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

Contents of /code/trunk/pcre_compile.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 349 - (hide annotations) (download)
Wed Jul 2 18:42:11 2008 UTC (6 years, 5 months ago) by ph10
File MIME type: text/plain
File size: 202638 byte(s)
Preliminary patch for upgraded Unicode Property support.

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 305 Copyright (c) 1997-2008 University of Cambridge
10 nigel 77
11     -----------------------------------------------------------------------------
12     Redistribution and use in source and binary forms, with or without
13     modification, are permitted provided that the following conditions are met:
14    
15     * Redistributions of source code must retain the above copyright notice,
16     this list of conditions and the following disclaimer.
17    
18     * Redistributions in binary form must reproduce the above copyright
19     notice, this list of conditions and the following disclaimer in the
20     documentation and/or other materials provided with the distribution.
21    
22     * Neither the name of the University of Cambridge nor the names of its
23     contributors may be used to endorse or promote products derived from
24     this software without specific prior written permission.
25    
26     THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
27     AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28     IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29     ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
30     LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31     CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32     SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33     INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34     CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35     ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36     POSSIBILITY OF SUCH DAMAGE.
37     -----------------------------------------------------------------------------
38     */
39    
40    
41     /* This module contains the external function pcre_compile(), along with
42     supporting internal functions that are not used by other modules. */
43    
44    
45 ph10 200 #ifdef HAVE_CONFIG_H
46 ph10 236 #include "config.h"
47 ph10 200 #endif
48 ph10 199
49 nigel 93 #define NLBLOCK cd /* Block containing newline information */
50     #define PSSTART start_pattern /* Field containing processed string start */
51     #define PSEND end_pattern /* Field containing processed string end */
52    
53 nigel 77 #include "pcre_internal.h"
54    
55    
56 nigel 85 /* When DEBUG is defined, we need the pcre_printint() function, which is also
57     used by pcretest. DEBUG is not defined when building a production library. */
58    
59     #ifdef DEBUG
60     #include "pcre_printint.src"
61     #endif
62    
63    
64 ph10 178 /* Macro for setting individual bits in class bitmaps. */
65    
66     #define SETBIT(a,b) a[b/8] |= (1 << (b%8))
67    
68 ph10 202 /* Maximum length value to check against when making sure that the integer that
69     holds the compiled pattern length does not overflow. We make it a bit less than
70     INT_MAX to allow for adding in group terminating bytes, so that we don't have
71     to check them every time. */
72 ph10 178
73 ph10 202 #define OFLOW_MAX (INT_MAX - 20)
74    
75    
76 nigel 77 /*************************************************
77     * Code parameters and static tables *
78     *************************************************/
79    
80 nigel 93 /* This value specifies the size of stack workspace that is used during the
81     first pre-compile phase that determines how much memory is required. The regex
82     is partly compiled into this space, but the compiled parts are discarded as
83     soon as they can be, so that hopefully there will never be an overrun. The code
84     does, however, check for an overrun. The largest amount I've seen used is 218,
85     so this number is very generous.
86 nigel 77
87 nigel 93 The same workspace is used during the second, actual compile phase for
88     remembering forward references to groups so that they can be filled in at the
89     end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE
90     is 4 there is plenty of room. */
91 nigel 77
92 nigel 93 #define COMPILE_WORK_SIZE (4096)
93 nigel 77
94 nigel 93
95 nigel 77 /* Table for handling escaped characters in the range '0'-'z'. Positive returns
96     are simple data values; negative values are for special things like \d and so
97     on. Zero means further processing is needed (for things like \x), or the escape
98     is invalid. */
99    
100 ph10 97 #ifndef EBCDIC /* This is the "normal" table for ASCII systems */
101 nigel 77 static const short int escapes[] = {
102     0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 7 */
103     0, 0, ':', ';', '<', '=', '>', '?', /* 8 - ? */
104     '@', -ESC_A, -ESC_B, -ESC_C, -ESC_D, -ESC_E, 0, -ESC_G, /* @ - G */
105 ph10 178 -ESC_H, 0, 0, -ESC_K, 0, 0, 0, 0, /* H - O */
106     -ESC_P, -ESC_Q, -ESC_R, -ESC_S, 0, 0, -ESC_V, -ESC_W, /* P - W */
107 nigel 77 -ESC_X, 0, -ESC_Z, '[', '\\', ']', '^', '_', /* X - _ */
108     '`', 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, /* ` - g */
109 ph10 178 -ESC_h, 0, 0, -ESC_k, 0, 0, ESC_n, 0, /* h - o */
110     -ESC_p, 0, ESC_r, -ESC_s, ESC_tee, 0, -ESC_v, -ESC_w, /* p - w */
111 nigel 77 0, 0, -ESC_z /* x - z */
112     };
113    
114 ph10 97 #else /* This is the "abnormal" table for EBCDIC systems */
115 nigel 77 static const short int escapes[] = {
116     /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|',
117     /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0,
118     /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~',
119     /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0,
120     /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?',
121     /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0,
122     /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"',
123     /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0,
124 ph10 178 /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0,
125 nigel 93 /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p,
126 nigel 77 /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0,
127 ph10 178 /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0,
128 nigel 77 /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0,
129     /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0,
130     /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-',
131     /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G,
132 ph10 178 /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0,
133 ph10 195 /* D0 */ '}', 0, -ESC_K, 0, 0, 0, 0, -ESC_P,
134 nigel 93 /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0,
135 ph10 178 /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X,
136 nigel 77 /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0,
137     /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0,
138     /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0
139     };
140     #endif
141    
142    
143 ph10 243 /* Table of special "verbs" like (*PRUNE). This is a short table, so it is
144     searched linearly. Put all the names into a single string, in order to reduce
145 ph10 240 the number of relocations when a shared library is dynamically linked. */
146 ph10 210
147     typedef struct verbitem {
148     int len;
149     int op;
150 ph10 211 } verbitem;
151 ph10 210
152 ph10 240 static const char verbnames[] =
153 ph10 243 "ACCEPT\0"
154     "COMMIT\0"
155     "F\0"
156     "FAIL\0"
157     "PRUNE\0"
158     "SKIP\0"
159     "THEN";
160 ph10 240
161 ph10 327 static const verbitem verbs[] = {
162 ph10 240 { 6, OP_ACCEPT },
163     { 6, OP_COMMIT },
164     { 1, OP_FAIL },
165     { 4, OP_FAIL },
166     { 5, OP_PRUNE },
167     { 4, OP_SKIP },
168     { 4, OP_THEN }
169 ph10 210 };
170    
171 ph10 327 static const int verbcount = sizeof(verbs)/sizeof(verbitem);
172 ph10 210
173    
174 ph10 243 /* Tables of names of POSIX character classes and their lengths. The names are
175     now all in a single string, to reduce the number of relocations when a shared
176 ph10 240 library is dynamically loaded. The list of lengths is terminated by a zero
177     length entry. The first three must be alpha, lower, upper, as this is assumed
178     for handling case independence. */
179 nigel 77
180 ph10 240 static const char posix_names[] =
181 ph10 243 "alpha\0" "lower\0" "upper\0" "alnum\0" "ascii\0" "blank\0"
182     "cntrl\0" "digit\0" "graph\0" "print\0" "punct\0" "space\0"
183 ph10 240 "word\0" "xdigit";
184 nigel 77
185     static const uschar posix_name_lengths[] = {
186     5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 };
187    
188 nigel 87 /* Table of class bit maps for each POSIX class. Each class is formed from a
189     base map, with an optional addition or removal of another map. Then, for some
190     classes, there is some additional tweaking: for [:blank:] the vertical space
191     characters are removed, and for [:alpha:] and [:alnum:] the underscore
192     character is removed. The triples in the table consist of the base map offset,
193     second map offset or -1 if no second map, and a non-negative value for map
194     addition or a negative value for map subtraction (if there are two maps). The
195     absolute value of the third field has these meanings: 0 => no tweaking, 1 =>
196     remove vertical space characters, 2 => remove underscore. */
197 nigel 77
198     static const int posix_class_maps[] = {
199 nigel 87 cbit_word, cbit_digit, -2, /* alpha */
200     cbit_lower, -1, 0, /* lower */
201     cbit_upper, -1, 0, /* upper */
202     cbit_word, -1, 2, /* alnum - word without underscore */
203     cbit_print, cbit_cntrl, 0, /* ascii */
204     cbit_space, -1, 1, /* blank - a GNU extension */
205     cbit_cntrl, -1, 0, /* cntrl */
206     cbit_digit, -1, 0, /* digit */
207     cbit_graph, -1, 0, /* graph */
208     cbit_print, -1, 0, /* print */
209     cbit_punct, -1, 0, /* punct */
210     cbit_space, -1, 0, /* space */
211     cbit_word, -1, 0, /* word - a Perl extension */
212     cbit_xdigit,-1, 0 /* xdigit */
213 nigel 77 };
214    
215    
216 nigel 93 #define STRING(a) # a
217     #define XSTRING(s) STRING(s)
218    
219 nigel 77 /* The texts of compile-time error messages. These are "char *" because they
220 nigel 93 are passed to the outside world. Do not ever re-use any error number, because
221     they are documented. Always add a new error instead. Messages marked DEAD below
222 ph10 243 are no longer used. This used to be a table of strings, but in order to reduce
223     the number of relocations needed when a shared library is loaded dynamically,
224     it is now one long string. We cannot use a table of offsets, because the
225     lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we
226     simply count through to the one we want - this isn't a performance issue
227 ph10 240 because these strings are used only when there is a compilation error. */
228 nigel 77
229 ph10 240 static const char error_texts[] =
230     "no error\0"
231     "\\ at end of pattern\0"
232     "\\c at end of pattern\0"
233     "unrecognized character follows \\\0"
234     "numbers out of order in {} quantifier\0"
235 nigel 77 /* 5 */
236 ph10 240 "number too big in {} quantifier\0"
237     "missing terminating ] for character class\0"
238     "invalid escape sequence in character class\0"
239     "range out of order in character class\0"
240     "nothing to repeat\0"
241 nigel 77 /* 10 */
242 ph10 240 "operand of unlimited repeat could match the empty string\0" /** DEAD **/
243     "internal error: unexpected repeat\0"
244 ph10 269 "unrecognized character after (? or (?-\0"
245 ph10 240 "POSIX named classes are supported only within a class\0"
246     "missing )\0"
247 nigel 77 /* 15 */
248 ph10 240 "reference to non-existent subpattern\0"
249     "erroffset passed as NULL\0"
250     "unknown option bit(s) set\0"
251     "missing ) after comment\0"
252     "parentheses nested too deeply\0" /** DEAD **/
253 nigel 77 /* 20 */
254 ph10 240 "regular expression is too large\0"
255     "failed to get memory\0"
256     "unmatched parentheses\0"
257     "internal error: code overflow\0"
258     "unrecognized character after (?<\0"
259 nigel 77 /* 25 */
260 ph10 240 "lookbehind assertion is not fixed length\0"
261     "malformed number or name after (?(\0"
262     "conditional group contains more than two branches\0"
263     "assertion expected after (?(\0"
264     "(?R or (?[+-]digits must be followed by )\0"
265 nigel 77 /* 30 */
266 ph10 240 "unknown POSIX class name\0"
267     "POSIX collating elements are not supported\0"
268     "this version of PCRE is not compiled with PCRE_UTF8 support\0"
269     "spare error\0" /** DEAD **/
270     "character value in \\x{...} sequence is too large\0"
271 nigel 77 /* 35 */
272 ph10 240 "invalid condition (?(0)\0"
273     "\\C not allowed in lookbehind assertion\0"
274     "PCRE does not support \\L, \\l, \\N, \\U, or \\u\0"
275     "number after (?C is > 255\0"
276     "closing ) for (?C expected\0"
277 nigel 77 /* 40 */
278 ph10 240 "recursive call could loop indefinitely\0"
279     "unrecognized character after (?P\0"
280     "syntax error in subpattern name (missing terminator)\0"
281     "two named subpatterns have the same name\0"
282     "invalid UTF-8 string\0"
283 nigel 77 /* 45 */
284 ph10 240 "support for \\P, \\p, and \\X has not been compiled\0"
285     "malformed \\P or \\p sequence\0"
286     "unknown property name after \\P or \\p\0"
287     "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0"
288     "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0"
289 nigel 91 /* 50 */
290 ph10 240 "repeated subpattern is too long\0" /** DEAD **/
291     "octal value is greater than \\377 (not in UTF-8 mode)\0"
292     "internal error: overran compiling workspace\0"
293     "internal error: previously-checked referenced subpattern not found\0"
294     "DEFINE group contains more than one branch\0"
295 nigel 93 /* 55 */
296 ph10 240 "repeating a DEFINE group is not allowed\0"
297     "inconsistent NEWLINE options\0"
298 ph10 333 "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0"
299     "a numbered reference must not be zero\0"
300 ph10 240 "(*VERB) with an argument is not supported\0"
301 ph10 211 /* 60 */
302 ph10 240 "(*VERB) not recognized\0"
303 ph10 268 "number is too big\0"
304 ph10 272 "subpattern name expected\0"
305 ph10 336 "digit expected after (?+\0"
306 ph10 345 "] is an invalid data character in JavaScript compatibility mode";
307 nigel 77
308    
309     /* Table to identify digits and hex digits. This is used when compiling
310     patterns. Note that the tables in chartables are dependent on the locale, and
311     may mark arbitrary characters as digits - but the PCRE compiling code expects
312     to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
313     a private table here. It costs 256 bytes, but it is a lot faster than doing
314     character value tests (at least in some simple cases I timed), and in some
315     applications one wants PCRE to compile efficiently as well as match
316     efficiently.
317    
318     For convenience, we use the same bit definitions as in chartables:
319    
320     0x04 decimal digit
321     0x08 hexadecimal digit
322    
323     Then we can use ctype_digit and ctype_xdigit in the code. */
324    
325 ph10 97 #ifndef EBCDIC /* This is the "normal" case, for ASCII systems */
326 nigel 77 static const unsigned char digitab[] =
327     {
328     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
329     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
330     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
331     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
332     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
333     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
334     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
335     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
336     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
337     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
338     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
339     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
340     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */
341     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */
342     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */
343     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */
344     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
345     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
346     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
347     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
348     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
349     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
350     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
351     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
352     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
353     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
354     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
355     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
356     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
357     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
358     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
359     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
360    
361 ph10 97 #else /* This is the "abnormal" case, for EBCDIC systems */
362 nigel 77 static const unsigned char digitab[] =
363     {
364     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
365     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
366     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
367     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
368     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
369     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
370     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
371     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
372     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
373     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
374     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
375 ph10 97 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
376 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
377     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
378     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
379     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
380     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
381     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
382     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
383     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
384     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
385     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
386     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
387     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
388     0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
389     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
390     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
391     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
392     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
393     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
394     0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
395     0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
396    
397     static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
398     0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
399     0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
400     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
401     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
402     0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
403     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
404     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
405     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
406     0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
407     0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
408     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
409 ph10 97 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
410 nigel 77 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
411     0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
412     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
413     0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
414     0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
415     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
416     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
417     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
418     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
419     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
420     0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
421     0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
422     0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
423     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
424     0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
425     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
426     0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
427     0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
428     0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
429     0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
430     #endif
431    
432    
433     /* Definition to allow mutual recursion */
434    
435     static BOOL
436 ph10 180 compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
437 ph10 175 int *, int *, branch_chain *, compile_data *, int *);
438 nigel 77
439    
440    
441     /*************************************************
442 ph10 240 * Find an error text *
443     *************************************************/
444    
445 ph10 243 /* The error texts are now all in one long string, to save on relocations. As
446     some of the text is of unknown length, we can't use a table of offsets.
447     Instead, just count through the strings. This is not a performance issue
448 ph10 240 because it happens only when there has been a compilation error.
449    
450     Argument: the error number
451     Returns: pointer to the error string
452     */
453    
454     static const char *
455     find_error_text(int n)
456     {
457     const char *s = error_texts;
458 ph10 243 for (; n > 0; n--) while (*s++ != 0);
459 ph10 240 return s;
460     }
461    
462    
463     /*************************************************
464 nigel 77 * Handle escapes *
465     *************************************************/
466    
467     /* This function is called when a \ has been encountered. It either returns a
468     positive value for a simple escape such as \n, or a negative value which
469 nigel 93 encodes one of the more complicated things such as \d. A backreference to group
470     n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
471     UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
472     ptr is pointing at the \. On exit, it is on the final character of the escape
473     sequence.
474 nigel 77
475     Arguments:
476     ptrptr points to the pattern position pointer
477     errorcodeptr points to the errorcode variable
478     bracount number of previous extracting brackets
479     options the options bits
480     isclass TRUE if inside a character class
481    
482     Returns: zero or positive => a data character
483     negative => a special escape sequence
484 ph10 213 on error, errorcodeptr is set
485 nigel 77 */
486    
487     static int
488     check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
489     int options, BOOL isclass)
490     {
491 nigel 87 BOOL utf8 = (options & PCRE_UTF8) != 0;
492     const uschar *ptr = *ptrptr + 1;
493 nigel 77 int c, i;
494    
495 nigel 87 GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
496     ptr--; /* Set pointer back to the last byte */
497    
498 nigel 77 /* If backslash is at the end of the pattern, it's an error. */
499    
500     if (c == 0) *errorcodeptr = ERR1;
501    
502 ph10 274 /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
503     in a table. A non-zero result is something that can be returned immediately.
504 nigel 77 Otherwise further processing may be required. */
505    
506 ph10 97 #ifndef EBCDIC /* ASCII coding */
507 ph10 274 else if (c < '0' || c > 'z') {} /* Not alphanumeric */
508 nigel 77 else if ((i = escapes[c - '0']) != 0) c = i;
509    
510 ph10 97 #else /* EBCDIC coding */
511 ph10 274 else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
512 nigel 77 else if ((i = escapes[c - 0x48]) != 0) c = i;
513     #endif
514    
515     /* Escapes that need further processing, or are illegal. */
516    
517     else
518     {
519     const uschar *oldptr;
520 nigel 93 BOOL braced, negated;
521    
522 nigel 77 switch (c)
523     {
524     /* A number of Perl escapes are not handled by PCRE. We give an explicit
525     error. */
526    
527     case 'l':
528     case 'L':
529     case 'N':
530     case 'u':
531     case 'U':
532     *errorcodeptr = ERR37;
533     break;
534    
535 ph10 333 /* \g must be followed by one of a number of specific things:
536 ph10 345
537 ph10 333 (1) A number, either plain or braced. If positive, it is an absolute
538     backreference. If negative, it is a relative backreference. This is a Perl
539     5.10 feature.
540 ph10 345
541 ph10 333 (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
542     is part of Perl's movement towards a unified syntax for back references. As
543     this is synonymous with \k{name}, we fudge it up by pretending it really
544     was \k.
545 ph10 345
546     (3) For Oniguruma compatibility we also support \g followed by a name or a
547     number either in angle brackets or in single quotes. However, these are
548     (possibly recursive) subroutine calls, _not_ backreferences. Just return
549 ph10 333 the -ESC_g code (cf \k). */
550 nigel 93
551     case 'g':
552 ph10 333 if (ptr[1] == '<' || ptr[1] == '\'')
553     {
554     c = -ESC_g;
555 ph10 345 break;
556     }
557 ph10 333
558     /* Handle the Perl-compatible cases */
559 ph10 345
560 nigel 93 if (ptr[1] == '{')
561     {
562 ph10 171 const uschar *p;
563     for (p = ptr+2; *p != 0 && *p != '}'; p++)
564     if (*p != '-' && (digitab[*p] & ctype_digit) == 0) break;
565 ph10 172 if (*p != 0 && *p != '}')
566 ph10 171 {
567     c = -ESC_k;
568     break;
569 ph10 172 }
570 nigel 93 braced = TRUE;
571     ptr++;
572     }
573     else braced = FALSE;
574    
575     if (ptr[1] == '-')
576     {
577     negated = TRUE;
578     ptr++;
579     }
580     else negated = FALSE;
581    
582     c = 0;
583     while ((digitab[ptr[1]] & ctype_digit) != 0)
584     c = c * 10 + *(++ptr) - '0';
585 ph10 220
586 ph10 333 if (c < 0) /* Integer overflow */
587 ph10 213 {
588     *errorcodeptr = ERR61;
589     break;
590 ph10 220 }
591 ph10 345
592 ph10 333 if (braced && *(++ptr) != '}')
593 nigel 93 {
594     *errorcodeptr = ERR57;
595 ph10 213 break;
596 nigel 93 }
597 ph10 345
598 ph10 333 if (c == 0)
599     {
600     *errorcodeptr = ERR58;
601     break;
602 ph10 345 }
603 nigel 93
604     if (negated)
605     {
606     if (c > bracount)
607     {
608     *errorcodeptr = ERR15;
609 ph10 213 break;
610 nigel 93 }
611     c = bracount - (c - 1);
612     }
613    
614     c = -(ESC_REF + c);
615     break;
616    
617 nigel 77 /* The handling of escape sequences consisting of a string of digits
618     starting with one that is not zero is not straightforward. By experiment,
619     the way Perl works seems to be as follows:
620    
621     Outside a character class, the digits are read as a decimal number. If the
622     number is less than 10, or if there are that many previous extracting
623     left brackets, then it is a back reference. Otherwise, up to three octal
624     digits are read to form an escaped byte. Thus \123 is likely to be octal
625     123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
626     value is greater than 377, the least significant 8 bits are taken. Inside a
627     character class, \ followed by a digit is always an octal number. */
628    
629     case '1': case '2': case '3': case '4': case '5':
630     case '6': case '7': case '8': case '9':
631    
632     if (!isclass)
633     {
634     oldptr = ptr;
635     c -= '0';
636     while ((digitab[ptr[1]] & ctype_digit) != 0)
637     c = c * 10 + *(++ptr) - '0';
638 ph10 333 if (c < 0) /* Integer overflow */
639 ph10 213 {
640     *errorcodeptr = ERR61;
641 ph10 220 break;
642     }
643 nigel 77 if (c < 10 || c <= bracount)
644     {
645     c = -(ESC_REF + c);
646     break;
647     }
648     ptr = oldptr; /* Put the pointer back and fall through */
649     }
650    
651     /* Handle an octal number following \. If the first digit is 8 or 9, Perl
652     generates a binary zero byte and treats the digit as a following literal.
653     Thus we have to pull back the pointer by one. */
654    
655     if ((c = *ptr) >= '8')
656     {
657     ptr--;
658     c = 0;
659     break;
660     }
661    
662     /* \0 always starts an octal number, but we may drop through to here with a
663 nigel 91 larger first octal digit. The original code used just to take the least
664     significant 8 bits of octal numbers (I think this is what early Perls used
665     to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
666     than 3 octal digits. */
667 nigel 77
668     case '0':
669     c -= '0';
670     while(i++ < 2 && ptr[1] >= '0' && ptr[1] <= '7')
671     c = c * 8 + *(++ptr) - '0';
672 nigel 91 if (!utf8 && c > 255) *errorcodeptr = ERR51;
673 nigel 77 break;
674    
675 nigel 87 /* \x is complicated. \x{ddd} is a character number which can be greater
676     than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
677     treated as a data character. */
678 nigel 77
679     case 'x':
680 nigel 87 if (ptr[1] == '{')
681 nigel 77 {
682     const uschar *pt = ptr + 2;
683 nigel 87 int count = 0;
684    
685 nigel 77 c = 0;
686     while ((digitab[*pt] & ctype_xdigit) != 0)
687     {
688 nigel 87 register int cc = *pt++;
689     if (c == 0 && cc == '0') continue; /* Leading zeroes */
690 nigel 77 count++;
691 nigel 87
692 ph10 97 #ifndef EBCDIC /* ASCII coding */
693 nigel 77 if (cc >= 'a') cc -= 32; /* Convert to upper case */
694 nigel 87 c = (c << 4) + cc - ((cc < 'A')? '0' : ('A' - 10));
695 ph10 97 #else /* EBCDIC coding */
696 nigel 77 if (cc >= 'a' && cc <= 'z') cc += 64; /* Convert to upper case */
697 nigel 87 c = (c << 4) + cc - ((cc >= '0')? '0' : ('A' - 10));
698 nigel 77 #endif
699     }
700 nigel 87
701 nigel 77 if (*pt == '}')
702     {
703 nigel 87 if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
704 nigel 77 ptr = pt;
705     break;
706     }
707 nigel 87
708 nigel 77 /* If the sequence of hex digits does not end with '}', then we don't
709     recognize this construct; fall through to the normal \x handling. */
710     }
711    
712 nigel 87 /* Read just a single-byte hex-defined char */
713 nigel 77
714     c = 0;
715     while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
716     {
717     int cc; /* Some compilers don't like ++ */
718     cc = *(++ptr); /* in initializers */
719 ph10 97 #ifndef EBCDIC /* ASCII coding */
720 nigel 77 if (cc >= 'a') cc -= 32; /* Convert to upper case */
721     c = c * 16 + cc - ((cc < 'A')? '0' : ('A' - 10));
722 ph10 97 #else /* EBCDIC coding */
723 nigel 77 if (cc <= 'z') cc += 64; /* Convert to upper case */
724     c = c * 16 + cc - ((cc >= '0')? '0' : ('A' - 10));
725     #endif
726     }
727     break;
728    
729 nigel 93 /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
730     This coding is ASCII-specific, but then the whole concept of \cx is
731     ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
732 nigel 77
733     case 'c':
734     c = *(++ptr);
735     if (c == 0)
736     {
737     *errorcodeptr = ERR2;
738 ph10 213 break;
739 nigel 77 }
740    
741 ph10 97 #ifndef EBCDIC /* ASCII coding */
742 nigel 77 if (c >= 'a' && c <= 'z') c -= 32;
743     c ^= 0x40;
744 ph10 97 #else /* EBCDIC coding */
745 nigel 77 if (c >= 'a' && c <= 'z') c += 64;
746     c ^= 0xC0;
747     #endif
748     break;
749    
750     /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
751 ph10 274 other alphanumeric following \ is an error if PCRE_EXTRA was set;
752     otherwise, for Perl compatibility, it is a literal. This code looks a bit
753     odd, but there used to be some cases other than the default, and there may
754     be again in future, so I haven't "optimized" it. */
755 nigel 77
756     default:
757     if ((options & PCRE_EXTRA) != 0) switch(c)
758     {
759     default:
760     *errorcodeptr = ERR3;
761     break;
762     }
763     break;
764     }
765     }
766    
767     *ptrptr = ptr;
768     return c;
769     }
770    
771    
772    
773     #ifdef SUPPORT_UCP
774     /*************************************************
775     * Handle \P and \p *
776     *************************************************/
777    
778     /* This function is called after \P or \p has been encountered, provided that
779     PCRE is compiled with support for Unicode properties. On entry, ptrptr is
780     pointing at the P or p. On exit, it is pointing at the final character of the
781     escape sequence.
782    
783     Argument:
784     ptrptr points to the pattern position pointer
785     negptr points to a boolean that is set TRUE for negation else FALSE
786 nigel 87 dptr points to an int that is set to the detailed property value
787 nigel 77 errorcodeptr points to the error code variable
788    
789 nigel 87 Returns: type value from ucp_type_table, or -1 for an invalid type
790 nigel 77 */
791    
792     static int
793 nigel 87 get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
794 nigel 77 {
795     int c, i, bot, top;
796     const uschar *ptr = *ptrptr;
797 nigel 87 char name[32];
798 nigel 77
799     c = *(++ptr);
800     if (c == 0) goto ERROR_RETURN;
801    
802     *negptr = FALSE;
803    
804 nigel 87 /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
805     negation. */
806 nigel 77
807     if (c == '{')
808     {
809     if (ptr[1] == '^')
810     {
811     *negptr = TRUE;
812     ptr++;
813     }
814 ph10 199 for (i = 0; i < (int)sizeof(name) - 1; i++)
815 nigel 77 {
816     c = *(++ptr);
817     if (c == 0) goto ERROR_RETURN;
818     if (c == '}') break;
819     name[i] = c;
820     }
821 nigel 87 if (c !='}') goto ERROR_RETURN;
822 nigel 77 name[i] = 0;
823     }
824    
825     /* Otherwise there is just one following character */
826    
827     else
828     {
829     name[0] = c;
830     name[1] = 0;
831     }
832    
833     *ptrptr = ptr;
834    
835     /* Search for a recognized property name using binary chop */
836    
837     bot = 0;
838     top = _pcre_utt_size;
839    
840     while (bot < top)
841     {
842 nigel 87 i = (bot + top) >> 1;
843 ph10 240 c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
844 nigel 87 if (c == 0)
845     {
846     *dptr = _pcre_utt[i].value;
847     return _pcre_utt[i].type;
848     }
849 nigel 77 if (c > 0) bot = i + 1; else top = i;
850     }
851    
852     *errorcodeptr = ERR47;
853     *ptrptr = ptr;
854     return -1;
855    
856     ERROR_RETURN:
857     *errorcodeptr = ERR46;
858     *ptrptr = ptr;
859     return -1;
860     }
861     #endif
862    
863    
864    
865    
866     /*************************************************
867     * Check for counted repeat *
868     *************************************************/
869    
870     /* This function is called when a '{' is encountered in a place where it might
871     start a quantifier. It looks ahead to see if it really is a quantifier or not.
872     It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
873     where the ddds are digits.
874    
875     Arguments:
876     p pointer to the first char after '{'
877    
878     Returns: TRUE or FALSE
879     */
880    
881     static BOOL
882     is_counted_repeat(const uschar *p)
883     {
884     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
885     while ((digitab[*p] & ctype_digit) != 0) p++;
886     if (*p == '}') return TRUE;
887    
888     if (*p++ != ',') return FALSE;
889     if (*p == '}') return TRUE;
890    
891     if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
892     while ((digitab[*p] & ctype_digit) != 0) p++;
893    
894     return (*p == '}');
895     }
896    
897    
898    
899     /*************************************************
900     * Read repeat counts *
901     *************************************************/
902    
903     /* Read an item of the form {n,m} and return the values. This is called only
904     after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
905     so the syntax is guaranteed to be correct, but we need to check the values.
906    
907     Arguments:
908     p pointer to first char after '{'
909     minp pointer to int for min
910     maxp pointer to int for max
911     returned as -1 if no max
912     errorcodeptr points to error code variable
913    
914     Returns: pointer to '}' on success;
915     current ptr on error, with errorcodeptr set non-zero
916     */
917    
918     static const uschar *
919     read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
920     {
921     int min = 0;
922     int max = -1;
923    
924 nigel 81 /* Read the minimum value and do a paranoid check: a negative value indicates
925     an integer overflow. */
926    
927 nigel 77 while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - '0';
928 nigel 81 if (min < 0 || min > 65535)
929     {
930     *errorcodeptr = ERR5;
931     return p;
932     }
933 nigel 77
934 nigel 81 /* Read the maximum value if there is one, and again do a paranoid on its size.
935     Also, max must not be less than min. */
936    
937 nigel 77 if (*p == '}') max = min; else
938     {
939     if (*(++p) != '}')
940     {
941     max = 0;
942     while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - '0';
943 nigel 81 if (max < 0 || max > 65535)
944     {
945     *errorcodeptr = ERR5;
946     return p;
947     }
948 nigel 77 if (max < min)
949     {
950     *errorcodeptr = ERR4;
951     return p;
952     }
953     }
954     }
955    
956 nigel 81 /* Fill in the required variables, and pass back the pointer to the terminating
957     '}'. */
958 nigel 77
959 nigel 81 *minp = min;
960     *maxp = max;
961 nigel 77 return p;
962     }
963    
964    
965    
966     /*************************************************
967 nigel 93 * Find forward referenced subpattern *
968 nigel 91 *************************************************/
969    
970 nigel 93 /* This function scans along a pattern's text looking for capturing
971     subpatterns, and counting them. If it finds a named pattern that matches the
972     name it is given, it returns its number. Alternatively, if the name is NULL, it
973     returns when it reaches a given numbered subpattern. This is used for forward
974     references to subpatterns. We know that if (?P< is encountered, the name will
975     be terminated by '>' because that is checked in the first pass.
976 nigel 91
977     Arguments:
978 nigel 93 ptr current position in the pattern
979 ph10 345 cd compile background data
980 nigel 93 name name to seek, or NULL if seeking a numbered subpattern
981     lorn name length, or subpattern number if name is NULL
982     xmode TRUE if we are in /x mode
983 nigel 91
984     Returns: the number of the named subpattern, or -1 if not found
985     */
986    
987     static int
988 ph10 341 find_parens(const uschar *ptr, compile_data *cd, const uschar *name, int lorn,
989 nigel 93 BOOL xmode)
990 nigel 91 {
991     const uschar *thisname;
992 ph10 341 int count = cd->bracount;
993 nigel 93
994 nigel 91 for (; *ptr != 0; ptr++)
995     {
996 nigel 93 int term;
997    
998     /* Skip over backslashed characters and also entire \Q...\E */
999    
1000     if (*ptr == '\\')
1001     {
1002     if (*(++ptr) == 0) return -1;
1003     if (*ptr == 'Q') for (;;)
1004     {
1005     while (*(++ptr) != 0 && *ptr != '\\');
1006     if (*ptr == 0) return -1;
1007     if (*(++ptr) == 'E') break;
1008     }
1009     continue;
1010     }
1011    
1012 ph10 340 /* Skip over character classes; this logic must be similar to the way they
1013     are handled for real. If the first character is '^', skip it. Also, if the
1014     first few characters (either before or after ^) are \Q\E or \E we skip them
1015     too. This makes for compatibility with Perl. */
1016 nigel 93
1017     if (*ptr == '[')
1018     {
1019 ph10 340 BOOL negate_class = FALSE;
1020     for (;;)
1021     {
1022     int c = *(++ptr);
1023     if (c == '\\')
1024     {
1025     if (ptr[1] == 'E') ptr++;
1026     else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
1027     else break;
1028     }
1029     else if (!negate_class && c == '^')
1030     negate_class = TRUE;
1031     else break;
1032     }
1033    
1034     /* If the next character is ']', it is a data character that must be
1035 ph10 341 skipped, except in JavaScript compatibility mode. */
1036 ph10 345
1037     if (ptr[1] == ']' && (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1038     ptr++;
1039    
1040 nigel 93 while (*(++ptr) != ']')
1041     {
1042 ph10 220 if (*ptr == 0) return -1;
1043 nigel 93 if (*ptr == '\\')
1044     {
1045     if (*(++ptr) == 0) return -1;
1046     if (*ptr == 'Q') for (;;)
1047     {
1048     while (*(++ptr) != 0 && *ptr != '\\');
1049     if (*ptr == 0) return -1;
1050     if (*(++ptr) == 'E') break;
1051     }
1052     continue;
1053     }
1054     }
1055     continue;
1056     }
1057    
1058     /* Skip comments in /x mode */
1059    
1060     if (xmode && *ptr == '#')
1061     {
1062     while (*(++ptr) != 0 && *ptr != '\n');
1063     if (*ptr == 0) return -1;
1064     continue;
1065     }
1066    
1067     /* An opening parens must now be a real metacharacter */
1068    
1069 nigel 91 if (*ptr != '(') continue;
1070 ph10 210 if (ptr[1] != '?' && ptr[1] != '*')
1071 nigel 93 {
1072     count++;
1073     if (name == NULL && count == lorn) return count;
1074     continue;
1075     }
1076    
1077     ptr += 2;
1078     if (*ptr == 'P') ptr++; /* Allow optional P */
1079    
1080     /* We have to disambiguate (?<! and (?<= from (?<name> */
1081    
1082     if ((*ptr != '<' || ptr[1] == '!' || ptr[1] == '=') &&
1083     *ptr != '\'')
1084     continue;
1085    
1086 nigel 91 count++;
1087 nigel 93
1088     if (name == NULL && count == lorn) return count;
1089     term = *ptr++;
1090     if (term == '<') term = '>';
1091 nigel 91 thisname = ptr;
1092 nigel 93 while (*ptr != term) ptr++;
1093     if (name != NULL && lorn == ptr - thisname &&
1094     strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1095 nigel 91 return count;
1096     }
1097 nigel 93
1098 nigel 91 return -1;
1099     }
1100    
1101    
1102    
1103     /*************************************************
1104 nigel 77 * Find first significant op code *
1105     *************************************************/
1106    
1107     /* This is called by several functions that scan a compiled expression looking
1108     for a fixed first character, or an anchoring op code etc. It skips over things
1109     that do not influence this. For some calls, a change of option is important.
1110     For some calls, it makes sense to skip negative forward and all backward
1111     assertions, and also the \b assertion; for others it does not.
1112    
1113     Arguments:
1114     code pointer to the start of the group
1115     options pointer to external options
1116     optbit the option bit whose changing is significant, or
1117     zero if none are
1118     skipassert TRUE if certain assertions are to be skipped
1119    
1120     Returns: pointer to the first significant opcode
1121     */
1122    
1123     static const uschar*
1124     first_significant_code(const uschar *code, int *options, int optbit,
1125     BOOL skipassert)
1126     {
1127     for (;;)
1128     {
1129     switch ((int)*code)
1130     {
1131     case OP_OPT:
1132     if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1133     *options = (int)code[1];
1134     code += 2;
1135     break;
1136    
1137     case OP_ASSERT_NOT:
1138     case OP_ASSERTBACK:
1139     case OP_ASSERTBACK_NOT:
1140     if (!skipassert) return code;
1141     do code += GET(code, 1); while (*code == OP_ALT);
1142     code += _pcre_OP_lengths[*code];
1143     break;
1144    
1145     case OP_WORD_BOUNDARY:
1146     case OP_NOT_WORD_BOUNDARY:
1147     if (!skipassert) return code;
1148     /* Fall through */
1149    
1150     case OP_CALLOUT:
1151     case OP_CREF:
1152 nigel 93 case OP_RREF:
1153     case OP_DEF:
1154 nigel 77 code += _pcre_OP_lengths[*code];
1155     break;
1156    
1157     default:
1158     return code;
1159     }
1160     }
1161     /* Control never reaches here */
1162     }
1163    
1164    
1165    
1166    
1167     /*************************************************
1168     * Find the fixed length of a pattern *
1169     *************************************************/
1170    
1171     /* Scan a pattern and compute the fixed length of subject that will match it,
1172     if the length is fixed. This is needed for dealing with backward assertions.
1173     In UTF8 mode, the result is in characters rather than bytes.
1174    
1175     Arguments:
1176     code points to the start of the pattern (the bracket)
1177     options the compiling options
1178    
1179     Returns: the fixed length, or -1 if there is no fixed length,
1180     or -2 if \C was encountered
1181     */
1182    
1183     static int
1184     find_fixedlength(uschar *code, int options)
1185     {
1186     int length = -1;
1187    
1188     register int branchlength = 0;
1189     register uschar *cc = code + 1 + LINK_SIZE;
1190    
1191     /* Scan along the opcodes for this branch. If we get to the end of the
1192     branch, check the length against that of the other branches. */
1193    
1194     for (;;)
1195     {
1196     int d;
1197     register int op = *cc;
1198     switch (op)
1199     {
1200 nigel 93 case OP_CBRA:
1201 nigel 77 case OP_BRA:
1202     case OP_ONCE:
1203     case OP_COND:
1204 nigel 93 d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options);
1205 nigel 77 if (d < 0) return d;
1206     branchlength += d;
1207     do cc += GET(cc, 1); while (*cc == OP_ALT);
1208     cc += 1 + LINK_SIZE;
1209     break;
1210    
1211     /* Reached end of a branch; if it's a ket it is the end of a nested
1212     call. If it's ALT it is an alternation in a nested call. If it is
1213     END it's the end of the outer call. All can be handled by the same code. */
1214    
1215     case OP_ALT:
1216     case OP_KET:
1217     case OP_KETRMAX:
1218     case OP_KETRMIN:
1219     case OP_END:
1220     if (length < 0) length = branchlength;
1221     else if (length != branchlength) return -1;
1222     if (*cc != OP_ALT) return length;
1223     cc += 1 + LINK_SIZE;
1224     branchlength = 0;
1225     break;
1226    
1227     /* Skip over assertive subpatterns */
1228    
1229     case OP_ASSERT:
1230     case OP_ASSERT_NOT:
1231     case OP_ASSERTBACK:
1232     case OP_ASSERTBACK_NOT:
1233     do cc += GET(cc, 1); while (*cc == OP_ALT);
1234     /* Fall through */
1235    
1236     /* Skip over things that don't match chars */
1237    
1238     case OP_REVERSE:
1239     case OP_CREF:
1240 nigel 93 case OP_RREF:
1241     case OP_DEF:
1242 nigel 77 case OP_OPT:
1243     case OP_CALLOUT:
1244     case OP_SOD:
1245     case OP_SOM:
1246     case OP_EOD:
1247     case OP_EODN:
1248     case OP_CIRC:
1249     case OP_DOLL:
1250     case OP_NOT_WORD_BOUNDARY:
1251     case OP_WORD_BOUNDARY:
1252     cc += _pcre_OP_lengths[*cc];
1253     break;
1254    
1255     /* Handle literal characters */
1256    
1257     case OP_CHAR:
1258     case OP_CHARNC:
1259 nigel 91 case OP_NOT:
1260 nigel 77 branchlength++;
1261     cc += 2;
1262     #ifdef SUPPORT_UTF8
1263     if ((options & PCRE_UTF8) != 0)
1264     {
1265     while ((*cc & 0xc0) == 0x80) cc++;
1266     }
1267     #endif
1268     break;
1269    
1270     /* Handle exact repetitions. The count is already in characters, but we
1271     need to skip over a multibyte character in UTF8 mode. */
1272    
1273     case OP_EXACT:
1274     branchlength += GET2(cc,1);
1275     cc += 4;
1276     #ifdef SUPPORT_UTF8
1277     if ((options & PCRE_UTF8) != 0)
1278     {
1279     while((*cc & 0x80) == 0x80) cc++;
1280     }
1281     #endif
1282     break;
1283    
1284     case OP_TYPEEXACT:
1285     branchlength += GET2(cc,1);
1286 ph10 220 if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1287 nigel 77 cc += 4;
1288     break;
1289    
1290     /* Handle single-char matchers */
1291    
1292     case OP_PROP:
1293     case OP_NOTPROP:
1294 nigel 87 cc += 2;
1295 nigel 77 /* Fall through */
1296    
1297     case OP_NOT_DIGIT:
1298     case OP_DIGIT:
1299     case OP_NOT_WHITESPACE:
1300     case OP_WHITESPACE:
1301     case OP_NOT_WORDCHAR:
1302     case OP_WORDCHAR:
1303     case OP_ANY:
1304 ph10 342 case OP_ALLANY:
1305 nigel 77 branchlength++;
1306     cc++;
1307     break;
1308    
1309     /* The single-byte matcher isn't allowed */
1310    
1311     case OP_ANYBYTE:
1312     return -2;
1313    
1314     /* Check a class for variable quantification */
1315    
1316     #ifdef SUPPORT_UTF8
1317     case OP_XCLASS:
1318     cc += GET(cc, 1) - 33;
1319     /* Fall through */
1320     #endif
1321    
1322     case OP_CLASS:
1323     case OP_NCLASS:
1324     cc += 33;
1325    
1326     switch (*cc)
1327     {
1328     case OP_CRSTAR:
1329     case OP_CRMINSTAR:
1330     case OP_CRQUERY:
1331     case OP_CRMINQUERY:
1332     return -1;
1333    
1334     case OP_CRRANGE:
1335     case OP_CRMINRANGE:
1336     if (GET2(cc,1) != GET2(cc,3)) return -1;
1337     branchlength += GET2(cc,1);
1338     cc += 5;
1339     break;
1340    
1341     default:
1342     branchlength++;
1343     }
1344     break;
1345    
1346     /* Anything else is variable length */
1347    
1348     default:
1349     return -1;
1350     }
1351     }
1352     /* Control never gets here */
1353     }
1354    
1355    
1356    
1357    
1358     /*************************************************
1359     * Scan compiled regex for numbered bracket *
1360     *************************************************/
1361    
1362     /* This little function scans through a compiled pattern until it finds a
1363     capturing bracket with the given number.
1364    
1365     Arguments:
1366     code points to start of expression
1367     utf8 TRUE in UTF-8 mode
1368     number the required bracket number
1369    
1370     Returns: pointer to the opcode for the bracket, or NULL if not found
1371     */
1372    
1373     static const uschar *
1374     find_bracket(const uschar *code, BOOL utf8, int number)
1375     {
1376     for (;;)
1377     {
1378     register int c = *code;
1379     if (c == OP_END) return NULL;
1380 nigel 91
1381     /* XCLASS is used for classes that cannot be represented just by a bit
1382     map. This includes negated single high-valued characters. The length in
1383     the table is zero; the actual length is stored in the compiled code. */
1384    
1385     if (c == OP_XCLASS) code += GET(code, 1);
1386    
1387 nigel 93 /* Handle capturing bracket */
1388 nigel 91
1389 nigel 93 else if (c == OP_CBRA)
1390 nigel 77 {
1391 nigel 93 int n = GET2(code, 1+LINK_SIZE);
1392 nigel 77 if (n == number) return (uschar *)code;
1393 nigel 93 code += _pcre_OP_lengths[c];
1394 nigel 77 }
1395 nigel 91
1396 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1397     repeated character types, we have to test for \p and \P, which have an extra
1398 ph10 218 two bytes of parameters. */
1399 nigel 91
1400 nigel 77 else
1401     {
1402 ph10 218 switch(c)
1403     {
1404     case OP_TYPESTAR:
1405     case OP_TYPEMINSTAR:
1406     case OP_TYPEPLUS:
1407     case OP_TYPEMINPLUS:
1408     case OP_TYPEQUERY:
1409     case OP_TYPEMINQUERY:
1410     case OP_TYPEPOSSTAR:
1411     case OP_TYPEPOSPLUS:
1412     case OP_TYPEPOSQUERY:
1413     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1414 ph10 220 break;
1415 ph10 221
1416     case OP_TYPEUPTO:
1417     case OP_TYPEMINUPTO:
1418     case OP_TYPEEXACT:
1419     case OP_TYPEPOSUPTO:
1420     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1421     break;
1422 ph10 220 }
1423    
1424 ph10 218 /* Add in the fixed length from the table */
1425 ph10 220
1426 nigel 77 code += _pcre_OP_lengths[c];
1427 ph10 220
1428 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1429     a multi-byte character. The length in the table is a minimum, so we have to
1430     arrange to skip the extra bytes. */
1431 ph10 220
1432 ph10 107 #ifdef SUPPORT_UTF8
1433 nigel 77 if (utf8) switch(c)
1434     {
1435     case OP_CHAR:
1436     case OP_CHARNC:
1437     case OP_EXACT:
1438     case OP_UPTO:
1439     case OP_MINUPTO:
1440 nigel 93 case OP_POSUPTO:
1441 nigel 77 case OP_STAR:
1442     case OP_MINSTAR:
1443 nigel 93 case OP_POSSTAR:
1444 nigel 77 case OP_PLUS:
1445     case OP_MINPLUS:
1446 nigel 93 case OP_POSPLUS:
1447 nigel 77 case OP_QUERY:
1448     case OP_MINQUERY:
1449 nigel 93 case OP_POSQUERY:
1450     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1451 nigel 77 break;
1452     }
1453 ph10 111 #endif
1454 nigel 77 }
1455     }
1456     }
1457    
1458    
1459    
1460     /*************************************************
1461     * Scan compiled regex for recursion reference *
1462     *************************************************/
1463    
1464     /* This little function scans through a compiled pattern until it finds an
1465     instance of OP_RECURSE.
1466    
1467     Arguments:
1468     code points to start of expression
1469     utf8 TRUE in UTF-8 mode
1470    
1471     Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1472     */
1473    
1474     static const uschar *
1475     find_recurse(const uschar *code, BOOL utf8)
1476     {
1477     for (;;)
1478     {
1479     register int c = *code;
1480     if (c == OP_END) return NULL;
1481 nigel 91 if (c == OP_RECURSE) return code;
1482 ph10 220
1483 nigel 91 /* XCLASS is used for classes that cannot be represented just by a bit
1484     map. This includes negated single high-valued characters. The length in
1485     the table is zero; the actual length is stored in the compiled code. */
1486    
1487     if (c == OP_XCLASS) code += GET(code, 1);
1488    
1489 ph10 220 /* Otherwise, we can get the item's length from the table, except that for
1490     repeated character types, we have to test for \p and \P, which have an extra
1491 ph10 218 two bytes of parameters. */
1492 nigel 91
1493 nigel 77 else
1494     {
1495 ph10 218 switch(c)
1496     {
1497     case OP_TYPESTAR:
1498     case OP_TYPEMINSTAR:
1499     case OP_TYPEPLUS:
1500     case OP_TYPEMINPLUS:
1501     case OP_TYPEQUERY:
1502     case OP_TYPEMINQUERY:
1503     case OP_TYPEPOSSTAR:
1504     case OP_TYPEPOSPLUS:
1505     case OP_TYPEPOSQUERY:
1506     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1507 ph10 220 break;
1508 ph10 221
1509     case OP_TYPEPOSUPTO:
1510     case OP_TYPEUPTO:
1511     case OP_TYPEMINUPTO:
1512     case OP_TYPEEXACT:
1513     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1514     break;
1515 ph10 220 }
1516    
1517 ph10 218 /* Add in the fixed length from the table */
1518    
1519 nigel 77 code += _pcre_OP_lengths[c];
1520 ph10 220
1521 ph10 218 /* In UTF-8 mode, opcodes that are followed by a character may be followed
1522     by a multi-byte character. The length in the table is a minimum, so we have
1523     to arrange to skip the extra bytes. */
1524 ph10 220
1525 ph10 107 #ifdef SUPPORT_UTF8
1526 nigel 77 if (utf8) switch(c)
1527     {
1528     case OP_CHAR:
1529     case OP_CHARNC:
1530     case OP_EXACT:
1531     case OP_UPTO:
1532     case OP_MINUPTO:
1533 nigel 93 case OP_POSUPTO:
1534 nigel 77 case OP_STAR:
1535     case OP_MINSTAR:
1536 nigel 93 case OP_POSSTAR:
1537 nigel 77 case OP_PLUS:
1538     case OP_MINPLUS:
1539 nigel 93 case OP_POSPLUS:
1540 nigel 77 case OP_QUERY:
1541     case OP_MINQUERY:
1542 nigel 93 case OP_POSQUERY:
1543     if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1544 nigel 77 break;
1545     }
1546 ph10 111 #endif
1547 nigel 77 }
1548     }
1549     }
1550    
1551    
1552    
1553     /*************************************************
1554     * Scan compiled branch for non-emptiness *
1555     *************************************************/
1556    
1557     /* This function scans through a branch of a compiled pattern to see whether it
1558 nigel 93 can match the empty string or not. It is called from could_be_empty()
1559     below and from compile_branch() when checking for an unlimited repeat of a
1560     group that can match nothing. Note that first_significant_code() skips over
1561 ph10 282 backward and negative forward assertions when its final argument is TRUE. If we
1562     hit an unclosed bracket, we return "empty" - this means we've struck an inner
1563     bracket whose current branch will already have been scanned.
1564 nigel 77
1565     Arguments:
1566     code points to start of search
1567     endcode points to where to stop
1568     utf8 TRUE if in UTF8 mode
1569    
1570     Returns: TRUE if what is matched could be empty
1571     */
1572    
1573     static BOOL
1574     could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8)
1575     {
1576     register int c;
1577 nigel 93 for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1578 nigel 77 code < endcode;
1579     code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1580     {
1581     const uschar *ccode;
1582    
1583     c = *code;
1584 ph10 286
1585     /* Skip over forward assertions; the other assertions are skipped by
1586 ph10 282 first_significant_code() with a TRUE final argument. */
1587 ph10 286
1588 ph10 282 if (c == OP_ASSERT)
1589 ph10 286 {
1590 ph10 282 do code += GET(code, 1); while (*code == OP_ALT);
1591     c = *code;
1592     continue;
1593 ph10 286 }
1594 ph10 172
1595 ph10 170 /* Groups with zero repeats can of course be empty; skip them. */
1596 nigel 77
1597 ph10 335 if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1598 ph10 170 {
1599 ph10 172 code += _pcre_OP_lengths[c];
1600 ph10 170 do code += GET(code, 1); while (*code == OP_ALT);
1601     c = *code;
1602     continue;
1603     }
1604    
1605     /* For other groups, scan the branches. */
1606 ph10 172
1607 ph10 206 if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1608 nigel 77 {
1609     BOOL empty_branch;
1610     if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1611    
1612     /* Scan a closed bracket */
1613    
1614     empty_branch = FALSE;
1615     do
1616     {
1617     if (!empty_branch && could_be_empty_branch(code, endcode, utf8))
1618     empty_branch = TRUE;
1619     code += GET(code, 1);
1620     }
1621     while (*code == OP_ALT);
1622     if (!empty_branch) return FALSE; /* All branches are non-empty */
1623 ph10 172 c = *code;
1624 nigel 93 continue;
1625 nigel 77 }
1626    
1627 nigel 93 /* Handle the other opcodes */
1628    
1629     switch (c)
1630 nigel 77 {
1631 ph10 216 /* Check for quantifiers after a class. XCLASS is used for classes that
1632     cannot be represented just by a bit map. This includes negated single
1633     high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1634 ph10 220 actual length is stored in the compiled code, so we must update "code"
1635 ph10 216 here. */
1636 nigel 77
1637     #ifdef SUPPORT_UTF8
1638     case OP_XCLASS:
1639 ph10 216 ccode = code += GET(code, 1);
1640 nigel 77 goto CHECK_CLASS_REPEAT;
1641     #endif
1642    
1643     case OP_CLASS:
1644     case OP_NCLASS:
1645     ccode = code + 33;
1646    
1647     #ifdef SUPPORT_UTF8
1648     CHECK_CLASS_REPEAT:
1649     #endif
1650    
1651     switch (*ccode)
1652     {
1653     case OP_CRSTAR: /* These could be empty; continue */
1654     case OP_CRMINSTAR:
1655     case OP_CRQUERY:
1656     case OP_CRMINQUERY:
1657     break;
1658    
1659     default: /* Non-repeat => class must match */
1660     case OP_CRPLUS: /* These repeats aren't empty */
1661     case OP_CRMINPLUS:
1662     return FALSE;
1663    
1664     case OP_CRRANGE:
1665     case OP_CRMINRANGE:
1666     if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
1667     break;
1668     }
1669     break;
1670    
1671     /* Opcodes that must match a character */
1672    
1673     case OP_PROP:
1674     case OP_NOTPROP:
1675     case OP_EXTUNI:
1676     case OP_NOT_DIGIT:
1677     case OP_DIGIT:
1678     case OP_NOT_WHITESPACE:
1679     case OP_WHITESPACE:
1680     case OP_NOT_WORDCHAR:
1681     case OP_WORDCHAR:
1682     case OP_ANY:
1683 ph10 345 case OP_ALLANY:
1684 nigel 77 case OP_ANYBYTE:
1685     case OP_CHAR:
1686     case OP_CHARNC:
1687     case OP_NOT:
1688     case OP_PLUS:
1689     case OP_MINPLUS:
1690 nigel 93 case OP_POSPLUS:
1691 nigel 77 case OP_EXACT:
1692     case OP_NOTPLUS:
1693     case OP_NOTMINPLUS:
1694 nigel 93 case OP_NOTPOSPLUS:
1695 nigel 77 case OP_NOTEXACT:
1696     case OP_TYPEPLUS:
1697     case OP_TYPEMINPLUS:
1698 nigel 93 case OP_TYPEPOSPLUS:
1699 nigel 77 case OP_TYPEEXACT:
1700     return FALSE;
1701 ph10 227
1702     /* These are going to continue, as they may be empty, but we have to
1703     fudge the length for the \p and \P cases. */
1704    
1705 ph10 224 case OP_TYPESTAR:
1706     case OP_TYPEMINSTAR:
1707     case OP_TYPEPOSSTAR:
1708     case OP_TYPEQUERY:
1709     case OP_TYPEMINQUERY:
1710     case OP_TYPEPOSQUERY:
1711     if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1712 ph10 227 break;
1713    
1714 ph10 224 /* Same for these */
1715 ph10 227
1716 ph10 224 case OP_TYPEUPTO:
1717     case OP_TYPEMINUPTO:
1718     case OP_TYPEPOSUPTO:
1719     if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1720     break;
1721 nigel 77
1722     /* End of branch */
1723    
1724     case OP_KET:
1725     case OP_KETRMAX:
1726     case OP_KETRMIN:
1727     case OP_ALT:
1728     return TRUE;
1729    
1730 nigel 93 /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
1731     MINUPTO, and POSUPTO may be followed by a multibyte character */
1732 nigel 77
1733     #ifdef SUPPORT_UTF8
1734     case OP_STAR:
1735     case OP_MINSTAR:
1736 nigel 93 case OP_POSSTAR:
1737 nigel 77 case OP_QUERY:
1738     case OP_MINQUERY:
1739 nigel 93 case OP_POSQUERY:
1740 nigel 77 case OP_UPTO:
1741     case OP_MINUPTO:
1742 nigel 93 case OP_POSUPTO:
1743 nigel 77 if (utf8) while ((code[2] & 0xc0) == 0x80) code++;
1744     break;
1745     #endif
1746     }
1747     }
1748    
1749     return TRUE;
1750     }
1751    
1752    
1753    
1754     /*************************************************
1755     * Scan compiled regex for non-emptiness *
1756     *************************************************/
1757    
1758     /* This function is called to check for left recursive calls. We want to check
1759     the current branch of the current pattern to see if it could match the empty
1760     string. If it could, we must look outwards for branches at other levels,
1761     stopping when we pass beyond the bracket which is the subject of the recursion.
1762    
1763     Arguments:
1764     code points to start of the recursion
1765     endcode points to where to stop (current RECURSE item)
1766     bcptr points to the chain of current (unclosed) branch starts
1767     utf8 TRUE if in UTF-8 mode
1768    
1769     Returns: TRUE if what is matched could be empty
1770     */
1771    
1772     static BOOL
1773     could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1774     BOOL utf8)
1775     {
1776     while (bcptr != NULL && bcptr->current >= code)
1777     {
1778     if (!could_be_empty_branch(bcptr->current, endcode, utf8)) return FALSE;
1779     bcptr = bcptr->outer;
1780     }
1781     return TRUE;
1782     }
1783    
1784    
1785    
1786     /*************************************************
1787     * Check for POSIX class syntax *
1788     *************************************************/
1789    
1790     /* This function is called when the sequence "[:" or "[." or "[=" is
1791 ph10 295 encountered in a character class. It checks whether this is followed by a
1792 ph10 298 sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
1793 ph10 295 reach an unescaped ']' without the special preceding character, return FALSE.
1794 nigel 77
1795 ph10 298 Originally, this function only recognized a sequence of letters between the
1796     terminators, but it seems that Perl recognizes any sequence of characters,
1797     though of course unknown POSIX names are subsequently rejected. Perl gives an
1798     "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
1799     didn't consider this to be a POSIX class. Likewise for [:1234:].
1800 ph10 295
1801 ph10 298 The problem in trying to be exactly like Perl is in the handling of escapes. We
1802     have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
1803     class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
1804     below handles the special case of \], but does not try to do any other escape
1805     processing. This makes it different from Perl for cases such as [:l\ower:]
1806 ph10 295 where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
1807 ph10 298 "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
1808 ph10 295 I think.
1809    
1810     Arguments:
1811 nigel 77 ptr pointer to the initial [
1812     endptr where to return the end pointer
1813    
1814     Returns: TRUE or FALSE
1815     */
1816    
1817     static BOOL
1818 ph10 295 check_posix_syntax(const uschar *ptr, const uschar **endptr)
1819 nigel 77 {
1820     int terminator; /* Don't combine these lines; the Solaris cc */
1821     terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
1822 ph10 295 for (++ptr; *ptr != 0; ptr++)
1823 nigel 77 {
1824 ph10 295 if (*ptr == '\\' && ptr[1] == ']') ptr++; else
1825 ph10 298 {
1826     if (*ptr == ']') return FALSE;
1827 ph10 295 if (*ptr == terminator && ptr[1] == ']')
1828     {
1829     *endptr = ptr;
1830     return TRUE;
1831 ph10 298 }
1832     }
1833     }
1834 nigel 77 return FALSE;
1835     }
1836    
1837    
1838    
1839    
1840     /*************************************************
1841     * Check POSIX class name *
1842     *************************************************/
1843    
1844     /* This function is called to check the name given in a POSIX-style class entry
1845     such as [:alnum:].
1846    
1847     Arguments:
1848     ptr points to the first letter
1849     len the length of the name
1850    
1851     Returns: a value representing the name, or -1 if unknown
1852     */
1853    
1854     static int
1855     check_posix_name(const uschar *ptr, int len)
1856     {
1857 ph10 240 const char *pn = posix_names;
1858 nigel 77 register int yield = 0;
1859     while (posix_name_lengths[yield] != 0)
1860     {
1861     if (len == posix_name_lengths[yield] &&
1862 ph10 240 strncmp((const char *)ptr, pn, len) == 0) return yield;
1863 ph10 243 pn += posix_name_lengths[yield] + 1;
1864 nigel 77 yield++;
1865     }
1866     return -1;
1867     }
1868    
1869    
1870     /*************************************************
1871     * Adjust OP_RECURSE items in repeated group *
1872     *************************************************/
1873    
1874     /* OP_RECURSE items contain an offset from the start of the regex to the group
1875     that is referenced. This means that groups can be replicated for fixed
1876     repetition simply by copying (because the recursion is allowed to refer to
1877     earlier groups that are outside the current group). However, when a group is
1878 ph10 335 optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
1879     inserted before it, after it has been compiled. This means that any OP_RECURSE
1880     items within it that refer to the group itself or any contained groups have to
1881     have their offsets adjusted. That one of the jobs of this function. Before it
1882     is called, the partially compiled regex must be temporarily terminated with
1883     OP_END.
1884 nigel 77
1885 nigel 93 This function has been extended with the possibility of forward references for
1886     recursions and subroutine calls. It must also check the list of such references
1887     for the group we are dealing with. If it finds that one of the recursions in
1888     the current group is on this list, it adjusts the offset in the list, not the
1889     value in the reference (which is a group number).
1890    
1891 nigel 77 Arguments:
1892     group points to the start of the group
1893     adjust the amount by which the group is to be moved
1894     utf8 TRUE in UTF-8 mode
1895     cd contains pointers to tables etc.
1896 nigel 93 save_hwm the hwm forward reference pointer at the start of the group
1897 nigel 77
1898     Returns: nothing
1899     */
1900    
1901     static void
1902 nigel 93 adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
1903     uschar *save_hwm)
1904 nigel 77 {
1905     uschar *ptr = group;
1906 ph10 224
1907 nigel 77 while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
1908     {
1909 nigel 93 int offset;
1910     uschar *hc;
1911    
1912     /* See if this recursion is on the forward reference list. If so, adjust the
1913     reference. */
1914 ph10 345
1915 nigel 93 for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
1916     {
1917     offset = GET(hc, 0);
1918     if (cd->start_code + offset == ptr + 1)
1919     {
1920     PUT(hc, 0, offset + adjust);
1921     break;
1922     }
1923     }
1924    
1925     /* Otherwise, adjust the recursion offset if it's after the start of this
1926     group. */
1927    
1928     if (hc >= cd->hwm)
1929     {
1930     offset = GET(ptr, 1);
1931     if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
1932     }
1933    
1934 nigel 77 ptr += 1 + LINK_SIZE;
1935     }
1936     }
1937    
1938    
1939    
1940     /*************************************************
1941     * Insert an automatic callout point *
1942     *************************************************/
1943    
1944     /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
1945     callout points before each pattern item.
1946    
1947     Arguments:
1948     code current code pointer
1949     ptr current pattern pointer
1950     cd pointers to tables etc
1951    
1952     Returns: new code pointer
1953     */
1954    
1955     static uschar *
1956     auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
1957     {
1958     *code++ = OP_CALLOUT;
1959     *code++ = 255;
1960     PUT(code, 0, ptr - cd->start_pattern); /* Pattern offset */
1961     PUT(code, LINK_SIZE, 0); /* Default length */
1962     return code + 2*LINK_SIZE;
1963     }
1964    
1965    
1966    
1967     /*************************************************
1968     * Complete a callout item *
1969     *************************************************/
1970    
1971     /* A callout item contains the length of the next item in the pattern, which
1972     we can't fill in till after we have reached the relevant point. This is used
1973     for both automatic and manual callouts.
1974    
1975     Arguments:
1976     previous_callout points to previous callout item
1977     ptr current pattern pointer
1978     cd pointers to tables etc
1979    
1980     Returns: nothing
1981     */
1982    
1983     static void
1984     complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
1985     {
1986     int length = ptr - cd->start_pattern - GET(previous_callout, 2);
1987     PUT(previous_callout, 2 + LINK_SIZE, length);
1988     }
1989    
1990    
1991    
1992     #ifdef SUPPORT_UCP
1993     /*************************************************
1994     * Get othercase range *
1995     *************************************************/
1996    
1997     /* This function is passed the start and end of a class range, in UTF-8 mode
1998     with UCP support. It searches up the characters, looking for internal ranges of
1999     characters in the "other" case. Each call returns the next one, updating the
2000     start address.
2001    
2002     Arguments:
2003     cptr points to starting character value; updated
2004     d end value
2005     ocptr where to put start of othercase range
2006     odptr where to put end of othercase range
2007    
2008     Yield: TRUE when range returned; FALSE when no more
2009     */
2010    
2011     static BOOL
2012 nigel 93 get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2013     unsigned int *odptr)
2014 nigel 77 {
2015 nigel 93 unsigned int c, othercase, next;
2016 nigel 77
2017     for (c = *cptr; c <= d; c++)
2018 ph10 349 { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
2019 nigel 77
2020     if (c > d) return FALSE;
2021    
2022     *ocptr = othercase;
2023     next = othercase + 1;
2024    
2025     for (++c; c <= d; c++)
2026     {
2027 ph10 349 if (UCD_OTHERCASE(c) != next) break;
2028 nigel 77 next++;
2029     }
2030    
2031     *odptr = next - 1;
2032     *cptr = c;
2033    
2034     return TRUE;
2035     }
2036     #endif /* SUPPORT_UCP */
2037    
2038    
2039 nigel 93
2040 nigel 77 /*************************************************
2041 nigel 93 * Check if auto-possessifying is possible *
2042     *************************************************/
2043    
2044     /* This function is called for unlimited repeats of certain items, to see
2045     whether the next thing could possibly match the repeated item. If not, it makes
2046     sense to automatically possessify the repeated item.
2047    
2048     Arguments:
2049     op_code the repeated op code
2050     this data for this item, depends on the opcode
2051     utf8 TRUE in UTF-8 mode
2052     utf8_char used for utf8 character bytes, NULL if not relevant
2053     ptr next character in pattern
2054     options options bits
2055     cd contains pointers to tables etc.
2056    
2057     Returns: TRUE if possessifying is wanted
2058     */
2059    
2060     static BOOL
2061     check_auto_possessive(int op_code, int item, BOOL utf8, uschar *utf8_char,
2062     const uschar *ptr, int options, compile_data *cd)
2063     {
2064     int next;
2065    
2066     /* Skip whitespace and comments in extended mode */
2067    
2068     if ((options & PCRE_EXTENDED) != 0)
2069     {
2070     for (;;)
2071     {
2072     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2073     if (*ptr == '#')
2074     {
2075     while (*(++ptr) != 0)
2076     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2077     }
2078     else break;
2079     }
2080     }
2081    
2082     /* If the next item is one that we can handle, get its value. A non-negative
2083     value is a character, a negative value is an escape value. */
2084    
2085     if (*ptr == '\\')
2086     {
2087     int temperrorcode = 0;
2088     next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2089     if (temperrorcode != 0) return FALSE;
2090     ptr++; /* Point after the escape sequence */
2091     }
2092    
2093     else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2094     {
2095     #ifdef SUPPORT_UTF8
2096     if (utf8) { GETCHARINC(next, ptr); } else
2097     #endif
2098     next = *ptr++;
2099     }
2100    
2101     else return FALSE;
2102    
2103     /* Skip whitespace and comments in extended mode */
2104    
2105     if ((options & PCRE_EXTENDED) != 0)
2106     {
2107     for (;;)
2108     {
2109     while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
2110     if (*ptr == '#')
2111     {
2112     while (*(++ptr) != 0)
2113     if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2114     }
2115     else break;
2116     }
2117     }
2118    
2119     /* If the next thing is itself optional, we have to give up. */
2120    
2121     if (*ptr == '*' || *ptr == '?' || strncmp((char *)ptr, "{0,", 3) == 0)
2122     return FALSE;
2123    
2124     /* Now compare the next item with the previous opcode. If the previous is a
2125     positive single character match, "item" either contains the character or, if
2126     "item" is greater than 127 in utf8 mode, the character's bytes are in
2127     utf8_char. */
2128    
2129    
2130     /* Handle cases when the next item is a character. */
2131    
2132     if (next >= 0) switch(op_code)
2133     {
2134     case OP_CHAR:
2135     #ifdef SUPPORT_UTF8
2136     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2137     #endif
2138     return item != next;
2139    
2140     /* For CHARNC (caseless character) we must check the other case. If we have
2141     Unicode property support, we can use it to test the other case of
2142     high-valued characters. */
2143    
2144     case OP_CHARNC:
2145     #ifdef SUPPORT_UTF8
2146     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2147     #endif
2148     if (item == next) return FALSE;
2149     #ifdef SUPPORT_UTF8
2150     if (utf8)
2151     {
2152     unsigned int othercase;
2153     if (next < 128) othercase = cd->fcc[next]; else
2154     #ifdef SUPPORT_UCP
2155 ph10 349 othercase = UCD_OTHERCASE((unsigned int)next);
2156 nigel 93 #else
2157     othercase = NOTACHAR;
2158     #endif
2159     return (unsigned int)item != othercase;
2160     }
2161     else
2162     #endif /* SUPPORT_UTF8 */
2163     return (item != cd->fcc[next]); /* Non-UTF-8 mode */
2164    
2165     /* For OP_NOT, "item" must be a single-byte character. */
2166    
2167     case OP_NOT:
2168     if (item == next) return TRUE;
2169     if ((options & PCRE_CASELESS) == 0) return FALSE;
2170     #ifdef SUPPORT_UTF8
2171     if (utf8)
2172     {
2173     unsigned int othercase;
2174     if (next < 128) othercase = cd->fcc[next]; else
2175     #ifdef SUPPORT_UCP
2176 ph10 349 othercase = UCD_OTHERCASE(next);
2177 nigel 93 #else
2178     othercase = NOTACHAR;
2179     #endif
2180     return (unsigned int)item == othercase;
2181     }
2182     else
2183     #endif /* SUPPORT_UTF8 */
2184     return (item == cd->fcc[next]); /* Non-UTF-8 mode */
2185    
2186     case OP_DIGIT:
2187     return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2188    
2189     case OP_NOT_DIGIT:
2190     return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2191    
2192     case OP_WHITESPACE:
2193     return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2194    
2195     case OP_NOT_WHITESPACE:
2196     return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2197    
2198     case OP_WORDCHAR:
2199     return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2200    
2201     case OP_NOT_WORDCHAR:
2202     return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2203    
2204 ph10 180 case OP_HSPACE:
2205     case OP_NOT_HSPACE:
2206     switch(next)
2207     {
2208     case 0x09:
2209     case 0x20:
2210     case 0xa0:
2211     case 0x1680:
2212     case 0x180e:
2213     case 0x2000:
2214     case 0x2001:
2215     case 0x2002:
2216     case 0x2003:
2217     case 0x2004:
2218     case 0x2005:
2219     case 0x2006:
2220     case 0x2007:
2221     case 0x2008:
2222     case 0x2009:
2223     case 0x200A:
2224     case 0x202f:
2225     case 0x205f:
2226     case 0x3000:
2227     return op_code != OP_HSPACE;
2228     default:
2229     return op_code == OP_HSPACE;
2230     }
2231    
2232     case OP_VSPACE:
2233     case OP_NOT_VSPACE:
2234     switch(next)
2235     {
2236     case 0x0a:
2237     case 0x0b:
2238     case 0x0c:
2239     case 0x0d:
2240     case 0x85:
2241     case 0x2028:
2242     case 0x2029:
2243     return op_code != OP_VSPACE;
2244     default:
2245     return op_code == OP_VSPACE;
2246     }
2247    
2248 nigel 93 default:
2249     return FALSE;
2250     }
2251    
2252    
2253     /* Handle the case when the next item is \d, \s, etc. */
2254    
2255     switch(op_code)
2256     {
2257     case OP_CHAR:
2258     case OP_CHARNC:
2259     #ifdef SUPPORT_UTF8
2260     if (utf8 && item > 127) { GETCHAR(item, utf8_char); }
2261     #endif
2262     switch(-next)
2263     {
2264     case ESC_d:
2265     return item > 127 || (cd->ctypes[item] & ctype_digit) == 0;
2266    
2267     case ESC_D:
2268     return item <= 127 && (cd->ctypes[item] & ctype_digit) != 0;
2269    
2270     case ESC_s:
2271     return item > 127 || (cd->ctypes[item] & ctype_space) == 0;
2272    
2273     case ESC_S:
2274     return item <= 127 && (cd->ctypes[item] & ctype_space) != 0;
2275    
2276     case ESC_w:
2277     return item > 127 || (cd->ctypes[item] & ctype_word) == 0;
2278    
2279     case ESC_W:
2280     return item <= 127 && (cd->ctypes[item] & ctype_word) != 0;
2281 ph10 182
2282 ph10 180 case ESC_h:
2283     case ESC_H:
2284     switch(item)
2285     {
2286     case 0x09:
2287     case 0x20:
2288     case 0xa0:
2289     case 0x1680:
2290     case 0x180e:
2291     case 0x2000:
2292     case 0x2001:
2293     case 0x2002:
2294     case 0x2003:
2295     case 0x2004:
2296     case 0x2005:
2297     case 0x2006:
2298     case 0x2007:
2299     case 0x2008:
2300     case 0x2009:
2301     case 0x200A:
2302     case 0x202f:
2303     case 0x205f:
2304     case 0x3000:
2305     return -next != ESC_h;
2306     default:
2307     return -next == ESC_h;
2308 ph10 182 }
2309    
2310 ph10 180 case ESC_v:
2311     case ESC_V:
2312     switch(item)
2313     {
2314     case 0x0a:
2315     case 0x0b:
2316     case 0x0c:
2317     case 0x0d:
2318     case 0x85:
2319     case 0x2028:
2320     case 0x2029:
2321     return -next != ESC_v;
2322     default:
2323     return -next == ESC_v;
2324 ph10 182 }
2325 nigel 93
2326     default:
2327     return FALSE;
2328     }
2329    
2330     case OP_DIGIT:
2331 ph10 180 return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
2332     next == -ESC_h || next == -ESC_v;
2333 nigel 93
2334     case OP_NOT_DIGIT:
2335     return next == -ESC_d;
2336    
2337     case OP_WHITESPACE:
2338     return next == -ESC_S || next == -ESC_d || next == -ESC_w;
2339    
2340     case OP_NOT_WHITESPACE:
2341 ph10 180 return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2342 nigel 93
2343 ph10 180 case OP_HSPACE:
2344     return next == -ESC_S || next == -ESC_H || next == -ESC_d || next == -ESC_w;
2345    
2346     case OP_NOT_HSPACE:
2347     return next == -ESC_h;
2348 ph10 182
2349 ph10 180 /* Can't have \S in here because VT matches \S (Perl anomaly) */
2350 ph10 182 case OP_VSPACE:
2351 ph10 180 return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2352    
2353     case OP_NOT_VSPACE:
2354 ph10 182 return next == -ESC_v;
2355 ph10 180
2356 nigel 93 case OP_WORDCHAR:
2357 ph10 180 return next == -ESC_W || next == -ESC_s || next == -ESC_h || next == -ESC_v;
2358 nigel 93
2359     case OP_NOT_WORDCHAR:
2360     return next == -ESC_w || next == -ESC_d;
2361 ph10 182
2362 nigel 93 default:
2363     return FALSE;
2364     }
2365    
2366     /* Control does not reach here */
2367     }
2368    
2369    
2370    
2371     /*************************************************
2372 nigel 77 * Compile one branch *
2373     *************************************************/
2374    
2375 nigel 93 /* Scan the pattern, compiling it into the a vector. If the options are
2376 nigel 77 changed during the branch, the pointer is used to change the external options
2377 nigel 93 bits. This function is used during the pre-compile phase when we are trying
2378     to find out the amount of memory needed, as well as during the real compile
2379     phase. The value of lengthptr distinguishes the two phases.
2380 nigel 77
2381     Arguments:
2382     optionsptr pointer to the option bits
2383     codeptr points to the pointer to the current code point
2384     ptrptr points to the current pattern pointer
2385     errorcodeptr points to error code variable
2386     firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2387     reqbyteptr set to the last literal character required, else < 0
2388     bcptr points to current branch chain
2389     cd contains pointers to tables etc.
2390 nigel 93 lengthptr NULL during the real compile phase
2391     points to length accumulator during pre-compile phase
2392 nigel 77
2393     Returns: TRUE on success
2394     FALSE, with *errorcodeptr set non-zero on error
2395     */
2396    
2397     static BOOL
2398 nigel 93 compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2399     int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2400     compile_data *cd, int *lengthptr)
2401 nigel 77 {
2402     int repeat_type, op_type;
2403     int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2404     int bravalue = 0;
2405     int greedy_default, greedy_non_default;
2406     int firstbyte, reqbyte;
2407     int zeroreqbyte, zerofirstbyte;
2408     int req_caseopt, reqvary, tempreqvary;
2409     int options = *optionsptr;
2410     int after_manual_callout = 0;
2411 nigel 93 int length_prevgroup = 0;
2412 nigel 77 register int c;
2413     register uschar *code = *codeptr;
2414 nigel 93 uschar *last_code = code;
2415     uschar *orig_code = code;
2416 nigel 77 uschar *tempcode;
2417     BOOL inescq = FALSE;
2418     BOOL groupsetfirstbyte = FALSE;
2419     const uschar *ptr = *ptrptr;
2420     const uschar *tempptr;
2421     uschar *previous = NULL;
2422     uschar *previous_callout = NULL;
2423 nigel 93 uschar *save_hwm = NULL;
2424 nigel 77 uschar classbits[32];
2425    
2426     #ifdef SUPPORT_UTF8
2427     BOOL class_utf8;
2428     BOOL utf8 = (options & PCRE_UTF8) != 0;
2429     uschar *class_utf8data;
2430 ph10 300 uschar *class_utf8data_base;
2431 nigel 77 uschar utf8_char[6];
2432     #else
2433     BOOL utf8 = FALSE;
2434 nigel 93 uschar *utf8_char = NULL;
2435 nigel 77 #endif
2436    
2437 nigel 93 #ifdef DEBUG
2438     if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2439     #endif
2440    
2441 nigel 77 /* Set up the default and non-default settings for greediness */
2442    
2443     greedy_default = ((options & PCRE_UNGREEDY) != 0);
2444     greedy_non_default = greedy_default ^ 1;
2445    
2446     /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2447     matching encountered yet". It gets changed to REQ_NONE if we hit something that
2448     matches a non-fixed char first char; reqbyte just remains unset if we never
2449     find one.
2450    
2451     When we hit a repeat whose minimum is zero, we may have to adjust these values
2452     to take the zero repeat into account. This is implemented by setting them to
2453     zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2454     item types that can be repeated set these backoff variables appropriately. */
2455    
2456     firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2457    
2458     /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2459     according to the current setting of the caseless flag. REQ_CASELESS is a bit
2460     value > 255. It is added into the firstbyte or reqbyte variables to record the
2461     case status of the value. This is used only for ASCII characters. */
2462    
2463     req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2464    
2465     /* Switch on next character until the end of the branch */
2466    
2467     for (;; ptr++)
2468     {
2469     BOOL negate_class;
2470 ph10 286 BOOL should_flip_negation;
2471 nigel 77 BOOL possessive_quantifier;
2472     BOOL is_quantifier;
2473 nigel 93 BOOL is_recurse;
2474 ph10 180 BOOL reset_bracount;
2475 nigel 77 int class_charcount;
2476     int class_lastchar;
2477     int newoptions;
2478     int recno;
2479 ph10 172 int refsign;
2480 nigel 77 int skipbytes;
2481     int subreqbyte;
2482     int subfirstbyte;
2483 nigel 93 int terminator;
2484 nigel 77 int mclength;
2485     uschar mcbuffer[8];
2486    
2487 nigel 93 /* Get next byte in the pattern */
2488 nigel 77
2489     c = *ptr;
2490 ph10 345
2491 nigel 93 /* If we are in the pre-compile phase, accumulate the length used for the
2492     previous cycle of this loop. */
2493    
2494     if (lengthptr != NULL)
2495     {
2496     #ifdef DEBUG
2497     if (code > cd->hwm) cd->hwm = code; /* High water info */
2498     #endif
2499     if (code > cd->start_workspace + COMPILE_WORK_SIZE) /* Check for overrun */
2500     {
2501     *errorcodeptr = ERR52;
2502     goto FAILED;
2503     }
2504    
2505     /* There is at least one situation where code goes backwards: this is the
2506     case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
2507     the class is simply eliminated. However, it is created first, so we have to
2508     allow memory for it. Therefore, don't ever reduce the length at this point.
2509     */
2510    
2511     if (code < last_code) code = last_code;
2512 ph10 202
2513     /* Paranoid check for integer overflow */
2514    
2515     if (OFLOW_MAX - *lengthptr < code - last_code)
2516     {
2517     *errorcodeptr = ERR20;
2518     goto FAILED;
2519     }
2520    
2521 nigel 93 *lengthptr += code - last_code;
2522     DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
2523    
2524     /* If "previous" is set and it is not at the start of the work space, move
2525     it back to there, in order to avoid filling up the work space. Otherwise,
2526     if "previous" is NULL, reset the current code pointer to the start. */
2527    
2528     if (previous != NULL)
2529     {
2530     if (previous > orig_code)
2531     {
2532     memmove(orig_code, previous, code - previous);
2533     code -= previous - orig_code;
2534     previous = orig_code;
2535     }
2536     }
2537     else code = orig_code;
2538    
2539     /* Remember where this code item starts so we can pick up the length
2540     next time round. */
2541    
2542     last_code = code;
2543     }
2544    
2545     /* In the real compile phase, just check the workspace used by the forward
2546     reference list. */
2547    
2548     else if (cd->hwm > cd->start_workspace + COMPILE_WORK_SIZE)
2549     {
2550     *errorcodeptr = ERR52;
2551     goto FAILED;
2552     }
2553    
2554 nigel 77 /* If in \Q...\E, check for the end; if not, we have a literal */
2555    
2556     if (inescq && c != 0)
2557     {
2558     if (c == '\\' && ptr[1] == 'E')
2559     {
2560     inescq = FALSE;
2561     ptr++;
2562     continue;
2563     }
2564     else
2565     {
2566     if (previous_callout != NULL)
2567     {
2568 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2569     complete_callout(previous_callout, ptr, cd);
2570 nigel 77 previous_callout = NULL;
2571     }
2572     if ((options & PCRE_AUTO_CALLOUT) != 0)
2573     {
2574     previous_callout = code;
2575     code = auto_callout(code, ptr, cd);
2576     }
2577     goto NORMAL_CHAR;
2578     }
2579     }
2580    
2581     /* Fill in length of a previous callout, except when the next thing is
2582     a quantifier. */
2583    
2584     is_quantifier = c == '*' || c == '+' || c == '?' ||
2585     (c == '{' && is_counted_repeat(ptr+1));
2586    
2587     if (!is_quantifier && previous_callout != NULL &&
2588     after_manual_callout-- <= 0)
2589     {
2590 nigel 93 if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
2591     complete_callout(previous_callout, ptr, cd);
2592 nigel 77 previous_callout = NULL;
2593     }
2594    
2595     /* In extended mode, skip white space and comments */
2596    
2597     if ((options & PCRE_EXTENDED) != 0)
2598     {
2599     if ((cd->ctypes[c] & ctype_space) != 0) continue;
2600     if (c == '#')
2601     {
2602 nigel 93 while (*(++ptr) != 0)
2603 nigel 91 {
2604 nigel 93 if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
2605 nigel 91 }
2606 nigel 93 if (*ptr != 0) continue;
2607    
2608 nigel 91 /* Else fall through to handle end of string */
2609     c = 0;
2610 nigel 77 }
2611     }
2612    
2613     /* No auto callout for quantifiers. */
2614    
2615     if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
2616     {
2617     previous_callout = code;
2618     code = auto_callout(code, ptr, cd);
2619     }
2620    
2621     switch(c)
2622     {
2623 nigel 93 /* ===================================================================*/
2624     case 0: /* The branch terminates at string end */
2625     case '|': /* or | or ) */
2626 nigel 77 case ')':
2627     *firstbyteptr = firstbyte;
2628     *reqbyteptr = reqbyte;
2629     *codeptr = code;
2630     *ptrptr = ptr;
2631 nigel 93 if (lengthptr != NULL)
2632     {
2633 ph10 202 if (OFLOW_MAX - *lengthptr < code - last_code)
2634     {
2635     *errorcodeptr = ERR20;
2636     goto FAILED;
2637     }
2638 nigel 93 *lengthptr += code - last_code; /* To include callout length */
2639     DPRINTF((">> end branch\n"));
2640     }
2641 nigel 77 return TRUE;
2642    
2643 nigel 93
2644     /* ===================================================================*/
2645 nigel 77 /* Handle single-character metacharacters. In multiline mode, ^ disables
2646     the setting of any following char as a first character. */
2647    
2648     case '^':
2649     if ((options & PCRE_MULTILINE) != 0)
2650     {
2651     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2652     }
2653     previous = NULL;
2654     *code++ = OP_CIRC;
2655     break;
2656    
2657     case '$':
2658     previous = NULL;
2659     *code++ = OP_DOLL;
2660     break;
2661    
2662     /* There can never be a first char if '.' is first, whatever happens about
2663     repeats. The value of reqbyte doesn't change either. */
2664    
2665     case '.':
2666     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2667     zerofirstbyte = firstbyte;
2668     zeroreqbyte = reqbyte;
2669     previous = code;
2670 ph10 342 *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
2671 nigel 77 break;
2672    
2673 nigel 93
2674     /* ===================================================================*/
2675 nigel 87 /* Character classes. If the included characters are all < 256, we build a
2676     32-byte bitmap of the permitted characters, except in the special case
2677     where there is only one such character. For negated classes, we build the
2678     map as usual, then invert it at the end. However, we use a different opcode
2679     so that data characters > 255 can be handled correctly.
2680 nigel 77
2681     If the class contains characters outside the 0-255 range, a different
2682     opcode is compiled. It may optionally have a bit map for characters < 256,
2683     but those above are are explicitly listed afterwards. A flag byte tells
2684     whether the bitmap is present, and whether this is a negated class or not.
2685 ph10 345
2686 ph10 336 In JavaScript compatibility mode, an isolated ']' causes an error. In
2687     default (Perl) mode, it is treated as a data character. */
2688 ph10 345
2689 ph10 336 case ']':
2690     if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2691     {
2692     *errorcodeptr = ERR64;
2693 ph10 345 goto FAILED;
2694 ph10 336 }
2695 ph10 345 goto NORMAL_CHAR;
2696 nigel 77
2697     case '[':
2698     previous = code;
2699    
2700     /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
2701     they are encountered at the top level, so we'll do that too. */
2702    
2703     if ((ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
2704 ph10 295 check_posix_syntax(ptr, &tempptr))
2705 nigel 77 {
2706     *errorcodeptr = (ptr[1] == ':')? ERR13 : ERR31;
2707     goto FAILED;
2708     }
2709    
2710 ph10 205 /* If the first character is '^', set the negation flag and skip it. Also,
2711 ph10 208 if the first few characters (either before or after ^) are \Q\E or \E we
2712 ph10 205 skip them too. This makes for compatibility with Perl. */
2713 ph10 208
2714 ph10 205 negate_class = FALSE;
2715     for (;;)
2716 nigel 77 {
2717     c = *(++ptr);
2718 ph10 205 if (c == '\\')
2719     {
2720 ph10 208 if (ptr[1] == 'E') ptr++;
2721 ph10 205 else if (strncmp((const char *)ptr+1, "Q\\E", 3) == 0) ptr += 3;
2722 ph10 208 else break;
2723 ph10 205 }
2724     else if (!negate_class && c == '^')
2725     negate_class = TRUE;
2726     else break;
2727 ph10 208 }
2728 ph10 345
2729     /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
2730     an initial ']' is taken as a data character -- the code below handles
2731 ph10 341 that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
2732     [^] must match any character, so generate OP_ALLANY. */
2733 ph10 345
2734 ph10 341 if (c ==']' && (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
2735     {
2736     *code++ = negate_class? OP_ALLANY : OP_FAIL;
2737     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
2738     zerofirstbyte = firstbyte;
2739     break;
2740 ph10 345 }
2741 nigel 77
2742 ph10 286 /* If a class contains a negative special such as \S, we need to flip the
2743     negation flag at the end, so that support for characters > 255 works
2744 ph10 264 correctly (they are all included in the class). */
2745    
2746     should_flip_negation = FALSE;
2747    
2748 nigel 77 /* Keep a count of chars with values < 256 so that we can optimize the case
2749 nigel 93 of just a single character (as long as it's < 256). However, For higher
2750     valued UTF-8 characters, we don't yet do any optimization. */
2751 nigel 77
2752     class_charcount = 0;
2753     class_lastchar = -1;
2754    
2755 nigel 93 /* Initialize the 32-char bit map to all zeros. We build the map in a
2756     temporary bit of memory, in case the class contains only 1 character (less
2757     than 256), because in that case the compiled code doesn't use the bit map.
2758     */
2759    
2760     memset(classbits, 0, 32 * sizeof(uschar));
2761    
2762 nigel 77 #ifdef SUPPORT_UTF8
2763     class_utf8 = FALSE; /* No chars >= 256 */
2764 nigel 93 class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
2765 ph10 309 class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
2766 nigel 77 #endif
2767    
2768     /* Process characters until ] is reached. By writing this as a "do" it
2769 nigel 93 means that an initial ] is taken as a data character. At the start of the
2770     loop, c contains the first byte of the character. */
2771 nigel 77
2772 nigel 93 if (c != 0) do
2773 nigel 77 {
2774 nigel 93 const uschar *oldptr;
2775    
2776 nigel 77 #ifdef SUPPORT_UTF8
2777     if (utf8 && c > 127)
2778     { /* Braces are required because the */
2779     GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
2780     }
2781 ph10 309
2782 ph10 300 /* In the pre-compile phase, accumulate the length of any UTF-8 extra
2783 ph10 309 data and reset the pointer. This is so that very large classes that
2784 ph10 300 contain a zillion UTF-8 characters no longer overwrite the work space
2785 ph10 309 (which is on the stack). */
2786    
2787 ph10 300 if (lengthptr != NULL)
2788     {
2789     *lengthptr += class_utf8data - class_utf8data_base;
2790 ph10 309 class_utf8data = class_utf8data_base;
2791     }
2792    
2793 nigel 77 #endif
2794    
2795     /* Inside \Q...\E everything is literal except \E */
2796    
2797     if (inescq)
2798     {
2799 nigel 93 if (c == '\\' && ptr[1] == 'E') /* If we are at \E */
2800 nigel 77 {
2801 nigel 93 inescq = FALSE; /* Reset literal state */
2802     ptr++; /* Skip the 'E' */
2803     continue; /* Carry on with next */
2804 nigel 77 }
2805 nigel 93 goto CHECK_RANGE; /* Could be range if \E follows */
2806 nigel 77 }
2807    
2808     /* Handle POSIX class names. Perl allows a negation extension of the
2809     form [:^name:]. A square bracket that doesn't match the syntax is
2810     treated as a literal. We also recognize the POSIX constructions
2811     [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
2812     5.6 and 5.8 do. */
2813    
2814     if (c == '[' &&
2815     (ptr[1] == ':' || ptr[1] == '.' || ptr[1] == '=') &&
2816 ph10 295 check_posix_syntax(ptr, &tempptr))
2817 nigel 77 {
2818     BOOL local_negate = FALSE;
2819 nigel 87 int posix_class, taboffset, tabopt;
2820 nigel 77 register const uschar *cbits = cd->cbits;
2821 nigel 87 uschar pbits[32];
2822 nigel 77
2823     if (ptr[1] != ':')
2824     {
2825     *errorcodeptr = ERR31;
2826     goto FAILED;
2827     }
2828    
2829     ptr += 2;
2830     if (*ptr == '^')
2831     {
2832     local_negate = TRUE;
2833 ph10 286 should_flip_negation = TRUE; /* Note negative special */
2834 nigel 77 ptr++;
2835     }
2836    
2837     posix_class = check_posix_name(ptr, tempptr - ptr);
2838     if (posix_class < 0)
2839     {
2840     *errorcodeptr = ERR30;
2841     goto FAILED;
2842     }
2843    
2844     /* If matching is caseless, upper and lower are converted to
2845     alpha. This relies on the fact that the class table starts with
2846     alpha, lower, upper as the first 3 entries. */
2847    
2848     if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
2849     posix_class = 0;
2850    
2851 nigel 87 /* We build the bit map for the POSIX class in a chunk of local store
2852     because we may be adding and subtracting from it, and we don't want to
2853     subtract bits that may be in the main map already. At the end we or the
2854     result into the bit map that is being built. */
2855 nigel 77
2856     posix_class *= 3;
2857 nigel 87
2858     /* Copy in the first table (always present) */
2859    
2860     memcpy(pbits, cbits + posix_class_maps[posix_class],
2861     32 * sizeof(uschar));
2862    
2863     /* If there is a second table, add or remove it as required. */
2864    
2865     taboffset = posix_class_maps[posix_class + 1];
2866     tabopt = posix_class_maps[posix_class + 2];
2867    
2868     if (taboffset >= 0)
2869 nigel 77 {
2870 nigel 87 if (tabopt >= 0)
2871     for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
2872 nigel 77 else
2873 nigel 87 for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
2874 nigel 77 }
2875    
2876 nigel 87 /* Not see if we need to remove any special characters. An option
2877     value of 1 removes vertical space and 2 removes underscore. */
2878    
2879     if (tabopt < 0) tabopt = -tabopt;
2880     if (tabopt == 1) pbits[1] &= ~0x3c;
2881     else if (tabopt == 2) pbits[11] &= 0x7f;
2882    
2883     /* Add the POSIX table or its complement into the main table that is
2884     being built and we are done. */
2885    
2886     if (local_negate)
2887     for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
2888     else
2889     for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
2890    
2891 nigel 77 ptr = tempptr + 1;
2892     class_charcount = 10; /* Set > 1; assumes more than 1 per class */
2893     continue; /* End of POSIX syntax handling */
2894     }
2895    
2896     /* Backslash may introduce a single character, or it may introduce one
2897 nigel 93 of the specials, which just set a flag. The sequence \b is a special
2898     case. Inside a class (and only there) it is treated as backspace.
2899     Elsewhere it marks a word boundary. Other escapes have preset maps ready
2900 ph10 205 to 'or' into the one we are building. We assume they have more than one
2901 nigel 77 character in them, so set class_charcount bigger than one. */
2902    
2903     if (c == '\\')
2904     {
2905 nigel 93 c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
2906     if (*errorcodeptr != 0) goto FAILED;
2907 nigel 77
2908 ph10 275 if (-c == ESC_b) c = '\b'; /* \b is backspace in a class */
2909 nigel 77 else if (-c == ESC_X) c = 'X'; /* \X is literal X in a class */
2910 nigel 93 else if (-c == ESC_R) c = 'R'; /* \R is literal R in a class */
2911 nigel 77 else if (-c == ESC_Q) /* Handle start of quoted string */
2912     {
2913     if (ptr[1] == '\\' && ptr[2] == 'E')
2914     {
2915     ptr += 2; /* avoid empty string */
2916     }
2917     else inescq = TRUE;
2918     continue;
2919     }
2920 ph10 220 else if (-c == ESC_E) continue; /* Ignore orphan \E */
2921 nigel 77
2922     if (c < 0)
2923     {
2924     register const uschar *cbits = cd->cbits;
2925     class_charcount += 2; /* Greater than 1 is what matters */
2926 nigel 93
2927     /* Save time by not doing this in the pre-compile phase. */
2928    
2929     if (lengthptr == NULL) switch (-c)
2930 nigel 77 {
2931     case ESC_d:
2932     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
2933     continue;
2934    
2935     case ESC_D:
2936 ph10 286 should_flip_negation = TRUE;
2937 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
2938     continue;
2939    
2940     case ESC_w:
2941     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
2942     continue;
2943    
2944     case ESC_W:
2945 ph10 286 should_flip_negation = TRUE;
2946 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
2947     continue;
2948    
2949     case ESC_s:
2950     for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
2951     classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
2952     continue;
2953    
2954     case ESC_S:
2955 ph10 286 should_flip_negation = TRUE;
2956 nigel 77 for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
2957     classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
2958     continue;
2959    
2960 nigel 93 default: /* Not recognized; fall through */
2961     break; /* Need "default" setting to stop compiler warning. */
2962     }
2963    
2964     /* In the pre-compile phase, just do the recognition. */
2965    
2966     else if (c == -ESC_d || c == -ESC_D || c == -ESC_w ||
2967     c == -ESC_W || c == -ESC_s || c == -ESC_S) continue;
2968 ph10 180
2969 ph10 178 /* We need to deal with \H, \h, \V, and \v in both phases because
2970     they use extra memory. */
2971 ph10 180
2972 ph10 178 if (-c == ESC_h)
2973     {
2974     SETBIT(classbits, 0x09); /* VT */
2975     SETBIT(classbits, 0x20); /* SPACE */
2976 ph10 180 SETBIT(classbits, 0xa0); /* NSBP */
2977 ph10 178 #ifdef SUPPORT_UTF8
2978     if (utf8)
2979 ph10 180 {
2980 ph10 178 class_utf8 = TRUE;
2981     *class_utf8data++ = XCL_SINGLE;
2982 ph10 180 class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
2983 ph10 178 *class_utf8data++ = XCL_SINGLE;
2984 ph10 180 class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
2985     *class_utf8data++ = XCL_RANGE;
2986     class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
2987     class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
2988 ph10 178 *class_utf8data++ = XCL_SINGLE;
2989 ph10 180 class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
2990 ph10 178 *class_utf8data++ = XCL_SINGLE;
2991 ph10 180 class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
2992 ph10 178 *class_utf8data++ = XCL_SINGLE;
2993 ph10 180 class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
2994     }
2995     #endif
2996     continue;
2997     }
2998 nigel 93
2999 ph10 178 if (-c == ESC_H)
3000     {
3001     for (c = 0; c < 32; c++)
3002     {
3003     int x = 0xff;
3004     switch (c)
3005 ph10 180 {
3006 ph10 178 case 0x09/8: x ^= 1 << (0x09%8); break;
3007     case 0x20/8: x ^= 1 << (0x20%8); break;
3008     case 0xa0/8: x ^= 1 << (0xa0%8); break;
3009     default: break;
3010     }
3011     classbits[c] |= x;
3012 ph10 180 }
3013    
3014 ph10 178 #ifdef SUPPORT_UTF8
3015     if (utf8)
3016 ph10 180 {
3017 ph10 178 class_utf8 = TRUE;
3018 ph10 180 *class_utf8data++ = XCL_RANGE;
3019     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3020     class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3021     *class_utf8data++ = XCL_RANGE;
3022     class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3023     class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3024     *class_utf8data++ = XCL_RANGE;
3025     class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3026     class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3027     *class_utf8data++ = XCL_RANGE;
3028     class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3029     class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3030     *class_utf8data++ = XCL_RANGE;
3031     class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3032     class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3033     *class_utf8data++ = XCL_RANGE;
3034     class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3035     class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3036     *class_utf8data++ = XCL_RANGE;
3037     class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3038     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3039     }
3040     #endif
3041     continue;
3042     }
3043 ph10 178
3044     if (-c == ESC_v)
3045     {
3046     SETBIT(classbits, 0x0a); /* LF */
3047     SETBIT(classbits, 0x0b); /* VT */
3048 ph10 180 SETBIT(classbits, 0x0c); /* FF */
3049     SETBIT(classbits, 0x0d); /* CR */
3050     SETBIT(classbits, 0x85); /* NEL */
3051 ph10 178 #ifdef SUPPORT_UTF8
3052     if (utf8)
3053 ph10 180 {
3054 ph10 178 class_utf8 = TRUE;
3055 ph10 180 *class_utf8data++ = XCL_RANGE;
3056     class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3057     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3058     }
3059     #endif
3060     continue;
3061     }
3062 ph10 178
3063     if (-c == ESC_V)
3064     {
3065     for (c = 0; c < 32; c++)
3066     {
3067     int x = 0xff;
3068     switch (c)
3069 ph10 180 {
3070 ph10 178 case 0x0a/8: x ^= 1 << (0x0a%8);
3071     x ^= 1 << (0x0b%8);
3072     x ^= 1 << (0x0c%8);
3073 ph10 180 x ^= 1 << (0x0d%8);
3074 ph10 178 break;
3075     case 0x85/8: x ^= 1 << (0x85%8); break;
3076     default: break;
3077     }
3078     classbits[c] |= x;
3079 ph10 180 }
3080    
3081 ph10 178 #ifdef SUPPORT_UTF8
3082     if (utf8)
3083 ph10 180 {
3084 ph10 178 class_utf8 = TRUE;
3085 ph10 180 *class_utf8data++ = XCL_RANGE;
3086     class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3087     class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3088     *class_utf8data++ = XCL_RANGE;
3089     class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3090     class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3091     }
3092     #endif
3093     continue;
3094     }
3095 ph10 178
3096 nigel 93 /* We need to deal with \P and \p in both phases. */
3097    
3098 nigel 77 #ifdef SUPPORT_UCP
3099 nigel 93 if (-c == ESC_p || -c == ESC_P)
3100     {
3101     BOOL negated;
3102     int pdata;
3103     int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
3104     if (ptype < 0) goto FAILED;
3105     class_utf8 = TRUE;
3106     *class_utf8data++ = ((-c == ESC_p) != negated)?
3107     XCL_PROP : XCL_NOTPROP;
3108     *class_utf8data++ = ptype;
3109     *class_utf8data++ = pdata;
3110     class_charcount -= 2; /* Not a < 256 character */
3111 nigel 77 continue;
3112 nigel 93 }
3113 nigel 77 #endif
3114 nigel 93 /* Unrecognized escapes are faulted if PCRE is running in its
3115     strict mode. By default, for compatibility with Perl, they are
3116     treated as literals. */
3117 nigel 77
3118 nigel 93 if ((options & PCRE_EXTRA) != 0)
3119     {
3120     *errorcodeptr = ERR7;
3121     goto FAILED;
3122     }
3123 nigel 77
3124 nigel 93 class_charcount -= 2; /* Undo the default count from above */
3125     c = *ptr; /* Get the final character and fall through */
3126 nigel 77 }
3127    
3128     /* Fall through if we have a single character (c >= 0). This may be
3129 nigel 93 greater than 256 in UTF-8 mode. */
3130 nigel 77
3131     } /* End of backslash handling */
3132    
3133     /* A single character may be followed by '-' to form a range. However,
3134     Perl does not permit ']' to be the end of the range. A '-' character
3135 nigel 93 at the end is treated as a literal. Perl ignores orphaned \E sequences
3136     entirely. The code for handling \Q and \E is messy. */
3137 nigel 77
3138 nigel 93 CHECK_RANGE:
3139     while (ptr[1] == '\\' && ptr[2] == 'E')
3140 nigel 77 {
3141 nigel 93 inescq = FALSE;
3142     ptr += 2;
3143     }
3144    
3145     oldptr = ptr;
3146 ph10 231
3147 ph10 230 /* Remember \r or \n */
3148 ph10 231
3149     if (c == '\r' || c == '\n') cd->external_flags |= PCRE_HASCRORLF;
3150    
3151 ph10 230 /* Check for range */
3152 nigel 93
3153     if (!inescq && ptr[1] == '-')
3154     {
3155 nigel 77 int d;
3156     ptr += 2;
3157 nigel 93 while (*ptr == '\\' && ptr[1] == 'E') ptr += 2;
3158 nigel 77
3159 nigel 93 /* If we hit \Q (not followed by \E) at this point, go into escaped
3160     mode. */
3161    
3162     while (*ptr == '\\' && ptr[1] == 'Q')
3163     {
3164     ptr += 2;
3165     if (*ptr == '\\' && ptr[1] == 'E') { ptr += 2; continue; }
3166     inescq = TRUE;
3167     break;
3168     }
3169    
3170     if (*ptr == 0 || (!inescq && *ptr == ']'))
3171     {
3172     ptr = oldptr;
3173     goto LONE_SINGLE_CHARACTER;
3174     }
3175    
3176 nigel 77 #ifdef SUPPORT_UTF8
3177     if (utf8)
3178     { /* Braces are required because the */
3179     GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3180     }
3181     else
3182     #endif
3183     d = *ptr; /* Not UTF-8 mode */
3184    
3185     /* The second part of a range can be a single-character escape, but
3186     not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3187     in such circumstances. */
3188    
3189 nigel 93 if (!inescq && d == '\\')
3190 nigel 77 {
3191 nigel 93 d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3192     if (*errorcodeptr != 0) goto FAILED;
3193 nigel 77
3194 ph10 275 /* \b is backspace; \X is literal X; \R is literal R; any other
3195 nigel 93 special means the '-' was literal */
3196 nigel 77
3197     if (d < 0)
3198     {
3199     if (d == -ESC_b) d = '\b';
3200 nigel 93 else if (d == -ESC_X) d = 'X';
3201     else if (d == -ESC_R) d = 'R'; else
3202 nigel 77 {
3203 nigel 93 ptr = oldptr;
3204 nigel 77 goto LONE_SINGLE_CHARACTER; /* A few lines below */
3205     }
3206     }
3207     }
3208    
3209 nigel 93 /* Check that the two values are in the correct order. Optimize
3210     one-character ranges */
3211 nigel 77
3212 nigel 93 if (d < c)
3213     {
3214     *errorcodeptr = ERR8;
3215     goto FAILED;
3216     }
3217    
3218 nigel 77 if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3219    
3220 ph10 230 /* Remember \r or \n */
3221 ph10 231
3222     if (d == '\r' || d == '\n') cd->external_flags |= PCRE_HASCRORLF;
3223    
3224 nigel 77 /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3225     matching, we have to use an XCLASS with extra data items. Caseless
3226     matching for characters > 127 is available only if UCP support is
3227     available. */
3228    
3229     #ifdef SUPPORT_UTF8
3230     if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3231     {
3232     class_utf8 = TRUE;
3233    
3234     /* With UCP support, we can find the other case equivalents of
3235     the relevant characters. There may be several ranges. Optimize how
3236     they fit with the basic range. */
3237    
3238     #ifdef SUPPORT_UCP
3239     if ((options & PCRE_CASELESS) != 0)
3240     {
3241 nigel 93 unsigned int occ, ocd;
3242     unsigned int cc = c;
3243     unsigned int origd = d;
3244 nigel 77 while (get_othercase_range(&cc, origd, &occ, &ocd))
3245     {
3246 ph10 180 if (occ >= (unsigned int)c &&
3247     ocd <= (unsigned int)d)
3248 ph10 176 continue; /* Skip embedded ranges */
3249 nigel 77
3250 ph10 180 if (occ < (unsigned int)c &&
3251 ph10 176 ocd >= (unsigned int)c - 1) /* Extend the basic range */
3252 nigel 77 { /* if there is overlap, */
3253     c = occ; /* noting that if occ < c */
3254     continue; /* we can't have ocd > d */
3255     } /* because a subrange is */
3256 ph10 180 if (ocd > (unsigned int)d &&
3257 ph10 176 occ <= (unsigned int)d + 1) /* always shorter than */
3258 nigel 77 { /* the basic range. */
3259     d = ocd;
3260     continue;
3261     }
3262    
3263     if (occ == ocd)
3264     {
3265     *class_utf8data++ = XCL_SINGLE;
3266     }
3267     else
3268     {
3269     *class_utf8data++ = XCL_RANGE;
3270     class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3271     }
3272     class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3273     }
3274     }
3275     #endif /* SUPPORT_UCP */
3276    
3277     /* Now record the original range, possibly modified for UCP caseless
3278     overlapping ranges. */
3279    
3280     *class_utf8data++ = XCL_RANGE;
3281     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3282     class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3283    
3284     /* With UCP support, we are done. Without UCP support, there is no
3285     caseless matching for UTF-8 characters > 127; we can use the bit map
3286     for the smaller ones. */
3287    
3288     #ifdef SUPPORT_UCP
3289     continue; /* With next character in the class */
3290     #else
3291     if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3292    
3293     /* Adjust upper limit and fall through to set up the map */
3294    
3295     d = 127;
3296    
3297     #endif /* SUPPORT_UCP */
3298     }
3299     #endif /* SUPPORT_UTF8 */
3300    
3301     /* We use the bit map for all cases when not in UTF-8 mode; else
3302     ranges that lie entirely within 0-127 when there is UCP support; else
3303     for partial ranges without UCP support. */
3304    
3305 nigel 93 class_charcount += d - c + 1;
3306     class_lastchar = d;
3307    
3308     /* We can save a bit of time by skipping this in the pre-compile. */
3309    
3310     if (lengthptr == NULL) for (; c <= d; c++)
3311 nigel 77 {
3312     classbits[c/8] |= (1 << (c&7));
3313     if ((options & PCRE_CASELESS) != 0)
3314     {
3315     int uc = cd->fcc[c]; /* flip case */
3316     classbits[uc/8] |= (1 << (uc&7));
3317     }
3318     }
3319    
3320     continue; /* Go get the next char in the class */
3321     }
3322    
3323     /* Handle a lone single character - we can get here for a normal
3324     non-escape char, or after \ that introduces a single character or for an
3325     apparent range that isn't. */
3326    
3327     LONE_SINGLE_CHARACTER:
3328 ph10 231
3329 nigel 77 /* Handle a character that cannot go in the bit map */
3330    
3331     #ifdef SUPPORT_UTF8
3332     if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3333     {
3334     class_utf8 = TRUE;
3335     *class_utf8data++ = XCL_SINGLE;
3336     class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3337    
3338     #ifdef SUPPORT_UCP
3339     if ((options & PCRE_CASELESS) != 0)
3340     {
3341 nigel 93 unsigned int othercase;
3342 ph10 349 if ((othercase = UCD_OTHERCASE(c)) != c)
3343 nigel 77 {
3344     *class_utf8data++ = XCL_SINGLE;
3345     class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3346     }
3347     }
3348     #endif /* SUPPORT_UCP */
3349    
3350     }
3351     else
3352     #endif /* SUPPORT_UTF8 */
3353    
3354     /* Handle a single-byte character */
3355     {
3356     classbits[c/8] |= (1 << (c&7));
3357     if ((options & PCRE_CASELESS) != 0)
3358     {
3359     c = cd->fcc[c]; /* flip case */
3360     classbits[c/8] |= (1 << (c&7));
3361     }
3362     class_charcount++;
3363     class_lastchar = c;
3364     }
3365     }
3366    
3367 nigel 93 /* Loop until ']' reached. This "while" is the end of the "do" above. */
3368 nigel 77
3369 nigel 93 while ((c = *(++ptr)) != 0 && (c != ']' || inescq));
3370 nigel 77
3371 nigel 93 if (c == 0) /* Missing terminating ']' */
3372     {
3373     *errorcodeptr = ERR6;
3374     goto FAILED;
3375     }
3376 ph10 231
3377    
3378 ph10 230 /* This code has been disabled because it would mean that \s counts as
3379     an explicit \r or \n reference, and that's not really what is wanted. Now
3380     we set the flag only if there is a literal "\r" or "\n" in the class. */
3381 ph10 227
3382 ph10 230 #if 0
3383 ph10 226 /* Remember whether \r or \n are in this class */
3384 ph10 227
3385 ph10 226 if (negate_class)
3386     {
3387 ph10 230 if ((classbits[1] & 0x24) != 0x24) cd->external_flags |= PCRE_HASCRORLF;
3388 ph10 226 }
3389     else
3390     {
3391 ph10 230 if ((classbits[1] & 0x24) != 0) cd->external_flags |= PCRE_HASCRORLF;
3392 ph10 227 }
3393 ph10 230 #endif
3394 ph10 227
3395 ph10 231
3396 nigel 77 /* If class_charcount is 1, we saw precisely one character whose value is
3397 ph10 227 less than 256. As long as there were no characters >= 128 and there was no
3398     use of \p or \P, in other words, no use of any XCLASS features, we can
3399     optimize.
3400    
3401 ph10 223 In UTF-8 mode, we can optimize the negative case only if there were no
3402     characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3403     operate on single-bytes only. This is an historical hangover. Maybe one day
3404     we can tidy these opcodes to handle multi-byte characters.
3405 nigel 77
3406     The optimization throws away the bit map. We turn the item into a
3407     1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3408     that OP_NOT does not support multibyte characters. In the positive case, it
3409     can cause firstbyte to be set. Otherwise, there can be no first char if
3410     this item is first, whatever repeat count may follow. In the case of
3411     reqbyte, save the previous value for reinstating. */
3412    
3413     #ifdef SUPPORT_UTF8
3414 ph10 227 if (class_charcount == 1 && !class_utf8 &&
3415 ph10 223 (!utf8 || !negate_class || class_lastchar < 128))
3416 nigel 77 #else
3417     if (class_charcount == 1)
3418     #endif
3419     {
3420     zeroreqbyte = reqbyte;
3421    
3422     /* The OP_NOT opcode works on one-byte characters only. */
3423    
3424     if (negate_class)
3425     {
3426     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3427     zerofirstbyte = firstbyte;
3428     *code++ = OP_NOT;
3429     *code++ = class_lastchar;
3430     break;
3431     }
3432    
3433     /* For a single, positive character, get the value into mcbuffer, and
3434     then we can handle this with the normal one-character code. */
3435    
3436     #ifdef SUPPORT_UTF8
3437     if (utf8 && class_lastchar > 127)
3438     mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3439     else
3440     #endif
3441     {
3442     mcbuffer[0] = class_lastchar;
3443     mclength = 1;
3444     }
3445     goto ONE_CHAR;
3446     } /* End of 1-char optimization */
3447    
3448     /* The general case - not the one-char optimization. If this is the first
3449     thing in the branch, there can be no first char setting, whatever the
3450     repeat count. Any reqbyte setting must remain unchanged after any kind of
3451     repeat. */
3452    
3453     if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3454     zerofirstbyte = firstbyte;
3455     zeroreqbyte = reqbyte;
3456    
3457     /* If there are characters with values > 255, we have to compile an
3458 ph10 286 extended class, with its own opcode, unless there was a negated special
3459     such as \S in the class, because in that case all characters > 255 are in
3460     the class, so any that were explicitly given as well can be ignored. If
3461 ph10 264 (when there are explicit characters > 255 that must be listed) there are no
3462     characters < 256, we can omit the bitmap in the actual compiled code. */
3463 nigel 77
3464     #ifdef SUPPORT_UTF8
3465 ph10 264 if (class_utf8 && !should_flip_negation)
3466 nigel 77 {
3467     *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3468     *code++ = OP_XCLASS;
3469     code += LINK_SIZE;
3470     *code = negate_class? XCL_NOT : 0;
3471    
3472 nigel 93 /* If the map is required, move up the extra data to make room for it;
3473     otherwise just move the code pointer to the end of the extra data. */
3474 nigel 77
3475     if (class_charcount > 0)
3476     {
3477     *code++ |= XCL_MAP;
3478 nigel 93 memmove(code + 32, code, class_utf8data - code);
3479 nigel 77 memcpy(code, classbits, 32);
3480 nigel 93 code = class_utf8data + 32;
3481 nigel 77 }
3482 nigel 93 else code = class_utf8data;
3483 nigel 77
3484     /* Now fill in the complete length of the item */
3485    
3486     PUT(previous, 1, code - previous);
3487     break; /* End of class handling */
3488     }
3489     #endif
3490    
3491 ph10 286 /* If there are no characters > 255, set the opcode to OP_CLASS or
3492     OP_NCLASS, depending on whether the whole class was negated and whether
3493     there were negative specials such as \S in the class. Then copy the 32-byte
3494 ph10 264 map into the code vector, negating it if necessary. */
3495 ph10 286
3496 ph10 264 *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
3497 nigel 77 if (negate_class)
3498     {
3499 nigel 93 if (lengthptr == NULL) /* Save time in the pre-compile phase */
3500     for (c = 0; c < 32; c++) code[c] = ~classbits[c];
3501 nigel 77 }
3502     else
3503     {
3504     memcpy(code, classbits, 32);
3505     }
3506     code += 32;
3507     break;
3508    
3509 nigel 93
3510     /* ===================================================================*/
3511 nigel 77 /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
3512     has been tested above. */
3513    
3514     case '{':
3515     if (!is_quantifier) goto NORMAL_CHAR;
3516     ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
3517     if (*errorcodeptr != 0) goto FAILED;
3518     goto REPEAT;
3519    
3520     case '*':
3521     repeat_min = 0;
3522     repeat_max = -1;
3523     goto REPEAT;
3524    
3525     case '+':
3526     repeat_min = 1;
3527     repeat_max = -1;
3528     goto REPEAT;
3529    
3530     case '?':
3531     repeat_min = 0;
3532     repeat_max = 1;
3533    
3534     REPEAT:
3535     if (previous == NULL)
3536     {
3537     *errorcodeptr = ERR9;
3538     goto FAILED;
3539     }
3540    
3541     if (repeat_min == 0)
3542     {
3543     firstbyte = zerofirstbyte; /* Adjust for zero repeat */
3544     reqbyte = zeroreqbyte; /* Ditto */
3545     }
3546    
3547     /* Remember whether this is a variable length repeat */
3548    
3549     reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
3550    
3551     op_type = 0; /* Default single-char op codes */
3552     possessive_quantifier = FALSE; /* Default not possessive quantifier */
3553    
3554     /* Save start of previous item, in case we have to move it up to make space
3555     for an inserted OP_ONCE for the additional '+' extension. */
3556    
3557     tempcode = previous;
3558    
3559     /* If the next character is '+', we have a possessive quantifier. This
3560     implies greediness, whatever the setting of the PCRE_UNGREEDY option.
3561     If the next character is '?' this is a minimizing repeat, by default,
3562     but if PCRE_UNGREEDY is set, it works the other way round. We change the
3563     repeat type to the non-default. */
3564    
3565     if (ptr[1] == '+')
3566     {
3567     repeat_type = 0; /* Force greedy */
3568     possessive_quantifier = TRUE;
3569     ptr++;
3570     }
3571     else if (ptr[1] == '?')
3572     {
3573     repeat_type = greedy_non_default;
3574     ptr++;
3575     }
3576     else repeat_type = greedy_default;
3577    
3578     /* If previous was a character match, abolish the item and generate a
3579     repeat item instead. If a char item has a minumum of more than one, ensure
3580     that it is set in reqbyte - it might not be if a sequence such as x{3} is
3581     the first thing in a branch because the x will have gone into firstbyte
3582     instead. */
3583    
3584     if (*previous == OP_CHAR || *previous == OP_CHARNC)
3585     {
3586     /* Deal with UTF-8 characters that take up more than one byte. It's
3587     easier to write this out separately than try to macrify it. Use c to
3588     hold the length of the character in bytes, plus 0x80 to flag that it's a
3589     length rather than a small character. */
3590    
3591     #ifdef SUPPORT_UTF8
3592     if (utf8 && (code[-1] & 0x80) != 0)
3593     {
3594     uschar *lastchar = code - 1;
3595     while((*lastchar & 0xc0) == 0x80) lastchar--;
3596     c = code - lastchar; /* Length of UTF-8 character */
3597     memcpy(utf8_char, lastchar, c); /* Save the char */
3598     c |= 0x80; /* Flag c as a length */
3599     }
3600     else
3601     #endif
3602    
3603     /* Handle the case of a single byte - either with no UTF8 support, or
3604     with UTF-8 disabled, or for a UTF-8 character < 128. */
3605    
3606     {
3607     c = code[-1];
3608     if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
3609     }
3610    
3611 nigel 93 /* If the repetition is unlimited, it pays to see if the next thing on
3612     the line is something that cannot possibly match this character. If so,
3613     automatically possessifying this item gains some performance in the case
3614     where the match fails. */
3615    
3616     if (!possessive_quantifier &&
3617     repeat_max < 0 &&
3618     check_auto_possessive(*previous, c, utf8, utf8_char, ptr + 1,
3619     options, cd))
3620     {
3621     repeat_type = 0; /* Force greedy */
3622     possessive_quantifier = TRUE;
3623     }
3624    
3625 nigel 77 goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
3626     }
3627    
3628     /* If previous was a single negated character ([^a] or similar), we use
3629     one of the special opcodes, replacing it. The code is shared with single-
3630     character repeats by setting opt_type to add a suitable offset into
3631 nigel 93 repeat_type. We can also test for auto-possessification. OP_NOT is
3632     currently used only for single-byte chars. */
3633 nigel 77
3634     else if (*previous == OP_NOT)
3635     {
3636     op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
3637     c = previous[1];
3638 nigel 93 if (!possessive_quantifier &&
3639     repeat_max < 0 &&
3640     check_auto_possessive(OP_NOT, c, utf8, NULL, ptr + 1, options, cd))
3641     {
3642     repeat_type = 0; /* Force greedy */
3643     possessive_quantifier = TRUE;
3644     }
3645 nigel 77 goto OUTPUT_SINGLE_REPEAT;
3646     }
3647    
3648     /* If previous was a character type match (\d or similar), abolish it and
3649     create a suitable repeat item. The code is shared with single-character
3650     repeats by setting op_type to add a suitable offset into repeat_type. Note
3651     the the Unicode property types will be present only when SUPPORT_UCP is
3652     defined, but we don't wrap the little bits of code here because it just
3653     makes it horribly messy. */
3654    
3655     else if (*previous < OP_EODN)
3656     {
3657     uschar *oldcode;
3658 nigel 87 int prop_type, prop_value;
3659 nigel 77 op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
3660     c = *previous;
3661    
3662 nigel 93 if (!possessive_quantifier &&
3663     repeat_max < 0 &&
3664     check_auto_possessive(c, 0, utf8, NULL, ptr + 1, options, cd))
3665     {
3666     repeat_type = 0; /* Force greedy */
3667     possessive_quantifier = TRUE;
3668     }
3669    
3670 nigel 77 OUTPUT_SINGLE_REPEAT:
3671 nigel 87 if (*previous == OP_PROP || *previous == OP_NOTPROP)
3672     {
3673     prop_type = previous[1];
3674     prop_value = previous[2];
3675     }
3676     else prop_type = prop_value = -1;
3677 nigel 77
3678     oldcode = code;
3679     code = previous; /* Usually overwrite previous item */
3680    
3681     /* If the maximum is zero then the minimum must also be zero; Perl allows
3682     this case, so we do too - by simply omitting the item altogether. */
3683    
3684     if (repeat_max == 0) goto END_REPEAT;
3685    
3686     /* All real repeats make it impossible to handle partial matching (maybe
3687     one day we will be able to remove this restriction). */
3688    
3689 ph10 230 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3690 nigel 77
3691     /* Combine the op_type with the repeat_type */
3692    
3693     repeat_type += op_type;
3694    
3695     /* A minimum of zero is handled either as the special case * or ?, or as
3696     an UPTO, with the maximum given. */
3697    
3698     if (repeat_min == 0)
3699     {
3700     if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
3701     else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
3702     else
3703     {
3704     *code++ = OP_UPTO + repeat_type;
3705     PUT2INC(code, 0, repeat_max);
3706     }
3707     }
3708    
3709     /* A repeat minimum of 1 is optimized into some special cases. If the
3710 nigel 93 maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
3711 nigel 77 left in place and, if the maximum is greater than 1, we use OP_UPTO with
3712     one less than the maximum. */
3713    
3714     else if (repeat_min == 1)
3715     {
3716     if (repeat_max == -1)
3717     *code++ = OP_PLUS + repeat_type;
3718     else
3719     {
3720     code = oldcode; /* leave previous item in place */
3721     if (repeat_max == 1) goto END_REPEAT;
3722     *code++ = OP_UPTO + repeat_type;
3723     PUT2INC(code, 0, repeat_max - 1);
3724     }
3725     }
3726    
3727     /* The case {n,n} is just an EXACT, while the general case {n,m} is
3728     handled as an EXACT followed by an UPTO. */
3729    
3730     else
3731     {
3732     *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
3733     PUT2INC(code, 0, repeat_min);
3734    
3735     /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
3736     we have to insert the character for the previous code. For a repeated
3737 nigel 87 Unicode property match, there are two extra bytes that define the
3738 nigel 77 required property. In UTF-8 mode, long characters have their length in
3739     c, with the 0x80 bit as a flag. */
3740    
3741     if (repeat_max < 0)
3742     {
3743     #ifdef SUPPORT_UTF8
3744     if (utf8 && c >= 128)
3745     {
3746     memcpy(code, utf8_char, c & 7);
3747     code += c & 7;
3748     }
3749     else
3750     #endif
3751     {
3752     *code++ = c;
3753 nigel 87 if (prop_type >= 0)
3754     {
3755     *code++ = prop_type;
3756     *code++ = prop_value;
3757     }
3758 nigel 77 }
3759     *code++ = OP_STAR + repeat_type;
3760     }
3761    
3762     /* Else insert an UPTO if the max is greater than the min, again
3763 nigel 93 preceded by the character, for the previously inserted code. If the
3764     UPTO is just for 1 instance, we can use QUERY instead. */
3765 nigel 77
3766     else if (repeat_max != repeat_min)
3767     {
3768     #ifdef SUPPORT_UTF8
3769     if (utf8 && c >= 128)
3770     {
3771     memcpy(code, utf8_char, c & 7);
3772     code += c & 7;
3773     }
3774     else
3775     #endif
3776     *code++ = c;
3777 nigel 87 if (prop_type >= 0)
3778     {
3779     *code++ = prop_type;
3780     *code++ = prop_value;
3781     }
3782 nigel 77 repeat_max -= repeat_min;
3783 nigel 93
3784     if (repeat_max == 1)
3785     {
3786     *code++ = OP_QUERY + repeat_type;
3787     }
3788     else
3789     {
3790     *code++ = OP_UPTO + repeat_type;
3791     PUT2INC(code, 0, repeat_max);
3792     }
3793 nigel 77 }
3794     }
3795    
3796     /* The character or character type itself comes last in all cases. */
3797    
3798     #ifdef SUPPORT_UTF8
3799     if (utf8 && c >= 128)
3800     {
3801     memcpy(code, utf8_char, c & 7);
3802     code += c & 7;
3803     }
3804     else
3805     #endif
3806     *code++ = c;
3807    
3808 nigel 87 /* For a repeated Unicode property match, there are two extra bytes that
3809     define the required property. */
3810 nigel 77
3811     #ifdef SUPPORT_UCP
3812 nigel 87 if (prop_type >= 0)
3813     {
3814     *code++ = prop_type;
3815     *code++ = prop_value;
3816     }
3817 nigel 77 #endif
3818     }
3819    
3820     /* If previous was a character class or a back reference, we put the repeat
3821     stuff after it, but just skip the item if the repeat was {0,0}. */
3822    
3823     else if (*previous == OP_CLASS ||
3824     *previous == OP_NCLASS ||
3825     #ifdef SUPPORT_UTF8
3826     *previous == OP_XCLASS ||
3827     #endif
3828     *previous == OP_REF)
3829     {
3830     if (repeat_max == 0)
3831     {
3832     code = previous;
3833     goto END_REPEAT;
3834     }
3835    
3836     /* All real repeats make it impossible to handle partial matching (maybe
3837     one day we will be able to remove this restriction). */
3838    
3839 ph10 230 if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL;
3840 nigel 77
3841     if (repeat_min == 0 && repeat_max == -1)
3842     *code++ = OP_CRSTAR + repeat_type;
3843     else if (repeat_min == 1 && repeat_max == -1)
3844     *code++ = OP_CRPLUS + repeat_type;
3845     else if (repeat_min == 0 && repeat_max == 1)
3846     *code++ = OP_CRQUERY + repeat_type;
3847     else
3848     {
3849     *code++ = OP_CRRANGE + repeat_type;
3850     PUT2INC(code, 0, repeat_min);
3851     if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
3852     PUT2INC(code, 0, repeat_max);
3853     }
3854     }
3855    
3856     /* If previous was a bracket group, we may have to replicate it in certain
3857     cases. */
3858    
3859 nigel 93 else if (*previous == OP_BRA || *previous == OP_CBRA ||
3860     *previous == OP_ONCE || *previous == OP_COND)
3861 nigel 77 {
3862     register int i;
3863     int ketoffset = 0;
3864     int len = code - previous;
3865     uschar *bralink = NULL;
3866    
3867 nigel 93 /* Repeating a DEFINE group is pointless */
3868    
3869     if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
3870     {
3871     *errorcodeptr = ERR55;
3872     goto FAILED;
3873     }
3874    
3875 nigel 77 /* If the maximum repeat count is unlimited, find the end of the bracket
3876     by scanning through from the start, and compute the offset back to it
3877     from the current code pointer. There may be an OP_OPT setting following
3878     the final KET, so we can't find the end just by going back from the code
3879     pointer. */
3880    
3881     if (repeat_max == -1)
3882     {
3883     register uschar *ket = previous;
3884     do ket += GET(ket, 1); while (*ket != OP_KET);
3885     ketoffset = code - ket;
3886     }
3887    
3888     /* The case of a zero minimum is special because of the need to stick
3889     OP_BRAZERO in front of it, and because the group appears once in the
3890     data, whereas in other cases it appears the minimum number of times. For
3891     this reason, it is simplest to treat this case separately, as otherwise
3892     the code gets far too messy. There are several special subcases when the
3893     minimum is zero. */
3894    
3895     if (repeat_min == 0)
3896     {
3897 ph10 335 /* If the maximum is also zero, we used to just omit the group from the
3898     output altogether, like this:
3899 nigel 77
3900 ph10 335 ** if (repeat_max == 0)
3901     ** {
3902     ** code = previous;
3903     ** goto END_REPEAT;
3904     ** }
3905 nigel 77
3906 ph10 345 However, that fails when a group is referenced as a subroutine from
3907     elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
3908     so that it is skipped on execution. As we don't have a list of which
3909     groups are referenced, we cannot do this selectively.
3910    
3911 ph10 335 If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
3912     and do no more at this point. However, we do need to adjust any
3913     OP_RECURSE calls inside the group that refer to the group itself or any
3914     internal or forward referenced group, because the offset is from the
3915     start of the whole regex. Temporarily terminate the pattern while doing
3916     this. */
3917 nigel 77
3918 ph10 335 if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
3919 nigel 77 {
3920     *code = OP_END;
3921 nigel 93 adjust_recurse(previous, 1, utf8, cd, save_hwm);
3922 nigel 77 memmove(previous+1, previous, len);
3923     code++;
3924 ph10 335 if (repeat_max == 0)
3925     {
3926     *previous++ = OP_SKIPZERO;
3927     goto END_REPEAT;
3928 ph10 345 }
3929 nigel 77 *previous++ = OP_BRAZERO + repeat_type;
3930     }
3931    
3932     /* If the maximum is greater than 1 and limited, we have to replicate
3933     in a nested fashion, sticking OP_BRAZERO before each set of brackets.
3934     The first one has to be handled carefully because it's the original
3935     copy, which has to be moved up. The remainder can be handled by code
3936     that is common with the non-zero minimum case below. We have to
3937     adjust the value or repeat_max, since one less copy is required. Once
3938     again, we may have to adjust any OP_RECURSE calls inside the group. */
3939    
3940     else
3941     {
3942     int offset;
3943     *code = OP_END;
3944 nigel 93 adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
3945 nigel 77 memmove(previous + 2 + LINK_SIZE, previous, len);
3946     code += 2 + LINK_SIZE;
3947     *previous++ = OP_BRAZERO + repeat_type;
3948     *previous++ = OP_BRA;
3949    
3950     /* We chain together the bracket offset fields that have to be
3951     filled in later when the ends of the brackets are reached. */
3952    
3953     offset = (bralink == NULL)? 0 : previous - bralink;
3954     bralink = previous;
3955     PUTINC(previous, 0, offset);
3956     }
3957    
3958     repeat_max--;
3959     }
3960    
3961     /* If the minimum is greater than zero, replicate the group as many
3962     times as necessary, and adjust the maximum to the number of subsequent
3963     copies that we need. If we set a first char from the group, and didn't
3964 nigel 93 set a required char, copy the latter from the former. If there are any
3965     forward reference subroutine calls in the group, there will be entries on
3966     the workspace list; replicate these with an appropriate increment. */
3967 nigel 77
3968     else
3969     {
3970     if (repeat_min > 1)
3971     {
3972 nigel 93 /* In the pre-compile phase, we don't actually do the replication. We
3973 ph10 202 just adjust the length as if we had. Do some paranoid checks for
3974     potential integer overflow. */
3975 nigel 93
3976     if (lengthptr != NULL)
3977 ph10 202 {
3978     int delta = (repeat_min - 1)*length_prevgroup;
3979     if ((double)(repeat_min - 1)*(double)length_prevgroup >
3980     (double)INT_MAX ||
3981     OFLOW_MAX - *lengthptr < delta)
3982     {
3983     *errorcodeptr = ERR20;
3984     goto FAILED;
3985     }
3986     *lengthptr += delta;
3987     }
3988 nigel 93
3989     /* This is compiling for real */
3990    
3991     else
3992 nigel 77 {
3993 nigel 93 if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
3994     for (i = 1; i < repeat_min; i++)
3995     {
3996     uschar *hc;
3997     uschar *this_hwm = cd->hwm;
3998     memcpy(code, previous, len);
3999     for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4000     {
4001     PUT(cd->hwm, 0, GET(hc, 0) + len);
4002     cd->hwm += LINK_SIZE;
4003     }
4004     save_hwm = this_hwm;
4005     code += len;
4006     }
4007 nigel 77 }
4008     }
4009 nigel 93
4010 nigel 77 if (repeat_max > 0) repeat_max -= repeat_min;
4011     }
4012    
4013     /* This code is common to both the zero and non-zero minimum cases. If
4014     the maximum is limited, it replicates the group in a nested fashion,
4015     remembering the bracket starts on a stack. In the case of a zero minimum,
4016     the first one was set up above. In all cases the repeat_max now specifies
4017 nigel 93 the number of additional copies needed. Again, we must remember to
4018     replicate entries on the forward reference list. */
4019 nigel 77
4020     if (repeat_max >= 0)
4021     {
4022 nigel 93 /* In the pre-compile phase, we don't actually do the replication. We
4023     just adjust the length as if we had. For each repetition we must add 1
4024     to the length for BRAZERO and for all but the last repetition we must
4025 ph10 202 add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
4026     paranoid checks to avoid integer overflow. */
4027 nigel 93
4028     if (lengthptr != NULL && repeat_max > 0)
4029 ph10 202 {
4030     int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4031     2 - 2*LINK_SIZE; /* Last one doesn't nest */
4032     if ((double)repeat_max *
4033     (double)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
4034     > (double)INT_MAX ||
4035     OFLOW_MAX - *lengthptr < delta)
4036     {
4037     *errorcodeptr = ERR20;
4038     goto FAILED;
4039     }
4040     *lengthptr += delta;
4041     }
4042 nigel 93
4043     /* This is compiling for real */
4044    
4045     else for (i = repeat_max - 1; i >= 0; i--)
4046 nigel 77 {
4047 nigel 93 uschar *hc;
4048     uschar *this_hwm = cd->hwm;
4049    
4050 nigel 77 *code++ = OP_BRAZERO + repeat_type;
4051    
4052     /* All but the final copy start a new nesting, maintaining the
4053     chain of brackets outstanding. */
4054    
4055     if (i != 0)
4056     {
4057     int offset;
4058     *code++ = OP_BRA;
4059     offset = (bralink == NULL)? 0 : code - bralink;
4060     bralink = code;
4061     PUTINC(code, 0, offset);
4062     }
4063    
4064 &nbs