3222     class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3223     *class_utf8data++ = XCL_RANGE;
3224     class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3225     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3226     }
3227     #endif
3228     continue;
3229     }
3230 ph10 178
3231     if (-c == ESC_v)
3232     {
3233     SETBIT(classbits, 0x0a); /* LF */
3234     SETBIT(classbits, 0x0b); /* VT */
3235 ph10 180 SETBIT(classbits, 0x0c); /* FF */
3236     SETBIT(classbits, 0x0d); /* CR */
3237     SETBIT(classbits, 0x85); /* NEL */
3238 ph10 178 #ifdef SUPPORT_UTF8
3239     if (utf8)
3240 ph10 180 {
3241 ph10 178 class_utf8 = TRUE;
3242 ph10 180 *class_utf8data++ = XCL_RANGE;
3243     class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3244     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3245     }
3246     #endif
3247     continue;
3248     }
3249 ph10 178
3250     if (-c == ESC_V)
3251     {
3252     for (c = 0; c < 32; c++)
3253     {
3254     int x = 0xff;
3255     switch (c)
3256 ph10 180 {
3257 ph10 178 case 0x0a/8: x ^= 1 << (0x0a%8);
3258     x ^= 1 << (0x0b%8);
3259     x ^= 1 << (0x0c%8);
3260 ph10 180 x ^= 1 << (0x0d%8);
3261 ph10 178 break;
3262     case 0x85/8: x ^= 1 << (0x85%8); break;
3263     default: break;
3264     }
3265     classbits[c] |= x;
3266 ph10 180 }
3267    
3268 ph10 178 #ifdef SUPPORT_UTF8
3269     if (utf8)
3270 ph10 180 {
3271 ph10 178 class_utf8 = TRUE;
3272 ph10 180 *class_utf8data++ = XCL_RANGE;
3273     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3274     class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3275     *class_utf8data++ = XCL_RANGE;
3276     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3277     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3278     }
3279     #endif
3280     continue;
3281     }
3282 ph10 178
3283 nigel 93 /* We need to deal with \P and \p in both phases. */
3284    
3285 nigel 77 #ifdef SUPPORT_UCP
3286 nigel 93 if (-c == ESC_p || -c == ESC_P)
3287     {
3288     BOOL negated;
3289     int pdata;
3290     int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3291     if (ptype < 0) goto FAILED;
3292     class_utf8 = TRUE;
3293     *class_utf8data++ = ((-c == ESC_p) != negated)?
3294     XCL_PROP : XCL_NOTPROP;
3295     *class_utf8data++ = ptype;
3296     *class_utf8data++ = pdata;
3297     class_charcount -= 2; /* Not a < 256 character */
3298 nigel 77 continue;
3299 nigel 93 }
3300 nigel 77 #endif
3301 nigel 93 /* Unrecognized escapes are faulted if PCRE is running in its
3302     strict mode. By default, for compatibility with Perl, they are
3303     treated as literals. */
3304 nigel 77
3305 nigel 93 if ((options & PCRE_EXTRA) != 0)
3306     {
3307     *errorcodeptr = ERR7;
3308     goto FAILED;
3309     }
3310 nigel 77
3311 nigel 93 class_charcount -= 2; /* Undo the default count from above */
3312     c = *ptr; /* Get the final character and fall through */
3313 nigel 77 }
3314    
3315     /* Fall through if we have a single character (c >= 0). This may be
3316 nigel 93 greater than 256 in UTF-8 mode. */
3317 nigel 77
3318     } /* End of backslash handling */
3319    
3320     /* A single character may be followed by '-' to form a range. However,
3321     Perl does not permit ']' to be the end of the range. A '-' character
3322 nigel 93 at the end is treated as a literal. Perl ignores orphaned \E sequences
3323     entirely. The code for handling \Q and \E is messy. */
3324 nigel 77
3325 nigel 93 CHECK_RANGE:
3326 ph10 391 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3327 nigel 77 {
3328 nigel 93 inescq = FALSE;
3329     ptr += 2;
3330     }
3331    
3332     oldptr = ptr;
3333 ph10 231
3334 ph10 230 /* Remember \r or \n */
3335 ph10 231
3336 ph10 391 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3337 ph10 231
3338 ph10 230 /* Check for range */
3339 nigel 93
3340 ph10 391 if (!inescq && ptr[1] == CHAR_MINUS)
3341 nigel 93 {
3342 nigel 77 int d;
3343     ptr += 2;
3344 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3345 nigel 77
3346 nigel 93 /* If we hit \Q (not followed by \E) at this point, go into escaped
3347     mode. */
3348    
3349 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3350 nigel 93 {
3351     ptr += 2;
3352 ph10 392 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3353 ph10 391 { ptr += 2; continue; }
3354 nigel 93 inescq = TRUE;
3355     break;
3356     }
3357    
3358 ph10 391 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3359 nigel 93 {
3360     ptr = oldptr;
3361     goto LONE_SINGLE_CHARACTER;
3362     }
3363    
3364 nigel 77 #ifdef SUPPORT_UTF8
3365     if (utf8)
3366     { /* Braces are required because the */
3367     GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3368     }
3369     else
3370     #endif
3371     d = *ptr; /* Not UTF-8 mode */
3372    
3373     /* The second part of a range can be a single-character escape, but
3374     not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3375     in such circumstances. */
3376    
3377 ph10 391 if (!inescq && d == CHAR_BACKSLASH)
3378 nigel 77 {
3379 nigel 93 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3380     if (*errorcodeptr != 0) goto FAILED;
3381 nigel 77
3382 ph10 275 /* \b is backspace; \X is literal X; \R is literal R; any other
3383 nigel 93 special means the '-' was literal */
3384 nigel 77
3385     if (d < 0)
3386     {
3387 ph10 391 if (d == -ESC_b) d = CHAR_BS;
3388     else if (d == -ESC_X) d = CHAR_X;
3389     else if (d == -ESC_R) d = CHAR_R; else
3390 nigel 77 {
3391 nigel 93 ptr = oldptr;
3392 nigel 77 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3393     }
3394     }
3395     }
3396    
3397 nigel 93 /* Check that the two values are in the correct order. Optimize
3398     one-character ranges */
3399 nigel 77
3400 nigel 93 if (d < c)
3401     {
3402     *errorcodeptr = ERR8;
3403     goto FAILED;
3404     }
3405    
3406 nigel 77 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3407    
3408 ph10 230 /* Remember \r or \n */
3409 ph10 231
3410 ph10 391 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3411 ph10 231
3412 nigel 77 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3413     matching, we have to use an XCLASS with extra data items. Caseless
3414     matching for characters > 127 is available only if UCP support is
3415     available. */
3416    
3417     #ifdef SUPPORT_UTF8
3418     if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3419     {
3420     class_utf8 = TRUE;
3421    
3422     /* With UCP support, we can find the other case equivalents of
3423     the relevant characters. There may be several ranges. Optimize how
3424     they fit with the basic range. */
3425    
3426     #ifdef SUPPORT_UCP
3427     if ((options & PCRE_CASELESS) != 0)
3428     {
3429 nigel 93 unsigned int occ, ocd;
3430     unsigned int cc = c;
3431     unsigned int origd = d;
3432 nigel 77 while (get_othercase_range(&cc, origd, &occ, &ocd))
3433     {
3434 ph10 180 if (occ >= (unsigned int)c &&
3435     ocd <= (unsigned int)d)
3436 ph10 176 continue; /* Skip embedded ranges */
3437 nigel 77
3438 ph10 180 if (occ < (unsigned int)c &&
3439 ph10 176 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3440 nigel 77 { /* if there is overlap, */
3441     c = occ; /* noting that if occ < c */
3442     continue; /* we can't have ocd > d */
3443     } /* because a subrange is */
3444 ph10 180 if (ocd > (unsigned int)d &&
3445 ph10 176 occ <= (unsigned int)d + 1) /* always shorter than */
3446 nigel 77 { /* the basic range. */
3447     d = ocd;
3448     continue;
3449     }
3450    
3451     if (occ == ocd)
3452     {
3453     *class_utf8data++ = XCL_SINGLE;
3454     }
3455     else
3456     {
3457     *class_utf8data++ = XCL_RANGE;
3458     class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3459     }
3460     class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3461     }
3462     }
3463     #endif /* SUPPORT_UCP */
3464    
3465     /* Now record the original range, possibly modified for UCP caseless
3466     overlapping ranges. */
3467    
3468     *class_utf8data++ = XCL_RANGE;
3469     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3470     class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3471    
3472     /* With UCP support, we are done. Without UCP support, there is no
3473     caseless matching for UTF-8 characters > 127; we can use the bit map
3474     for the smaller ones. */
3475    
3476     #ifdef SUPPORT_UCP
3477     continue; /* With next character in the class */
3478     #else
3479     if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3480    
3481     /* Adjust upper limit and fall through to set up the map */
3482    
3483     d = 127;
3484    
3485     #endif /* SUPPORT_UCP */
3486     }
3487     #endif /* SUPPORT_UTF8 */
3488    
3489     /* We use the bit map for all cases when not in UTF-8 mode; else
3490     ranges that lie entirely within 0-127 when there is UCP support; else
3491     for partial ranges without UCP support. */
3492    
3493 nigel 93 class_charcount += d - c + 1;
3494     class_lastchar = d;
3495    
3496     /* We can save a bit of time by skipping this in the pre-compile. */
3497    
3498     if (lengthptr == NULL) for (; c <= d; c++)
3499 nigel 77 {
3500     classbits[c/8] |= (1 << (c&7));
3501     if ((options & PCRE_CASELESS) != 0)
3502     {
3503     int uc = cd->fcc[c]; /* flip case */
3504     classbits[uc/8] |= (1 << (uc&7));
3505     }
3506     }
3507    
3508     continue; /* Go get the next char in the class */
3509     }
3510    
3511     /* Handle a lone single character - we can get here for a normal
3512     non-escape char, or after \ that introduces a single character or for an
3513     apparent range that isn't. */
3514    
3515     LONE_SINGLE_CHARACTER:
3516 ph10 231
3517 nigel 77 /* Handle a character that cannot go in the bit map */
3518    
3519     #ifdef SUPPORT_UTF8
3520     if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3521     {
3522     class_utf8 = TRUE;
3523     *class_utf8data++ = XCL_SINGLE;
3524     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3525    
3526     #ifdef SUPPORT_UCP
3527     if ((options & PCRE_CASELESS) != 0)
3528     {
3529 nigel 93 unsigned int othercase;
3530 ph10 349 if ((othercase = UCD_OTHERCASE(c)) != c)
3531 nigel 77 {
3532     *class_utf8data++ = XCL_SINGLE;
3533     class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3534     }
3535     }
3536     #endif /* SUPPORT_UCP */
3537    
3538     }
3539     else
3540     #endif /* SUPPORT_UTF8 */
3541    
3542     /* Handle a single-byte character */
3543     {
3544     classbits[c/8] |= (1 << (c&7));
3545     if ((options & PCRE_CASELESS) != 0)
3546     {
3547     c = cd->fcc[c]; /* flip case */
3548     classbits[c/8] |= (1 << (c&7));
3549     }
3550     class_charcount++;
3551     class_lastchar = c;
3552     }
3553     }
3554    
3555 nigel 93 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3556 nigel 77
3557 ph10 391 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3558 nigel 77
3559 nigel 93 if (c == 0) /* Missing terminating ']' */
3560     {
3561     *errorcodeptr = ERR6;
3562     goto FAILED;
3563     }
3564 ph10 231
3565    
3566 ph10 230 /* This code has been disabled because it would mean that \s counts as
3567     an explicit \r or \n reference, and that's not really what is wanted. Now
3568     we set the flag only if there is a literal "\r" or "\n" in the class. */
3569 ph10 227
3570 ph10 230 #if 0
3571 ph10 226 /* Remember whether \r or \n are in this class */
3572 ph10 227
3573 ph10 226 if (negate_class)
3574     {
3575 ph10 230 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3576 ph10 226 }
3577     else
3578     {
3579 ph10 230 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3580 ph10 227 }
3581 ph10 230 #endif
3582 ph10 227
3583 ph10 231
3584 nigel 77 /* If class_charcount is 1, we saw precisely one character whose value is
3585 ph10 227 less than 256. As long as there were no characters >= 128 and there was no
3586     use of \p or \P, in other words, no use of any XCLASS features, we can
3587     optimize.
3588    
3589 ph10 223 In UTF-8 mode, we can optimize the negative case only if there were no
3590     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3591     operate on single-bytes only. This is an historical hangover. Maybe one day
3592     we can tidy these opcodes to handle multi-byte characters.
3593 nigel 77
3594     The optimization throws away the bit map. We turn the item into a
3595     1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3596     that OP_NOT does not support multibyte characters. In the positive case, it
3597     can cause firstbyte to be set. Otherwise, there can be no first char if
3598     this item is first, whatever repeat count may follow. In the case of
3599     reqbyte, save the previous value for reinstating. */
3600    
3601     #ifdef SUPPORT_UTF8
3602 ph10 227 if (class_charcount == 1 && !class_utf8 &&
3603 ph10 223 (!utf8 || !negate_class || class_lastchar < 128))
3604 nigel 77 #else
3605     if (class_charcount == 1)
3606     #endif
3607     {
3608     zeroreqbyte = reqbyte;
3609    
3610     /* The OP_NOT opcode works on one-byte characters only. */
3611    
3612     if (negate_class)
3613     {
3614     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3615     zerofirstbyte = firstbyte;
3616     *code++ = OP_NOT;
3617     *code++ = class_lastchar;
3618     break;
3619     }
3620    
3621     /* For a single, positive character, get the value into mcbuffer, and
3622     then we can handle this with the normal one-character code. */
3623    
3624     #ifdef SUPPORT_UTF8
3625     if (utf8 && class_lastchar > 127)
3626     mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3627     else
3628     #endif
3629     {
3630     mcbuffer[0] = class_lastchar;
3631     mclength = 1;
3632     }
3633     goto ONE_CHAR;
3634     } /* End of 1-char optimization */
3635    
3636     /* The general case - not the one-char optimization. If this is the first
3637     thing in the branch, there can be no first char setting, whatever the
3638     repeat count. Any reqbyte setting must remain unchanged after any kind of
3639     repeat. */
3640    
3641     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3642     zerofirstbyte = firstbyte;
3643     zeroreqbyte = reqbyte;
3644    
3645     /* If there are characters with values > 255, we have to compile an
3646 ph10 286 extended class, with its own opcode, unless there was a negated special
3647     such as \S in the class, because in that case all characters > 255 are in
3648     the class, so any that were explicitly given as well can be ignored. If
3649 ph10 264 (when there are explicit characters > 255 that must be listed) there are no
3650     characters < 256, we can omit the bitmap in the actual compiled code. */
3651 nigel 77
3652     #ifdef SUPPORT_UTF8
3653 ph10 264 if (class_utf8 && !should_flip_negation)
3654 nigel 77 {
3655     *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3656     *code++ = OP_XCLASS;
3657     code += LINK_SIZE;
3658     *code = negate_class? XCL_NOT : 0;
3659    
3660 nigel 93 /* If the map is required, move up the extra data to make room for it;
3661     otherwise just move the code pointer to the end of the extra data. */
3662 nigel 77
3663     if (class_charcount > 0)
3664     {
3665     *code++ |= XCL_MAP;
3666 nigel 93 memmove(code + 32, code, class_utf8data - code);
3667 nigel 77 memcpy(code, classbits, 32);
3668 nigel 93 code = class_utf8data + 32;
3669 nigel 77 }
3670 nigel 93 else code = class_utf8data;
3671 nigel 77
3672     /* Now fill in the complete length of the item */
3673    
3674     PUT(previous, 1, code - previous);
3675     break; /* End of class handling */
3676     }
3677     #endif
3678    
3679 ph10 286 /* If there are no characters > 255, set the opcode to OP_CLASS or
3680     OP_NCLASS, depending on whether the whole class was negated and whether
3681     there were negative specials such as \S in the class. Then copy the 32-byte
3682 ph10 264 map into the code vector, negating it if necessary. */
3683 ph10 286
3684 ph10 264 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3685 nigel 77 if (negate_class)
3686     {
3687 nigel 93 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3688     for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3689 nigel 77 }
3690     else
3691     {
3692     memcpy(code, classbits, 32);
3693     }
3694     code += 32;
3695     break;
3696    
3697 nigel 93
3698     /* ===================================================================*/
3699 nigel 77 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3700     has been tested above. */
3701    
3702 ph10 391 case CHAR_LEFT_CURLY_BRACKET:
3703 nigel 77 if (!is_quantifier) goto NORMAL_CHAR;
3704     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3705     if (*errorcodeptr != 0) goto FAILED;
3706     goto REPEAT;
3707    
3708 ph10 391 case CHAR_ASTERISK:
3709 nigel 77 repeat_min = 0;
3710     repeat_max = -1;
3711     goto REPEAT;
3712    
3713 ph10 391 case CHAR_PLUS:
3714 nigel 77 repeat_min = 1;
3715     repeat_max = -1;
3716     goto REPEAT;
3717    
3718 ph10 391 case CHAR_QUESTION_MARK:
3719 nigel 77 repeat_min = 0;
3720     repeat_max = 1;
3721    
3722     REPEAT:
3723     if (previous == NULL)
3724     {
3725     *errorcodeptr = ERR9;
3726     goto FAILED;
3727     }
3728    
3729     if (repeat_min == 0)
3730     {
3731     firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3732     reqbyte = zeroreqbyte; /* Ditto */
3733     }
3734    
3735     /* Remember whether this is a variable length repeat */
3736    
3737     reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3738    
3739     op_type = 0; /* Default single-char op codes */
3740     possessive_quantifier = FALSE; /* Default not possessive quantifier */
3741    
3742     /* Save start of previous item, in case we have to move it up to make space
3743     for an inserted OP_ONCE for the additional '+' extension. */
3744    
3745     tempcode = previous;
3746    
3747     /* If the next character is '+', we have a possessive quantifier. This
3748     implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3749     If the next character is '?' this is a minimizing repeat, by default,
3750     but if PCRE_UNGREEDY is set, it works the other way round. We change the
3751     repeat type to the non-default. */
3752    
3753 ph10 391 if (ptr[1] == CHAR_PLUS)
3754 nigel 77 {
3755     repeat_type = 0; /* Force greedy */
3756     possessive_quantifier = TRUE;
3757     ptr++;
3758     }
3759 ph10 391 else if (ptr[1] == CHAR_QUESTION_MARK)
3760 nigel 77 {
3761     repeat_type = greedy_non_default;
3762     ptr++;
3763     }
3764     else repeat_type = greedy_default;
3765    
3766     /* If previous was a character match, abolish the item and generate a
3767     repeat item instead. If a char item has a minumum of more than one, ensure
3768     that it is set in reqbyte - it might not be if a sequence such as x{3} is
3769     the first thing in a branch because the x will have gone into firstbyte
3770     instead. */
3771    
3772     if (*previous == OP_CHAR || *previous == OP_CHARNC)
3773     {
3774     /* Deal with UTF-8 characters that take up more than one byte. It's
3775     easier to write this out separately than try to macrify it. Use c to
3776     hold the length of the character in bytes, plus 0x80 to flag that it's a
3777     length rather than a small character. */
3778    
3779     #ifdef SUPPORT_UTF8
3780     if (utf8 && (code[-1] & 0x80) != 0)
3781     {
3782     uschar *lastchar = code - 1;
3783     while((*lastchar & 0xc0) == 0x80) lastchar--;
3784     c = code - lastchar; /* Length of UTF-8 character */
3785     memcpy(utf8_char, lastchar, c); /* Save the char */
3786     c |= 0x80; /* Flag c as a length */
3787     }
3788     else
3789     #endif
3790    
3791     /* Handle the case of a single byte - either with no UTF8 support, or
3792     with UTF-8 disabled, or for a UTF-8 character < 128. */
3793    
3794     {
3795     c = code[-1];
3796     if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3797     }
3798    
3799 nigel 93 /* If the repetition is unlimited, it pays to see if the next thing on
3800     the line is something that cannot possibly match this character. If so,
3801     automatically possessifying this item gains some performance in the case
3802     where the match fails. */
3803    
3804     if (!possessive_quantifier &&
3805     repeat_max < 0 &&
3806     check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3807     options, cd))
3808     {
3809     repeat_type = 0; /* Force greedy */
3810     possessive_quantifier = TRUE;
3811     }
3812    
3813 nigel 77 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3814     }
3815    
3816     /* If previous was a single negated character ([^a] or similar), we use
3817     one of the special opcodes, replacing it. The code is shared with single-
3818     character repeats by setting opt_type to add a suitable offset into
3819 nigel 93 repeat_type. We can also test for auto-possessification. OP_NOT is
3820     currently used only for single-byte chars. */
3821 nigel 77
3822     else if (*previous == OP_NOT)
3823     {
3824     op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3825     c = previous[1];
3826 nigel 93 if (!possessive_quantifier &&
3827     repeat_max < 0 &&
3828     check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3829     {
3830     repeat_type = 0; /* Force greedy */
3831     possessive_quantifier = TRUE;
3832     }
3833 nigel 77 goto OUTPUT_SINGLE_REPEAT;
3834     }
3835    
3836     /* If previous was a character type match (\d or similar), abolish it and
3837     create a suitable repeat item. The code is shared with single-character
3838     repeats by setting op_type to add a suitable offset into repeat_type. Note
3839     the the Unicode property types will be present only when SUPPORT_UCP is
3840     defined, but we don't wrap the little bits of code here because it just
3841     makes it horribly messy. */
3842    
3843     else if (*previous < OP_EODN)
3844     {
3845     uschar *oldcode;
3846 nigel 87 int prop_type, prop_value;
3847 nigel 77 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3848     c = *previous;
3849    
3850 nigel 93 if (!possessive_quantifier &&
3851     repeat_max < 0 &&
3852     check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3853     {
3854     repeat_type = 0; /* Force greedy */
3855     possessive_quantifier = TRUE;
3856     }
3857    
3858 nigel 77 OUTPUT_SINGLE_REPEAT:
3859 nigel 87 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3860     {
3861     prop_type = previous[1];
3862     prop_value = previous[2];
3863     }
3864     else prop_type = prop_value = -1;
3865 nigel 77
3866     oldcode = code;
3867     code = previous; /* Usually overwrite previous item */
3868    
3869     /* If the maximum is zero then the minimum must also be zero; Perl allows
3870     this case, so we do too - by simply omitting the item altogether. */
3871    
3872     if (repeat_max == 0) goto END_REPEAT;
3873    
3874 ph10 426 /*--------------------------------------------------------------------*/
3875     /* This code is obsolete from release 8.00; the restriction was finally
3876     removed: */
3877    
3878 nigel 77 /* All real repeats make it impossible to handle partial matching (maybe
3879     one day we will be able to remove this restriction). */
3880 ph10 426
3881     /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3882     /*--------------------------------------------------------------------*/
3883 nigel 77
3884     /* Combine the op_type with the repeat_type */
3885    
3886     repeat_type += op_type;
3887    
3888     /* A minimum of zero is handled either as the special case * or ?, or as
3889     an UPTO, with the maximum given. */
3890    
3891     if (repeat_min == 0)
3892     {
3893     if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3894     else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3895     else
3896     {
3897     *code++ = OP_UPTO + repeat_type;
3898     PUT2INC(code, 0, repeat_max);
3899     }
3900     }
3901    
3902     /* A repeat minimum of 1 is optimized into some special cases. If the
3903 nigel 93 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3904 nigel 77 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3905     one less than the maximum. */
3906    
3907     else if (repeat_min == 1)
3908     {
3909     if (repeat_max == -1)
3910     *code++ = OP_PLUS + repeat_type;
3911     else
3912     {
3913     code = oldcode; /* leave previous item in place */
3914     if (repeat_max == 1) goto END_REPEAT;
3915     *code++ = OP_UPTO + repeat_type;
3916     PUT2INC(code, 0, repeat_max - 1);
3917     }
3918     }
3919    
3920     /* The case {n,n} is just an EXACT, while the general case {n,m} is
3921     handled as an EXACT followed by an UPTO. */
3922    
3923     else
3924     {
3925     *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3926     PUT2INC(code, 0, repeat_min);
3927    
3928     /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3929     we have to insert the character for the previous code. For a repeated
3930 nigel 87 Unicode property match, there are two extra bytes that define the
3931 nigel 77 required property. In UTF-8 mode, long characters have their length in
3932     c, with the 0x80 bit as a flag. */
3933    
3934     if (repeat_max < 0)
3935     {
3936     #ifdef SUPPORT_UTF8
3937     if (utf8 && c >= 128)
3938     {
3939     memcpy(code, utf8_char, c & 7);
3940     code += c & 7;
3941     }
3942     else
3943     #endif
3944     {
3945     *code++ = c;
3946 nigel 87 if (prop_type >= 0)
3947     {
3948     *code++ = prop_type;
3949     *code++ = prop_value;
3950     }
3951 nigel 77 }
3952     *code++ = OP_STAR + repeat_type;
3953     }
3954    
3955     /* Else insert an UPTO if the max is greater than the min, again
3956 nigel 93 preceded by the character, for the previously inserted code. If the
3957     UPTO is just for 1 instance, we can use QUERY instead. */
3958 nigel 77
3959     else if (repeat_max != repeat_min)
3960     {
3961     #ifdef SUPPORT_UTF8
3962     if (utf8 && c >= 128)
3963     {
3964     memcpy(code, utf8_char, c & 7);
3965     code += c & 7;
3966     }
3967