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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 167 - (hide annotations) (download)
Wed May 9 15:53:54 2007 UTC (7 years, 3 months ago) by ph10
File MIME type: text/plain
File size: 174513 byte(s)
Add (?(-n) and (?(+n) relative conditions.

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 117 Copyright (c) 1997-2007 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 nigel 93 #define NLBLOCK cd /* Block containing newline information */
46     #define PSSTART start_pattern /* Field containing processed string start */
47     #define PSEND end_pattern /* Field containing processed string end */
48    
49    
50 nigel 77 #include "pcre_internal.h"
51    
52    
53 nigel 85 /* When DEBUG is defined, we need the pcre_printint() function, which is also
54     used by pcretest. DEBUG is not defined when building a production library. */
55    
56     #ifdef DEBUG
57     #include "pcre_printint.src"
58     #endif
59    
60    
61 nigel 77 /*************************************************
62     * Code parameters and static tables *
63     *************************************************/
64    
65 nigel 93 /* This value specifies the size of stack workspace that is used during the
66     first pre-compile phase that determines how much memory is required. The regex
67     is partly compiled into this space, but the compiled parts are discarded as
68     soon as they can be, so that hopefully there will never be an overrun. The code
69     does, however, check for an overrun. The largest amount I've seen used is 218,
70     so this number is very generous.
71 nigel 77
72 nigel 93 The same workspace is used during the second, actual compile phase for
73     remembering forward references to groups so that they can be filled in at the
74     end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
75     is 4 there is plenty of room. */
76 nigel 77
77 nigel 93 #define COMPILE_WORK_SIZE (4096)
78 nigel 77
79 nigel 93
80 nigel 77 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
81     are simple data values; negative values are for special things like \d and so
82     on. Zero means further processing is needed (for things like \x), or the escape
83     is invalid. */
84    
85 ph10 97 #ifndef EBCDIC /* This is the "normal" table for ASCII systems */
86 nigel 77 static const short int escapes[] = {
87     0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 7 */
88     0, 0, ':', ';', '<', '=', '>', '?', /* 8 - ? */
89     '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E, 0, -ESC_G, /* @ - G */
90     0, 0, 0, 0, 0, 0, 0, 0, /* H - O */
91 nigel 93 -ESC_P, -ESC_Q, -ESC_R, -ESC_S, 0, 0, 0, -ESC_W, /* P - W */
92 nigel 77 -ESC_X, 0, -ESC_Z, '[', '\\', ']', '^', '_', /* X - _ */
93     '`', 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, /* ` - g */
94 nigel 93 0, 0, 0, -ESC_k, 0, 0, ESC_n, 0, /* h - o */
95 nigel 77 -ESC_p, 0, ESC_r, -ESC_s, ESC_tee, 0, 0, -ESC_w, /* p - w */
96     0, 0, -ESC_z /* x - z */
97     };
98    
99 ph10 97 #else /* This is the "abnormal" table for EBCDIC systems */
100 nigel 77 static const short int escapes[] = {
101     /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
102     /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
103     /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
104     /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
105     /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
106     /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
107     /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
108     /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
109     /* 88 */ 0, 0, 0, '{', 0, 0, 0, 0,
110 nigel 93 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
111 nigel 77 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
112     /* A0 */ 0, '~', -ESC_s, ESC_tee, 0, 0, -ESC_w, 0,
113     /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
114     /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
115     /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
116     /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
117     /* C8 */ 0, 0, 0, 0, 0, 0, 0, 0,
118     /* D0 */ '}', 0, 0, 0, 0, 0, 0, -ESC_P,
119 nigel 93 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
120 nigel 77 /* E0 */ '\\', 0, -ESC_S, 0, 0, 0, -ESC_W, -ESC_X,
121     /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
122     /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
123     /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
124     };
125     #endif
126    
127    
128     /* Tables of names of POSIX character classes and their lengths. The list is
129 nigel 87 terminated by a zero length entry. The first three must be alpha, lower, upper,
130 nigel 77 as this is assumed for handling case independence. */
131    
132     static const char *const posix_names[] = {
133     "alpha", "lower", "upper",
134     "alnum", "ascii", "blank", "cntrl", "digit", "graph",
135     "print", "punct", "space", "word", "xdigit" };
136    
137     static const uschar posix_name_lengths[] = {
138     5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
139    
140 nigel 87 /* Table of class bit maps for each POSIX class. Each class is formed from a
141     base map, with an optional addition or removal of another map. Then, for some
142     classes, there is some additional tweaking: for [:blank:] the vertical space
143     characters are removed, and for [:alpha:] and [:alnum:] the underscore
144     character is removed. The triples in the table consist of the base map offset,
145     second map offset or -1 if no second map, and a non-negative value for map
146     addition or a negative value for map subtraction (if there are two maps). The
147     absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
148     remove vertical space characters, 2 => remove underscore. */
149 nigel 77
150     static const int posix_class_maps[] = {
151 nigel 87 cbit_word, cbit_digit, -2, /* alpha */
152     cbit_lower, -1, 0, /* lower */
153     cbit_upper, -1, 0, /* upper */
154     cbit_word, -1, 2, /* alnum - word without underscore */
155     cbit_print, cbit_cntrl, 0, /* ascii */
156     cbit_space, -1, 1, /* blank - a GNU extension */
157     cbit_cntrl, -1, 0, /* cntrl */
158     cbit_digit, -1, 0, /* digit */
159     cbit_graph, -1, 0, /* graph */
160     cbit_print, -1, 0, /* print */
161     cbit_punct, -1, 0, /* punct */
162     cbit_space, -1, 0, /* space */
163     cbit_word, -1, 0, /* word - a Perl extension */
164     cbit_xdigit,-1, 0 /* xdigit */
165 nigel 77 };
166    
167    
168 nigel 93 #define STRING(a) # a
169     #define XSTRING(s) STRING(s)
170    
171 nigel 77 /* The texts of compile-time error messages. These are "char *" because they
172 nigel 93 are passed to the outside world. Do not ever re-use any error number, because
173     they are documented. Always add a new error instead. Messages marked DEAD below
174     are no longer used. */
175 nigel 77
176     static const char *error_texts[] = {
177     "no error",
178     "\\ at end of pattern",
179     "\\c at end of pattern",
180     "unrecognized character follows \\",
181     "numbers out of order in {} quantifier",
182     /* 5 */
183     "number too big in {} quantifier",
184     "missing terminating ] for character class",
185     "invalid escape sequence in character class",
186     "range out of order in character class",
187     "nothing to repeat",
188     /* 10 */
189 nigel 93 "operand of unlimited repeat could match the empty string", /** DEAD **/
190 nigel 77 "internal error: unexpected repeat",
191     "unrecognized character after (?",
192     "POSIX named classes are supported only within a class",
193     "missing )",
194     /* 15 */
195     "reference to non-existent subpattern",
196     "erroffset passed as NULL",
197     "unknown option bit(s) set",
198     "missing ) after comment",
199 nigel 93 "parentheses nested too deeply", /** DEAD **/
200 nigel 77 /* 20 */
201     "regular expression too large",
202     "failed to get memory",
203     "unmatched parentheses",
204     "internal error: code overflow",
205     "unrecognized character after (?<",
206     /* 25 */
207     "lookbehind assertion is not fixed length",
208 nigel 91 "malformed number or name after (?(",
209 nigel 77 "conditional group contains more than two branches",
210     "assertion expected after (?(",
211 ph10 166 "(?R or (?[+-]digits must be followed by )",
212 nigel 77 /* 30 */
213     "unknown POSIX class name",
214     "POSIX collating elements are not supported",
215     "this version of PCRE is not compiled with PCRE_UTF8 support",
216 nigel 93 "spare error", /** DEAD **/
217 nigel 77 "character value in \\x{...} sequence is too large",
218     /* 35 */
219     "invalid condition (?(0)",
220     "\\C not allowed in lookbehind assertion",
221     "PCRE does not support \\L, \\l, \\N, \\U, or \\u",
222     "number after (?C is > 255",
223     "closing ) for (?C expected",
224     /* 40 */
225     "recursive call could loop indefinitely",
226     "unrecognized character after (?P",
227 nigel 93 "syntax error in subpattern name (missing terminator)",
228 nigel 91 "two named subpatterns have the same name",
229 nigel 77 "invalid UTF-8 string",
230     /* 45 */
231     "support for \\P, \\p, and \\X has not been compiled",
232     "malformed \\P or \\p sequence",
233 nigel 91 "unknown property name after \\P or \\p",
234 nigel 93 "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)",
235     "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")",
236 nigel 91 /* 50 */
237     "repeated subpattern is too long",
238 nigel 93 "octal value is greater than \\377 (not in UTF-8 mode)",
239     "internal error: overran compiling workspace",
240     "internal error: previously-checked referenced subpattern not found",
241     "DEFINE group contains more than one branch",
242     /* 55 */
243     "repeating a DEFINE group is not allowed",
244     "inconsistent NEWLINE options",
245 ph10 166 "\\g is not followed by an (optionally braced) non-zero number",
246 ph10 167 "(?+ or (?- or (?(+ or (?(- must be followed by a non-zero number"
247 nigel 77 };
248    
249    
250     /* Table to identify digits and hex digits. This is used when compiling
251     patterns. Note that the tables in chartables are dependent on the locale, and
252     may mark arbitrary characters as digits - but the PCRE compiling code expects
253     to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
254     a private table here. It costs 256 bytes, but it is a lot faster than doing
255     character value tests (at least in some simple cases I timed), and in some
256     applications one wants PCRE to compile efficiently as well as match
257     efficiently.
258    
259     For convenience, we use the same bit definitions as in chartables:
260    
261     0x04 decimal digit
262     0x08 hexadecimal digit
263    
264     Then we can use ctype_digit and ctype_xdigit in the code. */
265    
266 ph10 97 #ifndef EBCDIC /* This is the "normal" case, for ASCII systems */
267 nigel 77 static const unsigned char digitab[] =
268     {
269     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
270     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
271     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
272     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
273     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
274     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
275     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
276     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
277     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
278     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
279     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
280     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
281     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
282     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
283     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
284     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
285     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
286     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
287     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
288     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
289     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
290     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
291     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
292     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
293     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
294     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
295     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
296     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
297     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
298     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
299     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
300     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
301    
302 ph10 97 #else /* This is the "abnormal" case, for EBCDIC systems */
303 nigel 77 static const unsigned char digitab[] =
304     {
305     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
306     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
307     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
308     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
309     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
310     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
311     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
312     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
313     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
314     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
315     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
316 ph10 97 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
317 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
318     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
319     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
320     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
321     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
322     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
323     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
324     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
325     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
326     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
327     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
328     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
329     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
330     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
331     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
332     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
333     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
334     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
335     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
336     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
337    
338     static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
339     0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
340     0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
341     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
342     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
343     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
344     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
345     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
346     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
347     0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
348     0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
349     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
350 ph10 97 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
351 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
352     0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
353     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
354     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
355     0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
356     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
357     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
358     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
359     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
360     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
361     0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
362     0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
363     0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
364     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
365     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
366     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
367     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
368     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
369     0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
370     0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
371     #endif
372    
373    
374     /* Definition to allow mutual recursion */
375    
376     static BOOL
377 nigel 93 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, int, int *,
378     int *, branch_chain *, compile_data *, int *);
379 nigel 77
380    
381    
382     /*************************************************
383     * Handle escapes *
384     *************************************************/
385    
386     /* This function is called when a \ has been encountered. It either returns a
387     positive value for a simple escape such as \n, or a negative value which
388 nigel 93 encodes one of the more complicated things such as \d. A backreference to group
389     n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
390     UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
391     ptr is pointing at the \. On exit, it is on the final character of the escape
392     sequence.
393 nigel 77
394     Arguments:
395     ptrptr points to the pattern position pointer
396     errorcodeptr points to the errorcode variable
397     bracount number of previous extracting brackets
398     options the options bits
399     isclass TRUE if inside a character class
400    
401     Returns: zero or positive => a data character
402     negative => a special escape sequence
403     on error, errorptr is set
404     */
405    
406     static int
407     check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
408     int options, BOOL isclass)
409     {
410 nigel 87 BOOL utf8 = (options & PCRE_UTF8) != 0;
411     const uschar *ptr = *ptrptr + 1;
412 nigel 77 int c, i;
413    
414 nigel 87 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
415     ptr--; /* Set pointer back to the last byte */
416    
417 nigel 77 /* If backslash is at the end of the pattern, it's an error. */
418    
419     if (c == 0) *errorcodeptr = ERR1;
420    
421     /* Non-alphamerics are literals. For digits or letters, do an initial lookup in
422     a table. A non-zero result is something that can be returned immediately.
423     Otherwise further processing may be required. */
424    
425 ph10 97 #ifndef EBCDIC /* ASCII coding */
426 nigel 77 else if (c < '0' || c > 'z') {} /* Not alphameric */
427     else if ((i = escapes[c - '0']) != 0) c = i;
428    
429 ph10 97 #else /* EBCDIC coding */
430 nigel 77 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphameric */
431     else if ((i = escapes[c - 0x48]) != 0) c = i;
432     #endif
433    
434     /* Escapes that need further processing, or are illegal. */
435    
436     else
437     {
438     const uschar *oldptr;
439 nigel 93 BOOL braced, negated;
440    
441 nigel 77 switch (c)
442     {
443     /* A number of Perl escapes are not handled by PCRE. We give an explicit
444     error. */
445    
446     case 'l':
447     case 'L':
448     case 'N':
449     case 'u':
450     case 'U':
451     *errorcodeptr = ERR37;
452     break;
453    
454 nigel 93 /* \g must be followed by a number, either plain or braced. If positive, it
455     is an absolute backreference. If negative, it is a relative backreference.
456     This is a Perl 5.10 feature. */
457    
458     case 'g':
459     if (ptr[1] == '{')
460     {
461     braced = TRUE;
462     ptr++;
463     }
464     else braced = FALSE;
465    
466     if (ptr[1] == '-')
467     {
468     negated = TRUE;
469     ptr++;
470     }
471     else negated = FALSE;
472    
473     c = 0;
474     while ((digitab[ptr[1]] & ctype_digit) != 0)
475     c = c * 10 + *(++ptr) - '0';
476    
477     if (c == 0 || (braced && *(++ptr) != '}'))
478     {
479     *errorcodeptr = ERR57;
480     return 0;
481     }
482    
483     if (negated)
484     {
485     if (c > bracount)
486     {
487     *errorcodeptr = ERR15;
488     return 0;
489     }
490     c = bracount - (c - 1);
491     }
492    
493     c = -(ESC_REF + c);
494     break;
495    
496 nigel 77 /* The handling of escape sequences consisting of a string of digits
497     starting with one that is not zero is not straightforward. By experiment,
498     the way Perl works seems to be as follows:
499    
500     Outside a character class, the digits are read as a decimal number. If the
501     number is less than 10, or if there are that many previous extracting
502     left brackets, then it is a back reference. Otherwise, up to three octal
503     digits are read to form an escaped byte. Thus \123 is likely to be octal
504     123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
505     value is greater than 377, the least significant 8 bits are taken. Inside a
506     character class, \ followed by a digit is always an octal number. */
507    
508     case '1': case '2': case '3': case '4': case '5':
509     case '6': case '7': case '8': case '9':
510    
511     if (!isclass)
512     {
513     oldptr = ptr;
514     c -= '0';
515     while ((digitab[ptr[1]] & ctype_digit) != 0)
516     c = c * 10 + *(++ptr) - '0';
517     if (c < 10 || c <= bracount)
518     {
519     c = -(ESC_REF + c);
520     break;
521     }
522     ptr = oldptr; /* Put the pointer back and fall through */
523     }
524    
525     /* Handle an octal number following \. If the first digit is 8 or 9, Perl
526     generates a binary zero byte and treats the digit as a following literal.
527     Thus we have to pull back the pointer by one. */
528    
529     if ((c = *ptr) >= '8')
530     {
531     ptr--;
532     c = 0;
533     break;
534     }
535    
536     /* \0 always starts an octal number, but we may drop through to here with a
537 nigel 91 larger first octal digit. The original code used just to take the least
538     significant 8 bits of octal numbers (I think this is what early Perls used
539     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
540     than 3 octal digits. */
541 nigel 77
542     case '0':
543     c -= '0';
544     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
545     c = c * 8 + *(++ptr) - '0';
546 nigel 91 if (!utf8 && c > 255) *errorcodeptr = ERR51;
547 nigel 77 break;
548    
549 nigel 87 /* \x is complicated. \x{ddd} is a character number which can be greater
550     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
551     treated as a data character. */
552 nigel 77
553     case 'x':
554 nigel 87 if (ptr[1] == '{')
555 nigel 77 {
556     const uschar *pt = ptr + 2;
557 nigel 87 int count = 0;
558    
559 nigel 77 c = 0;
560     while ((digitab[*pt] & ctype_xdigit) != 0)
561     {
562 nigel 87 register int cc = *pt++;
563     if (c == 0 && cc == '0') continue; /* Leading zeroes */
564 nigel 77 count++;
565 nigel 87
566 ph10 97 #ifndef EBCDIC /* ASCII coding */
567 nigel 77 if (cc >= 'a') cc -= 32; /* Convert to upper case */
568 nigel 87 c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
569 ph10 97 #else /* EBCDIC coding */
570 nigel 77 if (cc >= 'a' && cc <= 'z') cc += 64; /* Convert to upper case */
571 nigel 87 c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
572 nigel 77 #endif
573     }
574 nigel 87
575 nigel 77 if (*pt == '}')
576     {
577 nigel 87 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
578 nigel 77 ptr = pt;
579     break;
580     }
581 nigel 87
582 nigel 77 /* If the sequence of hex digits does not end with '}', then we don't
583     recognize this construct; fall through to the normal \x handling. */
584     }
585    
586 nigel 87 /* Read just a single-byte hex-defined char */
587 nigel 77
588     c = 0;
589     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
590     {
591     int cc; /* Some compilers don't like ++ */
592     cc = *(++ptr); /* in initializers */
593 ph10 97 #ifndef EBCDIC /* ASCII coding */
594 nigel 77 if (cc >= 'a') cc -= 32; /* Convert to upper case */
595     c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
596 ph10 97 #else /* EBCDIC coding */
597 nigel 77 if (cc <= 'z') cc += 64; /* Convert to upper case */
598     c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
599     #endif
600     }
601     break;
602    
603 nigel 93 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
604     This coding is ASCII-specific, but then the whole concept of \cx is
605     ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
606 nigel 77
607     case 'c':
608     c = *(++ptr);
609     if (c == 0)
610     {
611     *errorcodeptr = ERR2;
612     return 0;
613     }
614    
615 ph10 97 #ifndef EBCDIC /* ASCII coding */
616 nigel 77 if (c >= 'a' && c <= 'z') c -= 32;
617     c ^= 0x40;
618 ph10 97 #else /* EBCDIC coding */
619 nigel 77 if (c >= 'a' && c <= 'z') c += 64;
620     c ^= 0xC0;
621     #endif
622     break;
623    
624     /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
625     other alphameric following \ is an error if PCRE_EXTRA was set; otherwise,
626     for Perl compatibility, it is a literal. This code looks a bit odd, but
627     there used to be some cases other than the default, and there may be again
628     in future, so I haven't "optimized" it. */
629    
630     default:
631     if ((options & PCRE_EXTRA) != 0) switch(c)
632     {
633     default:
634     *errorcodeptr = ERR3;
635     break;
636     }
637     break;
638     }
639     }
640    
641     *ptrptr = ptr;
642     return c;
643     }
644    
645    
646    
647     #ifdef SUPPORT_UCP
648     /*************************************************
649     * Handle \P and \p *
650     *************************************************/
651    
652     /* This function is called after \P or \p has been encountered, provided that
653     PCRE is compiled with support for Unicode properties. On entry, ptrptr is
654     pointing at the P or p. On exit, it is pointing at the final character of the
655     escape sequence.
656    
657     Argument:
658     ptrptr points to the pattern position pointer
659     negptr points to a boolean that is set TRUE for negation else FALSE
660 nigel 87 dptr points to an int that is set to the detailed property value
661 nigel 77 errorcodeptr points to the error code variable
662    
663 nigel 87 Returns: type value from ucp_type_table, or -1 for an invalid type
664 nigel 77 */
665    
666     static int
667 nigel 87 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
668 nigel 77 {
669     int c, i, bot, top;
670     const uschar *ptr = *ptrptr;
671 nigel 87 char name[32];
672 nigel 77
673     c = *(++ptr);
674     if (c == 0) goto ERROR_RETURN;
675    
676     *negptr = FALSE;
677    
678 nigel 87 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
679     negation. */
680 nigel 77
681     if (c == '{')
682     {
683     if (ptr[1] == '^')
684     {
685     *negptr = TRUE;
686     ptr++;
687     }
688 nigel 87 for (i = 0; i < sizeof(name) - 1; i++)
689 nigel 77 {
690     c = *(++ptr);
691     if (c == 0) goto ERROR_RETURN;
692     if (c == '}') break;
693     name[i] = c;
694     }
695 nigel 87 if (c !='}') goto ERROR_RETURN;
696 nigel 77 name[i] = 0;
697     }
698    
699     /* Otherwise there is just one following character */
700    
701     else
702     {
703     name[0] = c;
704     name[1] = 0;
705     }
706    
707     *ptrptr = ptr;
708    
709     /* Search for a recognized property name using binary chop */
710    
711     bot = 0;
712     top = _pcre_utt_size;
713    
714     while (bot < top)
715     {
716 nigel 87 i = (bot + top) >> 1;
717 nigel 77 c = strcmp(name, _pcre_utt[i].name);
718 nigel 87 if (c == 0)
719     {
720     *dptr = _pcre_utt[i].value;
721     return _pcre_utt[i].type;
722     }
723 nigel 77 if (c > 0) bot = i + 1; else top = i;
724     }
725    
726     *errorcodeptr = ERR47;
727     *ptrptr = ptr;
728     return -1;
729    
730     ERROR_RETURN:
731     *errorcodeptr = ERR46;
732     *ptrptr = ptr;
733     return -1;
734     }
735     #endif
736    
737    
738    
739    
740     /*************************************************
741     * Check for counted repeat *
742     *************************************************/
743    
744     /* This function is called when a '{' is encountered in a place where it might
745     start a quantifier. It looks ahead to see if it really is a quantifier or not.
746     It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
747     where the ddds are digits.
748    
749     Arguments:
750     p pointer to the first char after '{'
751    
752     Returns: TRUE or FALSE
753     */
754    
755     static BOOL
756     is_counted_repeat(const uschar *p)
757     {
758     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
759     while ((digitab[*p] & ctype_digit) != 0) p++;
760     if (*p == '}') return TRUE;
761    
762     if (*p++ != ',') return FALSE;
763     if (*p == '}') return TRUE;
764    
765     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
766     while ((digitab[*p] & ctype_digit) != 0) p++;
767    
768     return (*p == '}');
769     }
770    
771    
772    
773     /*************************************************
774     * Read repeat counts *
775     *************************************************/
776    
777     /* Read an item of the form {n,m} and return the values. This is called only
778     after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
779     so the syntax is guaranteed to be correct, but we need to check the values.
780    
781     Arguments:
782     p pointer to first char after '{'
783     minp pointer to int for min
784     maxp pointer to int for max
785     returned as -1 if no max
786     errorcodeptr points to error code variable
787    
788     Returns: pointer to '}' on success;
789     current ptr on error, with errorcodeptr set non-zero
790     */
791    
792     static const uschar *
793     read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
794     {
795     int min = 0;
796     int max = -1;
797    
798 nigel 81 /* Read the minimum value and do a paranoid check: a negative value indicates
799     an integer overflow. */
800    
801 nigel 77 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
802 nigel 81 if (min < 0 || min > 65535)
803     {
804     *errorcodeptr = ERR5;
805     return p;
806     }
807 nigel 77
808 nigel 81 /* Read the maximum value if there is one, and again do a paranoid on its size.
809     Also, max must not be less than min. */
810    
811 nigel 77 if (*p == '}') max = min; else
812     {
813     if (*(++p) != '}')
814     {
815     max = 0;
816     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
817 nigel 81 if (max < 0 || max > 65535)
818     {
819     *errorcodeptr = ERR5;
820     return p;
821     }
822 nigel 77 if (max < min)
823     {
824     *errorcodeptr = ERR4;
825     return p;
826     }
827     }
828     }
829    
830 nigel 81 /* Fill in the required variables, and pass back the pointer to the terminating
831     '}'. */
832 nigel 77
833 nigel 81 *minp = min;
834     *maxp = max;
835 nigel 77 return p;
836     }
837    
838    
839    
840     /*************************************************
841 nigel 93 * Find forward referenced subpattern *
842 nigel 91 *************************************************/
843    
844 nigel 93 /* This function scans along a pattern's text looking for capturing
845     subpatterns, and counting them. If it finds a named pattern that matches the
846     name it is given, it returns its number. Alternatively, if the name is NULL, it
847     returns when it reaches a given numbered subpattern. This is used for forward
848     references to subpatterns. We know that if (?P< is encountered, the name will
849     be terminated by '>' because that is checked in the first pass.
850 nigel 91
851     Arguments:
852 nigel 93 ptr current position in the pattern
853     count current count of capturing parens so far encountered
854     name name to seek, or NULL if seeking a numbered subpattern
855     lorn name length, or subpattern number if name is NULL
856     xmode TRUE if we are in /x mode
857 nigel 91
858     Returns: the number of the named subpattern, or -1 if not found
859     */
860    
861     static int
862 nigel 93 find_parens(const uschar *ptr, int count, const uschar *name, int lorn,
863     BOOL xmode)
864 nigel 91 {
865     const uschar *thisname;
866 nigel 93
867 nigel 91 for (; *ptr != 0; ptr++)
868     {
869 nigel 93 int term;
870    
871     /* Skip over backslashed characters and also entire \Q...\E */
872    
873     if (*ptr == '\\')
874     {
875     if (*(++ptr) == 0) return -1;
876     if (*ptr == 'Q') for (;;)
877     {
878     while (*(++ptr) != 0 && *ptr != '\\');
879     if (*ptr == 0) return -1;
880     if (*(++ptr) == 'E') break;
881     }
882     continue;
883     }
884    
885     /* Skip over character classes */
886    
887     if (*ptr == '[')
888     {
889     while (*(++ptr) != ']')
890     {
891     if (*ptr == '\\')
892     {
893     if (*(++ptr) == 0) return -1;
894     if (*ptr == 'Q') for (;;)
895     {
896     while (*(++ptr) != 0 && *ptr != '\\');
897     if (*ptr == 0) return -1;
898     if (*(++ptr) == 'E') break;
899     }
900     continue;
901     }
902     }
903     continue;
904     }
905    
906     /* Skip comments in /x mode */
907    
908     if (xmode && *ptr == '#')
909     {
910     while (*(++ptr) != 0 && *ptr != '\n');
911     if (*ptr == 0) return -1;
912     continue;
913     }
914    
915     /* An opening parens must now be a real metacharacter */
916    
917 nigel 91 if (*ptr != '(') continue;
918 nigel 93 if (ptr[1] != '?')
919     {
920     count++;
921     if (name == NULL && count == lorn) return count;
922     continue;
923     }
924    
925     ptr += 2;
926     if (*ptr == 'P') ptr++; /* Allow optional P */
927    
928     /* We have to disambiguate (?<! and (?<= from (?<name> */
929    
930     if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') &&
931     *ptr != '\'')
932     continue;
933    
934 nigel 91 count++;
935 nigel 93
936     if (name == NULL && count == lorn) return count;
937     term = *ptr++;
938     if (term == '<') term = '>';
939 nigel 91 thisname = ptr;
940 nigel 93 while (*ptr != term) ptr++;
941     if (name != NULL && lorn == ptr - thisname &&
942     strncmp((const char *)name, (const char *)thisname, lorn) == 0)
943 nigel 91 return count;
944     }
945 nigel 93
946 nigel 91 return -1;
947     }
948    
949    
950    
951     /*************************************************
952 nigel 77 * Find first significant op code *
953     *************************************************/
954    
955     /* This is called by several functions that scan a compiled expression looking
956     for a fixed first character, or an anchoring op code etc. It skips over things
957     that do not influence this. For some calls, a change of option is important.
958     For some calls, it makes sense to skip negative forward and all backward
959     assertions, and also the \b assertion; for others it does not.
960    
961     Arguments:
962     code pointer to the start of the group
963     options pointer to external options
964     optbit the option bit whose changing is significant, or
965     zero if none are
966     skipassert TRUE if certain assertions are to be skipped
967    
968     Returns: pointer to the first significant opcode
969     */
970    
971     static const uschar*
972     first_significant_code(const uschar *code, int *options, int optbit,
973     BOOL skipassert)
974     {
975     for (;;)
976     {
977     switch ((int)*code)
978     {
979     case OP_OPT:
980     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
981     *options = (int)code[1];
982     code += 2;
983     break;
984    
985     case OP_ASSERT_NOT:
986     case OP_ASSERTBACK:
987     case OP_ASSERTBACK_NOT:
988     if (!skipassert) return code;
989     do code += GET(code, 1); while (*code == OP_ALT);
990     code += _pcre_OP_lengths[*code];
991     break;
992    
993     case OP_WORD_BOUNDARY:
994     case OP_NOT_WORD_BOUNDARY:
995     if (!skipassert) return code;
996     /* Fall through */
997    
998     case OP_CALLOUT:
999     case OP_CREF:
1000 nigel 93 case OP_RREF:
1001     case OP_DEF:
1002 nigel 77 code += _pcre_OP_lengths[*code];
1003     break;
1004    
1005     default:
1006     return code;
1007     }
1008     }
1009     /* Control never reaches here */
1010     }
1011    
1012    
1013    
1014    
1015     /*************************************************
1016     * Find the fixed length of a pattern *
1017     *************************************************/
1018    
1019     /* Scan a pattern and compute the fixed length of subject that will match it,
1020     if the length is fixed. This is needed for dealing with backward assertions.
1021     In UTF8 mode, the result is in characters rather than bytes.
1022    
1023     Arguments:
1024     code points to the start of the pattern (the bracket)
1025     options the compiling options
1026    
1027     Returns: the fixed length, or -1 if there is no fixed length,
1028     or -2 if \C was encountered
1029     */
1030    
1031     static int
1032     find_fixedlength(uschar *code, int options)
1033     {
1034     int length = -1;
1035    
1036     register int branchlength = 0;
1037     register uschar *cc = code + 1 + LINK_SIZE;
1038    
1039     /* Scan along the opcodes for this branch. If we get to the end of the
1040     branch, check the length against that of the other branches. */
1041    
1042     for (;;)
1043     {
1044     int d;
1045     register int op = *cc;
1046    
1047     switch (op)
1048     {
1049 nigel 93 case OP_CBRA:
1050 nigel 77 case OP_BRA:
1051     case OP_ONCE:
1052     case OP_COND:
1053 nigel 93 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1054 nigel 77 if (d < 0) return d;
1055     branchlength += d;
1056     do cc += GET(cc, 1); while (*cc == OP_ALT);
1057     cc += 1 + LINK_SIZE;
1058     break;
1059    
1060     /* Reached end of a branch; if it's a ket it is the end of a nested
1061     call. If it's ALT it is an alternation in a nested call. If it is
1062     END it's the end of the outer call. All can be handled by the same code. */
1063    
1064     case OP_ALT:
1065     case OP_KET:
1066     case OP_KETRMAX:
1067     case OP_KETRMIN:
1068     case OP_END:
1069     if (length < 0) length = branchlength;
1070     else if (length != branchlength) return -1;
1071     if (*cc != OP_ALT) return length;
1072     cc += 1 + LINK_SIZE;
1073     branchlength = 0;
1074     break;
1075    
1076     /* Skip over assertive subpatterns */
1077    
1078     case OP_ASSERT:
1079     case OP_ASSERT_NOT:
1080     case OP_ASSERTBACK:
1081     case OP_ASSERTBACK_NOT:
1082     do cc += GET(cc, 1); while (*cc == OP_ALT);
1083     /* Fall through */
1084    
1085     /* Skip over things that don't match chars */
1086    
1087     case OP_REVERSE:
1088     case OP_CREF:
1089 nigel 93 case OP_RREF:
1090     case OP_DEF:
1091 nigel 77 case OP_OPT:
1092     case OP_CALLOUT:
1093     case OP_SOD:
1094     case OP_SOM:
1095     case OP_EOD:
1096     case OP_EODN:
1097     case OP_CIRC:
1098     case OP_DOLL:
1099     case OP_NOT_WORD_BOUNDARY:
1100     case OP_WORD_BOUNDARY:
1101     cc += _pcre_OP_lengths[*cc];
1102     break;
1103    
1104     /* Handle literal characters */
1105    
1106     case OP_CHAR:
1107     case OP_CHARNC:
1108 nigel 91 case OP_NOT:
1109 nigel 77 branchlength++;
1110     cc += 2;
1111     #ifdef SUPPORT_UTF8
1112     if ((options & PCRE_UTF8) != 0)
1113     {
1114     while ((*cc & 0xc0) == 0x80) cc++;
1115     }
1116     #endif
1117     break;
1118    
1119     /* Handle exact repetitions. The count is already in characters, but we
1120     need to skip over a multibyte character in UTF8 mode. */
1121    
1122     case OP_EXACT:
1123     branchlength += GET2(cc,1);
1124     cc += 4;
1125     #ifdef SUPPORT_UTF8
1126     if ((options & PCRE_UTF8) != 0)
1127     {
1128     while((*cc & 0x80) == 0x80) cc++;
1129     }
1130     #endif
1131     break;
1132    
1133     case OP_TYPEEXACT:
1134     branchlength += GET2(cc,1);
1135     cc += 4;
1136     break;
1137    
1138     /* Handle single-char matchers */
1139    
1140     case OP_PROP:
1141     case OP_NOTPROP:
1142 nigel 87 cc += 2;
1143 nigel 77 /* Fall through */
1144    
1145     case OP_NOT_DIGIT:
1146     case OP_DIGIT:
1147     case OP_NOT_WHITESPACE:
1148     case OP_WHITESPACE:
1149     case OP_NOT_WORDCHAR:
1150     case OP_WORDCHAR:
1151     case OP_ANY:
1152     branchlength++;
1153     cc++;
1154     break;
1155    
1156     /* The single-byte matcher isn't allowed */
1157    
1158     case OP_ANYBYTE:
1159     return -2;
1160    
1161     /* Check a class for variable quantification */
1162    
1163     #ifdef SUPPORT_UTF8
1164     case OP_XCLASS:
1165     cc += GET(cc, 1) - 33;
1166     /* Fall through */
1167     #endif
1168    
1169     case OP_CLASS:
1170     case OP_NCLASS:
1171     cc += 33;
1172    
1173     switch (*cc)
1174     {
1175     case OP_CRSTAR:
1176     case OP_CRMINSTAR:
1177     case OP_CRQUERY:
1178     case OP_CRMINQUERY:
1179     return -1;
1180    
1181     case OP_CRRANGE:
1182     case OP_CRMINRANGE:
1183     if (GET2(cc,1) != GET2(cc,3)) return -1;
1184     branchlength += GET2(cc,1);
1185     cc += 5;
1186     break;
1187    
1188     default:
1189     branchlength++;
1190     }
1191     break;
1192    
1193     /* Anything else is variable length */
1194    
1195     default:
1196     return -1;
1197     }
1198     }
1199     /* Control never gets here */
1200     }
1201    
1202    
1203    
1204    
1205     /*************************************************
1206     * Scan compiled regex for numbered bracket *
1207     *************************************************/
1208    
1209     /* This little function scans through a compiled pattern until it finds a
1210     capturing bracket with the given number.
1211    
1212     Arguments:
1213     code points to start of expression
1214     utf8 TRUE in UTF-8 mode
1215     number the required bracket number
1216    
1217     Returns: pointer to the opcode for the bracket, or NULL if not found
1218     */
1219    
1220     static const uschar *
1221     find_bracket(const uschar *code, BOOL utf8, int number)
1222     {
1223     for (;;)
1224     {
1225     register int c = *code;
1226     if (c == OP_END) return NULL;
1227 nigel 91
1228     /* XCLASS is used for classes that cannot be represented just by a bit
1229     map. This includes negated single high-valued characters. The length in
1230     the table is zero; the actual length is stored in the compiled code. */
1231    
1232     if (c == OP_XCLASS) code += GET(code, 1);
1233    
1234 nigel 93 /* Handle capturing bracket */
1235 nigel 91
1236 nigel 93 else if (c == OP_CBRA)
1237 nigel 77 {
1238 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1239 nigel 77 if (n == number) return (uschar *)code;
1240 nigel 93 code += _pcre_OP_lengths[c];
1241 nigel 77 }
1242 nigel 91
1243 nigel 93 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1244     a multi-byte character. The length in the table is a minimum, so we have to
1245     arrange to skip the extra bytes. */
1246 nigel 91
1247 nigel 77 else
1248     {
1249     code += _pcre_OP_lengths[c];
1250 ph10 107 #ifdef SUPPORT_UTF8
1251 nigel 77 if (utf8) switch(c)
1252     {
1253     case OP_CHAR:
1254     case OP_CHARNC:
1255     case OP_EXACT:
1256     case OP_UPTO:
1257     case OP_MINUPTO:
1258 nigel 93 case OP_POSUPTO:
1259 nigel 77 case OP_STAR:
1260     case OP_MINSTAR:
1261 nigel 93 case OP_POSSTAR:
1262 nigel 77 case OP_PLUS:
1263     case OP_MINPLUS:
1264 nigel 93 case OP_POSPLUS:
1265 nigel 77 case OP_QUERY:
1266     case OP_MINQUERY:
1267 nigel 93 case OP_POSQUERY:
1268     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1269 nigel 77 break;
1270     }
1271 ph10 111 #endif
1272 nigel 77 }
1273     }
1274     }
1275    
1276    
1277    
1278     /*************************************************
1279     * Scan compiled regex for recursion reference *
1280     *************************************************/
1281    
1282     /* This little function scans through a compiled pattern until it finds an
1283     instance of OP_RECURSE.
1284    
1285     Arguments:
1286     code points to start of expression
1287     utf8 TRUE in UTF-8 mode
1288    
1289     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1290     */
1291    
1292     static const uschar *
1293     find_recurse(const uschar *code, BOOL utf8)
1294     {
1295     for (;;)
1296     {
1297     register int c = *code;
1298     if (c == OP_END) return NULL;
1299 nigel 91 if (c == OP_RECURSE) return code;
1300    
1301     /* XCLASS is used for classes that cannot be represented just by a bit
1302     map. This includes negated single high-valued characters. The length in
1303     the table is zero; the actual length is stored in the compiled code. */
1304    
1305     if (c == OP_XCLASS) code += GET(code, 1);
1306    
1307     /* Otherwise, we get the item's length from the table. In UTF-8 mode, opcodes
1308     that are followed by a character may be followed by a multi-byte character.
1309 nigel 93 The length in the table is a minimum, so we have to arrange to skip the extra
1310     bytes. */
1311 nigel 91
1312 nigel 77 else
1313     {
1314     code += _pcre_OP_lengths[c];
1315 ph10 107 #ifdef SUPPORT_UTF8
1316 nigel 77 if (utf8) switch(c)
1317     {
1318     case OP_CHAR:
1319     case OP_CHARNC:
1320     case OP_EXACT:
1321     case OP_UPTO:
1322     case OP_MINUPTO:
1323 nigel 93 case OP_POSUPTO:
1324 nigel 77 case OP_STAR:
1325     case OP_MINSTAR:
1326 nigel 93 case OP_POSSTAR:
1327 nigel 77 case OP_PLUS:
1328     case OP_MINPLUS:
1329 nigel 93 case OP_POSPLUS:
1330 nigel 77 case OP_QUERY:
1331     case OP_MINQUERY:
1332 nigel 93 case OP_POSQUERY:
1333     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1334 nigel 77 break;
1335     }
1336 ph10 111 #endif
1337 nigel 77 }
1338     }
1339     }
1340    
1341    
1342    
1343     /*************************************************
1344     * Scan compiled branch for non-emptiness *
1345     *************************************************/
1346    
1347     /* This function scans through a branch of a compiled pattern to see whether it
1348 nigel 93 can match the empty string or not. It is called from could_be_empty()
1349     below and from compile_branch() when checking for an unlimited repeat of a
1350     group that can match nothing. Note that first_significant_code() skips over
1351     assertions. If we hit an unclosed bracket, we return "empty" - this means we've
1352     struck an inner bracket whose current branch will already have been scanned.
1353 nigel 77
1354     Arguments:
1355     code points to start of search
1356     endcode points to where to stop
1357     utf8 TRUE if in UTF8 mode
1358    
1359     Returns: TRUE if what is matched could be empty
1360     */
1361    
1362     static BOOL
1363     could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1364     {
1365     register int c;
1366 nigel 93 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1367 nigel 77 code < endcode;
1368     code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1369     {
1370     const uschar *ccode;
1371    
1372     c = *code;
1373    
1374 nigel 93 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE)
1375 nigel 77 {
1376     BOOL empty_branch;
1377     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1378    
1379     /* Scan a closed bracket */
1380    
1381     empty_branch = FALSE;
1382     do
1383     {
1384     if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1385     empty_branch = TRUE;
1386     code += GET(code, 1);
1387     }
1388     while (*code == OP_ALT);
1389     if (!empty_branch) return FALSE; /* All branches are non-empty */
1390 nigel 93
1391     /* Move past the KET and fudge things so that the increment in the "for"
1392     above has no effect. */
1393    
1394     c = OP_END;
1395     code += 1 + LINK_SIZE - _pcre_OP_lengths[c];
1396     continue;
1397 nigel 77 }
1398    
1399 nigel 93 /* Handle the other opcodes */
1400    
1401     switch (c)
1402 nigel 77 {
1403     /* Check for quantifiers after a class */
1404    
1405     #ifdef SUPPORT_UTF8
1406     case OP_XCLASS:
1407     ccode = code + GET(code, 1);
1408     goto CHECK_CLASS_REPEAT;
1409     #endif
1410    
1411     case OP_CLASS:
1412     case OP_NCLASS:
1413     ccode = code + 33;
1414    
1415     #ifdef SUPPORT_UTF8
1416     CHECK_CLASS_REPEAT:
1417     #endif
1418    
1419     switch (*ccode)
1420     {
1421     case OP_CRSTAR: /* These could be empty; continue */
1422     case OP_CRMINSTAR:
1423     case OP_CRQUERY:
1424     case OP_CRMINQUERY:
1425     break;
1426    
1427     default: /* Non-repeat => class must match */
1428     case OP_CRPLUS: /* These repeats aren't empty */
1429     case OP_CRMINPLUS:
1430     return FALSE;
1431    
1432     case OP_CRRANGE:
1433     case OP_CRMINRANGE:
1434     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1435     break;
1436     }
1437     break;
1438    
1439     /* Opcodes that must match a character */
1440    
1441     case OP_PROP:
1442     case OP_NOTPROP:
1443     case OP_EXTUNI:
1444     case OP_NOT_DIGIT:
1445     case OP_DIGIT:
1446     case OP_NOT_WHITESPACE:
1447     case OP_WHITESPACE:
1448     case OP_NOT_WORDCHAR:
1449     case OP_WORDCHAR:
1450     case OP_ANY:
1451     case OP_ANYBYTE:
1452     case OP_CHAR:
1453     case OP_CHARNC:
1454     case OP_NOT:
1455     case OP_PLUS:
1456     case OP_MINPLUS:
1457 nigel 93 case OP_POSPLUS:
1458 nigel 77 case OP_EXACT:
1459     case OP_NOTPLUS:
1460     case OP_NOTMINPLUS:
1461 nigel 93 case OP_NOTPOSPLUS:
1462 nigel 77 case OP_NOTEXACT:
1463     case OP_TYPEPLUS:
1464     case OP_TYPEMINPLUS:
1465 nigel 93 case OP_TYPEPOSPLUS:
1466 nigel 77 case OP_TYPEEXACT:
1467     return FALSE;
1468    
1469     /* End of branch */
1470    
1471     case OP_KET:
1472     case OP_KETRMAX:
1473     case OP_KETRMIN:
1474     case OP_ALT:
1475     return TRUE;
1476    
1477 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1478     MINUPTO, and POSUPTO may be followed by a multibyte character */
1479 nigel 77
1480     #ifdef SUPPORT_UTF8
1481     case OP_STAR:
1482     case OP_MINSTAR:
1483 nigel 93 case OP_POSSTAR:
1484 nigel 77 case OP_QUERY:
1485     case OP_MINQUERY:
1486 nigel 93 case OP_POSQUERY:
1487 nigel 77 case OP_UPTO:
1488     case OP_MINUPTO:
1489 nigel 93 case OP_POSUPTO:
1490 nigel 77 if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
1491     break;
1492     #endif
1493     }
1494     }
1495    
1496     return TRUE;
1497     }
1498    
1499    
1500    
1501     /*************************************************
1502     * Scan compiled regex for non-emptiness *
1503     *************************************************/
1504    
1505     /* This function is called to check for left recursive calls. We want to check
1506     the current branch of the current pattern to see if it could match the empty
1507     string. If it could, we must look outwards for branches at other levels,
1508     stopping when we pass beyond the bracket which is the subject of the recursion.
1509    
1510     Arguments:
1511     code points to start of the recursion
1512     endcode points to where to stop (current RECURSE item)
1513     bcptr points to the chain of current (unclosed) branch starts
1514     utf8 TRUE if in UTF-8 mode
1515    
1516     Returns: TRUE if what is matched could be empty
1517     */
1518    
1519     static BOOL
1520     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1521     BOOL utf8)
1522     {
1523     while (bcptr != NULL && bcptr->current >= code)
1524     {
1525     if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1526     bcptr = bcptr->outer;
1527     }
1528     return TRUE;
1529     }
1530    
1531    
1532    
1533     /*************************************************
1534     * Check for POSIX class syntax *
1535     *************************************************/
1536    
1537     /* This function is called when the sequence "[:" or "[." or "[=" is
1538     encountered in a character class. It checks whether this is followed by an
1539     optional ^ and then a sequence of letters, terminated by a matching ":]" or
1540     ".]" or "=]".
1541    
1542     Argument:
1543     ptr pointer to the initial [
1544     endptr where to return the end pointer
1545     cd pointer to compile data
1546    
1547     Returns: TRUE or FALSE
1548     */
1549    
1550     static BOOL
1551     check_posix_syntax(const uschar *ptr, const uschar **endptr, compile_data *cd)
1552     {
1553     int terminator; /* Don't combine these lines; the Solaris cc */
1554     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1555     if (*(++ptr) == '^') ptr++;
1556     while ((cd->ctypes[*ptr] & ctype_letter) != 0) ptr++;
1557     if (*ptr == terminator && ptr[1] == ']')
1558     {
1559     *endptr = ptr;
1560     return TRUE;
1561     }
1562     return FALSE;
1563     }
1564    
1565    
1566    
1567    
1568     /*************************************************
1569     * Check POSIX class name *
1570     *************************************************/
1571    
1572     /* This function is called to check the name given in a POSIX-style class entry
1573     such as [:alnum:].
1574    
1575     Arguments:
1576     ptr points to the first letter
1577     len the length of the name
1578    
1579     Returns: a value representing the name, or -1 if unknown
1580     */
1581    
1582     static int
1583     check_posix_name(const uschar *ptr, int len)
1584     {
1585     register int yield = 0;
1586     while (posix_name_lengths[yield] != 0)
1587     {
1588     if (len == posix_name_lengths[yield] &&
1589     strncmp((const char *)ptr, posix_names[yield], len) == 0) return yield;
1590     yield++;
1591     }
1592     return -1;
1593     }
1594    
1595    
1596     /*************************************************
1597     * Adjust OP_RECURSE items in repeated group *
1598     *************************************************/
1599    
1600     /* OP_RECURSE items contain an offset from the start of the regex to the group
1601     that is referenced. This means that groups can be replicated for fixed
1602     repetition simply by copying (because the recursion is allowed to refer to
1603     earlier groups that are outside the current group). However, when a group is
1604     optional (i.e. the minimum quantifier is zero), OP_BRAZERO is inserted before
1605     it, after it has been compiled. This means that any OP_RECURSE items within it
1606     that refer to the group itself or any contained groups have to have their
1607 nigel 93 offsets adjusted. That one of the jobs of this function. Before it is called,
1608     the partially compiled regex must be temporarily terminated with OP_END.
1609 nigel 77
1610 nigel 93 This function has been extended with the possibility of forward references for
1611     recursions and subroutine calls. It must also check the list of such references
1612     for the group we are dealing with. If it finds that one of the recursions in
1613     the current group is on this list, it adjusts the offset in the list, not the
1614     value in the reference (which is a group number).
1615    
1616 nigel 77 Arguments:
1617     group points to the start of the group
1618     adjust the amount by which the group is to be moved
1619     utf8 TRUE in UTF-8 mode
1620     cd contains pointers to tables etc.
1621 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
1622 nigel 77
1623     Returns: nothing
1624     */
1625    
1626     static void
1627 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
1628     uschar *save_hwm)
1629 nigel 77 {
1630     uschar *ptr = group;
1631     while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
1632     {
1633 nigel 93 int offset;
1634     uschar *hc;
1635    
1636     /* See if this recursion is on the forward reference list. If so, adjust the
1637     reference. */
1638    
1639     for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
1640     {
1641     offset = GET(hc, 0);
1642     if (cd->start_code + offset == ptr + 1)
1643     {
1644     PUT(hc, 0, offset + adjust);
1645     break;
1646     }
1647     }
1648    
1649     /* Otherwise, adjust the recursion offset if it's after the start of this
1650     group. */
1651    
1652     if (hc >= cd->hwm)
1653     {
1654     offset = GET(ptr, 1);
1655     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
1656     }
1657    
1658 nigel 77 ptr += 1 + LINK_SIZE;
1659     }
1660     }
1661    
1662    
1663    
1664     /*************************************************
1665     * Insert an automatic callout point *
1666     *************************************************/
1667    
1668     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
1669     callout points before each pattern item.
1670    
1671     Arguments:
1672     code current code pointer
1673     ptr current pattern pointer
1674     cd pointers to tables etc
1675    
1676     Returns: new code pointer
1677     */
1678    
1679     static uschar *
1680     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
1681     {
1682     *code++ = OP_CALLOUT;
1683     *code++ = 255;
1684     PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
1685     PUT(code, LINK_SIZE, 0); /* Default length */
1686     return code + 2*LINK_SIZE;
1687     }
1688    
1689    
1690    
1691     /*************************************************
1692     * Complete a callout item *
1693     *************************************************/
1694    
1695     /* A callout item contains the length of the next item in the pattern, which
1696     we can't fill in till after we have reached the relevant point. This is used
1697     for both automatic and manual callouts.
1698    
1699     Arguments:
1700     previous_callout points to previous callout item
1701     ptr current pattern pointer
1702     cd pointers to tables etc
1703    
1704     Returns: nothing
1705     */
1706    
1707     static void
1708     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
1709     {
1710     int length = ptr - cd->start_pattern - GET(previous_callout, 2);
1711     PUT(previous_callout, 2 + LINK_SIZE, length);
1712     }
1713    
1714    
1715    
1716     #ifdef SUPPORT_UCP
1717     /*************************************************
1718     * Get othercase range *
1719     *************************************************/
1720    
1721     /* This function is passed the start and end of a class range, in UTF-8 mode
1722     with UCP support. It searches up the characters, looking for internal ranges of
1723     characters in the "other" case. Each call returns the next one, updating the
1724     start address.
1725    
1726     Arguments:
1727     cptr points to starting character value; updated
1728     d end value
1729     ocptr where to put start of othercase range
1730     odptr where to put end of othercase range
1731    
1732     Yield: TRUE when range returned; FALSE when no more
1733     */
1734    
1735     static BOOL
1736 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
1737     unsigned int *odptr)
1738 nigel 77 {
1739 nigel 93 unsigned int c, othercase, next;
1740 nigel 77
1741     for (c = *cptr; c <= d; c++)
1742 nigel 93 { if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR) break; }
1743 nigel 77
1744     if (c > d) return FALSE;
1745    
1746     *ocptr = othercase;
1747     next = othercase + 1;
1748    
1749     for (++c; c <= d; c++)
1750     {
1751 nigel 87 if (_pcre_ucp_othercase(c) != next) break;
1752 nigel 77 next++;
1753     }
1754    
1755     *odptr = next - 1;
1756     *cptr = c;
1757    
1758     return TRUE;
1759     }
1760     #endif /* SUPPORT_UCP */
1761    
1762    
1763 nigel 93
1764 nigel 77 /*************************************************
1765 nigel 93 * Check if auto-possessifying is possible *
1766     *************************************************/
1767    
1768     /* This function is called for unlimited repeats of certain items, to see
1769     whether the next thing could possibly match the repeated item. If not, it makes
1770     sense to automatically possessify the repeated item.
1771    
1772     Arguments:
1773     op_code the repeated op code
1774     this data for this item, depends on the opcode
1775     utf8 TRUE in UTF-8 mode
1776     utf8_char used for utf8 character bytes, NULL if not relevant
1777     ptr next character in pattern
1778     options options bits
1779     cd contains pointers to tables etc.
1780    
1781     Returns: TRUE if possessifying is wanted
1782     */
1783    
1784     static BOOL
1785     check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
1786     const uschar *ptr, int options, compile_data *cd)
1787     {
1788     int next;
1789    
1790     /* Skip whitespace and comments in extended mode */
1791    
1792     if ((options & PCRE_EXTENDED) != 0)
1793     {
1794     for (;;)
1795     {
1796     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
1797     if (*ptr == '#')
1798     {
1799     while (*(++ptr) != 0)
1800     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
1801     }
1802     else break;
1803     }
1804     }
1805    
1806     /* If the next item is one that we can handle, get its value. A non-negative
1807     value is a character, a negative value is an escape value. */
1808    
1809     if (*ptr == '\\')
1810     {
1811     int temperrorcode = 0;
1812     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
1813     if (temperrorcode != 0) return FALSE;
1814     ptr++; /* Point after the escape sequence */
1815     }
1816    
1817     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
1818     {
1819     #ifdef SUPPORT_UTF8
1820     if (utf8) { GETCHARINC(next, ptr); } else
1821     #endif
1822     next = *ptr++;
1823     }
1824    
1825     else return FALSE;
1826    
1827     /* Skip whitespace and comments in extended mode */
1828    
1829     if ((options & PCRE_EXTENDED) != 0)
1830     {
1831     for (;;)
1832     {
1833     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
1834     if (*ptr == '#')
1835     {
1836     while (*(++ptr) != 0)
1837     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
1838     }
1839     else break;
1840     }
1841     }
1842    
1843     /* If the next thing is itself optional, we have to give up. */
1844    
1845     if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
1846     return FALSE;
1847    
1848     /* Now compare the next item with the previous opcode. If the previous is a
1849     positive single character match, "item" either contains the character or, if
1850     "item" is greater than 127 in utf8 mode, the character's bytes are in
1851     utf8_char. */
1852    
1853    
1854     /* Handle cases when the next item is a character. */
1855    
1856     if (next >= 0) switch(op_code)
1857     {
1858     case OP_CHAR:
1859     #ifdef SUPPORT_UTF8
1860     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
1861     #endif
1862     return item != next;
1863    
1864     /* For CHARNC (caseless character) we must check the other case. If we have
1865     Unicode property support, we can use it to test the other case of
1866     high-valued characters. */
1867    
1868     case OP_CHARNC:
1869     #ifdef SUPPORT_UTF8
1870     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
1871     #endif
1872     if (item == next) return FALSE;
1873     #ifdef SUPPORT_UTF8
1874     if (utf8)
1875     {
1876     unsigned int othercase;
1877     if (next < 128) othercase = cd->fcc[next]; else
1878     #ifdef SUPPORT_UCP
1879     othercase = _pcre_ucp_othercase((unsigned int)next);
1880     #else
1881     othercase = NOTACHAR;
1882     #endif
1883     return (unsigned int)item != othercase;
1884     }
1885     else
1886     #endif /* SUPPORT_UTF8 */
1887     return (item != cd->fcc[next]); /* Non-UTF-8 mode */
1888    
1889     /* For OP_NOT, "item" must be a single-byte character. */
1890    
1891     case OP_NOT:
1892     if (next < 0) return FALSE; /* Not a character */
1893     if (item == next) return TRUE;
1894     if ((options & PCRE_CASELESS) == 0) return FALSE;
1895     #ifdef SUPPORT_UTF8
1896     if (utf8)
1897     {
1898     unsigned int othercase;
1899     if (next < 128) othercase = cd->fcc[next]; else
1900     #ifdef SUPPORT_UCP
1901     othercase = _pcre_ucp_othercase(next);
1902     #else
1903     othercase = NOTACHAR;
1904     #endif
1905     return (unsigned int)item == othercase;
1906     }
1907     else
1908     #endif /* SUPPORT_UTF8 */
1909     return (item == cd->fcc[next]); /* Non-UTF-8 mode */
1910    
1911     case OP_DIGIT:
1912     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
1913    
1914     case OP_NOT_DIGIT:
1915     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
1916    
1917     case OP_WHITESPACE:
1918     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
1919    
1920     case OP_NOT_WHITESPACE:
1921     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
1922    
1923     case OP_WORDCHAR:
1924     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
1925    
1926     case OP_NOT_WORDCHAR:
1927     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
1928    
1929     default:
1930     return FALSE;
1931     }
1932    
1933    
1934     /* Handle the case when the next item is \d, \s, etc. */
1935    
1936     switch(op_code)
1937     {
1938     case OP_CHAR:
1939     case OP_CHARNC:
1940     #ifdef SUPPORT_UTF8
1941     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
1942     #endif
1943     switch(-next)
1944     {
1945     case ESC_d:
1946     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
1947    
1948     case ESC_D:
1949     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
1950    
1951     case ESC_s:
1952     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
1953    
1954     case ESC_S:
1955     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
1956    
1957     case ESC_w:
1958     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
1959    
1960     case ESC_W:
1961     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
1962    
1963     default:
1964     return FALSE;
1965     }
1966    
1967     case OP_DIGIT:
1968     return next == -ESC_D || next == -ESC_s || next == -ESC_W;
1969    
1970     case OP_NOT_DIGIT:
1971     return next == -ESC_d;
1972    
1973     case OP_WHITESPACE:
1974     return next == -ESC_S || next == -ESC_d || next == -ESC_w;
1975    
1976     case OP_NOT_WHITESPACE:
1977     return next == -ESC_s;
1978    
1979     case OP_WORDCHAR:
1980     return next == -ESC_W || next == -ESC_s;
1981    
1982     case OP_NOT_WORDCHAR:
1983     return next == -ESC_w || next == -ESC_d;
1984    
1985     default:
1986     return FALSE;
1987     }
1988    
1989     /* Control does not reach here */
1990     }
1991    
1992    
1993    
1994     /*************************************************
1995 nigel 77 * Compile one branch *
1996     *************************************************/
1997    
1998 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
1999 nigel 77 changed during the branch, the pointer is used to change the external options
2000 nigel 93 bits. This function is used during the pre-compile phase when we are trying
2001     to find out the amount of memory needed, as well as during the real compile
2002     phase. The value of lengthptr distinguishes the two phases.
2003 nigel 77
2004     Arguments:
2005     optionsptr pointer to the option bits
2006     codeptr points to the pointer to the current code point
2007     ptrptr points to the current pattern pointer
2008     errorcodeptr points to error code variable
2009     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2010     reqbyteptr set to the last literal character required, else < 0
2011     bcptr points to current branch chain
2012     cd contains pointers to tables etc.
2013 nigel 93 lengthptr NULL during the real compile phase
2014     points to length accumulator during pre-compile phase
2015 nigel 77
2016     Returns: TRUE on success
2017     FALSE, with *errorcodeptr set non-zero on error
2018     */
2019    
2020     static BOOL
2021 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2022     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2023     compile_data *cd, int *lengthptr)
2024 nigel 77 {
2025     int repeat_type, op_type;
2026     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2027     int bravalue = 0;
2028     int greedy_default, greedy_non_default;
2029     int firstbyte, reqbyte;
2030     int zeroreqbyte, zerofirstbyte;
2031     int req_caseopt, reqvary, tempreqvary;
2032     int options = *optionsptr;
2033     int after_manual_callout = 0;
2034 nigel 93 int length_prevgroup = 0;
2035 nigel 77 register int c;
2036     register uschar *code = *codeptr;
2037 nigel 93 uschar *last_code = code;
2038     uschar *orig_code = code;
2039 nigel 77 uschar *tempcode;
2040     BOOL inescq = FALSE;
2041     BOOL groupsetfirstbyte = FALSE;
2042     const uschar *ptr = *ptrptr;
2043     const uschar *tempptr;
2044     uschar *previous = NULL;
2045     uschar *previous_callout = NULL;
2046 nigel 93 uschar *save_hwm = NULL;
2047 nigel 77 uschar classbits[32];
2048    
2049     #ifdef SUPPORT_UTF8
2050     BOOL class_utf8;
2051     BOOL utf8 = (options & PCRE_UTF8) != 0;
2052     uschar *class_utf8data;
2053     uschar utf8_char[6];
2054     #else
2055     BOOL utf8 = FALSE;
2056 nigel 93 uschar *utf8_char = NULL;
2057 nigel 77 #endif
2058    
2059 nigel 93 #ifdef DEBUG
2060     if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2061     #endif
2062    
2063 nigel 77 /* Set up the default and non-default settings for greediness */
2064    
2065     greedy_default = ((options & PCRE_UNGREEDY) != 0);
2066     greedy_non_default = greedy_default ^ 1;
2067    
2068     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2069     matching encountered yet". It gets changed to REQ_NONE if we hit something that
2070     matches a non-fixed char first char; reqbyte just remains unset if we never
2071     find one.
2072    
2073     When we hit a repeat whose minimum is zero, we may have to adjust these values
2074     to take the zero repeat into account. This is implemented by setting them to
2075     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2076     item types that can be repeated set these backoff variables appropriately. */
2077    
2078     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2079    
2080     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2081     according to the current setting of the caseless flag. REQ_CASELESS is a bit
2082     value > 255. It is added into the firstbyte or reqbyte variables to record the
2083     case status of the value. This is used only for ASCII characters. */
2084    
2085     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2086    
2087     /* Switch on next character until the end of the branch */
2088    
2089     for (;; ptr++)
2090     {
2091     BOOL negate_class;
2092     BOOL possessive_quantifier;
2093     BOOL is_quantifier;
2094 nigel 93 BOOL is_recurse;
2095 nigel 77 int class_charcount;
2096     int class_lastchar;
2097     int newoptions;
2098     int recno;
2099 ph10 167 int refsign;
2100 nigel 77 int skipbytes;
2101     int subreqbyte;
2102     int subfirstbyte;
2103 nigel 93 int terminator;
2104 nigel 77 int mclength;
2105     uschar mcbuffer[8];
2106    
2107 nigel 93 /* Get next byte in the pattern */
2108 nigel 77
2109     c = *ptr;
2110    
2111 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
2112     previous cycle of this loop. */
2113    
2114     if (lengthptr != NULL)
2115     {
2116     #ifdef DEBUG
2117     if (code > cd->hwm) cd->hwm = code; /* High water info */
2118     #endif
2119     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2120     {
2121     *errorcodeptr = ERR52;
2122     goto FAILED;
2123     }
2124    
2125     /* There is at least one situation where code goes backwards: this is the
2126     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2127     the class is simply eliminated. However, it is created first, so we have to
2128     allow memory for it. Therefore, don't ever reduce the length at this point.
2129     */
2130    
2131     if (code < last_code) code = last_code;
2132     *lengthptr += code - last_code;
2133     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2134    
2135     /* If "previous" is set and it is not at the start of the work space, move
2136     it back to there, in order to avoid filling up the work space. Otherwise,
2137     if "previous" is NULL, reset the current code pointer to the start. */
2138    
2139     if (previous != NULL)
2140     {
2141     if (previous > orig_code)
2142     {
2143     memmove(orig_code, previous, code - previous);
2144     code -= previous - orig_code;
2145     previous = orig_code;
2146     }
2147     }
2148     else code = orig_code;
2149    
2150     /* Remember where this code item starts so we can pick up the length
2151     next time round. */
2152    
2153     last_code = code;
2154     }
2155    
2156     /* In the real compile phase, just check the workspace used by the forward
2157     reference list. */
2158    
2159     else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2160     {
2161     *errorcodeptr = ERR52;
2162     goto FAILED;
2163     }
2164    
2165 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
2166    
2167     if (inescq && c != 0)
2168     {
2169     if (c == '\\' && ptr[1] == 'E')
2170     {
2171     inescq = FALSE;
2172     ptr++;
2173     continue;
2174     }
2175     else
2176     {
2177     if (previous_callout != NULL)
2178     {
2179 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2180     complete_callout(previous_callout, ptr, cd);
2181 nigel 77 previous_callout = NULL;
2182     }
2183     if ((options & PCRE_AUTO_CALLOUT) != 0)
2184     {
2185     previous_callout = code;
2186     code = auto_callout(code, ptr, cd);
2187     }
2188     goto NORMAL_CHAR;
2189     }
2190     }
2191    
2192     /* Fill in length of a previous callout, except when the next thing is
2193     a quantifier. */
2194    
2195     is_quantifier = c == '*' || c == '+' || c == '?' ||
2196     (c == '{' && is_counted_repeat(ptr+1));
2197    
2198     if (!is_quantifier && previous_callout != NULL &&
2199     after_manual_callout-- <= 0)
2200     {
2201 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2202     complete_callout(previous_callout, ptr, cd);
2203 nigel 77 previous_callout = NULL;
2204     }
2205    
2206     /* In extended mode, skip white space and comments */
2207    
2208     if ((options & PCRE_EXTENDED) != 0)
2209     {
2210     if ((cd->ctypes[c] & ctype_space) != 0) continue;
2211     if (c == '#')
2212     {
2213 nigel 93 while (*(++ptr) != 0)
2214 nigel 91 {
2215 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2216 nigel 91 }
2217 nigel 93 if (*ptr != 0) continue;
2218    
2219 nigel 91 /* Else fall through to handle end of string */
2220     c = 0;
2221 nigel 77 }
2222     }
2223    
2224     /* No auto callout for quantifiers. */
2225    
2226     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2227     {
2228     previous_callout = code;
2229     code = auto_callout(code, ptr, cd);
2230     }
2231    
2232     switch(c)
2233     {
2234 nigel 93 /* ===================================================================*/
2235     case 0: /* The branch terminates at string end */
2236     case '|': /* or | or ) */
2237 nigel 77 case ')':
2238     *firstbyteptr = firstbyte;
2239     *reqbyteptr = reqbyte;
2240     *codeptr = code;
2241     *ptrptr = ptr;
2242 nigel 93 if (lengthptr != NULL)
2243     {
2244     *lengthptr += code - last_code; /* To include callout length */
2245     DPRINTF((">> end branch\n"));
2246     }
2247 nigel 77 return TRUE;
2248    
2249 nigel 93
2250     /* ===================================================================*/
2251 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
2252     the setting of any following char as a first character. */
2253    
2254     case '^':
2255     if ((options & PCRE_MULTILINE) != 0)
2256     {
2257     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2258     }
2259     previous = NULL;
2260     *code++ = OP_CIRC;
2261     break;
2262    
2263     case '$':
2264     previous = NULL;
2265     *code++ = OP_DOLL;
2266     break;
2267    
2268     /* There can never be a first char if '.' is first, whatever happens about
2269     repeats. The value of reqbyte doesn't change either. */
2270    
2271     case '.':
2272     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2273     zerofirstbyte = firstbyte;
2274     zeroreqbyte = reqbyte;
2275     previous = code;
2276     *code++ = OP_ANY;
2277     break;
2278    
2279 nigel 93
2280     /* ===================================================================*/
2281 nigel 87 /* Character classes. If the included characters are all < 256, we build a
2282     32-byte bitmap of the permitted characters, except in the special case
2283     where there is only one such character. For negated classes, we build the
2284     map as usual, then invert it at the end. However, we use a different opcode
2285     so that data characters > 255 can be handled correctly.
2286 nigel 77
2287     If the class contains characters outside the 0-255 range, a different
2288     opcode is compiled. It may optionally have a bit map for characters < 256,
2289     but those above are are explicitly listed afterwards. A flag byte tells
2290     whether the bitmap is present, and whether this is a negated class or not.
2291     */
2292    
2293     case '[':
2294     previous = code;
2295    
2296     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2297     they are encountered at the top level, so we'll do that too. */
2298    
2299     if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
2300     check_posix_syntax(ptr, &tempptr, cd))
2301     {
2302     *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
2303     goto FAILED;
2304     }
2305    
2306     /* If the first character is '^', set the negation flag and skip it. */
2307    
2308     if ((c = *(++ptr)) == '^')
2309     {
2310     negate_class = TRUE;
2311     c = *(++ptr);
2312     }
2313     else
2314     {
2315     negate_class = FALSE;
2316     }
2317    
2318     /* Keep a count of chars with values < 256 so that we can optimize the case
2319 nigel 93 of just a single character (as long as it's < 256). However, For higher
2320     valued UTF-8 characters, we don't yet do any optimization. */
2321 nigel 77
2322     class_charcount = 0;
2323     class_lastchar = -1;
2324    
2325 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
2326     temporary bit of memory, in case the class contains only 1 character (less
2327     than 256), because in that case the compiled code doesn't use the bit map.
2328     */
2329    
2330     memset(classbits, 0, 32 * sizeof(uschar));
2331    
2332 nigel 77 #ifdef SUPPORT_UTF8
2333     class_utf8 = FALSE; /* No chars >= 256 */
2334 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2335 nigel 77 #endif
2336    
2337     /* Process characters until ] is reached. By writing this as a "do" it
2338 nigel 93 means that an initial ] is taken as a data character. At the start of the
2339     loop, c contains the first byte of the character. */
2340 nigel 77
2341 nigel 93 if (c != 0) do
2342 nigel 77 {
2343 nigel 93 const uschar *oldptr;
2344    
2345 nigel 77 #ifdef SUPPORT_UTF8
2346     if (utf8 && c > 127)
2347     { /* Braces are required because the */
2348     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2349     }
2350     #endif
2351    
2352     /* Inside \Q...\E everything is literal except \E */
2353    
2354     if (inescq)
2355     {
2356 nigel 93 if (c == '\\' && ptr[1] == 'E') /* If we are at \E */
2357 nigel 77 {
2358 nigel 93 inescq = FALSE; /* Reset literal state */
2359     ptr++; /* Skip the 'E' */
2360     continue; /* Carry on with next */
2361 nigel 77 }
2362 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
2363 nigel 77 }
2364    
2365     /* Handle POSIX class names. Perl allows a negation extension of the
2366     form [:^name:]. A square bracket that doesn't match the syntax is
2367     treated as a literal. We also recognize the POSIX constructions
2368     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2369     5.6 and 5.8 do. */
2370    
2371     if (c == '[' &&
2372     (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
2373     check_posix_syntax(ptr, &tempptr, cd))
2374     {
2375     BOOL local_negate = FALSE;
2376 nigel 87 int posix_class, taboffset, tabopt;
2377 nigel 77 register const uschar *cbits = cd->cbits;
2378 nigel 87 uschar pbits[32];
2379 nigel 77
2380     if (ptr[1] != ':')
2381     {
2382     *errorcodeptr = ERR31;
2383     goto FAILED;
2384     }
2385    
2386     ptr += 2;
2387     if (*ptr == '^')
2388     {
2389     local_negate = TRUE;
2390     ptr++;
2391     }
2392    
2393     posix_class = check_posix_name(ptr, tempptr - ptr);
2394     if (posix_class < 0)
2395     {
2396     *errorcodeptr = ERR30;
2397     goto FAILED;
2398     }
2399    
2400     /* If matching is caseless, upper and lower are converted to
2401     alpha. This relies on the fact that the class table starts with
2402     alpha, lower, upper as the first 3 entries. */
2403    
2404     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
2405     posix_class = 0;
2406    
2407 nigel 87 /* We build the bit map for the POSIX class in a chunk of local store
2408     because we may be adding and subtracting from it, and we don't want to
2409     subtract bits that may be in the main map already. At the end we or the
2410     result into the bit map that is being built. */
2411 nigel 77
2412     posix_class *= 3;
2413 nigel 87
2414     /* Copy in the first table (always present) */
2415    
2416     memcpy(pbits, cbits + posix_class_maps[posix_class],
2417     32 * sizeof(uschar));
2418    
2419     /* If there is a second table, add or remove it as required. */
2420    
2421     taboffset = posix_class_maps[posix_class + 1];
2422     tabopt = posix_class_maps[posix_class + 2];
2423    
2424     if (taboffset >= 0)
2425 nigel 77 {
2426 nigel 87 if (tabopt >= 0)
2427     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
2428 nigel 77 else
2429 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
2430 nigel 77 }
2431    
2432 nigel 87 /* Not see if we need to remove any special characters. An option
2433     value of 1 removes vertical space and 2 removes underscore. */
2434    
2435     if (tabopt < 0) tabopt = -tabopt;
2436     if (tabopt == 1) pbits[1] &= ~0x3c;
2437     else if (tabopt == 2) pbits[11] &= 0x7f;
2438    
2439     /* Add the POSIX table or its complement into the main table that is
2440     being built and we are done. */
2441    
2442     if (local_negate)
2443     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
2444     else
2445     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
2446    
2447 nigel 77 ptr = tempptr + 1;
2448     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
2449     continue; /* End of POSIX syntax handling */
2450     }
2451    
2452     /* Backslash may introduce a single character, or it may introduce one
2453 nigel 93 of the specials, which just set a flag. The sequence \b is a special
2454     case. Inside a class (and only there) it is treated as backspace.
2455     Elsewhere it marks a word boundary. Other escapes have preset maps ready
2456     to or into the one we are building. We assume they have more than one
2457 nigel 77 character in them, so set class_charcount bigger than one. */
2458    
2459     if (c == '\\')
2460     {
2461 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
2462     if (*errorcodeptr != 0) goto FAILED;
2463 nigel 77
2464     if (-c == ESC_b) c = '\b'; /* \b is backslash in a class */
2465     else if (-c == ESC_X) c = 'X'; /* \X is literal X in a class */
2466 nigel 93 else if (-c == ESC_R) c = 'R'; /* \R is literal R in a class */
2467 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
2468     {
2469     if (ptr[1] == '\\' && ptr[2] == 'E')
2470     {
2471     ptr += 2; /* avoid empty string */
2472     }
2473     else inescq = TRUE;
2474     continue;
2475     }
2476    
2477     if (c < 0)
2478     {
2479     register const uschar *cbits = cd->cbits;
2480     class_charcount += 2; /* Greater than 1 is what matters */
2481 nigel 93
2482     /* Save time by not doing this in the pre-compile phase. */
2483    
2484     if (lengthptr == NULL) switch (-c)
2485 nigel 77 {
2486     case ESC_d:
2487     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
2488     continue;
2489    
2490     case ESC_D:
2491     for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
2492     continue;
2493    
2494     case ESC_w:
2495     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
2496     continue;
2497    
2498     case ESC_W:
2499     for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
2500     continue;
2501    
2502     case ESC_s:
2503     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
2504     classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
2505     continue;
2506    
2507     case ESC_S:
2508     for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
2509     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
2510     continue;
2511    
2512 nigel 93 case ESC_E: /* Perl ignores an orphan \E */
2513     continue;
2514    
2515     default: /* Not recognized; fall through */
2516     break; /* Need "default" setting to stop compiler warning. */
2517     }
2518    
2519     /* In the pre-compile phase, just do the recognition. */
2520    
2521     else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
2522     c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
2523    
2524     /* We need to deal with \P and \p in both phases. */
2525    
2526 nigel 77 #ifdef SUPPORT_UCP
2527 nigel 93 if (-c == ESC_p || -c == ESC_P)
2528     {
2529     BOOL negated;
2530     int pdata;
2531     int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
2532     if (ptype < 0) goto FAILED;
2533     class_utf8 = TRUE;
2534     *class_utf8data++ = ((-c == ESC_p) != negated)?
2535     XCL_PROP : XCL_NOTPROP;
2536     *class_utf8data++ = ptype;
2537     *class_utf8data++ = pdata;
2538     class_charcount -= 2; /* Not a < 256 character */
2539 nigel 77 continue;
2540 nigel 93 }
2541 nigel 77 #endif
2542 nigel 93 /* Unrecognized escapes are faulted if PCRE is running in its
2543     strict mode. By default, for compatibility with Perl, they are
2544     treated as literals. */
2545 nigel 77
2546 nigel 93 if ((options & PCRE_EXTRA) != 0)
2547     {
2548     *errorcodeptr = ERR7;
2549     goto FAILED;
2550     }
2551 nigel 77
2552 nigel 93 class_charcount -= 2; /* Undo the default count from above */
2553     c = *ptr; /* Get the final character and fall through */
2554 nigel 77 }
2555    
2556     /* Fall through if we have a single character (c >= 0). This may be
2557 nigel 93 greater than 256 in UTF-8 mode. */
2558 nigel 77
2559     } /* End of backslash handling */
2560    
2561     /* A single character may be followed by '-' to form a range. However,
2562     Perl does not permit ']' to be the end of the range. A '-' character
2563 nigel 93 at the end is treated as a literal. Perl ignores orphaned \E sequences
2564     entirely. The code for handling \Q and \E is messy. */
2565 nigel 77
2566 nigel 93 CHECK_RANGE:
2567     while (ptr[1] == '\\' && ptr[2] == 'E')
2568 nigel 77 {
2569 nigel 93 inescq = FALSE;
2570     ptr += 2;
2571     }
2572    
2573     oldptr = ptr;
2574    
2575     if (!inescq && ptr[1] == '-')
2576     {
2577 nigel 77 int d;
2578     ptr += 2;
2579 nigel 93 while (*ptr == '\\' && ptr[1] == 'E') ptr += 2;
2580 nigel 77
2581 nigel 93 /* If we hit \Q (not followed by \E) at this point, go into escaped
2582     mode. */
2583    
2584     while (*ptr == '\\' && ptr[1] == 'Q')
2585     {
2586     ptr += 2;
2587     if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; }
2588     inescq = TRUE;
2589     break;
2590     }
2591    
2592     if (*ptr == 0 || (!inescq && *ptr == ']'))
2593     {
2594     ptr = oldptr;
2595     goto LONE_SINGLE_CHARACTER;
2596     }
2597    
2598 nigel 77 #ifdef SUPPORT_UTF8
2599     if (utf8)
2600     { /* Braces are required because the */
2601     GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
2602     }
2603     else
2604     #endif
2605     d = *ptr; /* Not UTF-8 mode */
2606    
2607     /* The second part of a range can be a single-character escape, but
2608     not any of the other escapes. Perl 5.6 treats a hyphen as a literal
2609     in such circumstances. */
2610    
2611 nigel 93 if (!inescq && d == '\\')
2612 nigel 77 {
2613 nigel 93 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
2614     if (*errorcodeptr != 0) goto FAILED;
2615 nigel 77
2616 nigel 93 /* \b is backslash; \X is literal X; \R is literal R; any other
2617     special means the '-' was literal */
2618 nigel 77
2619     if (d < 0)
2620     {
2621     if (d == -ESC_b) d = '\b';
2622 nigel 93 else if (d == -ESC_X) d = 'X';
2623     else if (d == -ESC_R) d = 'R'; else
2624 nigel 77 {
2625 nigel 93 ptr = oldptr;
2626 nigel 77 goto LONE_SINGLE_CHARACTER; /* A few lines below */
2627     }
2628     }
2629     }
2630    
2631 nigel 93 /* Check that the two values are in the correct order. Optimize
2632     one-character ranges */
2633 nigel 77
2634 nigel 93 if (d < c)
2635     {
2636     *errorcodeptr = ERR8;
2637     goto FAILED;
2638     }
2639    
2640 nigel 77 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
2641    
2642     /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
2643     matching, we have to use an XCLASS with extra data items. Caseless
2644     matching for characters > 127 is available only if UCP support is
2645     available. */
2646    
2647     #ifdef SUPPORT_UTF8
2648     if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
2649     {
2650     class_utf8 = TRUE;
2651    
2652     /* With UCP support, we can find the other case equivalents of
2653     the relevant characters. There may be several ranges. Optimize how
2654     they fit with the basic range. */
2655    
2656     #ifdef SUPPORT_UCP
2657     if ((options & PCRE_CASELESS) != 0)
2658     {
2659 nigel 93 unsigned int occ, ocd;
2660     unsigned int cc = c;
2661     unsigned int origd = d;
2662 nigel 77 while (get_othercase_range(&cc, origd, &occ, &ocd))
2663     {
2664     if (occ >= c && ocd <= d) continue; /* Skip embedded ranges */
2665    
2666     if (occ < c && ocd >= c - 1) /* Extend the basic range */
2667     { /* if there is overlap, */
2668     c = occ; /* noting that if occ < c */
2669     continue; /* we can't have ocd > d */
2670     } /* because a subrange is */
2671     if (ocd > d && occ <= d + 1) /* always shorter than */
2672     { /* the basic range. */
2673     d = ocd;
2674     continue;
2675     }
2676    
2677     if (occ == ocd)
2678     {
2679     *class_utf8data++ = XCL_SINGLE;
2680     }
2681     else
2682     {
2683     *class_utf8data++ = XCL_RANGE;
2684     class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
2685     }
2686     class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
2687     }
2688     }
2689     #endif /* SUPPORT_UCP */
2690    
2691     /* Now record the original range, possibly modified for UCP caseless
2692     overlapping ranges. */
2693    
2694     *class_utf8data++ = XCL_RANGE;
2695     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
2696     class_utf8data += _pcre_ord2utf8(d, class_utf8data);
2697    
2698     /* With UCP support, we are done. Without UCP support, there is no
2699     caseless matching for UTF-8 characters > 127; we can use the bit map
2700     for the smaller ones. */
2701    
2702     #ifdef SUPPORT_UCP
2703     continue; /* With next character in the class */
2704     #else
2705     if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
2706    
2707     /* Adjust upper limit and fall through to set up the map */
2708    
2709     d = 127;
2710    
2711     #endif /* SUPPORT_UCP */
2712     }
2713     #endif /* SUPPORT_UTF8 */
2714    
2715     /* We use the bit map for all cases when not in UTF-8 mode; else
2716     ranges that lie entirely within 0-127 when there is UCP support; else
2717     for partial ranges without UCP support. */
2718    
2719 nigel 93 class_charcount += d - c + 1;
2720     class_lastchar = d;
2721    
2722     /* We can save a bit of time by skipping this in the pre-compile. */
2723    
2724     if (lengthptr == NULL) for (; c <= d; c++)
2725 nigel 77 {
2726     classbits[c/8] |= (1 << (c&7));
2727     if ((options & PCRE_CASELESS) != 0)
2728     {
2729     int uc = cd->fcc[c]; /* flip case */
2730     classbits[uc/8] |= (1 << (uc&7));
2731     }
2732     }
2733    
2734     continue; /* Go get the next char in the class */
2735     }
2736    
2737     /* Handle a lone single character - we can get here for a normal
2738     non-escape char, or after \ that introduces a single character or for an
2739     apparent range that isn't. */
2740    
2741     LONE_SINGLE_CHARACTER:
2742    
2743     /* Handle a character that cannot go in the bit map */
2744    
2745     #ifdef SUPPORT_UTF8
2746     if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
2747     {
2748     class_utf8 = TRUE;
2749     *class_utf8data++ = XCL_SINGLE;
2750     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
2751    
2752     #ifdef SUPPORT_UCP
2753     if ((options & PCRE_CASELESS) != 0)
2754     {
2755 nigel 93 unsigned int othercase;
2756     if ((othercase = _pcre_ucp_othercase(c)) != NOTACHAR)
2757 nigel 77 {
2758     *class_utf8data++ = XCL_SINGLE;
2759     class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
2760     }
2761     }
2762     #endif /* SUPPORT_UCP */
2763    
2764     }
2765     else
2766     #endif /* SUPPORT_UTF8 */
2767    
2768     /* Handle a single-byte character */
2769     {
2770     classbits[c/8] |= (1 << (c&7));
2771     if ((options & PCRE_CASELESS) != 0)
2772     {
2773     c = cd->fcc[c]; /* flip case */
2774     classbits[c/8] |= (1 << (c&7));
2775     }
2776     class_charcount++;
2777     class_lastchar = c;
2778     }
2779     }
2780    
2781 nigel 93 /* Loop until ']' reached. This "while" is the end of the "do" above. */
2782 nigel 77
2783 nigel 93 while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
2784 nigel 77
2785 nigel 93 if (c == 0) /* Missing terminating ']' */
2786     {
2787     *errorcodeptr = ERR6;
2788     goto FAILED;
2789     }
2790    
2791 nigel 77 /* If class_charcount is 1, we saw precisely one character whose value is
2792     less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we
2793     can optimize the negative case only if there were no characters >= 128
2794     because OP_NOT and the related opcodes like OP_NOTSTAR operate on
2795     single-bytes only. This is an historical hangover. Maybe one day we can
2796     tidy these opcodes to handle multi-byte characters.
2797    
2798     The optimization throws away the bit map. We turn the item into a
2799     1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
2800     that OP_NOT does not support multibyte characters. In the positive case, it
2801     can cause firstbyte to be set. Otherwise, there can be no first char if
2802     this item is first, whatever repeat count may follow. In the case of
2803     reqbyte, save the previous value for reinstating. */
2804    
2805     #ifdef SUPPORT_UTF8
2806     if (class_charcount == 1 &&
2807     (!utf8 ||
2808     (!class_utf8 && (!negate_class || class_lastchar < 128))))
2809    
2810     #else
2811     if (class_charcount == 1)
2812     #endif
2813     {
2814     zeroreqbyte = reqbyte;
2815    
2816     /* The OP_NOT opcode works on one-byte characters only. */
2817    
2818     if (negate_class)
2819     {
2820     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2821     zerofirstbyte = firstbyte;
2822     *code++ = OP_NOT;
2823     *code++ = class_lastchar;
2824     break;
2825     }
2826    
2827     /* For a single, positive character, get the value into mcbuffer, and
2828     then we can handle this with the normal one-character code. */
2829    
2830     #ifdef SUPPORT_UTF8
2831     if (utf8 && class_lastchar > 127)
2832     mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
2833     else
2834     #endif
2835     {
2836     mcbuffer[0] = class_lastchar;
2837     mclength = 1;
2838     }
2839     goto ONE_CHAR;
2840     } /* End of 1-char optimization */
2841    
2842     /* The general case - not the one-char optimization. If this is the first
2843     thing in the branch, there can be no first char setting, whatever the
2844     repeat count. Any reqbyte setting must remain unchanged after any kind of
2845     repeat. */
2846    
2847     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2848     zerofirstbyte = firstbyte;
2849     zeroreqbyte = reqbyte;
2850    
2851     /* If there are characters with values > 255, we have to compile an
2852     extended class, with its own opcode. If there are no characters < 256,
2853 nigel 93 we can omit the bitmap in the actual compiled code. */
2854 nigel 77
2855     #ifdef SUPPORT_UTF8
2856     if (class_utf8)
2857     {
2858     *class_utf8data++ = XCL_END; /* Marks the end of extra data */
2859     *code++ = OP_XCLASS;
2860     code += LINK_SIZE;
2861     *code = negate_class? XCL_NOT : 0;
2862    
2863 nigel 93 /* If the map is required, move up the extra data to make room for it;
2864     otherwise just move the code pointer to the end of the extra data. */
2865 nigel 77
2866     if (class_charcount > 0)
2867     {
2868     *code++ |= XCL_MAP;
2869 nigel 93 memmove(code + 32, code, class_utf8data - code);
2870 nigel 77 memcpy(code, classbits, 32);
2871 nigel 93 code = class_utf8data + 32;
2872 nigel 77 }
2873 nigel 93 else code = class_utf8data;
2874 nigel 77
2875     /* Now fill in the complete length of the item */
2876    
2877     PUT(previous, 1, code - previous);
2878     break; /* End of class handling */
2879     }
2880     #endif
2881    
2882     /* If there are no characters > 255, negate the 32-byte map if necessary,
2883     and copy it into the code vector. If this is the first thing in the branch,
2884     there can be no first char setting, whatever the repeat count. Any reqbyte
2885     setting must remain unchanged after any kind of repeat. */
2886    
2887     if (negate_class)
2888     {
2889     *code++ = OP_NCLASS;
2890 nigel 93 if (lengthptr == NULL) /* Save time in the pre-compile phase */
2891     for (c = 0; c < 32; c++) code[c] = ~classbits[c];
2892 nigel 77 }
2893     else
2894     {
2895     *code++ = OP_CLASS;
2896     memcpy(code, classbits, 32);
2897     }
2898     code += 32;
2899     break;
2900    
2901 nigel 93
2902     /* ===================================================================*/
2903 nigel 77 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
2904     has been tested above. */
2905    
2906     case '{':
2907     if (!is_quantifier) goto NORMAL_CHAR;
2908     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
2909     if (*errorcodeptr != 0) goto FAILED;
2910     goto REPEAT;
2911    
2912     case '*':
2913     repeat_min = 0;
2914     repeat_max = -1;
2915     goto REPEAT;
2916    
2917     case '+':
2918     repeat_min = 1;
2919     repeat_max = -1;
2920     goto REPEAT;
2921    
2922     case '?':
2923     repeat_min = 0;
2924     repeat_max = 1;
2925    
2926     REPEAT:
2927     if (previous == NULL)
2928     {
2929     *errorcodeptr = ERR9;
2930     goto FAILED;
2931     }
2932    
2933     if (repeat_min == 0)
2934     {
2935     firstbyte = zerofirstbyte; /* Adjust for zero repeat */
2936     reqbyte = zeroreqbyte; /* Ditto */
2937     }
2938    
2939     /* Remember whether this is a variable length repeat */
2940    
2941     reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
2942    
2943     op_type = 0; /* Default single-char op codes */
2944     possessive_quantifier = FALSE; /* Default not possessive quantifier */
2945    
2946     /* Save start of previous item, in case we have to move it up to make space
2947     for an inserted OP_ONCE for the additional '+' extension. */
2948    
2949     tempcode = previous;
2950    
2951     /* If the next character is '+', we have a possessive quantifier. This
2952     implies greediness, whatever the setting of the PCRE_UNGREEDY option.
2953     If the next character is '?' this is a minimizing repeat, by default,
2954     but if PCRE_UNGREEDY is set, it works the other way round. We change the
2955     repeat type to the non-default. */
2956    
2957     if (ptr[1] == '+')
2958     {
2959     repeat_type = 0; /* Force greedy */
2960     possessive_quantifier = TRUE;
2961     ptr++;
2962     }
2963     else if (ptr[1] == '?')
2964     {
2965     repeat_type = greedy_non_default;
2966     ptr++;
2967     }
2968     else repeat_type = greedy_default;
2969    
2970     /* If previous was a character match, abolish the item and generate a
2971     repeat item instead. If a char item has a minumum of more than one, ensure
2972     that it is set in reqbyte - it might not be if a sequence such as x{3} is
2973     the first thing in a branch because the x will have gone into firstbyte
2974     instead. */
2975    
2976     if (*previous == OP_CHAR || *previous == OP_CHARNC)
2977     {
2978     /* Deal with UTF-8 characters that take up more than one byte. It's
2979     easier to write this out separately than try to macrify it. Use c to
2980     hold the length of the character in bytes, plus 0x80 to flag that it's a
2981     length rather than a small character. */
2982    
2983     #ifdef SUPPORT_UTF8
2984     if (utf8 && (code[-1] & 0x80) != 0)
2985     {
2986     uschar *lastchar = code - 1;
2987     while((*lastchar & 0xc0) == 0x80) lastchar--;
2988     c = code - lastchar; /* Length of UTF-8 character */
2989     memcpy(utf8_char, lastchar, c); /* Save the char */
2990     c |= 0x80; /* Flag c as a length */
2991     }
2992     else
2993     #endif
2994    
2995     /* Handle the case of a single byte - either with no UTF8 support, or
2996     with UTF-8 disabled, or for a UTF-8 character < 128. */
2997    
2998     {
2999     c = code[-1];
3000     if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3001     }
3002    
3003 nigel 93 /* If the repetition is unlimited, it pays to see if the next thing on
3004     the line is something that cannot possibly match this character. If so,
3005     automatically possessifying this item gains some performance in the case
3006     where the match fails. */
3007    
3008     if (!possessive_quantifier &&
3009     repeat_max < 0 &&
3010     check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3011     options, cd))
3012     {
3013     repeat_type = 0; /* Force greedy */
3014     possessive_quantifier = TRUE;
3015     }
3016    
3017 nigel 77 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3018     }
3019    
3020     /* If previous was a single negated character ([^a] or similar), we use
3021     one of the special opcodes, replacing it. The code is shared with single-
3022     character repeats by setting opt_type to add a suitable offset into
3023 nigel 93 repeat_type. We can also test for auto-possessification. OP_NOT is
3024     currently used only for single-byte chars. */
3025 nigel 77
3026     else if (*previous == OP_NOT)
3027     {
3028     op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3029     c = previous[1];
3030 nigel 93 if (!possessive_quantifier &&
3031     repeat_max < 0 &&
3032     check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3033     {
3034     repeat_type = 0; /* Force greedy */
3035     possessive_quantifier = TRUE;
3036     }
3037 nigel 77 goto OUTPUT_SINGLE_REPEAT;
3038     }
3039    
3040     /* If previous was a character type match (\d or similar), abolish it and
3041     create a suitable repeat item. The code is shared with single-character
3042     repeats by setting op_type to add a suitable offset into repeat_type. Note
3043     the the Unicode property types will be present only when SUPPORT_UCP is
3044     defined, but we don't wrap the little bits of code here because it just
3045     makes it horribly messy. */
3046    
3047     else if (*previous < OP_EODN)
3048     {
3049     uschar *oldcode;
3050 nigel 87 int prop_type, prop_value;
3051 nigel 77 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3052     c = *previous;
3053    
3054 nigel 93 if (!possessive_quantifier &&
3055     repeat_max < 0 &&
3056     check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3057     {
3058     repeat_type = 0; /* Force greedy */
3059     possessive_quantifier = TRUE;
3060     }
3061    
3062 nigel 77 OUTPUT_SINGLE_REPEAT:
3063 nigel 87 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3064     {
3065     prop_type = previous[1];
3066     prop_value = previous[2];
3067     }
3068     else prop_type = prop_value = -1;
3069 nigel 77
3070     oldcode = code;
3071     code = previous; /* Usually overwrite previous item */
3072    
3073     /* If the maximum is zero then the minimum must also be zero; Perl allows
3074     this case, so we do too - by simply omitting the item altogether. */
3075    
3076     if (repeat_max == 0) goto END_REPEAT;
3077    
3078     /* All real repeats make it impossible to handle partial matching (maybe
3079     one day we will be able to remove this restriction). */
3080    
3081     if (repeat_max != 1) cd->nopartial = TRUE;
3082    
3083     /* Combine the op_type with the repeat_type */
3084    
3085     repeat_type += op_type;
3086    
3087     /* A minimum of zero is handled either as the special case * or ?, or as
3088     an UPTO, with the maximum given. */
3089    
3090     if (repeat_min == 0)
3091     {
3092     if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3093     else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3094     else
3095     {
3096     *code++ = OP_UPTO + repeat_type;
3097     PUT2INC(code, 0, repeat_max);
3098     }
3099     }
3100    
3101     /* A repeat minimum of 1 is optimized into some special cases. If the
3102 nigel 93 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3103 nigel 77 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3104     one less than the maximum. */
3105    
3106     else if (repeat_min == 1)
3107     {
3108     if (repeat_max == -1)
3109     *code++ = OP_PLUS + repeat_type;
3110     else
3111     {
3112     code = oldcode; /* leave previous item in place */
3113     if (repeat_max == 1) goto END_REPEAT;
3114     *code++ = OP_UPTO + repeat_type;
3115     PUT2INC(code, 0, repeat_max - 1);
3116     }
3117     }
3118    
3119     /* The case {n,n} is just an EXACT, while the general case {n,m} is
3120     handled as an EXACT followed by an UPTO. */
3121    
3122     else
3123     {
3124     *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3125     PUT2INC(code, 0, repeat_min);
3126    
3127     /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3128     we have to insert the character for the previous code. For a repeated
3129 nigel 87 Unicode property match, there are two extra bytes that define the
3130 nigel 77 required property. In UTF-8 mode, long characters have their length in
3131     c, with the 0x80 bit as a flag. */
3132    
3133     if (repeat_max < 0)
3134     {
3135     #ifdef SUPPORT_UTF8
3136     if (utf8 && c >= 128)
3137     {
3138     memcpy(code, utf8_char, c & 7);
3139     code += c & 7;
3140     }
3141     else
3142     #endif
3143     {
3144     *code++ = c;
3145 nigel 87 if (prop_type >= 0)
3146     {
3147     *code++ = prop_type;
3148     *code++ = prop_value;
3149     }
3150 nigel 77 }
3151     *code++ = OP_STAR + repeat_type;
3152     }
3153    
3154     /* Else insert an UPTO if the max is greater than the min, again
3155 nigel 93 preceded by the character, for the previously inserted code. If the
3156     UPTO is just for 1 instance, we can use QUERY instead. */
3157 nigel 77
3158     else if (repeat_max != repeat_min)
3159     {
3160     #ifdef SUPPORT_UTF8
3161     if (utf8 && c >= 128)
3162     {
3163     memcpy(code, utf8_char, c & 7);
3164     code += c & 7;
3165     }
3166     else
3167     #endif
3168     *code++ = c;
3169 nigel 87 if (prop_type >= 0)
3170     {
3171     *code++ = prop_type;
3172     *code++ = prop_value;
3173     }
3174 nigel 77 repeat_max -= repeat_min;
3175 nigel 93
3176     if (repeat_max == 1)
3177     {
3178     *code++ = OP_QUERY + repeat_type;
3179     }
3180     else
3181     {
3182     *code++ = OP_UPTO + repeat_type;
3183     PUT2INC(code, 0, repeat_max);
3184     }
3185 nigel 77 }
3186     }
3187    
3188     /* The character or character type itself comes last in all cases. */
3189    
3190     #ifdef SUPPORT_UTF8
3191     if (utf8 && c >= 128)
3192     {
3193     memcpy(code, utf8_char, c & 7);
3194     code += c & 7;
3195     }
3196     else
3197     #endif
3198     *code++ = c;
3199    
3200 nigel 87 /* For a repeated Unicode property match, there are two extra bytes that
3201     define the required property. */
3202 nigel 77
3203     #ifdef SUPPORT_UCP
3204 nigel 87 if (prop_type >= 0)
3205     {
3206     *code++ = prop_type;
3207     *code++ = prop_value;
3208     }
3209 nigel 77 #endif
3210     }
3211    
3212     /* If previous was a character class or a back reference, we put the repeat
3213     stuff after it, but just skip the item if the repeat was {0,0}. */
3214    
3215     else if (*previous == OP_CLASS ||
3216     *previous == OP_NCLASS ||
3217     #ifdef SUPPORT_UTF8
3218     *previous == OP_XCLASS ||
3219     #endif
3220     *previous == OP_REF)
3221     {
3222     if (repeat_max == 0)
3223     {
3224     code = previous;
3225     goto END_REPEAT;
3226     }
3227    
3228     /* All real repeats make it impossible to handle partial matching (maybe
3229     one day we will be able to remove this restriction). */
3230    
3231     if (repeat_max != 1) cd->nopartial = TRUE;
3232    
3233     if (repeat_min == 0 && repeat_max == -1)
3234     *code++ = OP_CRSTAR + repeat_type;
3235     else if (repeat_min == 1 && repeat_max == -1)
3236     *code++ = OP_CRPLUS + repeat_type;
3237     else if (repeat_min == 0 && repeat_max == 1)
3238     *code++ = OP_CRQUERY + repeat_type;
3239     else
3240     {
3241     *code++ = OP_CRRANGE + repeat_type;
3242     PUT2INC(code, 0, repeat_min);
3243     if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
3244     PUT2INC(code, 0, repeat_max);
3245     }
3246     }
3247    
3248     /* If previous was a bracket group, we may have to replicate it in certain
3249     cases. */
3250    
3251 nigel 93 else if (*previous == OP_BRA || *previous == OP_CBRA ||
3252     *previous == OP_ONCE || *previous == OP_COND)
3253 nigel 77 {
3254     register int i;
3255     int ketoffset = 0;
3256     int len = code - previous;
3257     uschar *bralink = NULL;
3258    
3259 nigel 93 /* Repeating a DEFINE group is pointless */
3260    
3261     if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
3262     {
3263     *errorcodeptr = ERR55;
3264     goto FAILED;
3265     }
3266    
3267     /* This is a paranoid check to stop integer overflow later on */
3268    
3269     if (len > MAX_DUPLENGTH)
3270     {
3271     *errorcodeptr = ERR50;
3272     goto FAILED;
3273     }
3274    
3275 nigel 77 /* If the maximum repeat count is unlimited, find the end of the bracket
3276     by scanning through from the start, and compute the offset back to it
3277     from the current code pointer. There may be an OP_OPT setting following
3278     the final KET, so we can't find the end just by going back from the code
3279     pointer. */
3280    
3281     if (repeat_max == -1)
3282     {
3283     register uschar *ket = previous;
3284     do ket += GET(ket, 1); while (*ket != OP_KET);
3285     ketoffset = code - ket;
3286     }
3287    
3288     /* The case of a zero minimum is special because of the need to stick
3289     OP_BRAZERO in front of it, and because the group appears once in the
3290     data, whereas in other cases it appears the minimum number of times. For
3291     this reason, it is simplest to treat this case separately, as otherwise
3292     the code gets far too messy. There are several special subcases when the
3293     minimum is zero. */
3294    
3295     if (repeat_min == 0)
3296     {
3297     /* If the maximum is also zero, we just omit the group from the output
3298     altogether. */
3299    
3300     if (repeat_max == 0)
3301     {
3302     code = previous;
3303     goto END_REPEAT;
3304     }
3305    
3306     /* If the maximum is 1 or unlimited, we just have to stick in the
3307     BRAZERO and do no more at this point. However, we do need to adjust
3308     any OP_RECURSE calls inside the group that refer to the group itself or
3309 nigel 93 any internal or forward referenced group, because the offset is from
3310     the start of the whole regex. Temporarily terminate the pattern while
3311     doing this. */
3312 nigel 77
3313     if (repeat_max <= 1)
3314     {
3315     *code = OP_END;
3316 nigel 93 adjust_recurse(previous, 1, utf8, cd, save_hwm);
3317 nigel 77 memmove(previous+1, previous, len);
3318     code++;
3319     *previous++ = OP_BRAZERO + repeat_type;
3320     }
3321    
3322     /* If the maximum is greater than 1 and limited, we have to replicate
3323     in a nested fashion, sticking OP_BRAZERO before each set of brackets.
3324     The first one has to be handled carefully because it's the original
3325     copy, which has to be moved up. The remainder can be handled by code
3326     that is common with the non-zero minimum case below. We have to
3327     adjust the value or repeat_max, since one less copy is required. Once
3328     again, we may have to adjust any OP_RECURSE calls inside the group. */
3329    
3330     else
3331     {
3332     int offset;
3333     *code = OP_END;
3334 nigel 93 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
3335 nigel 77 memmove(previous + 2 + LINK_SIZE, previous, len);
3336     code += 2 + LINK_SIZE;
3337     *previous++ = OP_BRAZERO + repeat_type;
3338     *previous++ = OP_BRA;
3339    
3340     /* We chain together the bracket offset fields that have to be
3341     filled in later when the ends of the brackets are reached. */
3342    
3343     offset = (bralink == NULL)? 0 : previous - bralink;
3344     bralink = previous;
3345     PUTINC(previous, 0, offset);
3346     }
3347    
3348     repeat_max--;
3349     }
3350    
3351     /* If the minimum is greater than zero, replicate the group as many
3352     times as necessary, and adjust the maximum to the number of subsequent
3353     copies that we need. If we set a first char from the group, and didn't
3354 nigel 93 set a required char, copy the latter from the former. If there are any
3355     forward reference subroutine calls in the group, there will be entries on
3356     the workspace list; replicate these with an appropriate increment. */
3357 nigel 77
3358     else
3359     {
3360     if (repeat_min > 1)
3361     {
3362 nigel 93 /* In the pre-compile phase, we don't actually do the replication. We
3363     just adjust the length as if we had. */
3364    
3365     if (lengthptr != NULL)
3366     *lengthptr += (repeat_min - 1)*length_prevgroup;
3367    
3368     /* This is compiling for real */
3369    
3370     else
3371 nigel 77 {
3372 nigel 93 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
3373     for (i = 1; i < repeat_min; i++)
3374     {
3375     uschar *hc;
3376     uschar *this_hwm = cd->hwm;
3377     memcpy(code, previous, len);
3378     for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
3379     {
3380     PUT(cd->hwm, 0, GET(hc, 0) + len);
3381     cd->hwm += LINK_SIZE;
3382     }
3383     save_hwm = this_hwm;
3384     code += len;
3385     }
3386 nigel 77 }
3387     }
3388 nigel 93
3389 nigel 77 if (repeat_max > 0) repeat_max -= repeat_min;
3390     }
3391    
3392     /* This code is common to both the zero and non-zero minimum cases. If
3393     the maximum is limited, it replicates the group in a nested fashion,
3394     remembering the bracket starts on a stack. In the case of a zero minimum,
3395     the first one was set up above. In all cases the repeat_max now specifies
3396 nigel 93 the number of additional copies needed. Again, we must remember to
3397     replicate entries on the forward reference list. */
3398 nigel 77
3399     if (repeat_max >= 0)
3400     {
3401 nigel 93 /* In the pre-compile phase, we don't actually do the replication. We
3402     just adjust the length as if we had. For each repetition we must add 1
3403     to the length for BRAZERO and for all but the last repetition we must
3404     add 2 + 2*LINKSIZE to allow for the nesting that occurs. */
3405    
3406     if (lengthptr != NULL && repeat_max > 0)
3407     *lengthptr += repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
3408     2 - 2*LINK_SIZE; /* Last one doesn't nest */
3409    
3410     /* This is compiling for real */
3411    
3412     else for (i = repeat_max - 1; i >= 0; i--)
3413 nigel 77 {
3414 nigel 93 uschar *hc;
3415     uschar *this_hwm = cd->hwm;
3416    
3417 nigel 77 *code++ = OP_BRAZERO + repeat_type;
3418    
3419     /* All but the final copy start a new nesting, maintaining the
3420     chain of brackets outstanding. */
3421    
3422     if (i != 0)
3423     {
3424     int offset;
3425     *code++ = OP_BRA;
3426     offset = (bralink == NULL)? 0 : code - bralink;
3427     bralink = code;
3428     PUTINC(code, 0, offset);
3429     }
3430    
3431     memcpy(code, previous, len);
3432 nigel 93 for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
3433     {
3434     PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
3435     cd->hwm += LINK_SIZE;
3436     }
3437     save_hwm = this_hwm;
3438 nigel 77 code += len;
3439     }
3440    
3441     /* Now chain through the pending brackets, and fill in their length
3442     fields (which are holding the chain links pro tem). */
3443    
3444     while (bralink != NULL)
3445     {
3446     int oldlinkoffset;
3447     int offset = code - bralink + 1;
3448     uschar *bra = code - offset;
3449     oldlinkoffset = GET(bra, 1);
3450     bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
3451     *code++ = OP_KET;
3452     PUTINC(code, 0, offset);
3453     PUT(bra, 1, offset);
3454     }
3455     }
3456    
3457     /* If the maximum is unlimited, set a repeater in the final copy. We
3458     can't just offset backwards from the current code point, because we
3459     don't know if there's been an options resetting after the ket. The
3460 nigel 93 correct offset was computed above.
3461 nigel 77
3462 nigel 93 Then, when we are doing the actual compile phase, check to see whether
3463     this group is a non-atomic one that could match an empty string. If so,
3464     convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
3465     that runtime checking can be done. [This check is also applied to
3466     atomic groups at runtime, but in a different way.] */
3467    
3468     else
3469     {
3470     uschar *ketcode = code - ketoffset;
3471     uschar *bracode = ketcode - GET(ketcode, 1);
3472     *ketcode = OP_KETRMAX + repeat_type;
3473     if (lengthptr == NULL && *bracode != OP_ONCE)
3474     {
3475     uschar *scode = bracode;
3476     do
3477     {
3478     if (could_be_empty_branch(scode, ketcode, utf8))
3479     {
3480     *bracode += OP_SBRA - OP_BRA;
3481     break;
3482     }
3483     scode += GET(scode, 1);
3484     }
3485     while (*scode == OP_ALT);
3486     }
3487     }
3488 nigel 77 }
3489    
3490     /* Else there's some kind of shambles */
3491    
3492     else
3493     {
3494     *errorcodeptr = ERR11;
3495     goto FAILED;
3496     }
3497    
3498 nigel 93 /* If the character following a repeat is '+', or if certain optimization
3499     tests above succeeded, possessive_quantifier is TRUE. For some of the
3500     simpler opcodes, there is an special alternative opcode for this. For
3501     anything else, we wrap the entire repeated item inside OP_ONCE brackets.
3502     The '+' notation is just syntactic sugar, taken from Sun's Java package,
3503     but the special opcodes can optimize it a bit. The repeated item starts at
3504     tempcode, not at previous, which might be the first part of a string whose
3505     (former) last char we repeated.
3506 nigel 77
3507 nigel 93 Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
3508     an 'upto' may follow. We skip over an 'exact' item, and then test the
3509     length of what remains before proceeding. */
3510    
3511 nigel 77 if (possessive_quantifier)
3512     {
3513 nigel 93 int len;
3514     if (*tempcode == OP_EXACT || *tempcode == OP_TYPEEXACT ||
3515     *tempcode == OP_NOTEXACT)
3516     tempcode += _pcre_OP_lengths[*tempcode];
3517     len = code - tempcode;
3518     if (len > 0) switch (*tempcode)
3519     {
3520     case OP_STAR: *tempcode = OP_POSSTAR; break;
3521     case OP_PLUS: *tempcode = OP_POSPLUS; break;
3522     case OP_QUERY: *tempcode = OP_POSQUERY; break;
3523     case OP_UPTO: *tempcode = OP_POSUPTO; break;
3524    
3525     case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
3526     case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
3527     case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
3528     case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
3529    
3530     case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
3531     case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
3532     case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
3533     case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
3534    
3535     default:
3536     memmove(tempcode + 1+LINK_SIZE, tempcode, len);
3537     code += 1 + LINK_SIZE;
3538     len += 1 + LINK_SIZE;
3539     tempcode[0] = OP_ONCE;
3540     *code++ = OP_KET;
3541     PUTINC(code, 0, len);
3542     PUT(tempcode, 1, len);
3543     break;
3544     }
3545 nigel 77 }
3546    
3547     /* In all case we no longer have a previous item. We also set the
3548     "follows varying string" flag for subsequently encountered reqbytes if
3549     it isn't already set and we have just passed a varying length item. */
3550    
3551     END_REPEAT:
3552     previous = NULL;
3553     cd->req_varyopt |= reqvary;
3554     break;
3555    
3556    
3557 nigel 93 /* ===================================================================*/
3558     /* Start of nested parenthesized sub-expression, or comment or lookahead or
3559     lookbehind or option setting or condition or all the other extended
3560     parenthesis forms. First deal with the specials; all are introduced by ?,
3561     and the appearance of any of them means that this is not a capturing
3562     group. */
3563 nigel 77
3564     case '(':
3565     newoptions = options;
3566     skipbytes = 0;
3567 nigel 93 bravalue = OP_CBRA;
3568     save_hwm = cd->hwm;
3569 nigel 77
3570     if (*(++ptr) == '?')
3571     {
3572 nigel 93 int i, set, unset, namelen;
3573 nigel 77 int *optset;
3574 nigel 93 const uschar *name;
3575     uschar *slot;
3576 nigel 77
3577     switch (*(++ptr))
3578     {
3579     case '#': /* Comment; skip to ket */
3580     ptr++;
3581 nigel 93 while (*ptr != 0 && *ptr != ')') ptr++;
3582     if (*ptr == 0)
3583     {
3584     *errorcodeptr = ERR18;
3585     goto FAILED;
3586     }
3587 nigel 77 continue;
3588    
3589 nigel 93
3590     /* ------------------------------------------------------------ */
3591     case ':': /* Non-capturing bracket */
3592 nigel 77 bravalue = OP_BRA;
3593     ptr++;
3594     break;
3595    
3596 nigel 93
3597     /* ------------------------------------------------------------ */
3598 nigel 77 case '(':
3599     bravalue = OP_COND; /* Conditional group */
3600    
3601 nigel 93 /* A condition can be an assertion, a number (referring to a numbered
3602     group), a name (referring to a named group), or 'R', referring to
3603     recursion. R<digits> and R&name are also permitted for recursion tests.
3604 nigel 77
3605 nigel 93 There are several syntaxes for testing a named group: (?(name)) is used
3606     by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
3607    
3608     There are two unfortunate ambiguities, caused by history. (a) 'R' can
3609     be the recursive thing or the name 'R' (and similarly for 'R' followed
3610     by digits), and (b) a number could be a name that consists of digits.
3611     In both cases, we look for a name first; if not found, we try the other
3612     cases. */
3613    
3614     /* For conditions that are assertions, check the syntax, and then exit
3615     the switch. This will take control down to where bracketed groups,
3616     including assertions, are processed. */
3617    
3618     if (ptr[1] == '?' && (ptr[2] == '=' || ptr[2] == '!' || ptr[2] == '<'))
3619     break;
3620    
3621     /* Most other conditions use OP_CREF (a couple change to OP_RREF
3622     below), and all need to skip 3 bytes at the start of the group. */
3623    
3624     code[1+LINK_SIZE] = OP_CREF;
3625     skipbytes = 3;
3626 ph10 167 refsign = -1;
3627 nigel 93
3628     /* Check for a test for recursion in a named group. */
3629    
3630     if (ptr[1] == 'R' && ptr[2] == '&')
3631 nigel 77 {
3632 nigel 93 terminator = -1;
3633     ptr += 2;
3634     code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
3635     }
3636 nigel 91
3637 nigel 93 /* Check for a test for a named group's having been set, using the Perl
3638     syntax (?(<name>) or (?('name') */
3639 nigel 91
3640 nigel 93 else if (ptr[1] == '<')
3641     {
3642     terminator = '>';
3643     ptr++;
3644     }
3645     else if (ptr[1] == '\'')
3646     {
3647     terminator = '\'';
3648     ptr++;
3649     }
3650 ph10 167 else
3651     {
3652     terminator = 0;
3653     if (ptr[1] == '-' || ptr[1] == '+') refsign = *(++ptr);
3654     }
3655 nigel 77
3656 nigel 93 /* We now expect to read a name; any thing else is an error */
3657 nigel 77
3658 nigel 93 if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
3659     {
3660     ptr += 1; /* To get the right offset */
3661     *errorcodeptr = ERR28;
3662     goto FAILED;
3663     }
3664    
3665     /* Read the name, but also get it as a number if it's all digits */
3666    
3667     recno = 0;
3668     name = ++ptr;
3669     while ((cd->ctypes[*ptr] & ctype_word) != 0)
3670     {
3671     if (recno >= 0)
3672     recno = ((digitab[*ptr] & ctype_digit) != 0)?
3673     recno * 10 + *ptr - '0' : -1;
3674 nigel 91 ptr++;
3675 nigel 93 }
3676     namelen = ptr - name;
3677 nigel 91
3678 nigel 93 if ((terminator > 0 && *ptr++ != terminator) || *ptr++ != ')')
3679     {
3680     ptr--; /* Error offset */
3681     *errorcodeptr = ERR26;
3682     goto FAILED;
3683     }
3684 nigel 91
3685 nigel 93 /* Do no further checking in the pre-compile phase. */
3686 nigel 91
3687 nigel 93 if (lengthptr != NULL) break;
3688 nigel 91
3689 nigel 93 /* In the real compile we do the work of looking for the actual
3690 ph10 167 reference. If the string started with "+" or "-" we require the rest to
3691     be digits, in which case recno will be set. */
3692    
3693     if (refsign > 0)
3694     {
3695     if (recno <= 0)
3696     {
3697     *errorcodeptr = ERR58;
3698     goto FAILED;
3699     }
3700     if (refsign == '-')
3701     {
3702     recno = cd->bracount - recno + 1;
3703     if (recno <= 0)
3704     {
3705     *errorcodeptr = ERR15;
3706     goto FAILED;
3707     }
3708     }
3709     else recno += cd->bracount;
3710     PUT2(code, 2+LINK_SIZE, recno);
3711     break;
3712     }
3713 nigel 91
3714 ph10 167 /* Otherwise (did not start with "+" or "-"), start by looking for the
3715     name. */
3716    
3717 nigel 93 slot = cd->name_table;
3718     for (i = 0; i < cd->names_found; i++)
3719     {
3720     if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
3721     slot += cd->name_entry_size;
3722     }
3723 nigel 91
3724 nigel 93 /* Found a previous named subpattern */
3725 nigel 91
3726 nigel 93 if (i < cd->names_found)
3727     {
3728     recno = GET2(slot, 0);
3729     PUT2(code, 2+LINK_SIZE, recno);
3730     }
3731 nigel 91
3732 nigel 93 /* Search the pattern for a forward reference */
3733 nigel 91
3734 nigel 93 else if ((i = find_parens(ptr, cd->bracount, name, namelen,
3735     (options & PCRE_EXTENDED) != 0)) > 0)
3736     {
3737     PUT2(code, 2+LINK_SIZE, i);
3738     }
3739 nigel 91
3740 nigel 93 /* If terminator == 0 it means that the name followed directly after
3741     the opening parenthesis [e.g. (?(abc)...] and in this case there are
3742     some further alternatives to try. For the cases where terminator != 0
3743     [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
3744     now checked all the possibilities, so give an error. */
3745 nigel 91
3746 nigel 93 else if (terminator != 0)
3747     {
3748     *errorcodeptr = ERR15;
3749     goto FAILED;
3750     }
3751    
3752     /* Check for (?(R) for recursion. Allow digits after R to specify a
3753     specific group number. */
3754    
3755     else if (*name == 'R')
3756     {
3757     recno = 0;
3758     for (i = 1; i < namelen; i++)
3759 nigel 91 {
3760 nigel 93 if ((digitab[name[i]] & ctype_digit) == 0)
3761     {
3762     *errorcodeptr = ERR15;
3763     goto FAILED;
3764     }
3765     recno = recno * 10 + name[i] - '0';
3766 nigel 77 }
3767 nigel 93 if (recno == 0) recno = RREF_ANY;
3768     code[1+LINK_SIZE] = OP_RREF; /* Change test type */
3769     PUT2(code, 2+LINK_SIZE, recno);
3770 nigel 77 }
3771 nigel 91
3772 nigel 93 /* Similarly, check for the (?(DEFINE) "condition", which is always
3773     false. */
3774 nigel 91
3775 nigel 93 else if (namelen == 6 && strncmp((char *)name, "DEFINE", 6) == 0)
3776     {
3777     code[1+LINK_SIZE] = OP_DEF;
3778     skipbytes = 1;
3779     }
3780    
3781     /* Check for the "name" actually being a subpattern number. */
3782    
3783     else if (recno > 0)
3784     {
3785     PUT2(code, 2+LINK_SIZE, recno);
3786     }
3787    
3788     /* Either an unidentified subpattern, or a reference to (?(0) */
3789    
3790     else
3791     {
3792     *errorcodeptr = (recno == 0)? ERR35: ERR15;
3793     goto FAILED;
3794     }
3795 nigel 77 break;
3796    
3797 nigel 93
3798     /* ------------------------------------------------------------ */
3799 nigel 77 case '=': /* Positive lookahead */
3800     bravalue = OP_ASSERT;
3801     ptr++;
3802     break;
3803    
3804 nigel 93
3805     /* ------------------------------------------------------------ */
3806 nigel 77 case '!': /* Negative lookahead */
3807     bravalue = OP_ASSERT_NOT;
3808     ptr++;
3809     break;
3810    
3811 nigel 93
3812     /* ------------------------------------------------------------ */
3813     case '<': /* Lookbehind or named define */
3814     switch (ptr[1])
3815 nigel 77 {
3816     case '=': /* Positive lookbehind */
3817     bravalue = OP_ASSERTBACK;
3818 nigel 93 ptr += 2;
3819 nigel 77 break;
3820    
3821     case '!': /* Negative lookbehind */
3822     bravalue = OP_ASSERTBACK_NOT;
3823 nigel 93 ptr += 2;
3824 nigel 77 break;
3825 nigel 93
3826     default: /* Could be name define, else bad */
3827     if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
3828     ptr++; /* Correct offset for error */
3829     *errorcodeptr = ERR24;
3830     goto FAILED;
3831 nigel 77 }
3832     break;
3833    
3834 nigel 93
3835     /* ------------------------------------------------------------ */
3836 nigel 77 case '>': /* One-time brackets */
3837     bravalue = OP_ONCE;
3838     ptr++;
3839     break;
3840    
3841 nigel 93
3842     /* ------------------------------------------------------------ */
3843 nigel 77 case 'C': /* Callout - may be followed by digits; */
3844     previous_callout = code; /* Save for later completion */
3845     after_manual_callout = 1; /* Skip one item before completing */
3846 nigel 93 *code++ = OP_CALLOUT;
3847     {
3848 nigel 77 int n = 0;
3849     while ((digitab[*(++ptr)] & ctype_digit) != 0)
3850     n = n * 10 + *ptr - '0';
3851 nigel 93 if (*ptr != ')')
3852     {
3853     *errorcodeptr = ERR39;
3854     goto FAILED;
3855     }
3856 nigel 77 if (n > 255)
3857     {
3858     *errorcodeptr = ERR38;
3859     goto FAILED;
3860     }
3861     *code++ = n;
3862     PUT(code, 0, ptr - cd->start_pattern + 1); /* Pattern offset */
3863     PUT(code, LINK_SIZE, 0); /* Default length */
3864     code += 2 * LINK_SIZE;
3865     }
3866     previous = NULL;
3867     continue;
3868    
3869 nigel 93
3870     /* ------------------------------------------------------------ */
3871     case 'P': /* Python-style named subpattern handling */
3872     if (*(++ptr) == '=' || *ptr == '>') /* Reference or recursion */
3873 nigel 77 {
3874 nigel 93 is_recurse = *ptr == '>';
3875     terminator = ')';
3876     goto NAMED_REF_OR_RECURSE;
3877     }
3878     else if (*ptr != '<') /* Test for Python-style definition */
3879     {
3880     *errorcodeptr = ERR41;
3881     goto FAILED;
3882     }
3883     /* Fall through to handle (?P< as (?< is handled */
3884 nigel 77
3885    
3886 nigel 93 /* ------------------------------------------------------------ */
3887     DEFINE_NAME: /* Come here from (?< handling */
3888     case '\'':
3889     {
3890     terminator = (*ptr == '<')? '>' : '\'';
3891     name = ++ptr;
3892    
3893     while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
3894     namelen = ptr - name;
3895    
3896     /* In the pre-compile phase, just do a syntax check. */
3897    
3898     if (lengthptr != NULL)
3899 nigel 77 {
3900 nigel 93 if (*ptr != terminator)
3901 nigel 77 {
3902 nigel 93 *errorcodeptr = ERR42;
3903     goto FAILED;
3904     }
3905     if (cd->names_found >= MAX_NAME_COUNT)
3906     {
3907     *errorcodeptr = ERR49;
3908     goto FAILED;
3909     }
3910     if (namelen + 3 > cd->name_entry_size)
3911     {
3912     cd->name_entry_size = namelen + 3;
3913     if (namelen > MAX_NAME_SIZE)
3914 nigel 77 {
3915 nigel 93 *errorcodeptr = ERR48;
3916     goto FAILED;
3917     }
3918     }
3919     }
3920    
3921     /* In the real compile, create the entry in the table */
3922    
3923     else
3924     {
3925     slot = cd->name_table;
3926     for (i = 0; i < cd->names_found; i++)
3927     {
3928     int crc = memcmp(name, slot+2, namelen);
3929     if (crc == 0)
3930     {
3931     if (slot[2+namelen] == 0)
3932 nigel 91 {
3933 nigel 93 if ((options & PCRE_DUPNAMES) == 0)
3934     {
3935     *errorcodeptr = ERR43;
3936     goto FAILED;
3937     }
3938 nigel 91 }
3939 nigel 93 else crc = -1; /* Current name is substring */
3940 nigel 77 }
3941 nigel 93 if (crc < 0)
3942     {
3943     memmove(slot + cd->name_entry_size, slot,
3944     (cd->names_found - i) * cd->name_entry_size);
3945     break;
3946     }
3947     slot += cd->name_entry_size;
3948 nigel 77 }
3949 nigel 93
3950     PUT2(slot, 0, cd->bracount + 1);
3951     memcpy(slot + 2, name, namelen);
3952     slot[2+namelen] = 0;
3953 nigel 77 }
3954     }
3955    
3956 nigel 93 /* In both cases, count the number of names we've encountered. */
3957    
3958     ptr++; /* Move past > or ' */
3959     cd->names_found++;
3960     goto NUMBERED_GROUP;
3961    
3962    
3963     /* ------------------------------------------------------------ */
3964     case '&': /* Perl recursion/subroutine syntax */
3965     terminator = ')';
3966     is_recurse = TRUE;
3967     /* Fall through */
3968    
3969     /* We come here from the Python syntax above that handles both
3970     references (?P=name) and recursion (?P>name), as well as falling
3971     through from the Perl recursion syntax (?&name). */
3972    
3973     NAMED_REF_OR_RECURSE:
3974     name = ++ptr;
3975     while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
3976     namelen = ptr - name;
3977    
3978     /* In the pre-compile phase, do a syntax check and set a dummy
3979     reference number. */
3980    
3981     if (lengthptr != NULL)
3982 nigel 77 {
3983 nigel 93 if (*ptr != terminator)
3984     {
3985     *errorcodeptr = ERR42;
3986     goto FAILED;
3987     }
3988     if (namelen > MAX_NAME_SIZE)
3989     {
3990     *errorcodeptr = ERR48;
3991     goto FAILED;
3992     }
3993     recno = 0;
3994     }
3995 nigel 77
3996 nigel 93 /* In the real compile, seek the name in the table */
3997 nigel 77
3998 nigel 93 else
3999     {
4000     slot = cd->name_table;
4001 nigel 77 for (i = 0; i < cd->names_found; i++)
4002     {
4003     if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
4004     slot += cd->name_entry_size;
4005     }
4006 nigel 91
4007     if (i < cd->names_found) /* Back reference */
4008 nigel 77 {
4009 nigel 91 recno = GET2(slot, 0);
4010     }
4011     else if ((recno = /* Forward back reference */
4012 nigel 93 find_parens(ptr, cd->bracount, name, namelen,
4013     (options & PCRE_EXTENDED) != 0)) <= 0)
4014 nigel 91 {
4015 nigel 77 *errorcodeptr = ERR15;
4016     goto FAILED;
4017     }
4018 nigel 93 }
4019 nigel 77
4020 nigel 93 /* In both phases, we can now go to the code than handles numerical
4021     recursion or backreferences. */
4022 nigel 77
4023 nigel 93 if (is_recurse) goto HANDLE_RECURSION;
4024     else goto HANDLE_REFERENCE;
4025 nigel 77
4026    
4027 nigel 93 /* ------------------------------------------------------------ */
4028     case 'R': /* Recursion */
4029 nigel 77 ptr++; /* Same as (?0) */
4030     /* Fall through */
4031    
4032    
4033 nigel 93 /* ------------------------------------------------------------ */
4034 ph10 166 case '-': case '+':
4035 nigel 93 case '0': case '1': case '2': case '3': case '4': /* Recursion or */
4036     case '5': case '6': case '7': case '8': case '9': /* subroutine */
4037 nigel 77 {
4038     const uschar *called;
4039 ph10 166
4040 ph10 167 if ((refsign = *ptr) == '+') ptr++;
4041     else if (refsign == '-')
4042 ph10 166 {
4043     if ((digitab[ptr[1]] & ctype_digit) == 0)
4044     goto OTHER_CHAR_AFTER_QUERY;
4045     ptr++;
4046     }
4047    
4048 nigel 77 recno = 0;
4049     while((digitab[*ptr] & ctype_digit) != 0)
4050     recno = recno * 10 + *ptr++ - '0';
4051 ph10 166
4052 nigel 93 if (*ptr != ')')
4053     {
4054     *errorcodeptr = ERR29;
4055     goto FAILED;
4056     }
4057 ph10 166
4058 ph10 167 if (refsign == '-')
4059 ph10 166 {
4060     if (recno == 0)
4061     {
4062     *errorcodeptr = ERR58;
4063     goto FAILED;
4064     }
4065     recno = cd->bracount - recno + 1;
4066     if (recno <= 0)
4067     {
4068     *errorcodeptr = ERR15;
4069     goto FAILED;
4070     }
4071     }
4072 ph10 167 else if (refsign == '+')
4073 ph10 166 {
4074     if (recno == 0)
4075     {
4076     *errorcodeptr = ERR58;
4077     goto FAILED;
4078     }
4079     recno += cd->bracount;
4080     }
4081 nigel 77
4082     /* Come here from code above that handles a named recursion */
4083    
4084     HANDLE_RECURSION:
4085    
4086     previous = code;
4087 nigel 93 called = cd->start_code;
4088 nigel 77
4089 nigel 93 /* When we are actually compiling, find the bracket that is being
4090     referenced. Temporarily end the regex in case it doesn't exist before
4091     this point. If we end up with a forward reference, first check that
4092     the bracket does occur later so we can give the error (and position)
4093     now. Then remember this forward reference in the workspace so it can
4094     be filled in at the end. */
4095 nigel 77
4096 nigel 93 if (lengthptr == NULL)
4097 nigel 77 {
4098 nigel 93 *code = OP_END;
4099     if (recno != 0) called = find_bracket(cd->start_code, utf8, recno);
4100 nigel 77
4101 nigel 93 /* Forward reference */
4102 nigel 77
4103 nigel 93 if (called == NULL)
4104     {
4105     if (find_parens(ptr, cd->bracount, NULL, recno,
4106     (options & PCRE_EXTENDED) != 0) < 0)
4107     {
4108     *errorcodeptr = ERR15;
4109     goto FAILED;
4110     }
4111     called = cd->start_code + recno;
4112     PUTINC(cd->hwm, 0, code + 2 + LINK_SIZE - cd->start_code);
4113     }
4114    
4115     /* If not a forward reference, and the subpattern is still open,
4116     this is a recursive call. We check to see if this is a left
4117     recursion that could loop for ever, and diagnose that case. */
4118    
4119     else if (GET(called, 1) == 0 &&
4120     could_be_empty(called, code, bcptr, utf8))
4121     {
4122     *errorcodeptr = ERR40;
4123     goto FAILED;
4124     }
4125 nigel 77 }
4126    
4127 nigel 87 /* Insert the recursion/subroutine item, automatically wrapped inside
4128 nigel 93 "once" brackets. Set up a "previous group" length so that a
4129     subsequent quantifier will work. */
4130 nigel 77
4131 nigel 87 *code = OP_ONCE;
4132     PUT(code, 1, 2 + 2*LINK_SIZE);
4133     code += 1 + LINK_SIZE;
4134    
4135 nigel 77 *code = OP_RECURSE;
4136     PUT(code, 1, called - cd->start_code);
4137     code += 1 + LINK_SIZE;
4138 nigel 87
4139     *code = OP_KET;
4140     PUT(code, 1, 2 + 2*LINK_SIZE);
4141     code += 1 + LINK_SIZE;
4142 nigel 93
4143     length_prevgroup = 3 + 3*LINK_SIZE;
4144 nigel 77 }
4145 nigel 93
4146     /* Can't determine a first byte now */
4147    
4148     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
4149 nigel 77 continue;
4150    
4151    
4152 nigel 93 /* ------------------------------------------------------------ */
4153     default: /* Other characters: check option setting */
4154 ph10 166 OTHER_CHAR_AFTER_QUERY:
4155 nigel 77 set = unset = 0;
4156     optset = &set;
4157    
4158     while (*ptr != ')' && *ptr != ':')
4159     {
4160     switch (*ptr++)
4161     {
4162     case '-': optset = &unset; break;
4163    
4164 nigel 93 case 'J': /* Record that it changed in the external options */
4165     *optset |= PCRE_DUPNAMES;
4166     cd->external_options |= PCRE_JCHANGED;
4167     break;
4168    
4169 nigel 77 case 'i': *optset |= PCRE_CASELESS; break;
4170     case 'm': *optset |= PCRE_MULTILINE; break;
4171     case 's': *optset |= PCRE_DOTALL; break;
4172     case 'x': *optset |= PCRE_EXTENDED; break;
4173     case 'U': *optset |= PCRE_UNGREEDY; break;
4174     case 'X': *optset |= PCRE_EXTRA; break;
4175 nigel 93
4176     default: *errorcodeptr = ERR12;
4177     ptr--; /* Correct the offset */
4178     goto FAILED;
4179 nigel 77 }
4180     }
4181    
4182     /* Set up the changed option bits, but don't change anything yet. */
4183    
4184     newoptions = (options | set) & (~unset);
4185    
4186     /* If the options ended with ')' this is not the start of a nested
4187 nigel 93 group with option changes, so the options change at this level. If this
4188     item is right at the start of the pattern, the options can be