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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 426 - (hide annotations) (download)
Wed Aug 26 15:38:32 2009 UTC (4 years, 11 months ago) by ph10
File MIME type: text/plain
File size: 212760 byte(s)
Remove restrictions on pcre_exec() partial matching.

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 411 }
1104 ph10 408 }
1105 ph10 411 }
1106 ph10 408
1107 ph10 411 /* Past any initial parenthesis handling, scan for parentheses or vertical
1108 ph10 408 bars. */
1109    
1110 nigel 91 for (; *ptr != 0; ptr++)
1111     {
1112 nigel 93 /* Skip over backslashed characters and also entire \Q...\E */
1113    
1114 ph10 391 if (*ptr == CHAR_BACKSLASH)
1115 nigel 93 {
1116 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1117 ph10 391 if (*ptr == CHAR_Q) for (;;)
1118 nigel 93 {
1119 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1120 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1121 ph10 391 if (*(++ptr) == CHAR_E) break;
1122 nigel 93 }
1123     continue;
1124     }
1125    
1126 ph10 340 /* Skip over character classes; this logic must be similar to the way they
1127     are handled for real. If the first character is '^', skip it. Also, if the
1128     first few characters (either before or after ^) are \Q\E or \E we skip them
1129 ph10 392 too. This makes for compatibility with Perl. Note the use of STR macros to
1130 ph10 391 encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1131 nigel 93
1132 ph10 391 if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1133 nigel 93 {
1134 ph10 340 BOOL negate_class = FALSE;
1135     for (;;)
1136     {
1137     int c = *(++ptr);
1138 ph10 391 if (c == CHAR_BACKSLASH)
1139 ph10 340 {
1140 ph10 392 if (ptr[1] == CHAR_E)
1141 ph10 391 ptr++;
1142 ph10 392 else if (strncmp((const char *)ptr+1,
1143     STR_Q STR_BACKSLASH STR_E, 3) == 0)
1144 ph10 391 ptr += 3;
1145 ph10 392 else
1146 ph10 391 break;
1147 ph10 340 }
1148 ph10 391 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
1149 ph10 340 negate_class = TRUE;
1150     else break;
1151     }
1152    
1153     /* If the next character is ']', it is a data character that must be
1154 ph10 341 skipped, except in JavaScript compatibility mode. */
1155 ph10 345
1156 ph10 392 if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1157 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1158 ph10 345 ptr++;
1159    
1160 ph10 391 while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1161 nigel 93 {
1162 ph10 220 if (*ptr == 0) return -1;
1163 ph10 391 if (*ptr == CHAR_BACKSLASH)
1164 nigel 93 {
1165 ph10 408 if (*(++ptr) == 0) goto FAIL_EXIT;
1166 ph10 391 if (*ptr == CHAR_Q) for (;;)
1167 nigel 93 {
1168 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1169 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1170 ph10 391 if (*(++ptr) == CHAR_E) break;
1171 nigel 93 }
1172     continue;
1173     }
1174     }
1175     continue;
1176     }
1177    
1178     /* Skip comments in /x mode */
1179    
1180 ph10 391 if (xmode && *ptr == CHAR_NUMBER_SIGN)
1181 nigel 93 {
1182 ph10 391 while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1183 ph10 408 if (*ptr == 0) goto FAIL_EXIT;
1184 nigel 93 continue;
1185     }
1186    
1187 ph10 408 /* Check for the special metacharacters */
1188 ph10 411
1189 ph10 408 if (*ptr == CHAR_LEFT_PARENTHESIS)
1190 nigel 93 {
1191 ph10 408 int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1192     if (rc > 0) return rc;
1193     if (*ptr == 0) goto FAIL_EXIT;
1194 nigel 93 }
1195 ph10 411
1196 ph10 408 else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1197     {
1198     if (dup_parens && *count < hwm_count) *count = hwm_count;
1199     *ptrptr = ptr;
1200     return -1;
1201     }
1202 ph10 411
1203     else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1204 ph10 408 {
1205     if (*count > hwm_count) hwm_count = *count;
1206     *count = start_count;
1207 ph10 411 }
1208 ph10 408 }
1209 nigel 93
1210 ph10 408 FAIL_EXIT:
1211     *ptrptr = ptr;
1212     return -1;
1213     }
1214 nigel 93
1215    
1216    
1217    
1218 ph10 408 /*************************************************
1219     * Find forward referenced subpattern *
1220     *************************************************/
1221 nigel 93
1222 ph10 408 /* This function scans along a pattern's text looking for capturing
1223     subpatterns, and counting them. If it finds a named pattern that matches the
1224     name it is given, it returns its number. Alternatively, if the name is NULL, it
1225     returns when it reaches a given numbered subpattern. This is used for forward
1226     references to subpatterns. We used to be able to start this scan from the
1227     current compiling point, using the current count value from cd->bracount, and
1228     do it all in a single loop, but the addition of the possibility of duplicate
1229     subpattern numbers means that we have to scan from the very start, in order to
1230     take account of such duplicates, and to use a recursive function to keep track
1231     of the different types of group.
1232    
1233     Arguments:
1234     cd compile background data
1235     name name to seek, or NULL if seeking a numbered subpattern
1236     lorn name length, or subpattern number if name is NULL
1237     xmode TRUE if we are in /x mode
1238    
1239     Returns: the number of the found subpattern, or -1 if not found
1240     */
1241    
1242     static int
1243     find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1244     {
1245     uschar *ptr = (uschar *)cd->start_pattern;
1246     int count = 0;
1247     int rc;
1248    
1249     /* If the pattern does not start with an opening parenthesis, the first call
1250     to find_parens_sub() will scan right to the end (if necessary). However, if it
1251     does start with a parenthesis, find_parens_sub() will return when it hits the
1252     matching closing parens. That is why we have to have a loop. */
1253    
1254 ph10 411 for (;;)
1255     {
1256 ph10 408 rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1257 ph10 411 if (rc > 0 || *ptr++ == 0) break;
1258     }
1259    
1260 ph10 408 return rc;
1261 nigel 91 }
1262    
1263    
1264    
1265 ph10 408
1266 nigel 91 /*************************************************
1267 nigel 77 * Find first significant op code *
1268     *************************************************/
1269    
1270     /* This is called by several functions that scan a compiled expression looking
1271     for a fixed first character, or an anchoring op code etc. It skips over things
1272     that do not influence this. For some calls, a change of option is important.
1273     For some calls, it makes sense to skip negative forward and all backward
1274     assertions, and also the \b assertion; for others it does not.
1275    
1276     Arguments:
1277     code pointer to the start of the group
1278     options pointer to external options
1279     optbit the option bit whose changing is significant, or
1280     zero if none are
1281     skipassert TRUE if certain assertions are to be skipped
1282    
1283     Returns: pointer to the first significant opcode
1284     */
1285    
1286     static const uschar*
1287     first_significant_code(const uschar *code, int *options, int optbit,
1288     BOOL skipassert)
1289     {
1290     for (;;)
1291     {
1292     switch ((int)*code)
1293     {
1294     case OP_OPT:
1295     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1296     *options = (int)code[1];
1297     code += 2;
1298     break;
1299    
1300     case OP_ASSERT_NOT:
1301     case OP_ASSERTBACK:
1302     case OP_ASSERTBACK_NOT:
1303     if (!skipassert) return code;
1304     do code += GET(code, 1); while (*code == OP_ALT);
1305     code += _pcre_OP_lengths[*code];
1306     break;
1307    
1308     case OP_WORD_BOUNDARY:
1309     case OP_NOT_WORD_BOUNDARY:
1310     if (!skipassert) return code;
1311     /* Fall through */
1312    
1313     case OP_CALLOUT:
1314     case OP_CREF:
1315 nigel 93 case OP_RREF:
1316     case OP_DEF:
1317 nigel 77 code += _pcre_OP_lengths[*code];
1318     break;
1319    
1320     default:
1321     return code;
1322     }
1323     }
1324     /* Control never reaches here */
1325     }
1326    
1327    
1328    
1329    
1330     /*************************************************
1331     * Find the fixed length of a pattern *
1332     *************************************************/
1333    
1334     /* Scan a pattern and compute the fixed length of subject that will match it,
1335     if the length is fixed. This is needed for dealing with backward assertions.
1336     In UTF8 mode, the result is in characters rather than bytes.
1337    
1338     Arguments:
1339     code points to the start of the pattern (the bracket)
1340     options the compiling options
1341    
1342     Returns: the fixed length, or -1 if there is no fixed length,
1343     or -2 if \C was encountered
1344     */
1345    
1346     static int
1347     find_fixedlength(uschar *code, int options)
1348     {
1349     int length = -1;
1350    
1351     register int branchlength = 0;
1352     register uschar *cc = code + 1 + LINK_SIZE;
1353    
1354     /* Scan along the opcodes for this branch. If we get to the end of the
1355     branch, check the length against that of the other branches. */
1356    
1357     for (;;)
1358     {
1359     int d;
1360     register int op = *cc;
1361     switch (op)
1362     {
1363 nigel 93 case OP_CBRA:
1364 nigel 77 case OP_BRA:
1365     case OP_ONCE:
1366     case OP_COND:
1367 nigel 93 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1368 nigel 77 if (d < 0) return d;
1369     branchlength += d;
1370     do cc += GET(cc, 1); while (*cc == OP_ALT);
1371     cc += 1 + LINK_SIZE;
1372     break;
1373    
1374     /* Reached end of a branch; if it's a ket it is the end of a nested
1375     call. If it's ALT it is an alternation in a nested call. If it is
1376     END it's the end of the outer call. All can be handled by the same code. */
1377    
1378     case OP_ALT:
1379     case OP_KET:
1380     case OP_KETRMAX:
1381     case OP_KETRMIN:
1382     case OP_END:
1383     if (length < 0) length = branchlength;
1384     else if (length != branchlength) return -1;
1385     if (*cc != OP_ALT) return length;
1386     cc += 1 + LINK_SIZE;
1387     branchlength = 0;
1388     break;
1389    
1390     /* Skip over assertive subpatterns */
1391    
1392     case OP_ASSERT:
1393     case OP_ASSERT_NOT:
1394     case OP_ASSERTBACK:
1395     case OP_ASSERTBACK_NOT:
1396     do cc += GET(cc, 1); while (*cc == OP_ALT);
1397     /* Fall through */
1398    
1399     /* Skip over things that don't match chars */
1400    
1401     case OP_REVERSE:
1402     case OP_CREF:
1403 nigel 93 case OP_RREF:
1404     case OP_DEF:
1405 nigel 77 case OP_OPT:
1406     case OP_CALLOUT:
1407     case OP_SOD:
1408     case OP_SOM:
1409     case OP_EOD:
1410     case OP_EODN:
1411     case OP_CIRC:
1412     case OP_DOLL:
1413     case OP_NOT_WORD_BOUNDARY:
1414     case OP_WORD_BOUNDARY:
1415     cc += _pcre_OP_lengths[*cc];
1416     break;
1417    
1418     /* Handle literal characters */
1419    
1420     case OP_CHAR:
1421     case OP_CHARNC:
1422 nigel 91 case OP_NOT:
1423 nigel 77 branchlength++;
1424     cc += 2;
1425     #ifdef SUPPORT_UTF8
1426 ph10 426 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1427     cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1428 nigel 77 #endif
1429     break;
1430    
1431     /* Handle exact repetitions. The count is already in characters, but we
1432     need to skip over a multibyte character in UTF8 mode. */
1433    
1434     case OP_EXACT:
1435     branchlength += GET2(cc,1);
1436     cc += 4;
1437     #ifdef SUPPORT_UTF8
1438 ph10 426 if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
1439     cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1440 nigel 77 #endif
1441     break;
1442    
1443     case OP_TYPEEXACT:
1444     branchlength += GET2(cc,1);
1445 ph10 220 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1446 nigel 77 cc += 4;
1447     break;
1448    
1449     /* Handle single-char matchers */
1450    
1451     case OP_PROP:
1452     case OP_NOTPROP:
1453 nigel 87 cc += 2;
1454 nigel 77 /* Fall through */
1455    
1456     case OP_NOT_DIGIT:
1457     case OP_DIGIT:
1458     case OP_NOT_WHITESPACE:
1459     case OP_WHITESPACE:
1460     case OP_NOT_WORDCHAR:
1461     case OP_WORDCHAR:
1462     case OP_ANY:
1463 ph10 342 case OP_ALLANY:
1464 nigel 77 branchlength++;
1465     cc++;
1466     break;
1467    
1468     /* The single-byte matcher isn't allowed */
1469    
1470     case OP_ANYBYTE:
1471     return -2;
1472    
1473     /* Check a class for variable quantification */
1474    
1475     #ifdef SUPPORT_UTF8
1476     case OP_XCLASS:
1477     cc += GET(cc, 1) - 33;
1478     /* Fall through */
1479     #endif
1480    
1481     case OP_CLASS:
1482     case OP_NCLASS:
1483     cc += 33;
1484    
1485     switch (*cc)
1486     {
1487     case OP_CRSTAR:
1488     case OP_CRMINSTAR:
1489     case OP_CRQUERY:
1490     case OP_CRMINQUERY:
1491     return -1;
1492    
1493     case OP_CRRANGE:
1494     case OP_CRMINRANGE:
1495     if (GET2(cc,1) != GET2(cc,3)) return -1;
1496     branchlength += GET2(cc,1);
1497     cc += 5;
1498     break;
1499    
1500     default:
1501     branchlength++;
1502     }
1503     break;
1504    
1505     /* Anything else is variable length */
1506    
1507     default:
1508     return -1;
1509     }
1510     }
1511     /* Control never gets here */
1512     }
1513    
1514    
1515    
1516    
1517     /*************************************************
1518     * Scan compiled regex for numbered bracket *
1519     *************************************************/
1520    
1521     /* This little function scans through a compiled pattern until it finds a
1522     capturing bracket with the given number.
1523    
1524     Arguments:
1525     code points to start of expression
1526     utf8 TRUE in UTF-8 mode
1527     number the required bracket number
1528    
1529     Returns: pointer to the opcode for the bracket, or NULL if not found
1530     */
1531    
1532     static const uschar *
1533     find_bracket(const uschar *code, BOOL utf8, int number)
1534     {
1535     for (;;)
1536     {
1537     register int c = *code;
1538     if (c == OP_END) return NULL;
1539 nigel 91
1540     /* XCLASS is used for classes that cannot be represented just by a bit
1541     map. This includes negated single high-valued characters. The length in
1542     the table is zero; the actual length is stored in the compiled code. */
1543    
1544     if (c == OP_XCLASS) code += GET(code, 1);
1545    
1546 nigel 93 /* Handle capturing bracket */
1547 nigel 91
1548 nigel 93 else if (c == OP_CBRA)
1549 nigel 77 {
1550 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1551 nigel 77 if (n == number) return (uschar *)code;
1552 nigel 93 code += _pcre_OP_lengths[c];
1553 nigel 77 }
1554 nigel 91
1555 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1556     repeated character types, we have to test for \p and \P, which have an extra
1557 ph10 218 two bytes of parameters. */
1558 nigel 91
1559 nigel 77 else
1560     {
1561 ph10 218 switch(c)
1562     {
1563     case OP_TYPESTAR:
1564     case OP_TYPEMINSTAR:
1565     case OP_TYPEPLUS:
1566     case OP_TYPEMINPLUS:
1567     case OP_TYPEQUERY:
1568     case OP_TYPEMINQUERY:
1569     case OP_TYPEPOSSTAR:
1570     case OP_TYPEPOSPLUS:
1571     case OP_TYPEPOSQUERY:
1572     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1573 ph10 220 break;
1574 ph10 221
1575     case OP_TYPEUPTO:
1576     case OP_TYPEMINUPTO:
1577     case OP_TYPEEXACT:
1578     case OP_TYPEPOSUPTO:
1579     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1580     break;
1581 ph10 220 }
1582    
1583 ph10 218 /* Add in the fixed length from the table */
1584 ph10 220
1585 nigel 77 code += _pcre_OP_lengths[c];
1586 ph10 220
1587 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1588     a multi-byte character. The length in the table is a minimum, so we have to
1589     arrange to skip the extra bytes. */
1590 ph10 220
1591 ph10 107 #ifdef SUPPORT_UTF8
1592 nigel 77 if (utf8) switch(c)
1593     {
1594     case OP_CHAR:
1595     case OP_CHARNC:
1596     case OP_EXACT:
1597     case OP_UPTO:
1598     case OP_MINUPTO:
1599 nigel 93 case OP_POSUPTO:
1600 nigel 77 case OP_STAR:
1601     case OP_MINSTAR:
1602 nigel 93 case OP_POSSTAR:
1603 nigel 77 case OP_PLUS:
1604     case OP_MINPLUS:
1605 nigel 93 case OP_POSPLUS:
1606 nigel 77 case OP_QUERY:
1607     case OP_MINQUERY:
1608 nigel 93 case OP_POSQUERY:
1609     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1610 nigel 77 break;
1611     }
1612 ph10 369 #else
1613     (void)(utf8); /* Keep compiler happy by referencing function argument */
1614 ph10 111 #endif
1615 nigel 77 }
1616     }
1617     }
1618    
1619    
1620    
1621     /*************************************************
1622     * Scan compiled regex for recursion reference *
1623     *************************************************/
1624    
1625     /* This little function scans through a compiled pattern until it finds an
1626     instance of OP_RECURSE.
1627    
1628     Arguments:
1629     code points to start of expression
1630     utf8 TRUE in UTF-8 mode
1631    
1632     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1633     */
1634    
1635     static const uschar *
1636     find_recurse(const uschar *code, BOOL utf8)
1637     {
1638     for (;;)
1639     {
1640     register int c = *code;
1641     if (c == OP_END) return NULL;
1642 nigel 91 if (c == OP_RECURSE) return code;
1643 ph10 220
1644 nigel 91 /* XCLASS is used for classes that cannot be represented just by a bit
1645     map. This includes negated single high-valued characters. The length in
1646     the table is zero; the actual length is stored in the compiled code. */
1647    
1648     if (c == OP_XCLASS) code += GET(code, 1);
1649    
1650 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1651     repeated character types, we have to test for \p and \P, which have an extra
1652 ph10 218 two bytes of parameters. */
1653 nigel 91
1654 nigel 77 else
1655     {
1656 ph10 218 switch(c)
1657     {
1658     case OP_TYPESTAR:
1659     case OP_TYPEMINSTAR:
1660     case OP_TYPEPLUS:
1661     case OP_TYPEMINPLUS:
1662     case OP_TYPEQUERY:
1663     case OP_TYPEMINQUERY:
1664     case OP_TYPEPOSSTAR:
1665     case OP_TYPEPOSPLUS:
1666     case OP_TYPEPOSQUERY:
1667     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1668 ph10 220 break;
1669 ph10 221
1670     case OP_TYPEPOSUPTO:
1671     case OP_TYPEUPTO:
1672     case OP_TYPEMINUPTO:
1673     case OP_TYPEEXACT:
1674     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1675     break;
1676 ph10 220 }
1677    
1678 ph10 218 /* Add in the fixed length from the table */
1679    
1680 nigel 77 code += _pcre_OP_lengths[c];
1681 ph10 220
1682 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1683     by a multi-byte character. The length in the table is a minimum, so we have
1684     to arrange to skip the extra bytes. */
1685 ph10 220
1686 ph10 107 #ifdef SUPPORT_UTF8
1687 nigel 77 if (utf8) switch(c)
1688     {
1689     case OP_CHAR:
1690     case OP_CHARNC:
1691     case OP_EXACT:
1692     case OP_UPTO:
1693     case OP_MINUPTO:
1694 nigel 93 case OP_POSUPTO:
1695 nigel 77 case OP_STAR:
1696     case OP_MINSTAR:
1697 nigel 93 case OP_POSSTAR:
1698 nigel 77 case OP_PLUS:
1699     case OP_MINPLUS:
1700 nigel 93 case OP_POSPLUS:
1701 nigel 77 case OP_QUERY:
1702     case OP_MINQUERY:
1703 nigel 93 case OP_POSQUERY:
1704     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1705 nigel 77 break;
1706     }
1707 ph10 369 #else
1708     (void)(utf8); /* Keep compiler happy by referencing function argument */
1709 ph10 111 #endif
1710 nigel 77 }
1711     }
1712     }
1713    
1714    
1715    
1716     /*************************************************
1717     * Scan compiled branch for non-emptiness *
1718     *************************************************/
1719    
1720     /* This function scans through a branch of a compiled pattern to see whether it
1721 nigel 93 can match the empty string or not. It is called from could_be_empty()
1722     below and from compile_branch() when checking for an unlimited repeat of a
1723     group that can match nothing. Note that first_significant_code() skips over
1724 ph10 282 backward and negative forward assertions when its final argument is TRUE. If we
1725     hit an unclosed bracket, we return "empty" - this means we've struck an inner
1726     bracket whose current branch will already have been scanned.
1727 nigel 77
1728     Arguments:
1729     code points to start of search
1730     endcode points to where to stop
1731     utf8 TRUE if in UTF8 mode
1732    
1733     Returns: TRUE if what is matched could be empty
1734     */
1735    
1736     static BOOL
1737     could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1738     {
1739     register int c;
1740 nigel 93 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1741 nigel 77 code < endcode;
1742     code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1743     {
1744     const uschar *ccode;
1745    
1746     c = *code;
1747 ph10 286
1748     /* Skip over forward assertions; the other assertions are skipped by
1749 ph10 282 first_significant_code() with a TRUE final argument. */
1750 ph10 286
1751 ph10 282 if (c == OP_ASSERT)
1752 ph10 286 {
1753 ph10 282 do code += GET(code, 1); while (*code == OP_ALT);
1754     c = *code;
1755     continue;
1756 ph10 286 }
1757 ph10 172
1758 ph10 170 /* Groups with zero repeats can of course be empty; skip them. */
1759 nigel 77
1760 ph10 335 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1761 ph10 170 {
1762 ph10 172 code += _pcre_OP_lengths[c];
1763 ph10 170 do code += GET(code, 1); while (*code == OP_ALT);
1764     c = *code;
1765     continue;
1766     }
1767    
1768     /* For other groups, scan the branches. */
1769 ph10 172
1770 ph10 206 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1771 nigel 77 {
1772     BOOL empty_branch;
1773     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1774 ph10 406
1775     /* If a conditional group has only one branch, there is a second, implied,
1776 ph10 395 empty branch, so just skip over the conditional, because it could be empty.
1777     Otherwise, scan the individual branches of the group. */
1778 ph10 406
1779 ph10 395 if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1780 nigel 77 code += GET(code, 1);
1781 ph10 395 else
1782 ph10 406 {
1783 ph10 395 empty_branch = FALSE;
1784     do
1785     {
1786     if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1787     empty_branch = TRUE;
1788     code += GET(code, 1);
1789     }
1790     while (*code == OP_ALT);
1791     if (!empty_branch) return FALSE; /* All branches are non-empty */
1792 nigel 77 }
1793 ph10 406
1794 ph10 172 c = *code;
1795 nigel 93 continue;
1796 nigel 77 }
1797    
1798 nigel 93 /* Handle the other opcodes */
1799    
1800     switch (c)
1801 nigel 77 {
1802 ph10 216 /* Check for quantifiers after a class. XCLASS is used for classes that
1803     cannot be represented just by a bit map. This includes negated single
1804     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1805 ph10 220 actual length is stored in the compiled code, so we must update "code"
1806 ph10 216 here. */
1807 nigel 77
1808     #ifdef SUPPORT_UTF8
1809     case OP_XCLASS:
1810 ph10 216 ccode = code += GET(code, 1);
1811 nigel 77 goto CHECK_CLASS_REPEAT;
1812     #endif
1813    
1814     case OP_CLASS:
1815     case OP_NCLASS:
1816     ccode = code + 33;
1817    
1818     #ifdef SUPPORT_UTF8
1819     CHECK_CLASS_REPEAT:
1820     #endif
1821    
1822     switch (*ccode)
1823     {
1824     case OP_CRSTAR: /* These could be empty; continue */
1825     case OP_CRMINSTAR:
1826     case OP_CRQUERY:
1827     case OP_CRMINQUERY:
1828     break;
1829    
1830     default: /* Non-repeat => class must match */
1831     case OP_CRPLUS: /* These repeats aren't empty */
1832     case OP_CRMINPLUS:
1833     return FALSE;
1834    
1835     case OP_CRRANGE:
1836     case OP_CRMINRANGE:
1837     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1838     break;
1839     }
1840     break;
1841    
1842     /* Opcodes that must match a character */
1843    
1844     case OP_PROP:
1845     case OP_NOTPROP:
1846     case OP_EXTUNI:
1847     case OP_NOT_DIGIT:
1848     case OP_DIGIT:
1849     case OP_NOT_WHITESPACE:
1850     case OP_WHITESPACE:
1851     case OP_NOT_WORDCHAR:
1852     case OP_WORDCHAR:
1853     case OP_ANY:
1854 ph10 345 case OP_ALLANY:
1855 nigel 77 case OP_ANYBYTE:
1856     case OP_CHAR:
1857     case OP_CHARNC:
1858     case OP_NOT:
1859     case OP_PLUS:
1860     case OP_MINPLUS:
1861 nigel 93 case OP_POSPLUS:
1862 nigel 77 case OP_EXACT:
1863     case OP_NOTPLUS:
1864     case OP_NOTMINPLUS:
1865 nigel 93 case OP_NOTPOSPLUS:
1866 nigel 77 case OP_NOTEXACT:
1867     case OP_TYPEPLUS:
1868     case OP_TYPEMINPLUS:
1869 nigel 93 case OP_TYPEPOSPLUS:
1870 nigel 77 case OP_TYPEEXACT:
1871     return FALSE;
1872 ph10 227
1873     /* These are going to continue, as they may be empty, but we have to
1874     fudge the length for the \p and \P cases. */
1875    
1876 ph10 224 case OP_TYPESTAR:
1877     case OP_TYPEMINSTAR:
1878     case OP_TYPEPOSSTAR:
1879     case OP_TYPEQUERY:
1880     case OP_TYPEMINQUERY:
1881     case OP_TYPEPOSQUERY:
1882     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1883 ph10 227 break;
1884    
1885 ph10 224 /* Same for these */
1886 ph10 227
1887 ph10 224 case OP_TYPEUPTO:
1888     case OP_TYPEMINUPTO:
1889     case OP_TYPEPOSUPTO:
1890     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1891     break;
1892 nigel 77
1893     /* End of branch */
1894    
1895     case OP_KET:
1896     case OP_KETRMAX:
1897     case OP_KETRMIN:
1898     case OP_ALT:
1899     return TRUE;
1900    
1901 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1902     MINUPTO, and POSUPTO may be followed by a multibyte character */
1903 nigel 77
1904     #ifdef SUPPORT_UTF8
1905     case OP_STAR:
1906     case OP_MINSTAR:
1907 nigel 93 case OP_POSSTAR:
1908 nigel 77 case OP_QUERY:
1909     case OP_MINQUERY:
1910 nigel 93 case OP_POSQUERY:
1911 ph10 426 if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
1912     break;
1913    
1914 nigel 77 case OP_UPTO:
1915     case OP_MINUPTO:
1916 nigel 93 case OP_POSUPTO:
1917 ph10 426 if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1918 nigel 77 break;
1919     #endif
1920     }
1921     }
1922    
1923     return TRUE;
1924     }
1925    
1926    
1927    
1928     /*************************************************
1929     * Scan compiled regex for non-emptiness *
1930     *************************************************/
1931    
1932     /* This function is called to check for left recursive calls. We want to check
1933     the current branch of the current pattern to see if it could match the empty
1934     string. If it could, we must look outwards for branches at other levels,
1935     stopping when we pass beyond the bracket which is the subject of the recursion.
1936    
1937     Arguments:
1938     code points to start of the recursion
1939     endcode points to where to stop (current RECURSE item)
1940     bcptr points to the chain of current (unclosed) branch starts
1941     utf8 TRUE if in UTF-8 mode
1942    
1943     Returns: TRUE if what is matched could be empty
1944     */
1945    
1946     static BOOL
1947     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1948     BOOL utf8)
1949     {
1950     while (bcptr != NULL && bcptr->current >= code)
1951     {
1952     if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1953     bcptr = bcptr->outer;
1954     }
1955     return TRUE;
1956     }
1957    
1958    
1959    
1960     /*************************************************
1961     * Check for POSIX class syntax *
1962     *************************************************/
1963    
1964     /* This function is called when the sequence "[:" or "[." or "[=" is
1965 ph10 295 encountered in a character class. It checks whether this is followed by a
1966 ph10 298 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1967 ph10 295 reach an unescaped ']' without the special preceding character, return FALSE.
1968 nigel 77
1969 ph10 298 Originally, this function only recognized a sequence of letters between the
1970     terminators, but it seems that Perl recognizes any sequence of characters,
1971     though of course unknown POSIX names are subsequently rejected. Perl gives an
1972     "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1973     didn't consider this to be a POSIX class. Likewise for [:1234:].
1974 ph10 295
1975 ph10 298 The problem in trying to be exactly like Perl is in the handling of escapes. We
1976     have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1977     class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1978     below handles the special case of \], but does not try to do any other escape
1979     processing. This makes it different from Perl for cases such as [:l\ower:]
1980 ph10 295 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1981 ph10 298 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1982 ph10 295 I think.
1983    
1984     Arguments:
1985 nigel 77 ptr pointer to the initial [
1986     endptr where to return the end pointer
1987    
1988     Returns: TRUE or FALSE
1989     */
1990    
1991     static BOOL
1992 ph10 295 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1993 nigel 77 {
1994     int terminator; /* Don't combine these lines; the Solaris cc */
1995     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1996 ph10 295 for (++ptr; *ptr != 0; ptr++)
1997 nigel 77 {
1998 ph10 391 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
1999 ph10 298 {
2000 ph10 391 if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2001     if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
2002 ph10 295 {
2003     *endptr = ptr;
2004     return TRUE;
2005 ph10 298 }
2006     }
2007     }
2008 nigel 77 return FALSE;
2009     }
2010    
2011    
2012    
2013    
2014     /*************************************************
2015     * Check POSIX class name *
2016     *************************************************/
2017    
2018     /* This function is called to check the name given in a POSIX-style class entry
2019     such as [:alnum:].
2020    
2021     Arguments:
2022     ptr points to the first letter
2023     len the length of the name
2024    
2025     Returns: a value representing the name, or -1 if unknown
2026     */
2027    
2028     static int
2029     check_posix_name(const uschar *ptr, int len)
2030     {
2031 ph10 240 const char *pn = posix_names;
2032 nigel 77 register int yield = 0;
2033     while (posix_name_lengths[yield] != 0)
2034     {
2035     if (len == posix_name_lengths[yield] &&
2036 ph10 240 strncmp((const char *)ptr, pn, len) == 0) return yield;
2037 ph10 243 pn += posix_name_lengths[yield] + 1;
2038 nigel 77 yield++;
2039     }
2040     return -1;
2041     }
2042    
2043    
2044     /*************************************************
2045     * Adjust OP_RECURSE items in repeated group *
2046     *************************************************/
2047    
2048     /* OP_RECURSE items contain an offset from the start of the regex to the group
2049     that is referenced. This means that groups can be replicated for fixed
2050     repetition simply by copying (because the recursion is allowed to refer to
2051     earlier groups that are outside the current group). However, when a group is
2052 ph10 335 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2053     inserted before it, after it has been compiled. This means that any OP_RECURSE
2054     items within it that refer to the group itself or any contained groups have to
2055     have their offsets adjusted. That one of the jobs of this function. Before it
2056     is called, the partially compiled regex must be temporarily terminated with
2057     OP_END.
2058 nigel 77
2059 nigel 93 This function has been extended with the possibility of forward references for
2060     recursions and subroutine calls. It must also check the list of such references
2061     for the group we are dealing with. If it finds that one of the recursions in
2062     the current group is on this list, it adjusts the offset in the list, not the
2063     value in the reference (which is a group number).
2064    
2065 nigel 77 Arguments:
2066     group points to the start of the group
2067     adjust the amount by which the group is to be moved
2068     utf8 TRUE in UTF-8 mode
2069     cd contains pointers to tables etc.
2070 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
2071 nigel 77
2072     Returns: nothing
2073     */
2074    
2075     static void
2076 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2077     uschar *save_hwm)
2078 nigel 77 {
2079     uschar *ptr = group;
2080 ph10 224
2081 nigel 77 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2082     {
2083 nigel 93 int offset;
2084     uschar *hc;
2085    
2086     /* See if this recursion is on the forward reference list. If so, adjust the
2087     reference. */
2088 ph10 345
2089 nigel 93 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2090     {
2091     offset = GET(hc, 0);
2092     if (cd->start_code + offset == ptr + 1)
2093     {
2094     PUT(hc, 0, offset + adjust);
2095     break;
2096     }
2097     }
2098    
2099     /* Otherwise, adjust the recursion offset if it's after the start of this
2100     group. */
2101    
2102     if (hc >= cd->hwm)
2103     {
2104     offset = GET(ptr, 1);
2105     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2106     }
2107    
2108 nigel 77 ptr += 1 + LINK_SIZE;
2109     }
2110     }
2111    
2112    
2113    
2114     /*************************************************
2115     * Insert an automatic callout point *
2116     *************************************************/
2117    
2118     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2119     callout points before each pattern item.
2120    
2121     Arguments:
2122     code current code pointer
2123     ptr current pattern pointer
2124     cd pointers to tables etc
2125    
2126     Returns: new code pointer
2127     */
2128    
2129     static uschar *
2130     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2131     {
2132     *code++ = OP_CALLOUT;
2133     *code++ = 255;
2134     PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
2135     PUT(code, LINK_SIZE, 0); /* Default length */
2136     return code + 2*LINK_SIZE;
2137     }
2138    
2139    
2140    
2141     /*************************************************
2142     * Complete a callout item *
2143     *************************************************/
2144    
2145     /* A callout item contains the length of the next item in the pattern, which
2146     we can't fill in till after we have reached the relevant point. This is used
2147     for both automatic and manual callouts.
2148    
2149     Arguments:
2150     previous_callout points to previous callout item
2151     ptr current pattern pointer
2152     cd pointers to tables etc
2153    
2154     Returns: nothing
2155     */
2156    
2157     static void
2158     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2159     {
2160     int length = ptr - cd->start_pattern - GET(previous_callout, 2);
2161     PUT(previous_callout, 2 + LINK_SIZE, length);
2162     }
2163    
2164    
2165    
2166     #ifdef SUPPORT_UCP
2167     /*************************************************
2168     * Get othercase range *
2169     *************************************************/
2170    
2171     /* This function is passed the start and end of a class range, in UTF-8 mode
2172     with UCP support. It searches up the characters, looking for internal ranges of
2173     characters in the "other" case. Each call returns the next one, updating the
2174     start address.
2175    
2176     Arguments:
2177     cptr points to starting character value; updated
2178     d end value
2179     ocptr where to put start of othercase range
2180     odptr where to put end of othercase range
2181    
2182     Yield: TRUE when range returned; FALSE when no more
2183     */
2184    
2185     static BOOL
2186 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2187     unsigned int *odptr)
2188 nigel 77 {
2189 nigel 93 unsigned int c, othercase, next;
2190 nigel 77
2191     for (c = *cptr; c <= d; c++)
2192 ph10 349 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2193 nigel 77
2194     if (c > d) return FALSE;
2195    
2196     *ocptr = othercase;
2197     next = othercase + 1;
2198    
2199     for (++c; c <= d; c++)
2200     {
2201 ph10 349 if (UCD_OTHERCASE(c) != next) break;
2202 nigel 77 next++;
2203     }
2204    
2205     *odptr = next - 1;
2206     *cptr = c;
2207    
2208     return TRUE;
2209     }
2210     #endif /* SUPPORT_UCP */
2211    
2212    
2213 nigel 93
2214 nigel 77 /*************************************************
2215 nigel 93 * Check if auto-possessifying is possible *
2216     *************************************************/
2217    
2218     /* This function is called for unlimited repeats of certain items, to see
2219     whether the next thing could possibly match the repeated item. If not, it makes
2220     sense to automatically possessify the repeated item.
2221    
2222     Arguments:
2223     op_code the repeated op code
2224     this data for this item, depends on the opcode
2225     utf8 TRUE in UTF-8 mode
2226     utf8_char used for utf8 character bytes, NULL if not relevant
2227     ptr next character in pattern
2228     options options bits
2229     cd contains pointers to tables etc.
2230    
2231     Returns: TRUE if possessifying is wanted
2232     */
2233    
2234     static BOOL
2235     check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2236     const uschar *ptr, int options, compile_data *cd)
2237     {
2238     int next;
2239    
2240     /* Skip whitespace and comments in extended mode */
2241    
2242     if ((options & PCRE_EXTENDED) != 0)
2243     {
2244     for (;;)
2245     {
2246     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2247 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2248 nigel 93 {
2249     while (*(++ptr) != 0)
2250     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2251     }
2252     else break;
2253     }
2254     }
2255    
2256     /* If the next item is one that we can handle, get its value. A non-negative
2257     value is a character, a negative value is an escape value. */
2258    
2259 ph10 391 if (*ptr == CHAR_BACKSLASH)
2260 nigel 93 {
2261     int temperrorcode = 0;
2262     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2263     if (temperrorcode != 0) return FALSE;
2264     ptr++; /* Point after the escape sequence */
2265     }
2266    
2267     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2268     {
2269     #ifdef SUPPORT_UTF8
2270     if (utf8) { GETCHARINC(next, ptr); } else
2271     #endif
2272     next = *ptr++;
2273     }
2274    
2275     else return FALSE;
2276    
2277     /* Skip whitespace and comments in extended mode */
2278    
2279     if ((options & PCRE_EXTENDED) != 0)
2280     {
2281     for (;;)
2282     {
2283     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2284 ph10 391 if (*ptr == CHAR_NUMBER_SIGN)
2285 nigel 93 {
2286     while (*(++ptr) != 0)
2287     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2288     }
2289     else break;
2290     }
2291     }
2292    
2293     /* If the next thing is itself optional, we have to give up. */
2294    
2295 ph10 392 if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2296 ph10 391 strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2297     return FALSE;
2298 nigel 93
2299     /* Now compare the next item with the previous opcode. If the previous is a
2300     positive single character match, "item" either contains the character or, if
2301     "item" is greater than 127 in utf8 mode, the character's bytes are in
2302     utf8_char. */
2303    
2304    
2305     /* Handle cases when the next item is a character. */
2306    
2307     if (next >= 0) switch(op_code)
2308     {
2309     case OP_CHAR:
2310     #ifdef SUPPORT_UTF8
2311     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2312 ph10 369 #else
2313     (void)(utf8_char); /* Keep compiler happy by referencing function argument */
2314 nigel 93 #endif
2315     return item != next;
2316    
2317     /* For CHARNC (caseless character) we must check the other case. If we have
2318     Unicode property support, we can use it to test the other case of
2319     high-valued characters. */
2320    
2321     case OP_CHARNC:
2322     #ifdef SUPPORT_UTF8
2323     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2324     #endif
2325     if (item == next) return FALSE;
2326     #ifdef SUPPORT_UTF8
2327     if (utf8)
2328     {
2329     unsigned int othercase;
2330     if (next < 128) othercase = cd->fcc[next]; else
2331     #ifdef SUPPORT_UCP
2332 ph10 349 othercase = UCD_OTHERCASE((unsigned int)next);
2333 nigel 93 #else
2334     othercase = NOTACHAR;
2335     #endif
2336     return (unsigned int)item != othercase;
2337     }
2338     else
2339     #endif /* SUPPORT_UTF8 */
2340     return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2341    
2342     /* For OP_NOT, "item" must be a single-byte character. */
2343    
2344     case OP_NOT:
2345     if (item == next) return TRUE;
2346     if ((options & PCRE_CASELESS) == 0) return FALSE;
2347     #ifdef SUPPORT_UTF8
2348     if (utf8)
2349     {
2350     unsigned int othercase;
2351     if (next < 128) othercase = cd->fcc[next]; else
2352     #ifdef SUPPORT_UCP
2353 ph10 349 othercase = UCD_OTHERCASE(next);
2354 nigel 93 #else
2355     othercase = NOTACHAR;
2356     #endif
2357     return (unsigned int)item == othercase;
2358     }
2359     else
2360     #endif /* SUPPORT_UTF8 */
2361     return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2362    
2363     case OP_DIGIT:
2364     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2365    
2366     case OP_NOT_DIGIT:
2367     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2368    
2369     case OP_WHITESPACE:
2370     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2371    
2372     case OP_NOT_WHITESPACE:
2373     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2374    
2375     case OP_WORDCHAR:
2376     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2377    
2378     case OP_NOT_WORDCHAR:
2379     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2380    
2381 ph10 180 case OP_HSPACE:
2382     case OP_NOT_HSPACE:
2383     switch(next)
2384     {
2385     case 0x09:
2386     case 0x20:
2387     case 0xa0:
2388     case 0x1680:
2389     case 0x180e:
2390     case 0x2000:
2391     case 0x2001:
2392     case 0x2002:
2393     case 0x2003:
2394     case 0x2004:
2395     case 0x2005:
2396     case 0x2006:
2397     case 0x2007:
2398     case 0x2008:
2399     case 0x2009:
2400     case 0x200A:
2401     case 0x202f:
2402     case 0x205f:
2403     case 0x3000:
2404     return op_code != OP_HSPACE;
2405     default:
2406     return op_code == OP_HSPACE;
2407     }
2408    
2409     case OP_VSPACE:
2410     case OP_NOT_VSPACE:
2411     switch(next)
2412     {
2413     case 0x0a:
2414     case 0x0b:
2415     case 0x0c:
2416     case 0x0d:
2417     case 0x85:
2418     case 0x2028:
2419     case 0x2029:
2420     return op_code != OP_VSPACE;
2421     default:
2422     return op_code == OP_VSPACE;
2423     }
2424    
2425 nigel 93 default:
2426     return FALSE;
2427     }
2428    
2429    
2430     /* Handle the case when the next item is \d, \s, etc. */
2431    
2432     switch(op_code)
2433     {
2434     case OP_CHAR:
2435     case OP_CHARNC:
2436     #ifdef SUPPORT_UTF8
2437     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2438     #endif
2439     switch(-next)
2440     {
2441     case ESC_d:
2442     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2443    
2444     case ESC_D:
2445     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2446    
2447     case ESC_s:
2448     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2449    
2450     case ESC_S:
2451     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2452    
2453     case ESC_w:
2454     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2455    
2456     case ESC_W:
2457     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2458 ph10 182
2459 ph10 180 case ESC_h:
2460     case ESC_H:
2461     switch(item)
2462     {
2463     case 0x09:
2464     case 0x20:
2465     case 0xa0:
2466     case 0x1680:
2467     case 0x180e:
2468     case 0x2000:
2469     case 0x2001:
2470     case 0x2002:
2471     case 0x2003:
2472     case 0x2004:
2473     case 0x2005:
2474     case 0x2006:
2475     case 0x2007:
2476     case 0x2008:
2477     case 0x2009:
2478     case 0x200A:
2479     case 0x202f:
2480     case 0x205f:
2481     case 0x3000:
2482     return -next != ESC_h;
2483     default:
2484     return -next == ESC_h;
2485 ph10 182 }
2486    
2487 ph10 180 case ESC_v:
2488     case ESC_V:
2489     switch(item)
2490     {
2491     case 0x0a:
2492     case 0x0b:
2493     case 0x0c:
2494     case 0x0d:
2495     case 0x85:
2496     case 0x2028:
2497     case 0x2029:
2498     return -next != ESC_v;
2499     default:
2500     return -next == ESC_v;
2501 ph10 182 }
2502 nigel 93
2503     default:
2504     return FALSE;
2505     }
2506    
2507     case OP_DIGIT:
2508 ph10 180 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2509     next == -ESC_h || next == -ESC_v;
2510 nigel 93
2511     case OP_NOT_DIGIT:
2512     return next == -ESC_d;
2513    
2514     case OP_WHITESPACE:
2515     return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2516    
2517     case OP_NOT_WHITESPACE:
2518 ph10 180 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2519 nigel 93
2520 ph10 180 case OP_HSPACE:
2521     return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2522    
2523     case OP_NOT_HSPACE:
2524     return next == -ESC_h;
2525 ph10 182
2526 ph10 180 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2527 ph10 182 case OP_VSPACE:
2528 ph10 180 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2529    
2530     case OP_NOT_VSPACE:
2531 ph10 182 return next == -ESC_v;
2532 ph10 180
2533 nigel 93 case OP_WORDCHAR:
2534 ph10 180 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2535 nigel 93
2536     case OP_NOT_WORDCHAR:
2537     return next == -ESC_w || next == -ESC_d;
2538 ph10 182
2539 nigel 93 default:
2540     return FALSE;
2541     }
2542    
2543     /* Control does not reach here */
2544     }
2545    
2546    
2547    
2548     /*************************************************
2549 nigel 77 * Compile one branch *
2550     *************************************************/
2551    
2552 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
2553 nigel 77 changed during the branch, the pointer is used to change the external options
2554 nigel 93 bits. This function is used during the pre-compile phase when we are trying
2555     to find out the amount of memory needed, as well as during the real compile
2556     phase. The value of lengthptr distinguishes the two phases.
2557 nigel 77
2558     Arguments:
2559     optionsptr pointer to the option bits
2560     codeptr points to the pointer to the current code point
2561     ptrptr points to the current pattern pointer
2562     errorcodeptr points to error code variable
2563     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2564     reqbyteptr set to the last literal character required, else < 0
2565     bcptr points to current branch chain
2566     cd contains pointers to tables etc.
2567 nigel 93 lengthptr NULL during the real compile phase
2568     points to length accumulator during pre-compile phase
2569 nigel 77
2570     Returns: TRUE on success
2571     FALSE, with *errorcodeptr set non-zero on error
2572     */
2573    
2574     static BOOL
2575 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2576     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2577     compile_data *cd, int *lengthptr)
2578 nigel 77 {
2579     int repeat_type, op_type;
2580     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2581     int bravalue = 0;
2582     int greedy_default, greedy_non_default;
2583     int firstbyte, reqbyte;
2584     int zeroreqbyte, zerofirstbyte;
2585     int req_caseopt, reqvary, tempreqvary;
2586     int options = *optionsptr;
2587     int after_manual_callout = 0;
2588 nigel 93 int length_prevgroup = 0;
2589 nigel 77 register int c;
2590     register uschar *code = *codeptr;
2591 nigel 93 uschar *last_code = code;
2592     uschar *orig_code = code;
2593 nigel 77 uschar *tempcode;
2594     BOOL inescq = FALSE;
2595     BOOL groupsetfirstbyte = FALSE;
2596     const uschar *ptr = *ptrptr;
2597     const uschar *tempptr;
2598     uschar *previous = NULL;
2599     uschar *previous_callout = NULL;
2600 nigel 93 uschar *save_hwm = NULL;
2601 nigel 77 uschar classbits[32];
2602    
2603     #ifdef SUPPORT_UTF8
2604     BOOL class_utf8;
2605     BOOL utf8 = (options & PCRE_UTF8) != 0;
2606     uschar *class_utf8data;
2607 ph10 300 uschar *class_utf8data_base;
2608 nigel 77 uschar utf8_char[6];
2609     #else
2610     BOOL utf8 = FALSE;
2611 nigel 93 uschar *utf8_char = NULL;
2612 nigel 77 #endif
2613    
2614 nigel 93 #ifdef DEBUG
2615     if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2616     #endif
2617    
2618 nigel 77 /* Set up the default and non-default settings for greediness */
2619    
2620     greedy_default = ((options & PCRE_UNGREEDY) != 0);
2621     greedy_non_default = greedy_default ^ 1;
2622    
2623     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2624     matching encountered yet". It gets changed to REQ_NONE if we hit something that
2625     matches a non-fixed char first char; reqbyte just remains unset if we never
2626     find one.
2627    
2628     When we hit a repeat whose minimum is zero, we may have to adjust these values
2629     to take the zero repeat into account. This is implemented by setting them to
2630     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2631     item types that can be repeated set these backoff variables appropriately. */
2632    
2633     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2634    
2635     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2636     according to the current setting of the caseless flag. REQ_CASELESS is a bit
2637     value > 255. It is added into the firstbyte or reqbyte variables to record the
2638     case status of the value. This is used only for ASCII characters. */
2639    
2640     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2641    
2642     /* Switch on next character until the end of the branch */
2643    
2644     for (;; ptr++)
2645     {
2646     BOOL negate_class;
2647 ph10 286 BOOL should_flip_negation;
2648 nigel 77 BOOL possessive_quantifier;
2649     BOOL is_quantifier;
2650 nigel 93 BOOL is_recurse;
2651 ph10 180 BOOL reset_bracount;
2652 nigel 77 int class_charcount;
2653     int class_lastchar;
2654     int newoptions;
2655     int recno;
2656 ph10 172 int refsign;
2657 nigel 77 int skipbytes;
2658     int subreqbyte;
2659     int subfirstbyte;
2660 nigel 93 int terminator;
2661 nigel 77 int mclength;
2662     uschar mcbuffer[8];
2663    
2664 nigel 93 /* Get next byte in the pattern */
2665 nigel 77
2666     c = *ptr;
2667 ph10 345
2668 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
2669     previous cycle of this loop. */
2670    
2671     if (lengthptr != NULL)
2672     {
2673     #ifdef DEBUG
2674     if (code > cd->hwm) cd->hwm = code; /* High water info */
2675     #endif
2676     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2677     {
2678     *errorcodeptr = ERR52;
2679     goto FAILED;
2680     }
2681    
2682     /* There is at least one situation where code goes backwards: this is the
2683     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2684     the class is simply eliminated. However, it is created first, so we have to
2685     allow memory for it. Therefore, don't ever reduce the length at this point.
2686     */
2687    
2688     if (code < last_code) code = last_code;
2689 ph10 202
2690     /* Paranoid check for integer overflow */
2691    
2692     if (OFLOW_MAX - *lengthptr < code - last_code)
2693     {
2694     *errorcodeptr = ERR20;
2695     goto FAILED;
2696     }
2697    
2698 nigel 93 *lengthptr += code - last_code;
2699     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2700    
2701     /* If "previous" is set and it is not at the start of the work space, move
2702     it back to there, in order to avoid filling up the work space. Otherwise,
2703     if "previous" is NULL, reset the current code pointer to the start. */
2704    
2705     if (previous != NULL)
2706     {
2707     if (previous > orig_code)
2708     {
2709     memmove(orig_code, previous, code - previous);
2710     code -= previous - orig_code;
2711     previous = orig_code;
2712     }
2713     }
2714     else code = orig_code;
2715    
2716     /* Remember where this code item starts so we can pick up the length
2717     next time round. */
2718    
2719     last_code = code;
2720     }
2721    
2722     /* In the real compile phase, just check the workspace used by the forward
2723     reference list. */
2724    
2725     else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2726     {
2727     *errorcodeptr = ERR52;
2728     goto FAILED;
2729     }
2730    
2731 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
2732    
2733     if (inescq && c != 0)
2734     {
2735 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
2736 nigel 77 {
2737     inescq = FALSE;
2738     ptr++;
2739     continue;
2740     }
2741     else
2742     {
2743     if (previous_callout != NULL)
2744     {
2745 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2746     complete_callout(previous_callout, ptr, cd);
2747 nigel 77 previous_callout = NULL;
2748     }
2749     if ((options & PCRE_AUTO_CALLOUT) != 0)
2750     {
2751     previous_callout = code;
2752     code = auto_callout(code, ptr, cd);
2753     }
2754     goto NORMAL_CHAR;
2755     }
2756     }
2757    
2758     /* Fill in length of a previous callout, except when the next thing is
2759     a quantifier. */
2760    
2761 ph10 392 is_quantifier =
2762 ph10 391 c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
2763     (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
2764 nigel 77
2765     if (!is_quantifier && previous_callout != NULL &&
2766     after_manual_callout-- <= 0)
2767     {
2768 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2769     complete_callout(previous_callout, ptr, cd);
2770 nigel 77 previous_callout = NULL;
2771     }
2772    
2773     /* In extended mode, skip white space and comments */
2774    
2775     if ((options & PCRE_EXTENDED) != 0)
2776     {
2777     if ((cd->ctypes[c] & ctype_space) != 0) continue;
2778 ph10 391 if (c == CHAR_NUMBER_SIGN)
2779 nigel 77 {
2780 nigel 93 while (*(++ptr) != 0)
2781 nigel 91 {
2782 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2783 nigel 91 }
2784 nigel 93 if (*ptr != 0) continue;
2785    
2786 nigel 91 /* Else fall through to handle end of string */
2787     c = 0;
2788 nigel 77 }
2789     }
2790    
2791     /* No auto callout for quantifiers. */
2792    
2793     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2794     {
2795     previous_callout = code;
2796     code = auto_callout(code, ptr, cd);
2797     }
2798    
2799     switch(c)
2800     {
2801 nigel 93 /* ===================================================================*/
2802     case 0: /* The branch terminates at string end */
2803 ph10 391 case CHAR_VERTICAL_LINE: /* or | or ) */
2804     case CHAR_RIGHT_PARENTHESIS:
2805 nigel 77 *firstbyteptr = firstbyte;
2806     *reqbyteptr = reqbyte;
2807     *codeptr = code;
2808     *ptrptr = ptr;
2809 nigel 93 if (lengthptr != NULL)
2810     {
2811 ph10 202 if (OFLOW_MAX - *lengthptr < code - last_code)
2812     {
2813     *errorcodeptr = ERR20;
2814     goto FAILED;
2815     }
2816 nigel 93 *lengthptr += code - last_code; /* To include callout length */
2817     DPRINTF((">> end branch\n"));
2818     }
2819 nigel 77 return TRUE;
2820    
2821 nigel 93
2822     /* ===================================================================*/
2823 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
2824     the setting of any following char as a first character. */
2825    
2826 ph10 391 case CHAR_CIRCUMFLEX_ACCENT:
2827 nigel 77 if ((options & PCRE_MULTILINE) != 0)
2828     {
2829     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2830     }
2831     previous = NULL;
2832     *code++ = OP_CIRC;
2833     break;
2834    
2835 ph10 391 case CHAR_DOLLAR_SIGN:
2836 nigel 77 previous = NULL;
2837     *code++ = OP_DOLL;
2838     break;
2839    
2840     /* There can never be a first char if '.' is first, whatever happens about
2841     repeats. The value of reqbyte doesn't change either. */
2842    
2843 ph10 391 case CHAR_DOT:
2844 nigel 77 if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2845     zerofirstbyte = firstbyte;
2846     zeroreqbyte = reqbyte;
2847     previous = code;
2848 ph10 342 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2849 nigel 77 break;
2850    
2851 nigel 93
2852     /* ===================================================================*/
2853 nigel 87 /* Character classes. If the included characters are all < 256, we build a
2854     32-byte bitmap of the permitted characters, except in the special case
2855     where there is only one such character. For negated classes, we build the
2856     map as usual, then invert it at the end. However, we use a different opcode
2857     so that data characters > 255 can be handled correctly.
2858 nigel 77
2859     If the class contains characters outside the 0-255 range, a different
2860     opcode is compiled. It may optionally have a bit map for characters < 256,
2861     but those above are are explicitly listed afterwards. A flag byte tells
2862     whether the bitmap is present, and whether this is a negated class or not.
2863 ph10 345
2864 ph10 336 In JavaScript compatibility mode, an isolated ']' causes an error. In
2865     default (Perl) mode, it is treated as a data character. */
2866 ph10 345
2867 ph10 391 case CHAR_RIGHT_SQUARE_BRACKET:
2868 ph10 336 if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2869     {
2870     *errorcodeptr = ERR64;
2871 ph10 345 goto FAILED;
2872 ph10 336 }
2873 ph10 345 goto NORMAL_CHAR;
2874 nigel 77
2875 ph10 391 case CHAR_LEFT_SQUARE_BRACKET:
2876 nigel 77 previous = code;
2877    
2878     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2879     they are encountered at the top level, so we'll do that too. */
2880    
2881 ph10 392 if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
2882 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) &&
2883 ph10 295 check_posix_syntax(ptr, &tempptr))
2884 nigel 77 {
2885 ph10 391 *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
2886 nigel 77 goto FAILED;
2887     }
2888    
2889 ph10 205 /* If the first character is '^', set the negation flag and skip it. Also,
2890 ph10 208 if the first few characters (either before or after ^) are \Q\E or \E we
2891 ph10 205 skip them too. This makes for compatibility with Perl. */
2892 ph10 208
2893 ph10 205 negate_class = FALSE;
2894     for (;;)
2895 nigel 77 {
2896     c = *(++ptr);
2897 ph10 391 if (c == CHAR_BACKSLASH)
2898 ph10 205 {
2899 ph10 392 if (ptr[1] == CHAR_E)
2900 ph10 391 ptr++;
2901 ph10 392 else if (strncmp((const char *)ptr+1,
2902     STR_Q STR_BACKSLASH STR_E, 3) == 0)
2903 ph10 391 ptr += 3;
2904 ph10 392 else
2905 ph10 391 break;
2906 ph10 205 }
2907 ph10 391 else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
2908 ph10 205 negate_class = TRUE;
2909     else break;
2910 ph10 208 }
2911 ph10 345
2912     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2913     an initial ']' is taken as a data character -- the code below handles
2914 ph10 341 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2915     [^] must match any character, so generate OP_ALLANY. */
2916 ph10 345
2917 ph10 392 if (c == CHAR_RIGHT_SQUARE_BRACKET &&
2918 ph10 391 (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2919 ph10 341 {
2920     *code++ = negate_class? OP_ALLANY : OP_FAIL;
2921     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2922     zerofirstbyte = firstbyte;
2923     break;
2924 ph10 345 }
2925 nigel 77
2926 ph10 286 /* If a class contains a negative special such as \S, we need to flip the
2927     negation flag at the end, so that support for characters > 255 works
2928 ph10 264 correctly (they are all included in the class). */
2929    
2930     should_flip_negation = FALSE;
2931    
2932 nigel 77 /* Keep a count of chars with values < 256 so that we can optimize the case
2933 nigel 93 of just a single character (as long as it's < 256). However, For higher
2934     valued UTF-8 characters, we don't yet do any optimization. */
2935 nigel 77
2936     class_charcount = 0;
2937     class_lastchar = -1;
2938    
2939 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
2940     temporary bit of memory, in case the class contains only 1 character (less
2941     than 256), because in that case the compiled code doesn't use the bit map.
2942     */
2943    
2944     memset(classbits, 0, 32 * sizeof(uschar));
2945    
2946 nigel 77 #ifdef SUPPORT_UTF8
2947     class_utf8 = FALSE; /* No chars >= 256 */
2948 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2949 ph10 309 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2950 nigel 77 #endif
2951    
2952     /* Process characters until ] is reached. By writing this as a "do" it
2953 nigel 93 means that an initial ] is taken as a data character. At the start of the
2954     loop, c contains the first byte of the character. */
2955 nigel 77
2956 nigel 93 if (c != 0) do
2957 nigel 77 {
2958 nigel 93 const uschar *oldptr;
2959    
2960 nigel 77 #ifdef SUPPORT_UTF8
2961     if (utf8 && c > 127)
2962     { /* Braces are required because the */
2963     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2964     }
2965 ph10 309
2966 ph10 300 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2967 ph10 309 data and reset the pointer. This is so that very large classes that
2968 ph10 300 contain a zillion UTF-8 characters no longer overwrite the work space
2969 ph10 309 (which is on the stack). */
2970    
2971 ph10 300 if (lengthptr != NULL)
2972     {
2973     *lengthptr += class_utf8data - class_utf8data_base;
2974 ph10 309 class_utf8data = class_utf8data_base;
2975     }
2976    
2977 nigel 77 #endif
2978    
2979     /* Inside \Q...\E everything is literal except \E */
2980    
2981     if (inescq)
2982     {
2983 ph10 391 if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
2984 nigel 77 {
2985 nigel 93 inescq = FALSE; /* Reset literal state */
2986     ptr++; /* Skip the 'E' */
2987     continue; /* Carry on with next */
2988 nigel 77 }
2989 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
2990 nigel 77 }
2991    
2992     /* Handle POSIX class names. Perl allows a negation extension of the
2993     form [:^name:]. A square bracket that doesn't match the syntax is
2994     treated as a literal. We also recognize the POSIX constructions
2995     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2996     5.6 and 5.8 do. */
2997    
2998 ph10 391 if (c == CHAR_LEFT_SQUARE_BRACKET &&
2999 ph10 392 (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3000 ph10 391 ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
3001 nigel 77 {
3002     BOOL local_negate = FALSE;
3003 nigel 87 int posix_class, taboffset, tabopt;
3004 nigel 77 register const uschar *cbits = cd->cbits;
3005 nigel 87 uschar pbits[32];
3006 nigel 77
3007 ph10 391 if (ptr[1] != CHAR_COLON)
3008 nigel 77 {
3009     *errorcodeptr = ERR31;
3010     goto FAILED;
3011     }
3012    
3013     ptr += 2;
3014 ph10 391 if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
3015 nigel 77 {
3016     local_negate = TRUE;
3017 ph10 286 should_flip_negation = TRUE; /* Note negative special */
3018 nigel 77 ptr++;
3019     }
3020    
3021     posix_class = check_posix_name(ptr, tempptr - ptr);
3022     if (posix_class < 0)
3023     {
3024     *errorcodeptr = ERR30;
3025     goto FAILED;
3026     }
3027    
3028     /* If matching is caseless, upper and lower are converted to
3029     alpha. This relies on the fact that the class table starts with
3030     alpha, lower, upper as the first 3 entries. */
3031    
3032     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3033     posix_class = 0;
3034    
3035 nigel 87 /* We build the bit map for the POSIX class in a chunk of local store
3036     because we may be adding and subtracting from it, and we don't want to
3037     subtract bits that may be in the main map already. At the end we or the
3038     result into the bit map that is being built. */
3039 nigel 77
3040     posix_class *= 3;
3041 nigel 87
3042     /* Copy in the first table (always present) */
3043    
3044     memcpy(pbits, cbits + posix_class_maps[posix_class],
3045     32 * sizeof(uschar));
3046    
3047     /* If there is a second table, add or remove it as required. */
3048    
3049     taboffset = posix_class_maps[posix_class + 1];
3050     tabopt = posix_class_maps[posix_class + 2];
3051    
3052     if (taboffset >= 0)
3053 nigel 77 {
3054 nigel 87 if (tabopt >= 0)
3055     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3056 nigel 77 else
3057 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3058 nigel 77 }
3059    
3060 nigel 87 /* Not see if we need to remove any special characters. An option
3061     value of 1 removes vertical space and 2 removes underscore. */
3062    
3063     if (tabopt < 0) tabopt = -tabopt;
3064     if (tabopt == 1) pbits[1] &= ~0x3c;
3065     else if (tabopt == 2) pbits[11] &= 0x7f;
3066    
3067     /* Add the POSIX table or its complement into the main table that is
3068     being built and we are done. */
3069    
3070     if (local_negate)
3071     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3072     else
3073     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3074    
3075 nigel 77 ptr = tempptr + 1;
3076     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3077     continue; /* End of POSIX syntax handling */
3078     }
3079    
3080     /* Backslash may introduce a single character, or it may introduce one
3081 nigel 93 of the specials, which just set a flag. The sequence \b is a special
3082     case. Inside a class (and only there) it is treated as backspace.
3083     Elsewhere it marks a word boundary. Other escapes have preset maps ready
3084 ph10 205 to 'or' into the one we are building. We assume they have more than one
3085 nigel 77 character in them, so set class_charcount bigger than one. */
3086    
3087 ph10 391 if (c == CHAR_BACKSLASH)
3088 nigel 77 {
3089 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3090     if (*errorcodeptr != 0) goto FAILED;
3091 nigel 77
3092 ph10 391 if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
3093     else if (-c == ESC_X) c = CHAR_X; /* \X is literal X in a class */
3094     else if (-c == ESC_R) c = CHAR_R; /* \R is literal R in a class */
3095 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
3096     {
3097 ph10 391 if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3098 nigel 77 {
3099     ptr += 2; /* avoid empty string */
3100     }
3101     else inescq = TRUE;
3102     continue;
3103     }
3104 ph10 220 else if (-c == ESC_E) continue; /* Ignore orphan \E */
3105 nigel 77
3106     if (c < 0)
3107     {
3108     register const uschar *cbits = cd->cbits;
3109     class_charcount += 2; /* Greater than 1 is what matters */
3110 nigel 93
3111     /* Save time by not doing this in the pre-compile phase. */
3112    
3113     if (lengthptr == NULL) switch (-c)
3114 nigel 77 {
3115     case ESC_d:
3116     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3117     continue;
3118    
3119     case ESC_D:
3120 ph10 286 should_flip_negation = TRUE;
3121 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3122     continue;
3123    
3124     case ESC_w:
3125     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3126     continue;
3127    
3128     case ESC_W:
3129 ph10 286 should_flip_negation = TRUE;
3130 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3131     continue;
3132    
3133     case ESC_s:
3134     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3135     classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3136     continue;
3137    
3138     case ESC_S:
3139 ph10 286 should_flip_negation = TRUE;
3140 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3141     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3142     continue;
3143    
3144 nigel 93 default: /* Not recognized; fall through */
3145     break; /* Need "default" setting to stop compiler warning. */
3146     }
3147    
3148     /* In the pre-compile phase, just do the recognition. */
3149    
3150     else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
3151     c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
3152 ph10 180
3153 ph10 178 /* We need to deal with \H, \h, \V, and \v in both phases because
3154     they use extra memory. */
3155 ph10 180
3156 ph10 178 if (-c == ESC_h)
3157     {
3158     SETBIT(classbits, 0x09); /* VT */
3159     SETBIT(classbits, 0x20); /* SPACE */
3160 ph10 180 SETBIT(classbits, 0xa0); /* NSBP */
3161 ph10 178 #ifdef SUPPORT_UTF8
3162     if (utf8)
3163 ph10 180 {
3164 ph10 178 class_utf8 = TRUE;
3165     *class_utf8data++ = XCL_SINGLE;
3166 ph10 180 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3167 ph10 178 *class_utf8data++ = XCL_SINGLE;
3168 ph10 180 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3169     *class_utf8data++ = XCL_RANGE;
3170     class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3171     class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3172 ph10 178 *class_utf8data++ = XCL_SINGLE;
3173 ph10 180 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3174 ph10 178 *class_utf8data++ = XCL_SINGLE;
3175 ph10 180 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3176 ph10 178 *class_utf8data++ = XCL_SINGLE;
3177 ph10 180 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3178     }
3179     #endif
3180     continue;
3181     }
3182 nigel 93
3183 ph10 178 if (-c == ESC_H)
3184     {
3185     for (c = 0; c < 32; c++)
3186     {
3187     int x = 0xff;
3188     switch (c)
3189 ph10 180 {
3190 ph10 178 case 0x09/8: x ^= 1 << (0x09%8); break;
3191     case 0x20/8: x ^= 1 << (0x20%8); break;
3192     case 0xa0/8: x ^= 1 << (0xa0%8); break;
3193     default: break;
3194     }
3195     classbits[c] |= x;
3196 ph10 180 }
3197    
3198 ph10 178 #ifdef SUPPORT_UTF8
3199     if (utf8)
3200 ph10 180 {
3201 ph10 178 class_utf8 = TRUE;
3202 ph10 180 *class_utf8data++ = XCL_RANGE;
3203     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3204     class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3205     *class_utf8data++ = XCL_RANGE;
3206     class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3207     class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3208     *class_utf8data++ = XCL_RANGE;
3209     class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3210     class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3211     *class_utf8data++ = XCL_RANGE;
3212     class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3213     class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3214     *class_utf8data++ = XCL_RANGE;
3215     class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3216     class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3217     *class_utf8data++ = XCL_RANGE;
3218     class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3219     class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3220     *class_utf8data++ = XCL_RANGE;
3221     class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3222     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3223     }
3224     #endif
3225     continue;
3226     }
3227 ph10 178
3228     if (-c == ESC_v)
3229     {
3230     SETBIT(classbits, 0x0a); /* LF */
3231     SETBIT(classbits, 0x0b); /* VT */
3232 ph10 180 SETBIT(classbits, 0x0c); /* FF */
3233     SETBIT(classbits, 0x0d); /* CR */
3234     SETBIT(classbits, 0x85); /* NEL */
3235 ph10 178 #ifdef SUPPORT_UTF8
3236     if (utf8)
3237 ph10 180 {
3238 ph10 178 class_utf8 = TRUE;
3239 ph10 180 *class_utf8data++ = XCL_RANGE;
3240     class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3241     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3242     }
3243     #endif
3244     continue;
3245     }
3246 ph10 178
3247     if (-c == ESC_V)
3248     {
3249     for (c = 0; c < 32; c++)
3250     {
3251     int x = 0xff;
3252     switch (c)
3253 ph10 180 {
3254 ph10 178 case 0x0a/8: x ^= 1 << (0x0a%8);
3255     x ^= 1 << (0x0b%8);
3256     x ^= 1 << (0x0c%8);
3257 ph10 180 x ^= 1 << (0x0d%8);
3258 ph10 178 break;
3259     case 0x85/8: x ^= 1 << (0x85%8); break;
3260     default: break;
3261     }
3262     classbits[c] |= x;
3263 ph10 180 }
3264    
3265 ph10 178 #ifdef SUPPORT_UTF8
3266     if (utf8)
3267 ph10 180 {
3268 ph10 178 class_utf8 = TRUE;
3269 ph10 180 *class_utf8data++ = XCL_RANGE;
3270     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3271     class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3272     *class_utf8data++ = XCL_RANGE;
3273     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3274     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3275     }
3276     #endif
3277     continue;
3278     }
3279 ph10 178
3280 nigel 93 /* We need to deal with \P and \p in both phases. */
3281    
3282 nigel 77 #ifdef SUPPORT_UCP
3283 nigel 93 if (-c == ESC_p || -c == ESC_P)
3284     {
3285     BOOL negated;
3286     int pdata;
3287     int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3288     if (ptype < 0) goto FAILED;
3289     class_utf8 = TRUE;
3290     *class_utf8data++ = ((-c == ESC_p) != negated)?
3291     XCL_PROP : XCL_NOTPROP;
3292     *class_utf8data++ = ptype;
3293     *class_utf8data++ = pdata;
3294     class_charcount -= 2; /* Not a < 256 character */
3295 nigel 77 continue;
3296 nigel 93 }
3297 nigel 77 #endif
3298 nigel 93 /* Unrecognized escapes are faulted if PCRE is running in its
3299     strict mode. By default, for compatibility with Perl, they are
3300     treated as literals. */
3301 nigel 77
3302 nigel 93 if ((options & PCRE_EXTRA) != 0)
3303     {
3304     *errorcodeptr = ERR7;
3305     goto FAILED;
3306     }
3307 nigel 77
3308 nigel 93 class_charcount -= 2; /* Undo the default count from above */
3309     c = *ptr; /* Get the final character and fall through */
3310 nigel 77 }
3311    
3312     /* Fall through if we have a single character (c >= 0). This may be
3313 nigel 93 greater than 256 in UTF-8 mode. */
3314 nigel 77
3315     } /* End of backslash handling */
3316    
3317     /* A single character may be followed by '-' to form a range. However,
3318     Perl does not permit ']' to be the end of the range. A '-' character
3319 nigel 93 at the end is treated as a literal. Perl ignores orphaned \E sequences
3320     entirely. The code for handling \Q and \E is messy. */
3321 nigel 77
3322 nigel 93 CHECK_RANGE:
3323 ph10 391 while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
3324 nigel 77 {
3325 nigel 93 inescq = FALSE;
3326     ptr += 2;
3327     }
3328    
3329     oldptr = ptr;
3330 ph10 231
3331 ph10 230 /* Remember \r or \n */
3332 ph10 231
3333 ph10 391 if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3334 ph10 231
3335 ph10 230 /* Check for range */
3336 nigel 93
3337 ph10 391 if (!inescq && ptr[1] == CHAR_MINUS)
3338 nigel 93 {
3339 nigel 77 int d;
3340     ptr += 2;
3341 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
3342 nigel 77
3343 nigel 93 /* If we hit \Q (not followed by \E) at this point, go into escaped
3344     mode. */
3345    
3346 ph10 391 while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
3347 nigel 93 {
3348     ptr += 2;
3349 ph10 392 if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3350 ph10 391 { ptr += 2; continue; }
3351 nigel 93 inescq = TRUE;
3352     break;
3353     }
3354    
3355 ph10 391 if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
3356 nigel 93 {
3357     ptr = oldptr;
3358     goto LONE_SINGLE_CHARACTER;
3359     }
3360    
3361 nigel 77 #ifdef SUPPORT_UTF8
3362     if (utf8)
3363     { /* Braces are required because the */
3364     GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3365     }
3366     else
3367     #endif
3368     d = *ptr; /* Not UTF-8 mode */
3369    
3370     /* The second part of a range can be a single-character escape, but
3371     not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3372     in such circumstances. */
3373    
3374 ph10 391 if (!inescq && d == CHAR_BACKSLASH)
3375 nigel 77 {
3376 nigel 93 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3377     if (*errorcodeptr != 0) goto FAILED;
3378 nigel 77
3379 ph10 275 /* \b is backspace; \X is literal X; \R is literal R; any other
3380 nigel 93 special means the '-' was literal */
3381 nigel 77
3382     if (d < 0)
3383     {
3384 ph10 391 if (d == -ESC_b) d = CHAR_BS;
3385     else if (d == -ESC_X) d = CHAR_X;
3386     else if (d == -ESC_R) d = CHAR_R; else
3387 nigel 77 {
3388 nigel 93 ptr = oldptr;
3389 nigel 77 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3390     }
3391     }
3392     }
3393    
3394 nigel 93 /* Check that the two values are in the correct order. Optimize
3395     one-character ranges */
3396 nigel 77
3397 nigel 93 if (d < c)
3398     {
3399     *errorcodeptr = ERR8;
3400     goto FAILED;
3401     }
3402    
3403 nigel 77 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3404    
3405 ph10 230 /* Remember \r or \n */
3406 ph10 231
3407 ph10 391 if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
3408 ph10 231
3409 nigel 77 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3410     matching, we have to use an XCLASS with extra data items. Caseless
3411     matching for characters > 127 is available only if UCP support is
3412     available. */
3413    
3414     #ifdef SUPPORT_UTF8
3415     if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3416     {
3417     class_utf8 = TRUE;
3418    
3419     /* With UCP support, we can find the other case equivalents of
3420     the relevant characters. There may be several ranges. Optimize how
3421     they fit with the basic range. */
3422    
3423     #ifdef SUPPORT_UCP
3424     if ((options & PCRE_CASELESS) != 0)
3425     {
3426 nigel 93 unsigned int occ, ocd;
3427     unsigned int cc = c;
3428     unsigned int origd = d;
3429 nigel 77 while (get_othercase_range(&cc, origd, &occ, &ocd))
3430     {
3431 ph10 180 if (occ >= (unsigned int)c &&
3432     ocd <= (unsigned int)d)
3433 ph10 176 continue; /* Skip embedded ranges */
3434 nigel 77
3435 ph10 180 if (occ < (unsigned int)c &&
3436 ph10 176 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3437 nigel 77 { /* if there is overlap, */
3438     c = occ; /* noting that if occ < c */
3439     continue; /* we can't have ocd > d */
3440     } /* because a subrange is */
3441 ph10 180 if (ocd > (unsigned int)d &&
3442 ph10 176 occ <= (unsigned int)d + 1) /* always shorter than */
3443 nigel 77 { /* the basic range. */
3444     d = ocd;
3445     continue;
3446     }
3447    
3448     if (occ == ocd)
3449     {
3450     *class_utf8data++ = XCL_SINGLE;
3451     }
3452     else
3453     {
3454     *class_utf8data++ = XCL_RANGE;
3455     class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3456     }
3457     class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3458     }
3459     }
3460     #endif /* SUPPORT_UCP */
3461    
3462     /* Now record the original range, possibly modified for UCP caseless
3463     overlapping ranges. */
3464    
3465     *class_utf8data++ = XCL_RANGE;
3466     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3467     class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3468    
3469     /* With UCP support, we are done. Without UCP support, there is no
3470     caseless matching for UTF-8 characters > 127; we can use the bit map
3471     for the smaller ones. */
3472    
3473     #ifdef SUPPORT_UCP
3474     continue; /* With next character in the class */
3475     #else
3476     if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3477    
3478     /* Adjust upper limit and fall through to set up the map */
3479    
3480     d = 127;
3481    
3482     #endif /* SUPPORT_UCP */
3483     }
3484     #endif /* SUPPORT_UTF8 */
3485    
3486     /* We use the bit map for all cases when not in UTF-8 mode; else
3487     ranges that lie entirely within 0-127 when there is UCP support; else
3488     for partial ranges without UCP support. */
3489    
3490 nigel 93 class_charcount += d - c + 1;
3491     class_lastchar = d;
3492    
3493     /* We can save a bit of time by skipping this in the pre-compile. */
3494    
3495     if (lengthptr == NULL) for (; c <= d; c++)
3496 nigel 77 {
3497     classbits[c/8] |= (1 << (c&7));
3498     if ((options & PCRE_CASELESS) != 0)
3499     {
3500     int uc = cd->fcc[c]; /* flip case */
3501     classbits[uc/8] |= (1 << (uc&7));
3502     }
3503     }
3504    
3505     continue; /* Go get the next char in the class */
3506     }
3507    
3508     /* Handle a lone single character - we can get here for a normal
3509     non-escape char, or after \ that introduces a single character or for an
3510     apparent range that isn't. */
3511    
3512     LONE_SINGLE_CHARACTER:
3513 ph10 231
3514 nigel 77 /* Handle a character that cannot go in the bit map */
3515    
3516     #ifdef SUPPORT_UTF8
3517     if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3518     {
3519     class_utf8 = TRUE;
3520     *class_utf8data++ = XCL_SINGLE;
3521     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3522    
3523     #ifdef SUPPORT_UCP
3524     if ((options & PCRE_CASELESS) != 0)
3525     {
3526 nigel 93 unsigned int othercase;
3527 ph10 349 if ((othercase = UCD_OTHERCASE(c)) != c)
3528 nigel 77 {
3529     *class_utf8data++ = XCL_SINGLE;
3530     class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3531     }
3532     }
3533     #endif /* SUPPORT_UCP */
3534    
3535     }
3536     else
3537     #endif /* SUPPORT_UTF8 */
3538    
3539     /* Handle a single-byte character */
3540     {
3541     classbits[c/8] |= (1 << (c&7));
3542     if ((options & PCRE_CASELESS) != 0)
3543     {
3544     c = cd->fcc[c]; /* flip case */
3545     classbits[c/8] |= (1 << (c&7));
3546     }
3547     class_charcount++;
3548     class_lastchar = c;
3549     }
3550     }
3551    
3552 nigel 93 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3553 nigel 77
3554 ph10 391 while ((c = *(++ptr)) != 0 && (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
3555 nigel 77
3556 nigel 93 if (c == 0) /* Missing terminating ']' */
3557     {
3558     *errorcodeptr = ERR6;
3559     goto FAILED;
3560     }
3561 ph10 231
3562    
3563 ph10 230 /* This code has been disabled because it would mean that \s counts as
3564     an explicit \r or \n reference, and that's not really what is wanted. Now
3565     we set the flag only if there is a literal "\r" or "\n" in the class. */
3566 ph10 227
3567 ph10 230 #if 0
3568 ph10 226 /* Remember whether \r or \n are in this class */
3569 ph10 227
3570 ph10 226 if (negate_class)
3571     {
3572 ph10 230 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3573 ph10 226 }
3574     else
3575     {
3576 ph10 230 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3577 ph10 227 }
3578 ph10 230 #endif
3579 ph10 227
3580 ph10 231
3581 nigel 77 /* If class_charcount is 1, we saw precisely one character whose value is
3582 ph10 227 less than 256. As long as there were no characters >= 128 and there was no
3583     use of \p or \P, in other words, no use of any XCLASS features, we can
3584     optimize.
3585    
3586 ph10 223 In UTF-8 mode, we can optimize the negative case only if there were no
3587     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3588     operate on single-bytes only. This is an historical hangover. Maybe one day
3589     we can tidy these opcodes to handle multi-byte characters.
3590 nigel 77
3591     The optimization throws away the bit map. We turn the item into a
3592     1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3593     that OP_NOT does not support multibyte characters. In the positive case, it
3594     can cause firstbyte to be set. Otherwise, there can be no first char if
3595     this item is first, whatever repeat count may follow. In the case of
3596     reqbyte, save the previous value for reinstating. */
3597    
3598     #ifdef SUPPORT_UTF8
3599 ph10 227 if (class_charcount == 1 && !class_utf8 &&
3600 ph10 223 (!utf8 || !negate_class || class_lastchar < 128))
3601 nigel 77 #else
3602     if (class_charcount == 1)
3603     #endif
3604     {
3605     zeroreqbyte = reqbyte;
3606    
3607     /* The OP_NOT opcode works on one-byte characters only. */
3608    
3609     if (negate_class)
3610     {
3611     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3612     zerofirstbyte = firstbyte;
3613     *code++ = OP_NOT;
3614     *code++ = class_lastchar;
3615     break;
3616     }
3617    
3618     /* For a single, positive character, get the value into mcbuffer, and
3619     then we can handle this with the normal one-character code. */
3620    
3621     #ifdef SUPPORT_UTF8
3622     if (utf8 && class_lastchar > 127)
3623     mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3624     else
3625     #endif
3626     {
3627     mcbuffer[0] = class_lastchar;
3628     mclength = 1;
3629     }
3630     goto ONE_CHAR;
3631     } /* End of 1-char optimization */
3632    
3633     /* The general case - not the one-char optimization. If this is the first
3634     thing in the branch, there can be no first char setting, whatever the
3635     repeat count. Any reqbyte setting must remain unchanged after any kind of
3636     repeat. */
3637    
3638     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3639     zerofirstbyte = firstbyte;
3640     zeroreqbyte = reqbyte;
3641    
3642     /* If there are characters with values > 255, we have to compile an
3643 ph10 286 extended class, with its own opcode, unless there was a negated special
3644     such as \S in the class, because in that case all characters > 255 are in
3645     the class, so any that were explicitly given as well can be ignored. If
3646 ph10 264 (when there are explicit characters > 255 that must be listed) there are no
3647     characters < 256, we can omit the bitmap in the actual compiled code. */
3648 nigel 77
3649     #ifdef SUPPORT_UTF8
3650 ph10 264 if (class_utf8 && !should_flip_negation)
3651 nigel 77 {
3652     *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3653     *code++ = OP_XCLASS;
3654     code += LINK_SIZE;
3655     *code = negate_class? XCL_NOT : 0;
3656    
3657 nigel 93 /* If the map is required, move up the extra data to make room for it;
3658     otherwise just move the code pointer to the end of the extra data. */
3659 nigel 77
3660     if (class_charcount > 0)
3661     {
3662     *code++ |= XCL_MAP;
3663 nigel 93 memmove(code + 32, code, class_utf8data - code);
3664 nigel 77 memcpy(code, classbits, 32);
3665 nigel 93 code = class_utf8data + 32;
3666 nigel 77 }
3667 nigel 93 else code = class_utf8data;
3668 nigel 77
3669     /* Now fill in the complete length of the item */
3670    
3671     PUT(previous, 1, code - previous);
3672     break; /* End of class handling */
3673     }
3674     #endif
3675    
3676 ph10 286 /* If there are no characters > 255, set the opcode to OP_CLASS or
3677     OP_NCLASS, depending on whether the whole class was negated and whether
3678     there were negative specials such as \S in the class. Then copy the 32-byte
3679 ph10 264 map into the code vector, negating it if necessary. */
3680 ph10 286
3681 ph10 264 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3682 nigel 77 if (negate_class)
3683     {
3684 nigel 93 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3685     for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3686 nigel 77 }
3687     else
3688     {
3689     memcpy(code, classbits, 32);
3690     }
3691     code += 32;
3692     break;
3693    
3694 nigel 93
3695     /* ===================================================================*/
3696 nigel 77 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3697     has been tested above. */
3698    
3699 ph10 391 case CHAR_LEFT_CURLY_BRACKET:
3700 nigel 77 if (!is_quantifier) goto NORMAL_CHAR;
3701     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3702     if (*errorcodeptr != 0) goto FAILED;
3703     goto REPEAT;
3704    
3705 ph10 391 case CHAR_ASTERISK:
3706 nigel 77 repeat_min = 0;
3707     repeat_max = -1;
3708     goto REPEAT;
3709    
3710 ph10 391 case CHAR_PLUS:
3711 nigel 77 repeat_min = 1;
3712     repeat_max = -1;
3713     goto REPEAT;
3714    
3715 ph10 391 case CHAR_QUESTION_MARK:
3716 nigel 77 repeat_min = 0;
3717     repeat_max = 1;
3718    
3719     REPEAT:
3720     if (previous == NULL)
3721     {
3722     *errorcodeptr = ERR9;
3723     goto FAILED;
3724     }
3725    
3726     if (repeat_min == 0)
3727     {
3728     firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3729     reqbyte = zeroreqbyte; /* Ditto */
3730     }
3731    
3732     /* Remember whether this is a variable length repeat */
3733    
3734     reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3735    
3736     op_type = 0; /* Default single-char op codes */
3737     possessive_quantifier = FALSE; /* Default not possessive quantifier */
3738    
3739     /* Save start of previous item, in case we have to move it up to make space
3740     for an inserted OP_ONCE for the additional '+' extension. */
3741    
3742     tempcode = previous;
3743    
3744     /* If the next character is '+', we have a possessive quantifier. This
3745     implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3746     If the next character is '?' this is a minimizing repeat, by default,
3747     but if PCRE_UNGREEDY is set, it works the other way round. We change the
3748     repeat type to the non-default. */
3749    
3750 ph10 391 if (ptr[1] == CHAR_PLUS)
3751 nigel 77 {
3752     repeat_type = 0; /* Force greedy */
3753     possessive_quantifier = TRUE;
3754     ptr++;
3755     }
3756 ph10 391 else if (ptr[1] == CHAR_QUESTION_MARK)
3757 nigel 77 {
3758     repeat_type = greedy_non_default;
3759     ptr++;
3760     }
3761     else repeat_type = greedy_default;
3762    
3763     /* If previous was a character match, abolish the item and generate a
3764     repeat item instead. If a char item has a minumum of more than one, ensure
3765     that it is set in reqbyte - it might not be if a sequence such as x{3} is
3766     the first thing in a branch because the x will have gone into firstbyte
3767     instead. */
3768    
3769     if (*previous == OP_CHAR || *previous == OP_CHARNC)
3770     {
3771     /* Deal with UTF-8 characters that take up more than one byte. It's
3772     easier to write this out separately than try to macrify it. Use c to
3773     hold the length of the character in bytes, plus 0x80 to flag that it's a
3774     length rather than a small character. */
3775    
3776     #ifdef SUPPORT_UTF8
3777     if (utf8 && (code[-1] & 0x80) != 0)
3778     {
3779     uschar *lastchar = code - 1;
3780     while((*lastchar & 0xc0) == 0x80) lastchar--;
3781     c = code - lastchar; /* Length of UTF-8 character */
3782     memcpy(utf8_char, lastchar, c); /* Save the char */
3783     c |= 0x80; /* Flag c as a length */
3784     }
3785     else
3786     #endif
3787    
3788     /* Handle the case of a single byte - either with no UTF8 support, or
3789     with UTF-8 disabled, or for a UTF-8 character < 128. */
3790    
3791     {
3792     c = code[-1];
3793     if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3794     }
3795    
3796 nigel 93 /* If the repetition is unlimited, it pays to see if the next thing on
3797     the line is something that cannot possibly match this character. If so,
3798     automatically possessifying this item gains some performance in the case
3799     where the match fails. */
3800    
3801     if (!possessive_quantifier &&
3802     repeat_max < 0 &&
3803     check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3804     options, cd))
3805     {
3806     repeat_type = 0; /* Force greedy */
3807     possessive_quantifier = TRUE;
3808     }
3809    
3810 nigel 77 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3811     }
3812    
3813     /* If previous was a single negated character ([^a] or similar), we use
3814     one of the special opcodes, replacing it. The code is shared with single-
3815     character repeats by setting opt_type to add a suitable offset into
3816 nigel 93 repeat_type. We can also test for auto-possessification. OP_NOT is
3817     currently used only for single-byte chars. */
3818 nigel 77
3819     else if (*previous == OP_NOT)
3820     {
3821     op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3822     c = previous[1];
3823 nigel 93 if (!possessive_quantifier &&
3824     repeat_max < 0 &&
3825     check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3826     {
3827     repeat_type = 0; /* Force greedy */
3828     possessive_quantifier = TRUE;
3829     }
3830 nigel 77 goto OUTPUT_SINGLE_REPEAT;
3831     }
3832    
3833     /* If previous was a character type match (\d or similar), abolish it and
3834     create a suitable repeat item. The code is shared with single-character
3835     repeats by setting op_type to add a suitable offset into repeat_type. Note
3836     the the Unicode property types will be present only when SUPPORT_UCP is
3837     defined, but we don't wrap the little bits of code here because it just
3838     makes it horribly messy. */
3839    
3840     else if (*previous < OP_EODN)
3841     {
3842     uschar *oldcode;
3843 nigel 87 int prop_type, prop_value;
3844 nigel 77 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3845     c = *previous;
3846    
3847 nigel 93 if (!possessive_quantifier &&
3848     repeat_max < 0 &&
3849     check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3850     {
3851     repeat_type = 0; /* Force greedy */
3852     possessive_quantifier = TRUE;
3853     }
3854    
3855 nigel 77 OUTPUT_SINGLE_REPEAT:
3856 nigel 87 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3857     {
3858     prop_type = previous[1];
3859     prop_value = previous[2];
3860     }
3861     else prop_type = prop_value = -1;
3862 nigel 77
3863     oldcode = code;
3864     code = previous; /* Usually overwrite previous item */
3865    
3866     /* If the maximum is zero then the minimum must also be zero; Perl allows
3867     this case, so we do too - by simply omitting the item altogether. */
3868    
3869     if (repeat_max == 0) goto END_REPEAT;
3870    
3871 ph10 426 /*--------------------------------------------------------------------*/
3872     /* This code is obsolete from release 8.00; the restriction was finally
3873     removed: */
3874    
3875 nigel 77 /* All real repeats make it impossible to handle partial matching (maybe
3876     one day we will be able to remove this restriction). */
3877 ph10 426
3878     /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
3879     /*--------------------------------------------------------------------*/
3880 nigel 77
3881     /* Combine the op_type with the repeat_type */
3882    
3883     repeat_type += op_type;
3884    
3885     /* A minimum of zero is handled either as the special case * or ?, or as
3886     an UPTO, with the maximum given. */
3887    
3888     if (repeat_min == 0)
3889     {
3890     if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3891     else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3892     else
3893     {
3894     *code++ = OP_UPTO + repeat_type;
3895     PUT2INC(code, 0, repeat_max);
3896     }
3897     }
3898    
3899     /* A repeat minimum of 1 is optimized into some special cases. If the
3900 nigel 93 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3901 nigel 77 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3902     one less than the maximum. */
3903    
3904     else if (repeat_min == 1)
3905     {
3906     if (repeat_max == -1)
3907     *code++ = OP_PLUS + repeat_type;
3908     else
3909     {
3910     code = oldcode; /* leave previous item in place */
3911     if (repeat_max == 1) goto END_REPEAT;
3912     *code++ = OP_UPTO + repeat_type;
3913     PUT2INC(code, 0, repeat_max - 1);
3914     }
3915     }
3916    
3917     /* The case {n,n} is just an EXACT, while the general case {n,m} is
3918     handled as an EXACT followed by an UPTO. */
3919    
3920     else
3921     {
3922     *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3923     PUT2INC(code, 0, repeat_min);
3924    
3925     /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3926     we have to insert the character for the previous code. For a repeated
3927 nigel 87 Unicode property match, there are two extra bytes that define the
3928 nigel 77 required property. In UTF-8 mode, long characters have their length in
3929     c, with the 0x80 bit as a flag. */
3930    
3931     if (repeat_max < 0)
3932     {
3933     #ifdef SUPPORT_UTF8
3934     if (utf8 && c >= 128)
3935     {
3936     memcpy(code, utf8_char, c & 7);
3937     code += c & 7;
3938     }
3939     else
3940     #endif
3941     {
3942     *code++ = c;
3943 nigel 87 if (prop_type >= 0)
3944     {
3945     *code++ = prop_type;
3946     *code++ = prop_value;
3947     }
3948 nigel 77 }
3949     *code++ = OP_STAR + repeat_type;
3950     }
3951    
3952     /* Else insert an UPTO if the max is greater than the min, again
3953 nigel 93 preceded by the character, for the previously inserted code. If the
3954     UPTO is just for 1 instance, we can use QUERY instead. */
3955 nigel 77
3956     else if (repeat_max != repeat_min)
3957     {
3958     #ifdef SUPPORT_UTF8
3959     if (utf8 && c >= 128)
3960     {
3961     memcpy(code, utf8_char, c & 7);
3962     code += c & 7;
3963     }
3964     else
3965     #endif
3966     *code++ = c;
3967 nigel 87<