Parent Directory
|
Revision Log
Fix problem with possessively repeated groups with minima greater than one.
1 | /************************************************* |
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 | Copyright (c) 1997-2011 University of Cambridge |
10 | |
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 | #ifdef HAVE_CONFIG_H |
46 | #include "config.h" |
47 | #endif |
48 | |
49 | #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 | #include "pcre_internal.h" |
54 | |
55 | |
56 | /* When PCRE_DEBUG is defined, we need the pcre_printint() function, which is |
57 | also used by pcretest. PCRE_DEBUG is not defined when building a production |
58 | library. */ |
59 | |
60 | #ifdef PCRE_DEBUG |
61 | #include "pcre_printint.src" |
62 | #endif |
63 | |
64 | |
65 | /* Macro for setting individual bits in class bitmaps. */ |
66 | |
67 | #define SETBIT(a,b) a[b/8] |= (1 << (b%8)) |
68 | |
69 | /* Maximum length value to check against when making sure that the integer that |
70 | holds the compiled pattern length does not overflow. We make it a bit less than |
71 | INT_MAX to allow for adding in group terminating bytes, so that we don't have |
72 | to check them every time. */ |
73 | |
74 | #define OFLOW_MAX (INT_MAX - 20) |
75 | |
76 | |
77 | /************************************************* |
78 | * Code parameters and static tables * |
79 | *************************************************/ |
80 | |
81 | /* This value specifies the size of stack workspace that is used during the |
82 | first pre-compile phase that determines how much memory is required. The regex |
83 | is partly compiled into this space, but the compiled parts are discarded as |
84 | soon as they can be, so that hopefully there will never be an overrun. The code |
85 | does, however, check for an overrun. The largest amount I've seen used is 218, |
86 | so this number is very generous. |
87 | |
88 | The same workspace is used during the second, actual compile phase for |
89 | remembering forward references to groups so that they can be filled in at the |
90 | end. Each entry in this list occupies LINK_SIZE bytes, so even when LINK_SIZE |
91 | is 4 there is plenty of room. */ |
92 | |
93 | #define COMPILE_WORK_SIZE (4096) |
94 | |
95 | /* The overrun tests check for a slightly smaller size so that they detect the |
96 | overrun before it actually does run off the end of the data block. */ |
97 | |
98 | #define WORK_SIZE_CHECK (COMPILE_WORK_SIZE - 100) |
99 | |
100 | |
101 | /* Table for handling escaped characters in the range '0'-'z'. Positive returns |
102 | are simple data values; negative values are for special things like \d and so |
103 | on. Zero means further processing is needed (for things like \x), or the escape |
104 | is invalid. */ |
105 | |
106 | #ifndef EBCDIC |
107 | |
108 | /* This is the "normal" table for ASCII systems or for EBCDIC systems running |
109 | in UTF-8 mode. */ |
110 | |
111 | static const short int escapes[] = { |
112 | 0, 0, |
113 | 0, 0, |
114 | 0, 0, |
115 | 0, 0, |
116 | 0, 0, |
117 | CHAR_COLON, CHAR_SEMICOLON, |
118 | CHAR_LESS_THAN_SIGN, CHAR_EQUALS_SIGN, |
119 | CHAR_GREATER_THAN_SIGN, CHAR_QUESTION_MARK, |
120 | CHAR_COMMERCIAL_AT, -ESC_A, |
121 | -ESC_B, -ESC_C, |
122 | -ESC_D, -ESC_E, |
123 | 0, -ESC_G, |
124 | -ESC_H, 0, |
125 | 0, -ESC_K, |
126 | 0, 0, |
127 | -ESC_N, 0, |
128 | -ESC_P, -ESC_Q, |
129 | -ESC_R, -ESC_S, |
130 | 0, 0, |
131 | -ESC_V, -ESC_W, |
132 | -ESC_X, 0, |
133 | -ESC_Z, CHAR_LEFT_SQUARE_BRACKET, |
134 | CHAR_BACKSLASH, CHAR_RIGHT_SQUARE_BRACKET, |
135 | CHAR_CIRCUMFLEX_ACCENT, CHAR_UNDERSCORE, |
136 | CHAR_GRAVE_ACCENT, 7, |
137 | -ESC_b, 0, |
138 | -ESC_d, ESC_e, |
139 | ESC_f, 0, |
140 | -ESC_h, 0, |
141 | 0, -ESC_k, |
142 | 0, 0, |
143 | ESC_n, 0, |
144 | -ESC_p, 0, |
145 | ESC_r, -ESC_s, |
146 | ESC_tee, 0, |
147 | -ESC_v, -ESC_w, |
148 | 0, 0, |
149 | -ESC_z |
150 | }; |
151 | |
152 | #else |
153 | |
154 | /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */ |
155 | |
156 | static const short int escapes[] = { |
157 | /* 48 */ 0, 0, 0, '.', '<', '(', '+', '|', |
158 | /* 50 */ '&', 0, 0, 0, 0, 0, 0, 0, |
159 | /* 58 */ 0, 0, '!', '$', '*', ')', ';', '~', |
160 | /* 60 */ '-', '/', 0, 0, 0, 0, 0, 0, |
161 | /* 68 */ 0, 0, '|', ',', '%', '_', '>', '?', |
162 | /* 70 */ 0, 0, 0, 0, 0, 0, 0, 0, |
163 | /* 78 */ 0, '`', ':', '#', '@', '\'', '=', '"', |
164 | /* 80 */ 0, 7, -ESC_b, 0, -ESC_d, ESC_e, ESC_f, 0, |
165 | /* 88 */-ESC_h, 0, 0, '{', 0, 0, 0, 0, |
166 | /* 90 */ 0, 0, -ESC_k, 'l', 0, ESC_n, 0, -ESC_p, |
167 | /* 98 */ 0, ESC_r, 0, '}', 0, 0, 0, 0, |
168 | /* A0 */ 0, '~', -ESC_s, ESC_tee, 0,-ESC_v, -ESC_w, 0, |
169 | /* A8 */ 0,-ESC_z, 0, 0, 0, '[', 0, 0, |
170 | /* B0 */ 0, 0, 0, 0, 0, 0, 0, 0, |
171 | /* B8 */ 0, 0, 0, 0, 0, ']', '=', '-', |
172 | /* C0 */ '{',-ESC_A, -ESC_B, -ESC_C, -ESC_D,-ESC_E, 0, -ESC_G, |
173 | /* C8 */-ESC_H, 0, 0, 0, 0, 0, 0, 0, |
174 | /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P, |
175 | /* D8 */-ESC_Q,-ESC_R, 0, 0, 0, 0, 0, 0, |
176 | /* E0 */ '\\', 0, -ESC_S, 0, 0,-ESC_V, -ESC_W, -ESC_X, |
177 | /* E8 */ 0,-ESC_Z, 0, 0, 0, 0, 0, 0, |
178 | /* F0 */ 0, 0, 0, 0, 0, 0, 0, 0, |
179 | /* F8 */ 0, 0, 0, 0, 0, 0, 0, 0 |
180 | }; |
181 | #endif |
182 | |
183 | |
184 | /* Table of special "verbs" like (*PRUNE). This is a short table, so it is |
185 | searched linearly. Put all the names into a single string, in order to reduce |
186 | the number of relocations when a shared library is dynamically linked. The |
187 | string is built from string macros so that it works in UTF-8 mode on EBCDIC |
188 | platforms. */ |
189 | |
190 | typedef struct verbitem { |
191 | int len; /* Length of verb name */ |
192 | int op; /* Op when no arg, or -1 if arg mandatory */ |
193 | int op_arg; /* Op when arg present, or -1 if not allowed */ |
194 | } verbitem; |
195 | |
196 | static const char verbnames[] = |
197 | "\0" /* Empty name is a shorthand for MARK */ |
198 | STRING_MARK0 |
199 | STRING_ACCEPT0 |
200 | STRING_COMMIT0 |
201 | STRING_F0 |
202 | STRING_FAIL0 |
203 | STRING_PRUNE0 |
204 | STRING_SKIP0 |
205 | STRING_THEN; |
206 | |
207 | static const verbitem verbs[] = { |
208 | { 0, -1, OP_MARK }, |
209 | { 4, -1, OP_MARK }, |
210 | { 6, OP_ACCEPT, -1 }, |
211 | { 6, OP_COMMIT, -1 }, |
212 | { 1, OP_FAIL, -1 }, |
213 | { 4, OP_FAIL, -1 }, |
214 | { 5, OP_PRUNE, OP_PRUNE_ARG }, |
215 | { 4, OP_SKIP, OP_SKIP_ARG }, |
216 | { 4, OP_THEN, OP_THEN_ARG } |
217 | }; |
218 | |
219 | static const int verbcount = sizeof(verbs)/sizeof(verbitem); |
220 | |
221 | |
222 | /* Tables of names of POSIX character classes and their lengths. The names are |
223 | now all in a single string, to reduce the number of relocations when a shared |
224 | library is dynamically loaded. The list of lengths is terminated by a zero |
225 | length entry. The first three must be alpha, lower, upper, as this is assumed |
226 | for handling case independence. */ |
227 | |
228 | static const char posix_names[] = |
229 | STRING_alpha0 STRING_lower0 STRING_upper0 STRING_alnum0 |
230 | STRING_ascii0 STRING_blank0 STRING_cntrl0 STRING_digit0 |
231 | STRING_graph0 STRING_print0 STRING_punct0 STRING_space0 |
232 | STRING_word0 STRING_xdigit; |
233 | |
234 | static const uschar posix_name_lengths[] = { |
235 | 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 4, 6, 0 }; |
236 | |
237 | /* Table of class bit maps for each POSIX class. Each class is formed from a |
238 | base map, with an optional addition or removal of another map. Then, for some |
239 | classes, there is some additional tweaking: for [:blank:] the vertical space |
240 | characters are removed, and for [:alpha:] and [:alnum:] the underscore |
241 | character is removed. The triples in the table consist of the base map offset, |
242 | second map offset or -1 if no second map, and a non-negative value for map |
243 | addition or a negative value for map subtraction (if there are two maps). The |
244 | absolute value of the third field has these meanings: 0 => no tweaking, 1 => |
245 | remove vertical space characters, 2 => remove underscore. */ |
246 | |
247 | static const int posix_class_maps[] = { |
248 | cbit_word, cbit_digit, -2, /* alpha */ |
249 | cbit_lower, -1, 0, /* lower */ |
250 | cbit_upper, -1, 0, /* upper */ |
251 | cbit_word, -1, 2, /* alnum - word without underscore */ |
252 | cbit_print, cbit_cntrl, 0, /* ascii */ |
253 | cbit_space, -1, 1, /* blank - a GNU extension */ |
254 | cbit_cntrl, -1, 0, /* cntrl */ |
255 | cbit_digit, -1, 0, /* digit */ |
256 | cbit_graph, -1, 0, /* graph */ |
257 | cbit_print, -1, 0, /* print */ |
258 | cbit_punct, -1, 0, /* punct */ |
259 | cbit_space, -1, 0, /* space */ |
260 | cbit_word, -1, 0, /* word - a Perl extension */ |
261 | cbit_xdigit,-1, 0 /* xdigit */ |
262 | }; |
263 | |
264 | /* Table of substitutes for \d etc when PCRE_UCP is set. The POSIX class |
265 | substitutes must be in the order of the names, defined above, and there are |
266 | both positive and negative cases. NULL means no substitute. */ |
267 | |
268 | #ifdef SUPPORT_UCP |
269 | static const uschar *substitutes[] = { |
270 | (uschar *)"\\P{Nd}", /* \D */ |
271 | (uschar *)"\\p{Nd}", /* \d */ |
272 | (uschar *)"\\P{Xsp}", /* \S */ /* NOTE: Xsp is Perl space */ |
273 | (uschar *)"\\p{Xsp}", /* \s */ |
274 | (uschar *)"\\P{Xwd}", /* \W */ |
275 | (uschar *)"\\p{Xwd}" /* \w */ |
276 | }; |
277 | |
278 | static const uschar *posix_substitutes[] = { |
279 | (uschar *)"\\p{L}", /* alpha */ |
280 | (uschar *)"\\p{Ll}", /* lower */ |
281 | (uschar *)"\\p{Lu}", /* upper */ |
282 | (uschar *)"\\p{Xan}", /* alnum */ |
283 | NULL, /* ascii */ |
284 | (uschar *)"\\h", /* blank */ |
285 | NULL, /* cntrl */ |
286 | (uschar *)"\\p{Nd}", /* digit */ |
287 | NULL, /* graph */ |
288 | NULL, /* print */ |
289 | NULL, /* punct */ |
290 | (uschar *)"\\p{Xps}", /* space */ /* NOTE: Xps is POSIX space */ |
291 | (uschar *)"\\p{Xwd}", /* word */ |
292 | NULL, /* xdigit */ |
293 | /* Negated cases */ |
294 | (uschar *)"\\P{L}", /* ^alpha */ |
295 | (uschar *)"\\P{Ll}", /* ^lower */ |
296 | (uschar *)"\\P{Lu}", /* ^upper */ |
297 | (uschar *)"\\P{Xan}", /* ^alnum */ |
298 | NULL, /* ^ascii */ |
299 | (uschar *)"\\H", /* ^blank */ |
300 | NULL, /* ^cntrl */ |
301 | (uschar *)"\\P{Nd}", /* ^digit */ |
302 | NULL, /* ^graph */ |
303 | NULL, /* ^print */ |
304 | NULL, /* ^punct */ |
305 | (uschar *)"\\P{Xps}", /* ^space */ /* NOTE: Xps is POSIX space */ |
306 | (uschar *)"\\P{Xwd}", /* ^word */ |
307 | NULL /* ^xdigit */ |
308 | }; |
309 | #define POSIX_SUBSIZE (sizeof(posix_substitutes)/sizeof(uschar *)) |
310 | #endif |
311 | |
312 | #define STRING(a) # a |
313 | #define XSTRING(s) STRING(s) |
314 | |
315 | /* The texts of compile-time error messages. These are "char *" because they |
316 | are passed to the outside world. Do not ever re-use any error number, because |
317 | they are documented. Always add a new error instead. Messages marked DEAD below |
318 | are no longer used. This used to be a table of strings, but in order to reduce |
319 | the number of relocations needed when a shared library is loaded dynamically, |
320 | it is now one long string. We cannot use a table of offsets, because the |
321 | lengths of inserts such as XSTRING(MAX_NAME_SIZE) are not known. Instead, we |
322 | simply count through to the one we want - this isn't a performance issue |
323 | because these strings are used only when there is a compilation error. |
324 | |
325 | Each substring ends with \0 to insert a null character. This includes the final |
326 | substring, so that the whole string ends with \0\0, which can be detected when |
327 | counting through. */ |
328 | |
329 | static const char error_texts[] = |
330 | "no error\0" |
331 | "\\ at end of pattern\0" |
332 | "\\c at end of pattern\0" |
333 | "unrecognized character follows \\\0" |
334 | "numbers out of order in {} quantifier\0" |
335 | /* 5 */ |
336 | "number too big in {} quantifier\0" |
337 | "missing terminating ] for character class\0" |
338 | "invalid escape sequence in character class\0" |
339 | "range out of order in character class\0" |
340 | "nothing to repeat\0" |
341 | /* 10 */ |
342 | "operand of unlimited repeat could match the empty string\0" /** DEAD **/ |
343 | "internal error: unexpected repeat\0" |
344 | "unrecognized character after (? or (?-\0" |
345 | "POSIX named classes are supported only within a class\0" |
346 | "missing )\0" |
347 | /* 15 */ |
348 | "reference to non-existent subpattern\0" |
349 | "erroffset passed as NULL\0" |
350 | "unknown option bit(s) set\0" |
351 | "missing ) after comment\0" |
352 | "parentheses nested too deeply\0" /** DEAD **/ |
353 | /* 20 */ |
354 | "regular expression is too large\0" |
355 | "failed to get memory\0" |
356 | "unmatched parentheses\0" |
357 | "internal error: code overflow\0" |
358 | "unrecognized character after (?<\0" |
359 | /* 25 */ |
360 | "lookbehind assertion is not fixed length\0" |
361 | "malformed number or name after (?(\0" |
362 | "conditional group contains more than two branches\0" |
363 | "assertion expected after (?(\0" |
364 | "(?R or (?[+-]digits must be followed by )\0" |
365 | /* 30 */ |
366 | "unknown POSIX class name\0" |
367 | "POSIX collating elements are not supported\0" |
368 | "this version of PCRE is not compiled with PCRE_UTF8 support\0" |
369 | "spare error\0" /** DEAD **/ |
370 | "character value in \\x{...} sequence is too large\0" |
371 | /* 35 */ |
372 | "invalid condition (?(0)\0" |
373 | "\\C not allowed in lookbehind assertion\0" |
374 | "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0" |
375 | "number after (?C is > 255\0" |
376 | "closing ) for (?C expected\0" |
377 | /* 40 */ |
378 | "recursive call could loop indefinitely\0" |
379 | "unrecognized character after (?P\0" |
380 | "syntax error in subpattern name (missing terminator)\0" |
381 | "two named subpatterns have the same name\0" |
382 | "invalid UTF-8 string\0" |
383 | /* 45 */ |
384 | "support for \\P, \\p, and \\X has not been compiled\0" |
385 | "malformed \\P or \\p sequence\0" |
386 | "unknown property name after \\P or \\p\0" |
387 | "subpattern name is too long (maximum " XSTRING(MAX_NAME_SIZE) " characters)\0" |
388 | "too many named subpatterns (maximum " XSTRING(MAX_NAME_COUNT) ")\0" |
389 | /* 50 */ |
390 | "repeated subpattern is too long\0" /** DEAD **/ |
391 | "octal value is greater than \\377 (not in UTF-8 mode)\0" |
392 | "internal error: overran compiling workspace\0" |
393 | "internal error: previously-checked referenced subpattern not found\0" |
394 | "DEFINE group contains more than one branch\0" |
395 | /* 55 */ |
396 | "repeating a DEFINE group is not allowed\0" /** DEAD **/ |
397 | "inconsistent NEWLINE options\0" |
398 | "\\g is not followed by a braced, angle-bracketed, or quoted name/number or by a plain number\0" |
399 | "a numbered reference must not be zero\0" |
400 | "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0" |
401 | /* 60 */ |
402 | "(*VERB) not recognized\0" |
403 | "number is too big\0" |
404 | "subpattern name expected\0" |
405 | "digit expected after (?+\0" |
406 | "] is an invalid data character in JavaScript compatibility mode\0" |
407 | /* 65 */ |
408 | "different names for subpatterns of the same number are not allowed\0" |
409 | "(*MARK) must have an argument\0" |
410 | "this version of PCRE is not compiled with PCRE_UCP support\0" |
411 | "\\c must be followed by an ASCII character\0" |
412 | "\\k is not followed by a braced, angle-bracketed, or quoted name\0" |
413 | /* 70 */ |
414 | "internal error: unknown opcode in find_fixedlength()\0" |
415 | ; |
416 | |
417 | /* Table to identify digits and hex digits. This is used when compiling |
418 | patterns. Note that the tables in chartables are dependent on the locale, and |
419 | may mark arbitrary characters as digits - but the PCRE compiling code expects |
420 | to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have |
421 | a private table here. It costs 256 bytes, but it is a lot faster than doing |
422 | character value tests (at least in some simple cases I timed), and in some |
423 | applications one wants PCRE to compile efficiently as well as match |
424 | efficiently. |
425 | |
426 | For convenience, we use the same bit definitions as in chartables: |
427 | |
428 | 0x04 decimal digit |
429 | 0x08 hexadecimal digit |
430 | |
431 | Then we can use ctype_digit and ctype_xdigit in the code. */ |
432 | |
433 | #ifndef EBCDIC |
434 | |
435 | /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in |
436 | UTF-8 mode. */ |
437 | |
438 | static const unsigned char digitab[] = |
439 | { |
440 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */ |
441 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */ |
442 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */ |
443 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */ |
444 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */ |
445 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */ |
446 | 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */ |
447 | 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */ |
448 | 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */ |
449 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */ |
450 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */ |
451 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */ |
452 | 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* ` - g */ |
453 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h - o */ |
454 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* p - w */ |
455 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* x -127 */ |
456 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */ |
457 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */ |
458 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */ |
459 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */ |
460 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */ |
461 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */ |
462 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */ |
463 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */ |
464 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */ |
465 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */ |
466 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */ |
467 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */ |
468 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */ |
469 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */ |
470 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */ |
471 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */ |
472 | |
473 | #else |
474 | |
475 | /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */ |
476 | |
477 | static const unsigned char digitab[] = |
478 | { |
479 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */ |
480 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */ |
481 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */ |
482 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */ |
483 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */ |
484 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */ |
485 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */ |
486 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */ |
487 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */ |
488 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */ |
489 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */ |
490 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */ |
491 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */ |
492 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */ |
493 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */ |
494 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */ |
495 | 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */ |
496 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */ |
497 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */ |
498 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */ |
499 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */ |
500 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */ |
501 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */ |
502 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */ |
503 | 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */ |
504 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */ |
505 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */ |
506 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */ |
507 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */ |
508 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */ |
509 | 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */ |
510 | 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */ |
511 | |
512 | static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */ |
513 | 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */ |
514 | 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */ |
515 | 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */ |
516 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */ |
517 | 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */ |
518 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */ |
519 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */ |
520 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */ |
521 | 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */ |
522 | 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */ |
523 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */ |
524 | 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */ |
525 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */ |
526 | 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */ |
527 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */ |
528 | 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */ |
529 | 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */ |
530 | 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */ |
531 | 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */ |
532 | 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */ |
533 | 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */ |
534 | 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */ |
535 | 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */ |
536 | 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */ |
537 | 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */ |
538 | 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */ |
539 | 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */ |
540 | 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */ |
541 | 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */ |
542 | 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */ |
543 | 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */ |
544 | 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */ |
545 | #endif |
546 | |
547 | |
548 | /* Definition to allow mutual recursion */ |
549 | |
550 | static BOOL |
551 | compile_regex(int, uschar **, const uschar **, int *, BOOL, BOOL, int, int, |
552 | int *, int *, branch_chain *, compile_data *, int *); |
553 | |
554 | |
555 | |
556 | /************************************************* |
557 | * Find an error text * |
558 | *************************************************/ |
559 | |
560 | /* The error texts are now all in one long string, to save on relocations. As |
561 | some of the text is of unknown length, we can't use a table of offsets. |
562 | Instead, just count through the strings. This is not a performance issue |
563 | because it happens only when there has been a compilation error. |
564 | |
565 | Argument: the error number |
566 | Returns: pointer to the error string |
567 | */ |
568 | |
569 | static const char * |
570 | find_error_text(int n) |
571 | { |
572 | const char *s = error_texts; |
573 | for (; n > 0; n--) |
574 | { |
575 | while (*s++ != 0) {}; |
576 | if (*s == 0) return "Error text not found (please report)"; |
577 | } |
578 | return s; |
579 | } |
580 | |
581 | |
582 | /************************************************* |
583 | * Check for counted repeat * |
584 | *************************************************/ |
585 | |
586 | /* This function is called when a '{' is encountered in a place where it might |
587 | start a quantifier. It looks ahead to see if it really is a quantifier or not. |
588 | It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd} |
589 | where the ddds are digits. |
590 | |
591 | Arguments: |
592 | p pointer to the first char after '{' |
593 | |
594 | Returns: TRUE or FALSE |
595 | */ |
596 | |
597 | static BOOL |
598 | is_counted_repeat(const uschar *p) |
599 | { |
600 | if ((digitab[*p++] & ctype_digit) == 0) return FALSE; |
601 | while ((digitab[*p] & ctype_digit) != 0) p++; |
602 | if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE; |
603 | |
604 | if (*p++ != CHAR_COMMA) return FALSE; |
605 | if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE; |
606 | |
607 | if ((digitab[*p++] & ctype_digit) == 0) return FALSE; |
608 | while ((digitab[*p] & ctype_digit) != 0) p++; |
609 | |
610 | return (*p == CHAR_RIGHT_CURLY_BRACKET); |
611 | } |
612 | |
613 | |
614 | |
615 | /************************************************* |
616 | * Handle escapes * |
617 | *************************************************/ |
618 | |
619 | /* This function is called when a \ has been encountered. It either returns a |
620 | positive value for a simple escape such as \n, or a negative value which |
621 | encodes one of the more complicated things such as \d. A backreference to group |
622 | n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When |
623 | UTF-8 is enabled, a positive value greater than 255 may be returned. On entry, |
624 | ptr is pointing at the \. On exit, it is on the final character of the escape |
625 | sequence. |
626 | |
627 | Arguments: |
628 | ptrptr points to the pattern position pointer |
629 | errorcodeptr points to the errorcode variable |
630 | bracount number of previous extracting brackets |
631 | options the options bits |
632 | isclass TRUE if inside a character class |
633 | |
634 | Returns: zero or positive => a data character |
635 | negative => a special escape sequence |
636 | on error, errorcodeptr is set |
637 | */ |
638 | |
639 | static int |
640 | check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount, |
641 | int options, BOOL isclass) |
642 | { |
643 | BOOL utf8 = (options & PCRE_UTF8) != 0; |
644 | const uschar *ptr = *ptrptr + 1; |
645 | int c, i; |
646 | |
647 | GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */ |
648 | ptr--; /* Set pointer back to the last byte */ |
649 | |
650 | /* If backslash is at the end of the pattern, it's an error. */ |
651 | |
652 | if (c == 0) *errorcodeptr = ERR1; |
653 | |
654 | /* Non-alphanumerics are literals. For digits or letters, do an initial lookup |
655 | in a table. A non-zero result is something that can be returned immediately. |
656 | Otherwise further processing may be required. */ |
657 | |
658 | #ifndef EBCDIC /* ASCII/UTF-8 coding */ |
659 | else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */ |
660 | else if ((i = escapes[c - CHAR_0]) != 0) c = i; |
661 | |
662 | #else /* EBCDIC coding */ |
663 | else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */ |
664 | else if ((i = escapes[c - 0x48]) != 0) c = i; |
665 | #endif |
666 | |
667 | /* Escapes that need further processing, or are illegal. */ |
668 | |
669 | else |
670 | { |
671 | const uschar *oldptr; |
672 | BOOL braced, negated; |
673 | |
674 | switch (c) |
675 | { |
676 | /* A number of Perl escapes are not handled by PCRE. We give an explicit |
677 | error. */ |
678 | |
679 | case CHAR_l: |
680 | case CHAR_L: |
681 | *errorcodeptr = ERR37; |
682 | break; |
683 | |
684 | case CHAR_u: |
685 | if ((options & PCRE_JAVASCRIPT_COMPAT) != 0) |
686 | { |
687 | /* In JavaScript, \u must be followed by four hexadecimal numbers. |
688 | Otherwise it is a lowercase u letter. */ |
689 | if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0 |
690 | && (digitab[ptr[3]] & ctype_xdigit) != 0 && (digitab[ptr[4]] & ctype_xdigit) != 0) |
691 | { |
692 | c = 0; |
693 | for (i = 0; i < 4; ++i) |
694 | { |
695 | register int cc = *(++ptr); |
696 | #ifndef EBCDIC /* ASCII/UTF-8 coding */ |
697 | if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ |
698 | c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); |
699 | #else /* EBCDIC coding */ |
700 | if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */ |
701 | c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); |
702 | #endif |
703 | } |
704 | } |
705 | } |
706 | else |
707 | *errorcodeptr = ERR37; |
708 | break; |
709 | |
710 | case CHAR_U: |
711 | /* In JavaScript, \U is an uppercase U letter. */ |
712 | if ((options & PCRE_JAVASCRIPT_COMPAT) == 0) *errorcodeptr = ERR37; |
713 | break; |
714 | |
715 | /* In a character class, \g is just a literal "g". Outside a character |
716 | class, \g must be followed by one of a number of specific things: |
717 | |
718 | (1) A number, either plain or braced. If positive, it is an absolute |
719 | backreference. If negative, it is a relative backreference. This is a Perl |
720 | 5.10 feature. |
721 | |
722 | (2) Perl 5.10 also supports \g{name} as a reference to a named group. This |
723 | is part of Perl's movement towards a unified syntax for back references. As |
724 | this is synonymous with \k{name}, we fudge it up by pretending it really |
725 | was \k. |
726 | |
727 | (3) For Oniguruma compatibility we also support \g followed by a name or a |
728 | number either in angle brackets or in single quotes. However, these are |
729 | (possibly recursive) subroutine calls, _not_ backreferences. Just return |
730 | the -ESC_g code (cf \k). */ |
731 | |
732 | case CHAR_g: |
733 | if (isclass) break; |
734 | if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE) |
735 | { |
736 | c = -ESC_g; |
737 | break; |
738 | } |
739 | |
740 | /* Handle the Perl-compatible cases */ |
741 | |
742 | if (ptr[1] == CHAR_LEFT_CURLY_BRACKET) |
743 | { |
744 | const uschar *p; |
745 | for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++) |
746 | if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break; |
747 | if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET) |
748 | { |
749 | c = -ESC_k; |
750 | break; |
751 | } |
752 | braced = TRUE; |
753 | ptr++; |
754 | } |
755 | else braced = FALSE; |
756 | |
757 | if (ptr[1] == CHAR_MINUS) |
758 | { |
759 | negated = TRUE; |
760 | ptr++; |
761 | } |
762 | else negated = FALSE; |
763 | |
764 | c = 0; |
765 | while ((digitab[ptr[1]] & ctype_digit) != 0) |
766 | c = c * 10 + *(++ptr) - CHAR_0; |
767 | |
768 | if (c < 0) /* Integer overflow */ |
769 | { |
770 | *errorcodeptr = ERR61; |
771 | break; |
772 | } |
773 | |
774 | if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET) |
775 | { |
776 | *errorcodeptr = ERR57; |
777 | break; |
778 | } |
779 | |
780 | if (c == 0) |
781 | { |
782 | *errorcodeptr = ERR58; |
783 | break; |
784 | } |
785 | |
786 | if (negated) |
787 | { |
788 | if (c > bracount) |
789 | { |
790 | *errorcodeptr = ERR15; |
791 | break; |
792 | } |
793 | c = bracount - (c - 1); |
794 | } |
795 | |
796 | c = -(ESC_REF + c); |
797 | break; |
798 | |
799 | /* The handling of escape sequences consisting of a string of digits |
800 | starting with one that is not zero is not straightforward. By experiment, |
801 | the way Perl works seems to be as follows: |
802 | |
803 | Outside a character class, the digits are read as a decimal number. If the |
804 | number is less than 10, or if there are that many previous extracting |
805 | left brackets, then it is a back reference. Otherwise, up to three octal |
806 | digits are read to form an escaped byte. Thus \123 is likely to be octal |
807 | 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal |
808 | value is greater than 377, the least significant 8 bits are taken. Inside a |
809 | character class, \ followed by a digit is always an octal number. */ |
810 | |
811 | case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5: |
812 | case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9: |
813 | |
814 | if (!isclass) |
815 | { |
816 | oldptr = ptr; |
817 | c -= CHAR_0; |
818 | while ((digitab[ptr[1]] & ctype_digit) != 0) |
819 | c = c * 10 + *(++ptr) - CHAR_0; |
820 | if (c < 0) /* Integer overflow */ |
821 | { |
822 | *errorcodeptr = ERR61; |
823 | break; |
824 | } |
825 | if (c < 10 || c <= bracount) |
826 | { |
827 | c = -(ESC_REF + c); |
828 | break; |
829 | } |
830 | ptr = oldptr; /* Put the pointer back and fall through */ |
831 | } |
832 | |
833 | /* Handle an octal number following \. If the first digit is 8 or 9, Perl |
834 | generates a binary zero byte and treats the digit as a following literal. |
835 | Thus we have to pull back the pointer by one. */ |
836 | |
837 | if ((c = *ptr) >= CHAR_8) |
838 | { |
839 | ptr--; |
840 | c = 0; |
841 | break; |
842 | } |
843 | |
844 | /* \0 always starts an octal number, but we may drop through to here with a |
845 | larger first octal digit. The original code used just to take the least |
846 | significant 8 bits of octal numbers (I think this is what early Perls used |
847 | to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more |
848 | than 3 octal digits. */ |
849 | |
850 | case CHAR_0: |
851 | c -= CHAR_0; |
852 | while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7) |
853 | c = c * 8 + *(++ptr) - CHAR_0; |
854 | if (!utf8 && c > 255) *errorcodeptr = ERR51; |
855 | break; |
856 | |
857 | /* \x is complicated. \x{ddd} is a character number which can be greater |
858 | than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is |
859 | treated as a data character. */ |
860 | |
861 | case CHAR_x: |
862 | if ((options & PCRE_JAVASCRIPT_COMPAT) != 0) |
863 | { |
864 | /* In JavaScript, \x must be followed by two hexadecimal numbers. |
865 | Otherwise it is a lowercase x letter. */ |
866 | if ((digitab[ptr[1]] & ctype_xdigit) != 0 && (digitab[ptr[2]] & ctype_xdigit) != 0) |
867 | { |
868 | c = 0; |
869 | for (i = 0; i < 2; ++i) |
870 | { |
871 | register int cc = *(++ptr); |
872 | #ifndef EBCDIC /* ASCII/UTF-8 coding */ |
873 | if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ |
874 | c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); |
875 | #else /* EBCDIC coding */ |
876 | if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */ |
877 | c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); |
878 | #endif |
879 | } |
880 | } |
881 | break; |
882 | } |
883 | |
884 | if (ptr[1] == CHAR_LEFT_CURLY_BRACKET) |
885 | { |
886 | const uschar *pt = ptr + 2; |
887 | int count = 0; |
888 | |
889 | c = 0; |
890 | while ((digitab[*pt] & ctype_xdigit) != 0) |
891 | { |
892 | register int cc = *pt++; |
893 | if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */ |
894 | count++; |
895 | |
896 | #ifndef EBCDIC /* ASCII/UTF-8 coding */ |
897 | if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ |
898 | c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); |
899 | #else /* EBCDIC coding */ |
900 | if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */ |
901 | c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); |
902 | #endif |
903 | } |
904 | |
905 | if (*pt == CHAR_RIGHT_CURLY_BRACKET) |
906 | { |
907 | if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34; |
908 | ptr = pt; |
909 | break; |
910 | } |
911 | |
912 | /* If the sequence of hex digits does not end with '}', then we don't |
913 | recognize this construct; fall through to the normal \x handling. */ |
914 | } |
915 | |
916 | /* Read just a single-byte hex-defined char */ |
917 | |
918 | c = 0; |
919 | while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0) |
920 | { |
921 | int cc; /* Some compilers don't like */ |
922 | cc = *(++ptr); /* ++ in initializers */ |
923 | #ifndef EBCDIC /* ASCII/UTF-8 coding */ |
924 | if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */ |
925 | c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10)); |
926 | #else /* EBCDIC coding */ |
927 | if (cc <= CHAR_z) cc += 64; /* Convert to upper case */ |
928 | c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10)); |
929 | #endif |
930 | } |
931 | break; |
932 | |
933 | /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped. |
934 | An error is given if the byte following \c is not an ASCII character. This |
935 | coding is ASCII-specific, but then the whole concept of \cx is |
936 | ASCII-specific. (However, an EBCDIC equivalent has now been added.) */ |
937 | |
938 | case CHAR_c: |
939 | c = *(++ptr); |
940 | if (c == 0) |
941 | { |
942 | *errorcodeptr = ERR2; |
943 | break; |
944 | } |
945 | #ifndef EBCDIC /* ASCII/UTF-8 coding */ |
946 | if (c > 127) /* Excludes all non-ASCII in either mode */ |
947 | { |
948 | *errorcodeptr = ERR68; |
949 | break; |
950 | } |
951 | if (c >= CHAR_a && c <= CHAR_z) c -= 32; |
952 | c ^= 0x40; |
953 | #else /* EBCDIC coding */ |
954 | if (c >= CHAR_a && c <= CHAR_z) c += 64; |
955 | c ^= 0xC0; |
956 | #endif |
957 | break; |
958 | |
959 | /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any |
960 | other alphanumeric following \ is an error if PCRE_EXTRA was set; |
961 | otherwise, for Perl compatibility, it is a literal. This code looks a bit |
962 | odd, but there used to be some cases other than the default, and there may |
963 | be again in future, so I haven't "optimized" it. */ |
964 | |
965 | default: |
966 | if ((options & PCRE_EXTRA) != 0) switch(c) |
967 | { |
968 | default: |
969 | *errorcodeptr = ERR3; |
970 | break; |
971 | } |
972 | break; |
973 | } |
974 | } |
975 | |
976 | /* Perl supports \N{name} for character names, as well as plain \N for "not |
977 | newline". PCRE does not support \N{name}. However, it does support |
978 | quantification such as \N{2,3}. */ |
979 | |
980 | if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET && |
981 | !is_counted_repeat(ptr+2)) |
982 | *errorcodeptr = ERR37; |
983 | |
984 | /* If PCRE_UCP is set, we change the values for \d etc. */ |
985 | |
986 | if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w) |
987 | c -= (ESC_DU - ESC_D); |
988 | |
989 | /* Set the pointer to the final character before returning. */ |
990 | |
991 | *ptrptr = ptr; |
992 | return c; |
993 | } |
994 | |
995 | |
996 | |
997 | #ifdef SUPPORT_UCP |
998 | /************************************************* |
999 | * Handle \P and \p * |
1000 | *************************************************/ |
1001 | |
1002 | /* This function is called after \P or \p has been encountered, provided that |
1003 | PCRE is compiled with support for Unicode properties. On entry, ptrptr is |
1004 | pointing at the P or p. On exit, it is pointing at the final character of the |
1005 | escape sequence. |
1006 | |
1007 | Argument: |
1008 | ptrptr points to the pattern position pointer |
1009 | negptr points to a boolean that is set TRUE for negation else FALSE |
1010 | dptr points to an int that is set to the detailed property value |
1011 | errorcodeptr points to the error code variable |
1012 | |
1013 | Returns: type value from ucp_type_table, or -1 for an invalid type |
1014 | */ |
1015 | |
1016 | static int |
1017 | get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr) |
1018 | { |
1019 | int c, i, bot, top; |
1020 | const uschar *ptr = *ptrptr; |
1021 | char name[32]; |
1022 | |
1023 | c = *(++ptr); |
1024 | if (c == 0) goto ERROR_RETURN; |
1025 | |
1026 | *negptr = FALSE; |
1027 | |
1028 | /* \P or \p can be followed by a name in {}, optionally preceded by ^ for |
1029 | negation. */ |
1030 | |
1031 | if (c == CHAR_LEFT_CURLY_BRACKET) |
1032 | { |
1033 | if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT) |
1034 | { |
1035 | *negptr = TRUE; |
1036 | ptr++; |
1037 | } |
1038 | for (i = 0; i < (int)sizeof(name) - 1; i++) |
1039 | { |
1040 | c = *(++ptr); |
1041 | if (c == 0) goto ERROR_RETURN; |
1042 | if (c == CHAR_RIGHT_CURLY_BRACKET) break; |
1043 | name[i] = c; |
1044 | } |
1045 | if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN; |
1046 | name[i] = 0; |
1047 | } |
1048 | |
1049 | /* Otherwise there is just one following character */ |
1050 | |
1051 | else |
1052 | { |
1053 | name[0] = c; |
1054 | name[1] = 0; |
1055 | } |
1056 | |
1057 | *ptrptr = ptr; |
1058 | |
1059 | /* Search for a recognized property name using binary chop */ |
1060 | |
1061 | bot = 0; |
1062 | top = _pcre_utt_size; |
1063 | |
1064 | while (bot < top) |
1065 | { |
1066 | i = (bot + top) >> 1; |
1067 | c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset); |
1068 | if (c == 0) |
1069 | { |
1070 | *dptr = _pcre_utt[i].value; |
1071 | return _pcre_utt[i].type; |
1072 | } |
1073 | if (c > 0) bot = i + 1; else top = i; |
1074 | } |
1075 | |
1076 | *errorcodeptr = ERR47; |
1077 | *ptrptr = ptr; |
1078 | return -1; |
1079 | |
1080 | ERROR_RETURN: |
1081 | *errorcodeptr = ERR46; |
1082 | *ptrptr = ptr; |
1083 | return -1; |
1084 | } |
1085 | #endif |
1086 | |
1087 | |
1088 | |
1089 | |
1090 | /************************************************* |
1091 | * Read repeat counts * |
1092 | *************************************************/ |
1093 | |
1094 | /* Read an item of the form {n,m} and return the values. This is called only |
1095 | after is_counted_repeat() has confirmed that a repeat-count quantifier exists, |
1096 | so the syntax is guaranteed to be correct, but we need to check the values. |
1097 | |
1098 | Arguments: |
1099 | p pointer to first char after '{' |
1100 | minp pointer to int for min |
1101 | maxp pointer to int for max |
1102 | returned as -1 if no max |
1103 | errorcodeptr points to error code variable |
1104 | |
1105 | Returns: pointer to '}' on success; |
1106 | current ptr on error, with errorcodeptr set non-zero |
1107 | */ |
1108 | |
1109 | static const uschar * |
1110 | read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr) |
1111 | { |
1112 | int min = 0; |
1113 | int max = -1; |
1114 | |
1115 | /* Read the minimum value and do a paranoid check: a negative value indicates |
1116 | an integer overflow. */ |
1117 | |
1118 | while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0; |
1119 | if (min < 0 || min > 65535) |
1120 | { |
1121 | *errorcodeptr = ERR5; |
1122 | return p; |
1123 | } |
1124 | |
1125 | /* Read the maximum value if there is one, and again do a paranoid on its size. |
1126 | Also, max must not be less than min. */ |
1127 | |
1128 | if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else |
1129 | { |
1130 | if (*(++p) != CHAR_RIGHT_CURLY_BRACKET) |
1131 | { |
1132 | max = 0; |
1133 | while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0; |
1134 | if (max < 0 || max > 65535) |
1135 | { |
1136 | *errorcodeptr = ERR5; |
1137 | return p; |
1138 | } |
1139 | if (max < min) |
1140 | { |
1141 | *errorcodeptr = ERR4; |
1142 | return p; |
1143 | } |
1144 | } |
1145 | } |
1146 | |
1147 | /* Fill in the required variables, and pass back the pointer to the terminating |
1148 | '}'. */ |
1149 | |
1150 | *minp = min; |
1151 | *maxp = max; |
1152 | return p; |
1153 | } |
1154 | |
1155 | |
1156 | |
1157 | /************************************************* |
1158 | * Subroutine for finding forward reference * |
1159 | *************************************************/ |
1160 | |
1161 | /* This recursive function is called only from find_parens() below. The |
1162 | top-level call starts at the beginning of the pattern. All other calls must |
1163 | start at a parenthesis. It scans along a pattern's text looking for capturing |
1164 | subpatterns, and counting them. If it finds a named pattern that matches the |
1165 | name it is given, it returns its number. Alternatively, if the name is NULL, it |
1166 | returns when it reaches a given numbered subpattern. Recursion is used to keep |
1167 | track of subpatterns that reset the capturing group numbers - the (?| feature. |
1168 | |
1169 | This function was originally called only from the second pass, in which we know |
1170 | that if (?< or (?' or (?P< is encountered, the name will be correctly |
1171 | terminated because that is checked in the first pass. There is now one call to |
1172 | this function in the first pass, to check for a recursive back reference by |
1173 | name (so that we can make the whole group atomic). In this case, we need check |
1174 | only up to the current position in the pattern, and that is still OK because |
1175 | and previous occurrences will have been checked. To make this work, the test |
1176 | for "end of pattern" is a check against cd->end_pattern in the main loop, |
1177 | instead of looking for a binary zero. This means that the special first-pass |
1178 | call can adjust cd->end_pattern temporarily. (Checks for binary zero while |
1179 | processing items within the loop are OK, because afterwards the main loop will |
1180 | terminate.) |
1181 | |
1182 | Arguments: |
1183 | ptrptr address of the current character pointer (updated) |
1184 | cd compile background data |
1185 | name name to seek, or NULL if seeking a numbered subpattern |
1186 | lorn name length, or subpattern number if name is NULL |
1187 | xmode TRUE if we are in /x mode |
1188 | utf8 TRUE if we are in UTF-8 mode |
1189 | count pointer to the current capturing subpattern number (updated) |
1190 | |
1191 | Returns: the number of the named subpattern, or -1 if not found |
1192 | */ |
1193 | |
1194 | static int |
1195 | find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn, |
1196 | BOOL xmode, BOOL utf8, int *count) |
1197 | { |
1198 | uschar *ptr = *ptrptr; |
1199 | int start_count = *count; |
1200 | int hwm_count = start_count; |
1201 | BOOL dup_parens = FALSE; |
1202 | |
1203 | /* If the first character is a parenthesis, check on the type of group we are |
1204 | dealing with. The very first call may not start with a parenthesis. */ |
1205 | |
1206 | if (ptr[0] == CHAR_LEFT_PARENTHESIS) |
1207 | { |
1208 | /* Handle specials such as (*SKIP) or (*UTF8) etc. */ |
1209 | |
1210 | if (ptr[1] == CHAR_ASTERISK) ptr += 2; |
1211 | |
1212 | /* Handle a normal, unnamed capturing parenthesis. */ |
1213 | |
1214 | else if (ptr[1] != CHAR_QUESTION_MARK) |
1215 | { |
1216 | *count += 1; |
1217 | if (name == NULL && *count == lorn) return *count; |
1218 | ptr++; |
1219 | } |
1220 | |
1221 | /* All cases now have (? at the start. Remember when we are in a group |
1222 | where the parenthesis numbers are duplicated. */ |
1223 | |
1224 | else if (ptr[2] == CHAR_VERTICAL_LINE) |
1225 | { |
1226 | ptr += 3; |
1227 | dup_parens = TRUE; |
1228 | } |
1229 | |
1230 | /* Handle comments; all characters are allowed until a ket is reached. */ |
1231 | |
1232 | else if (ptr[2] == CHAR_NUMBER_SIGN) |
1233 | { |
1234 | for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break; |
1235 | goto FAIL_EXIT; |
1236 | } |
1237 | |
1238 | /* Handle a condition. If it is an assertion, just carry on so that it |
1239 | is processed as normal. If not, skip to the closing parenthesis of the |
1240 | condition (there can't be any nested parens). */ |
1241 | |
1242 | else if (ptr[2] == CHAR_LEFT_PARENTHESIS) |
1243 | { |
1244 | ptr += 2; |
1245 | if (ptr[1] != CHAR_QUESTION_MARK) |
1246 | { |
1247 | while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; |
1248 | if (*ptr != 0) ptr++; |
1249 | } |
1250 | } |
1251 | |
1252 | /* Start with (? but not a condition. */ |
1253 | |
1254 | else |
1255 | { |
1256 | ptr += 2; |
1257 | if (*ptr == CHAR_P) ptr++; /* Allow optional P */ |
1258 | |
1259 | /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */ |
1260 | |
1261 | if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK && |
1262 | ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE) |
1263 | { |
1264 | int term; |
1265 | const uschar *thisname; |
1266 | *count += 1; |
1267 | if (name == NULL && *count == lorn) return *count; |
1268 | term = *ptr++; |
1269 | if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN; |
1270 | thisname = ptr; |
1271 | while (*ptr != term) ptr++; |
1272 | if (name != NULL && lorn == ptr - thisname && |
1273 | strncmp((const char *)name, (const char *)thisname, lorn) == 0) |
1274 | return *count; |
1275 | term++; |
1276 | } |
1277 | } |
1278 | } |
1279 | |
1280 | /* Past any initial parenthesis handling, scan for parentheses or vertical |
1281 | bars. Stop if we get to cd->end_pattern. Note that this is important for the |
1282 | first-pass call when this value is temporarily adjusted to stop at the current |
1283 | position. So DO NOT change this to a test for binary zero. */ |
1284 | |
1285 | for (; ptr < cd->end_pattern; ptr++) |
1286 | { |
1287 | /* Skip over backslashed characters and also entire \Q...\E */ |
1288 | |
1289 | if (*ptr == CHAR_BACKSLASH) |
1290 | { |
1291 | if (*(++ptr) == 0) goto FAIL_EXIT; |
1292 | if (*ptr == CHAR_Q) for (;;) |
1293 | { |
1294 | while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {}; |
1295 | if (*ptr == 0) goto FAIL_EXIT; |
1296 | if (*(++ptr) == CHAR_E) break; |
1297 | } |
1298 | continue; |
1299 | } |
1300 | |
1301 | /* Skip over character classes; this logic must be similar to the way they |
1302 | are handled for real. If the first character is '^', skip it. Also, if the |
1303 | first few characters (either before or after ^) are \Q\E or \E we skip them |
1304 | too. This makes for compatibility with Perl. Note the use of STR macros to |
1305 | encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */ |
1306 | |
1307 | if (*ptr == CHAR_LEFT_SQUARE_BRACKET) |
1308 | { |
1309 | BOOL negate_class = FALSE; |
1310 | for (;;) |
1311 | { |
1312 | if (ptr[1] == CHAR_BACKSLASH) |
1313 | { |
1314 | if (ptr[2] == CHAR_E) |
1315 | ptr+= 2; |
1316 | else if (strncmp((const char *)ptr+2, |
1317 | STR_Q STR_BACKSLASH STR_E, 3) == 0) |
1318 | ptr += 4; |
1319 | else |
1320 | break; |
1321 | } |
1322 | else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT) |
1323 | { |
1324 | negate_class = TRUE; |
1325 | ptr++; |
1326 | } |
1327 | else break; |
1328 | } |
1329 | |
1330 | /* If the next character is ']', it is a data character that must be |
1331 | skipped, except in JavaScript compatibility mode. */ |
1332 | |
1333 | if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET && |
1334 | (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0) |
1335 | ptr++; |
1336 | |
1337 | while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET) |
1338 | { |
1339 | if (*ptr == 0) return -1; |
1340 | if (*ptr == CHAR_BACKSLASH) |
1341 | { |
1342 | if (*(++ptr) == 0) goto FAIL_EXIT; |
1343 | if (*ptr == CHAR_Q) for (;;) |
1344 | { |
1345 | while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {}; |
1346 | if (*ptr == 0) goto FAIL_EXIT; |
1347 | if (*(++ptr) == CHAR_E) break; |
1348 | } |
1349 | continue; |
1350 | } |
1351 | } |
1352 | continue; |
1353 | } |
1354 | |
1355 | /* Skip comments in /x mode */ |
1356 | |
1357 | if (xmode && *ptr == CHAR_NUMBER_SIGN) |
1358 | { |
1359 | ptr++; |
1360 | while (*ptr != 0) |
1361 | { |
1362 | if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; } |
1363 | ptr++; |
1364 | #ifdef SUPPORT_UTF8 |
1365 | if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++; |
1366 | #endif |
1367 | } |
1368 | if (*ptr == 0) goto FAIL_EXIT; |
1369 | continue; |
1370 | } |
1371 | |
1372 | /* Check for the special metacharacters */ |
1373 | |
1374 | if (*ptr == CHAR_LEFT_PARENTHESIS) |
1375 | { |
1376 | int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, count); |
1377 | if (rc > 0) return rc; |
1378 | if (*ptr == 0) goto FAIL_EXIT; |
1379 | } |
1380 | |
1381 | else if (*ptr == CHAR_RIGHT_PARENTHESIS) |
1382 | { |
1383 | if (dup_parens && *count < hwm_count) *count = hwm_count; |
1384 | goto FAIL_EXIT; |
1385 | } |
1386 | |
1387 | else if (*ptr == CHAR_VERTICAL_LINE && dup_parens) |
1388 | { |
1389 | if (*count > hwm_count) hwm_count = *count; |
1390 | *count = start_count; |
1391 | } |
1392 | } |
1393 | |
1394 | FAIL_EXIT: |
1395 | *ptrptr = ptr; |
1396 | return -1; |
1397 | } |
1398 | |
1399 | |
1400 | |
1401 | |
1402 | /************************************************* |
1403 | * Find forward referenced subpattern * |
1404 | *************************************************/ |
1405 | |
1406 | /* This function scans along a pattern's text looking for capturing |
1407 | subpatterns, and counting them. If it finds a named pattern that matches the |
1408 | name it is given, it returns its number. Alternatively, if the name is NULL, it |
1409 | returns when it reaches a given numbered subpattern. This is used for forward |
1410 | references to subpatterns. We used to be able to start this scan from the |
1411 | current compiling point, using the current count value from cd->bracount, and |
1412 | do it all in a single loop, but the addition of the possibility of duplicate |
1413 | subpattern numbers means that we have to scan from the very start, in order to |
1414 | take account of such duplicates, and to use a recursive function to keep track |
1415 | of the different types of group. |
1416 | |
1417 | Arguments: |
1418 | cd compile background data |
1419 | name name to seek, or NULL if seeking a numbered subpattern |
1420 | lorn name length, or subpattern number if name is NULL |
1421 | xmode TRUE if we are in /x mode |
1422 | utf8 TRUE if we are in UTF-8 mode |
1423 | |
1424 | Returns: the number of the found subpattern, or -1 if not found |
1425 | */ |
1426 | |
1427 | static int |
1428 | find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode, |
1429 | BOOL utf8) |
1430 | { |
1431 | uschar *ptr = (uschar *)cd->start_pattern; |
1432 | int count = 0; |
1433 | int rc; |
1434 | |
1435 | /* If the pattern does not start with an opening parenthesis, the first call |
1436 | to find_parens_sub() will scan right to the end (if necessary). However, if it |
1437 | does start with a parenthesis, find_parens_sub() will return when it hits the |
1438 | matching closing parens. That is why we have to have a loop. */ |
1439 | |
1440 | for (;;) |
1441 | { |
1442 | rc = find_parens_sub(&ptr, cd, name, lorn, xmode, utf8, &count); |
1443 | if (rc > 0 || *ptr++ == 0) break; |
1444 | } |
1445 | |
1446 | return rc; |
1447 | } |
1448 | |
1449 | |
1450 | |
1451 | |
1452 | /************************************************* |
1453 | * Find first significant op code * |
1454 | *************************************************/ |
1455 | |
1456 | /* This is called by several functions that scan a compiled expression looking |
1457 | for a fixed first character, or an anchoring op code etc. It skips over things |
1458 | that do not influence this. For some calls, it makes sense to skip negative |
1459 | forward and all backward assertions, and also the \b assertion; for others it |
1460 | does not. |
1461 | |
1462 | Arguments: |
1463 | code pointer to the start of the group |
1464 | skipassert TRUE if certain assertions are to be skipped |
1465 | |
1466 | Returns: pointer to the first significant opcode |
1467 | */ |
1468 | |
1469 | static const uschar* |
1470 | first_significant_code(const uschar *code, BOOL skipassert) |
1471 | { |
1472 | for (;;) |
1473 | { |
1474 | switch ((int)*code) |
1475 | { |
1476 | case OP_ASSERT_NOT: |
1477 | case OP_ASSERTBACK: |
1478 | case OP_ASSERTBACK_NOT: |
1479 | if (!skipassert) return code; |
1480 | do code += GET(code, 1); while (*code == OP_ALT); |
1481 | code += _pcre_OP_lengths[*code]; |
1482 | break; |
1483 | |
1484 | case OP_WORD_BOUNDARY: |
1485 | case OP_NOT_WORD_BOUNDARY: |
1486 | if (!skipassert) return code; |
1487 | /* Fall through */ |
1488 | |
1489 | case OP_CALLOUT: |
1490 | case OP_CREF: |
1491 | case OP_NCREF: |
1492 | case OP_RREF: |
1493 | case OP_NRREF: |
1494 | case OP_DEF: |
1495 | code += _pcre_OP_lengths[*code]; |
1496 | break; |
1497 | |
1498 | default: |
1499 | return code; |
1500 | } |
1501 | } |
1502 | /* Control never reaches here */ |
1503 | } |
1504 | |
1505 | |
1506 | |
1507 | |
1508 | /************************************************* |
1509 | * Find the fixed length of a branch * |
1510 | *************************************************/ |
1511 | |
1512 | /* Scan a branch and compute the fixed length of subject that will match it, |
1513 | if the length is fixed. This is needed for dealing with backward assertions. |
1514 | In UTF8 mode, the result is in characters rather than bytes. The branch is |
1515 | temporarily terminated with OP_END when this function is called. |
1516 | |
1517 | This function is called when a backward assertion is encountered, so that if it |
1518 | fails, the error message can point to the correct place in the pattern. |
1519 | However, we cannot do this when the assertion contains subroutine calls, |
1520 | because they can be forward references. We solve this by remembering this case |
1521 | and doing the check at the end; a flag specifies which mode we are running in. |
1522 | |
1523 | Arguments: |
1524 | code points to the start of the pattern (the bracket) |
1525 | utf8 TRUE in UTF-8 mode |
1526 | atend TRUE if called when the pattern is complete |
1527 | cd the "compile data" structure |
1528 | |
1529 | Returns: the fixed length, |
1530 | or -1 if there is no fixed length, |
1531 | or -2 if \C was encountered |
1532 | or -3 if an OP_RECURSE item was encountered and atend is FALSE |
1533 | or -4 if an unknown opcode was encountered (internal error) |
1534 | */ |
1535 | |
1536 | static int |
1537 | find_fixedlength(uschar *code, BOOL utf8, BOOL atend, compile_data *cd) |
1538 | { |
1539 | int length = -1; |
1540 | |
1541 | register int branchlength = 0; |
1542 | register uschar *cc = code + 1 + LINK_SIZE; |
1543 | |
1544 | /* Scan along the opcodes for this branch. If we get to the end of the |
1545 | branch, check the length against that of the other branches. */ |
1546 | |
1547 | for (;;) |
1548 | { |
1549 | int d; |
1550 | uschar *ce, *cs; |
1551 | register int op = *cc; |
1552 | switch (op) |
1553 | { |
1554 | /* We only need to continue for OP_CBRA (normal capturing bracket) and |
1555 | OP_BRA (normal non-capturing bracket) because the other variants of these |
1556 | opcodes are all concerned with unlimited repeated groups, which of course |
1557 | are not of fixed length. */ |
1558 | |
1559 | case OP_CBRA: |
1560 | case OP_BRA: |
1561 | case OP_ONCE: |
1562 | case OP_ONCE_NC: |
1563 | case OP_COND: |
1564 | d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), utf8, atend, cd); |
1565 | if (d < 0) return d; |
1566 | branchlength += d; |
1567 | do cc += GET(cc, 1); while (*cc == OP_ALT); |
1568 | cc += 1 + LINK_SIZE; |
1569 | break; |
1570 | |
1571 | /* Reached end of a branch; if it's a ket it is the end of a nested call. |
1572 | If it's ALT it is an alternation in a nested call. An ACCEPT is effectively |
1573 | an ALT. If it is END it's the end of the outer call. All can be handled by |
1574 | the same code. Note that we must not include the OP_KETRxxx opcodes here, |
1575 | because they all imply an unlimited repeat. */ |
1576 | |
1577 | case OP_ALT: |
1578 | case OP_KET: |
1579 | case OP_END: |
1580 | case OP_ACCEPT: |
1581 | case OP_ASSERT_ACCEPT: |
1582 | if (length < 0) length = branchlength; |
1583 | else if (length != branchlength) return -1; |
1584 | if (*cc != OP_ALT) return length; |
1585 | cc += 1 + LINK_SIZE; |
1586 | branchlength = 0; |
1587 | break; |
1588 | |
1589 | /* A true recursion implies not fixed length, but a subroutine call may |
1590 | be OK. If the subroutine is a forward reference, we can't deal with |
1591 | it until the end of the pattern, so return -3. */ |
1592 | |
1593 | case OP_RECURSE: |
1594 | if (!atend) return -3; |
1595 | cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */ |
1596 | do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */ |
1597 | if (cc > cs && cc < ce) return -1; /* Recursion */ |
1598 | d = find_fixedlength(cs + 2, utf8, atend, cd); |
1599 | if (d < 0) return d; |
1600 | branchlength += d; |
1601 | cc += 1 + LINK_SIZE; |
1602 | break; |
1603 | |
1604 | /* Skip over assertive subpatterns */ |
1605 | |
1606 | case OP_ASSERT: |
1607 | case OP_ASSERT_NOT: |
1608 | case OP_ASSERTBACK: |
1609 | case OP_ASSERTBACK_NOT: |
1610 | do cc += GET(cc, 1); while (*cc == OP_ALT); |
1611 | /* Fall through */ |
1612 | |
1613 | /* Skip over things that don't match chars */ |
1614 | |
1615 | case OP_MARK: |
1616 | case OP_PRUNE_ARG: |
1617 | case OP_SKIP_ARG: |
1618 | case OP_THEN_ARG: |
1619 | cc += cc[1] + _pcre_OP_lengths[*cc]; |
1620 | break; |
1621 | |
1622 | case OP_CALLOUT: |
1623 | case OP_CIRC: |
1624 | case OP_CIRCM: |
1625 | case OP_CLOSE: |
1626 | case OP_COMMIT: |
1627 | case OP_CREF: |
1628 | case OP_DEF: |
1629 | case OP_DOLL: |
1630 | case OP_DOLLM: |
1631 | case OP_EOD: |
1632 | case OP_EODN: |
1633 | case OP_FAIL: |
1634 | case OP_NCREF: |
1635 | case OP_NRREF: |
1636 | case OP_NOT_WORD_BOUNDARY: |
1637 | case OP_PRUNE: |
1638 | case OP_REVERSE: |
1639 | case OP_RREF: |
1640 | case OP_SET_SOM: |
1641 | case OP_SKIP: |
1642 | case OP_SOD: |
1643 | case OP_SOM: |
1644 | case OP_THEN: |
1645 | case OP_WORD_BOUNDARY: |
1646 | cc += _pcre_OP_lengths[*cc]; |
1647 | break; |
1648 | |
1649 | /* Handle literal characters */ |
1650 | |
1651 | case OP_CHAR: |
1652 | case OP_CHARI: |
1653 | case OP_NOT: |
1654 | case OP_NOTI: |
1655 | branchlength++; |
1656 | cc += 2; |
1657 | #ifdef SUPPORT_UTF8 |
1658 | if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f]; |
1659 | #endif |
1660 | break; |
1661 | |
1662 | /* Handle exact repetitions. The count is already in characters, but we |
1663 | need to skip over a multibyte character in UTF8 mode. */ |
1664 | |
1665 | case OP_EXACT: |
1666 | case OP_EXACTI: |
1667 | case OP_NOTEXACT: |
1668 | case OP_NOTEXACTI: |
1669 | branchlength += GET2(cc,1); |
1670 | cc += 4; |
1671 | #ifdef SUPPORT_UTF8 |
1672 | if (utf8 && cc[-1] >= 0xc0) cc += _pcre_utf8_table4[cc[-1] & 0x3f]; |
1673 | #endif |
1674 | break; |
1675 | |
1676 | case OP_TYPEEXACT: |
1677 | branchlength += GET2(cc,1); |
1678 | if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2; |
1679 | cc += 4; |
1680 | break; |
1681 | |
1682 | /* Handle single-char matchers */ |
1683 | |
1684 | case OP_PROP: |
1685 | case OP_NOTPROP: |
1686 | cc += 2; |
1687 | /* Fall through */ |
1688 | |
1689 | case OP_HSPACE: |
1690 | case OP_VSPACE: |
1691 | case OP_NOT_HSPACE: |
1692 | case OP_NOT_VSPACE: |
1693 | case OP_NOT_DIGIT: |
1694 | case OP_DIGIT: |
1695 | case OP_NOT_WHITESPACE: |
1696 | case OP_WHITESPACE: |
1697 | case OP_NOT_WORDCHAR: |
1698 | case OP_WORDCHAR: |
1699 | case OP_ANY: |
1700 | case OP_ALLANY: |
1701 | branchlength++; |
1702 | cc++; |
1703 | break; |
1704 | |
1705 | /* The single-byte matcher isn't allowed */ |
1706 | |
1707 | case OP_ANYBYTE: |
1708 | return -2; |
1709 | |
1710 | /* Check a class for variable quantification */ |
1711 | |
1712 | #ifdef SUPPORT_UTF8 |
1713 | case OP_XCLASS: |
1714 | cc += GET(cc, 1) - 33; |
1715 | /* Fall through */ |
1716 | #endif |
1717 | |
1718 | case OP_CLASS: |
1719 | case OP_NCLASS: |
1720 | cc += 33; |
1721 | |
1722 | switch (*cc) |
1723 | { |
1724 | case OP_CRPLUS: |
1725 | case OP_CRMINPLUS: |
1726 | case OP_CRSTAR: |
1727 | case OP_CRMINSTAR: |
1728 | case OP_CRQUERY: |
1729 | case OP_CRMINQUERY: |
1730 | return -1; |
1731 | |
1732 | case OP_CRRANGE: |
1733 | case OP_CRMINRANGE: |
1734 | if (GET2(cc,1) != GET2(cc,3)) return -1; |
1735 | branchlength += GET2(cc,1); |
1736 | cc += 5; |
1737 | break; |
1738 | |
1739 | default: |
1740 | branchlength++; |
1741 | } |
1742 | break; |
1743 | |
1744 | /* Anything else is variable length */ |
1745 | |
1746 | case OP_ANYNL: |
1747 | case OP_BRAMINZERO: |
1748 | case OP_BRAPOS: |
1749 | case OP_BRAPOSZERO: |
1750 | case OP_BRAZERO: |
1751 | case OP_CBRAPOS: |
1752 | case OP_EXTUNI: |
1753 | case OP_KETRMAX: |
1754 | case OP_KETRMIN: |
1755 | case OP_KETRPOS: |
1756 | case OP_MINPLUS: |
1757 | case OP_MINPLUSI: |
1758 | case OP_MINQUERY: |
1759 | case OP_MINQUERYI: |
1760 | case OP_MINSTAR: |
1761 | case OP_MINSTARI: |
1762 | case OP_MINUPTO: |
1763 | case OP_MINUPTOI: |
1764 | case OP_NOTMINPLUS: |
1765 | case OP_NOTMINPLUSI: |
1766 | case OP_NOTMINQUERY: |
1767 | case OP_NOTMINQUERYI: |
1768 | case OP_NOTMINSTAR: |
1769 | case OP_NOTMINSTARI: |
1770 | case OP_NOTMINUPTO: |
1771 | case OP_NOTMINUPTOI: |
1772 | case OP_NOTPLUS: |
1773 | case OP_NOTPLUSI: |
1774 | case OP_NOTPOSPLUS: |
1775 | case OP_NOTPOSPLUSI: |
1776 | case OP_NOTPOSQUERY: |
1777 | case OP_NOTPOSQUERYI: |
1778 | case OP_NOTPOSSTAR: |
1779 | case OP_NOTPOSSTARI: |
1780 | case OP_NOTPOSUPTO: |
1781 | case OP_NOTPOSUPTOI: |
1782 | case OP_NOTQUERY: |
1783 | case OP_NOTQUERYI: |
1784 | case OP_NOTSTAR: |
1785 | case OP_NOTSTARI: |
1786 | case OP_NOTUPTO: |
1787 | case OP_NOTUPTOI: |
1788 | case OP_PLUS: |
1789 | case OP_PLUSI: |
1790 | case OP_POSPLUS: |
1791 | case OP_POSPLUSI: |
1792 | case OP_POSQUERY: |
1793 | case OP_POSQUERYI: |
1794 | case OP_POSSTAR: |
1795 | case OP_POSSTARI: |
1796 | case OP_POSUPTO: |
1797 | case OP_POSUPTOI: |
1798 | case OP_QUERY: |
1799 | case OP_QUERYI: |
1800 | case OP_REF: |
1801 | case OP_REFI: |
1802 | case OP_SBRA: |
1803 | case OP_SBRAPOS: |
1804 | case OP_SCBRA: |
1805 | case OP_SCBRAPOS: |
1806 | case OP_SCOND: |
1807 | case OP_SKIPZERO: |
1808 | case OP_STAR: |
1809 | case OP_STARI: |
1810 | case OP_TYPEMINPLUS: |
1811 | case OP_TYPEMINQUERY: |
1812 | case OP_TYPEMINSTAR: |
1813 | case OP_TYPEMINUPTO: |
1814 | case OP_TYPEPLUS: |
1815 | case OP_TYPEPOSPLUS: |
1816 | case OP_TYPEPOSQUERY: |
1817 | case OP_TYPEPOSSTAR: |
1818 | case OP_TYPEPOSUPTO: |
1819 | case OP_TYPEQUERY: |
1820 | case OP_TYPESTAR: |
1821 | case OP_TYPEUPTO: |
1822 | case OP_UPTO: |
1823 | case OP_UPTOI: |
1824 | return -1; |
1825 | |
1826 | /* Catch unrecognized opcodes so that when new ones are added they |
1827 | are not forgotten, as has happened in the past. */ |
1828 | |
1829 | default: |
1830 | return -4; |
1831 | } |
1832 | } |
1833 | /* Control never gets here */ |
1834 | } |
1835 | |
1836 | |
1837 | |
1838 | |
1839 | /************************************************* |
1840 | * Scan compiled regex for specific bracket * |
1841 | *************************************************/ |
1842 | |
1843 | /* This little function scans through a compiled pattern until it finds a |
1844 | capturing bracket with the given number, or, if the number is negative, an |
1845 | instance of OP_REVERSE for a lookbehind. The function is global in the C sense |
1846 | so that it can be called from pcre_study() when finding the minimum matching |
1847 | length. |
1848 | |
1849 | Arguments: |
1850 | code points to start of expression |
1851 | utf8 TRUE in UTF-8 mode |
1852 | number the required bracket number or negative to find a lookbehind |
1853 | |
1854 | Returns: pointer to the opcode for the bracket, or NULL if not found |
1855 | */ |
1856 | |
1857 | const uschar * |
1858 | _pcre_find_bracket(const uschar *code, BOOL utf8, int number) |
1859 | { |
1860 | for (;;) |
1861 | { |
1862 | register int c = *code; |
1863 | |
1864 | if (c == OP_END) return NULL; |
1865 | |
1866 | /* XCLASS is used for classes that cannot be represented just by a bit |
1867 | map. This includes negated single high-valued characters. The length in |
1868 | the table is zero; the actual length is stored in the compiled code. */ |
1869 | |
1870 | if (c == OP_XCLASS) code += GET(code, 1); |
1871 | |
1872 | /* Handle recursion */ |
1873 | |
1874 | else if (c == OP_REVERSE) |
1875 | { |
1876 | if (number < 0) return (uschar *)code; |
1877 | code += _pcre_OP_lengths[c]; |
1878 | } |
1879 | |
1880 | /* Handle capturing bracket */ |
1881 | |
1882 | else if (c == OP_CBRA || c == OP_SCBRA || |
1883 | c == OP_CBRAPOS || c == OP_SCBRAPOS) |
1884 | { |
1885 | int n = GET2(code, 1+LINK_SIZE); |
1886 | if (n == number) return (uschar *)code; |
1887 | code += _pcre_OP_lengths[c]; |
1888 | } |
1889 | |
1890 | /* Otherwise, we can get the item's length from the table, except that for |
1891 | repeated character types, we have to test for \p and \P, which have an extra |
1892 | two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we |
1893 | must add in its length. */ |
1894 | |
1895 | else |
1896 | { |
1897 | switch(c) |
1898 | { |
1899 | case OP_TYPESTAR: |
1900 | case OP_TYPEMINSTAR: |
1901 | case OP_TYPEPLUS: |
1902 | case OP_TYPEMINPLUS: |
1903 | case OP_TYPEQUERY: |
1904 | case OP_TYPEMINQUERY: |
1905 | case OP_TYPEPOSSTAR: |
1906 | case OP_TYPEPOSPLUS: |
1907 | case OP_TYPEPOSQUERY: |
1908 | if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2; |
1909 | break; |
1910 | |
1911 | case OP_TYPEUPTO: |
1912 | case OP_TYPEMINUPTO: |
1913 | case OP_TYPEEXACT: |
1914 | case OP_TYPEPOSUPTO: |
1915 | if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2; |
1916 | break; |
1917 | |
1918 | case OP_MARK: |
1919 | case OP_PRUNE_ARG: |
1920 | case OP_SKIP_ARG: |
1921 | code += code[1]; |
1922 | break; |
1923 | |
1924 | case OP_THEN_ARG: |
1925 | code += code[1]; |
1926 | break; |
1927 | } |
1928 | |
1929 | /* Add in the fixed length from the table */ |
1930 | |
1931 | code += _pcre_OP_lengths[c]; |
1932 | |
1933 | /* In UTF-8 mode, opcodes that are followed by a character may be followed by |
1934 | a multi-byte character. The length in the table is a minimum, so we have to |
1935 | arrange to skip the extra bytes. */ |
1936 | |
1937 | #ifdef SUPPORT_UTF8 |
1938 | if (utf8) switch(c) |
1939 | { |
1940 | case OP_CHAR: |
1941 | case OP_CHARI: |
1942 | case OP_EXACT: |
1943 | case OP_EXACTI: |
1944 | case OP_UPTO: |
1945 | case OP_UPTOI: |
1946 | case OP_MINUPTO: |
1947 | case OP_MINUPTOI: |
1948 | case OP_POSUPTO: |
1949 | case OP_POSUPTOI: |
1950 | case OP_STAR: |
1951 | case OP_STARI: |
1952 | case OP_MINSTAR: |
1953 | case OP_MINSTARI: |
1954 | case OP_POSSTAR: |
1955 | case OP_POSSTARI: |
1956 | case OP_PLUS: |
1957 | case OP_PLUSI: |
1958 | case OP_MINPLUS: |
1959 | case OP_MINPLUSI: |
1960 | case OP_POSPLUS: |
1961 | case OP_POSPLUSI: |
1962 | case OP_QUERY: |
1963 | case OP_QUERYI: |
1964 | case OP_MINQUERY: |
1965 | case OP_MINQUERYI: |
1966 | case OP_POSQUERY: |
1967 | case OP_POSQUERYI: |
1968 | if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f]; |
1969 | break; |
1970 | } |
1971 | #else |
1972 | (void)(utf8); /* Keep compiler happy by referencing function argument */ |
1973 | #endif |
1974 | } |
1975 | } |
1976 | } |
1977 | |
1978 | |
1979 | |
1980 | /************************************************* |
1981 | * Scan compiled regex for recursion reference * |
1982 | *************************************************/ |
1983 | |
1984 | /* This little function scans through a compiled pattern until it finds an |
1985 | instance of OP_RECURSE. |
1986 | |
1987 | Arguments: |
1988 | code points to start of expression |
1989 | utf8 TRUE in UTF-8 mode |
1990 | |
1991 | Returns: pointer to the opcode for OP_RECURSE, or NULL if not found |
1992 | */ |
1993 | |
1994 | static const uschar * |
1995 | find_recurse(const uschar *code, BOOL utf8) |
1996 | { |
1997 | for (;;) |
1998 | { |
1999 | register int c = *code; |
2000 | if (c == OP_END) return NULL; |
2001 | if (c == OP_RECURSE) return code; |
2002 | |
2003 | /* XCLASS is used for classes that cannot be represented just by a bit |
2004 | map. This includes negated single high-valued characters. The length in |
2005 | the table is zero; the actual length is stored in the compiled code. */ |
2006 | |
2007 | if (c == OP_XCLASS) code += GET(code, 1); |
2008 | |
2009 | /* Otherwise, we can get the item's length from the table, except that for |
2010 | repeated character types, we have to test for \p and \P, which have an extra |
2011 | two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we |
2012 | must add in its length. */ |
2013 | |
2014 | else |
2015 | { |
2016 | switch(c) |
2017 | { |
2018 | case OP_TYPESTAR: |
2019 | case OP_TYPEMINSTAR: |
2020 | case OP_TYPEPLUS: |
2021 | case OP_TYPEMINPLUS: |
2022 | case OP_TYPEQUERY: |
2023 | case OP_TYPEMINQUERY: |
2024 | case OP_TYPEPOSSTAR: |
2025 | case OP_TYPEPOSPLUS: |
2026 | case OP_TYPEPOSQUERY: |
2027 | if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2; |
2028 | break; |
2029 | |
2030 | case OP_TYPEPOSUPTO: |
2031 | case OP_TYPEUPTO: |
2032 | case OP_TYPEMINUPTO: |
2033 | case OP_TYPEEXACT: |
2034 | if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2; |
2035 | break; |
2036 | |
2037 | case OP_MARK: |
2038 | case OP_PRUNE_ARG: |
2039 | case OP_SKIP_ARG: |
2040 | code += code[1]; |
2041 | break; |
2042 | |
2043 | case OP_THEN_ARG: |
2044 | code += code[1]; |
2045 | break; |
2046 | } |
2047 | |
2048 | /* Add in the fixed length from the table */ |
2049 | |
2050 | code += _pcre_OP_lengths[c]; |
2051 | |
2052 | /* In UTF-8 mode, opcodes that are followed by a character may be followed |
2053 | by a multi-byte character. The length in the table is a minimum, so we have |
2054 | to arrange to skip the extra bytes. */ |
2055 | |
2056 | #ifdef SUPPORT_UTF8 |
2057 | if (utf8) switch(c) |
2058 | { |
2059 | case OP_CHAR: |
2060 | case OP_CHARI: |
2061 | case OP_EXACT: |
2062 | case OP_EXACTI: |
2063 | case OP_UPTO: |
2064 | case OP_UPTOI: |
2065 | case OP_MINUPTO: |
2066 | case OP_MINUPTOI: |
2067 | case OP_POSUPTO: |
2068 | case OP_POSUPTOI: |
2069 | case OP_STAR: |
2070 | case OP_STARI: |
2071 | case OP_MINSTAR: |
2072 | case OP_MINSTARI: |
2073 | case OP_POSSTAR: |
2074 | case OP_POSSTARI: |
2075 | case OP_PLUS: |
2076 | case OP_PLUSI: |
2077 | case OP_MINPLUS: |
2078 | case OP_MINPLUSI: |
2079 | case OP_POSPLUS: |
2080 | case OP_POSPLUSI: |
2081 | case OP_QUERY: |
2082 | case OP_QUERYI: |
2083 | case OP_MINQUERY: |
2084 | case OP_MINQUERYI: |
2085 | case OP_POSQUERY: |
2086 | case OP_POSQUERYI: |
2087 | if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f]; |
2088 | break; |
2089 | } |
2090 | #else |
2091 | (void)(utf8); /* Keep compiler happy by referencing function argument */ |
2092 | #endif |
2093 | } |
2094 | } |
2095 | } |
2096 | |
2097 | |
2098 | |
2099 | /************************************************* |
2100 | * Scan compiled branch for non-emptiness * |
2101 | *************************************************/ |
2102 | |
2103 | /* This function scans through a branch of a compiled pattern to see whether it |
2104 | can match the empty string or not. It is called from could_be_empty() |
2105 | below and from compile_branch() when checking for an unlimited repeat of a |
2106 | group that can match nothing. Note that first_significant_code() skips over |
2107 | backward and negative forward assertions when its final argument is TRUE. If we |
2108 | hit an unclosed bracket, we return "empty" - this means we've struck an inner |
2109 | bracket whose current branch will already have been scanned. |
2110 | |
2111 | Arguments: |
2112 | code points to start of search |
2113 | endcode points to where to stop |
2114 | utf8 TRUE if in UTF8 mode |
2115 | cd contains pointers to tables etc. |
2116 | |
2117 | Returns: TRUE if what is matched could be empty |
2118 | */ |
2119 | |
2120 | static BOOL |
2121 | could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8, |
2122 | compile_data *cd) |
2123 | { |
2124 | register int c; |
2125 | for (code = first_significant_code(code + _pcre_OP_lengths[*code], TRUE); |
2126 | code < endcode; |
2127 | code = first_significant_code(code + _pcre_OP_lengths[c], TRUE)) |
2128 | { |
2129 | const uschar *ccode; |
2130 | |
2131 | c = *code; |
2132 | |
2133 | /* Skip over forward assertions; the other assertions are skipped by |
2134 | first_significant_code() with a TRUE final argument. */ |
2135 | |
2136 | if (c == OP_ASSERT) |
2137 | { |
2138 | do code += GET(code, 1); while (*code == OP_ALT); |
2139 | c = *code; |
2140 | continue; |
2141 | } |
2142 | |
2143 | /* For a recursion/subroutine call, if its end has been reached, which |
2144 | implies a backward reference subroutine call, we can scan it. If it's a |
2145 | forward reference subroutine call, we can't. To detect forward reference |
2146 | we have to scan up the list that is kept in the workspace. This function is |
2147 | called only when doing the real compile, not during the pre-compile that |
2148 | measures the size of the compiled pattern. */ |
2149 | |
2150 | if (c == OP_RECURSE) |
2151 | { |
2152 | const uschar *scode; |
2153 | BOOL empty_branch; |
2154 | |
2155 | /* Test for forward reference */ |
2156 | |
2157 | for (scode = cd->start_workspace; scode < cd->hwm; scode += LINK_SIZE) |
2158 | if (GET(scode, 0) == code + 1 - cd->start_code) return TRUE; |
2159 | |
2160 | /* Not a forward reference, test for completed backward reference */ |
2161 | |
2162 | empty_branch = FALSE; |
2163 | scode = cd->start_code + GET(code, 1); |
2164 | if (GET(scode, 1) == 0) return TRUE; /* Unclosed */ |
2165 | |
2166 | /* Completed backwards reference */ |
2167 | |
2168 | do |
2169 | { |
2170 | if (could_be_empty_branch(scode, endcode, utf8, cd)) |
2171 | { |
2172 | empty_branch = TRUE; |
2173 | break; |
2174 | } |
2175 | scode += GET(scode, 1); |
2176 | } |
2177 | while (*scode == OP_ALT); |
2178 | |
2179 | if (!empty_branch) return FALSE; /* All branches are non-empty */ |
2180 | continue; |
2181 | } |
2182 | |
2183 | /* Groups with zero repeats can of course be empty; skip them. */ |
2184 | |
2185 | if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO || |
2186 | c == OP_BRAPOSZERO) |
2187 | { |
2188 | code += _pcre_OP_lengths[c]; |
2189 | do code += GET(code, 1); while (*code == OP_ALT); |
2190 | c = *code; |
2191 | continue; |
2192 | } |
2193 | |
2194 | /* A nested group that is already marked as "could be empty" can just be |
2195 | skipped. */ |
2196 | |
2197 | if (c == OP_SBRA || c == OP_SBRAPOS || |
2198 | c == OP_SCBRA || c == OP_SCBRAPOS) |
2199 | { |
2200 | do code += GET(code, 1); while (*code == OP_ALT); |
2201 | c = *code; |
2202 | continue; |
2203 | } |
2204 | |
2205 | /* For other groups, scan the branches. */ |
2206 | |
2207 | if (c == OP_BRA || c == OP_BRAPOS || |
2208 | c == OP_CBRA || c == OP_CBRAPOS || |
2209 | c == OP_ONCE || c == OP_ONCE_NC || |
2210 | c == OP_COND) |
2211 | { |
2212 | BOOL empty_branch; |
2213 | if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */ |
2214 | |
2215 | /* If a conditional group has only one branch, there is a second, implied, |
2216 | empty branch, so just skip over the conditional, because it could be empty. |
2217 | Otherwise, scan the individual branches of the group. */ |
2218 | |
2219 | if (c == OP_COND && code[GET(code, 1)] != OP_ALT) |
2220 | code += GET(code, 1); |
2221 | else |
2222 | { |
2223 | empty_branch = FALSE; |
2224 | do |
2225 | { |
2226 | if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd)) |
2227 | empty_branch = TRUE; |
2228 | code += GET(code, 1); |
2229 | } |
2230 | while (*code == OP_ALT); |
2231 | if (!empty_branch) return FALSE; /* All branches are non-empty */ |
2232 | } |
2233 | |
2234 | c = *code; |
2235 | continue; |
2236 | } |
2237 | |
2238 | /* Handle the other opcodes */ |
2239 | |
2240 | switch (c) |
2241 | { |
2242 | /* Check for quantifiers after a class. XCLASS is used for classes that |
2243 | cannot be represented just by a bit map. This includes negated single |
2244 | high-valued characters. The length in _pcre_OP_lengths[] is zero; the |
2245 | actual length is stored in the compiled code, so we must update "code" |
2246 | here. */ |
2247 | |
2248 | #ifdef SUPPORT_UTF8 |
2249 | case OP_XCLASS: |
2250 | ccode = code += GET(code, 1); |
2251 | goto CHECK_CLASS_REPEAT; |
2252 | #endif |
2253 | |
2254 | case OP_CLASS: |
2255 | case OP_NCLASS: |
2256 | ccode = code + 33; |
2257 | |
2258 | #ifdef SUPPORT_UTF8 |
2259 | CHECK_CLASS_REPEAT: |
2260 | #endif |
2261 | |
2262 | switch (*ccode) |
2263 | { |
2264 | case OP_CRSTAR: /* These could be empty; continue */ |
2265 | case OP_CRMINSTAR: |
2266 | case OP_CRQUERY: |
2267 | case OP_CRMINQUERY: |
2268 | break; |
2269 | |
2270 | default: /* Non-repeat => class must match */ |
2271 | case OP_CRPLUS: /* These repeats aren't empty */ |
2272 | case OP_CRMINPLUS: |
2273 | return FALSE; |
2274 | |
2275 | case OP_CRRANGE: |
2276 | case OP_CRMINRANGE: |
2277 | if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */ |
2278 | break; |
2279 | } |
2280 | break; |
2281 | |
2282 | /* Opcodes that must match a character */ |
2283 | |
2284 | case OP_PROP: |
2285 | case OP_NOTPROP: |
2286 | case OP_EXTUNI: |
2287 | case OP_NOT_DIGIT: |
2288 | case OP_DIGIT: |
2289 | case OP_NOT_WHITESPACE: |
2290 | case OP_WHITESPACE: |
2291 | case OP_NOT_WORDCHAR: |
2292 | case OP_WORDCHAR: |
2293 | case OP_ANY: |
2294 | case OP_ALLANY: |
2295 | case OP_ANYBYTE: |
2296 | case OP_CHAR: |
2297 | case OP_CHARI: |
2298 | case OP_NOT: |
2299 | case OP_NOTI: |
2300 | case OP_PLUS: |
2301 | case OP_MINPLUS: |
2302 | case OP_POSPLUS: |
2303 | case OP_EXACT: |
2304 | case OP_NOTPLUS: |
2305 | case OP_NOTMINPLUS: |
2306 | case OP_NOTPOSPLUS: |
2307 | case OP_NOTEXACT: |
2308 | case OP_TYPEPLUS: |
2309 | case OP_TYPEMINPLUS: |
2310 | case OP_TYPEPOSPLUS: |
2311 | case OP_TYPEEXACT: |
2312 | return FALSE; |
2313 | |
2314 | /* These are going to continue, as they may be empty, but we have to |
2315 | fudge the length for the \p and \P cases. */ |
2316 | |
2317 | case OP_TYPESTAR: |
2318 | case OP_TYPEMINSTAR: |
2319 | case OP_TYPEPOSSTAR: |
2320 | case OP_TYPEQUERY: |
2321 | case OP_TYPEMINQUERY: |
2322 | case OP_TYPEPOSQUERY: |
2323 | if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2; |
2324 | break; |
2325 | |
2326 | /* Same for these */ |
2327 | |
2328 | case OP_TYPEUPTO: |
2329 | case OP_TYPEMINUPTO: |
2330 | case OP_TYPEPOSUPTO: |
2331 | if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2; |
2332 | break; |
2333 | |
2334 | /* End of branch */ |
2335 | |
2336 | case OP_KET: |
2337 | case OP_KETRMAX: |
2338 | case OP_KETRMIN: |
2339 | case OP_KETRPOS: |
2340 | case OP_ALT: |
2341 | return TRUE; |
2342 | |
2343 | /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO, |
2344 | MINUPTO, and POSUPTO may be followed by a multibyte character */ |
2345 | |
2346 | #ifdef SUPPORT_UTF8 |
2347 | case OP_STAR: |
2348 | case OP_STARI: |
2349 | case OP_MINSTAR: |
2350 | case OP_MINSTARI: |
2351 | case OP_POSSTAR: |
2352 | case OP_POSSTARI: |
2353 | case OP_QUERY: |
2354 | case OP_QUERYI: |
2355 | case OP_MINQUERY: |
2356 | case OP_MINQUERYI: |
2357 | case OP_POSQUERY: |
2358 | case OP_POSQUERYI: |
2359 | if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f]; |
2360 | break; |
2361 | |
2362 | case OP_UPTO: |
2363 | case OP_UPTOI: |
2364 | case OP_MINUPTO: |
2365 | case OP_MINUPTOI: |
2366 | case OP_POSUPTO: |
2367 | case OP_POSUPTOI: |
2368 | if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f]; |
2369 | break; |
2370 | #endif |
2371 | |
2372 | /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument |
2373 | string. */ |
2374 | |
2375 | case OP_MARK: |
2376 | case OP_PRUNE_ARG: |
2377 | case OP_SKIP_ARG: |
2378 | code += code[1]; |
2379 | break; |
2380 | |
2381 | case OP_THEN_ARG: |
2382 | code += code[1]; |
2383 | break; |
2384 | |
2385 | /* None of the remaining opcodes are required to match a character. */ |
2386 | |
2387 | default: |
2388 | break; |
2389 | } |
2390 | } |
2391 | |
2392 | return TRUE; |
2393 | } |
2394 | |
2395 | |
2396 | |
2397 | /************************************************* |
2398 | * Scan compiled regex for non-emptiness * |
2399 | *************************************************/ |
2400 | |
2401 | /* This function is called to check for left recursive calls. We want to check |
2402 | the current branch of the current pattern to see if it could match the empty |
2403 | string. If it could, we must look outwards for branches at other levels, |
2404 | stopping when we pass beyond the bracket which is the subject of the recursion. |
2405 | This function is called only during the real compile, not during the |
2406 | pre-compile. |
2407 | |
2408 | Arguments: |
2409 | code points to start of the recursion |
2410 | endcode points to where to stop (current RECURSE item) |
2411 | bcptr points to the chain of current (unclosed) branch starts |
2412 | utf8 TRUE if in UTF-8 mode |
2413 | cd pointers to tables etc |
2414 | |
2415 | Returns: TRUE if what is matched could be empty |
2416 | */ |
2417 | |
2418 | static BOOL |
2419 | could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr, |
2420 | BOOL utf8, compile_data *cd) |
2421 | { |
2422 | while (bcptr != NULL && bcptr->current_branch >= code) |
2423 | { |
2424 | if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd)) |
2425 | return FALSE; |
2426 | bcptr = bcptr->outer; |
2427 | } |
2428 | return TRUE; |
2429 | } |
2430 | |
2431 | |
2432 | |
2433 | /************************************************* |
2434 | * Check for POSIX class syntax * |
2435 | *************************************************/ |
2436 | |
2437 | /* This function is called when the sequence "[:" or "[." or "[=" is |
2438 | encountered in a character class. It checks whether this is followed by a |
2439 | sequence of characters terminated by a matching ":]" or ".]" or "=]". If we |
2440 | reach an unescaped ']' without the special preceding character, return FALSE. |
2441 | |
2442 | Originally, this function only recognized a sequence of letters between the |
2443 | terminators, but it seems that Perl recognizes any sequence of characters, |
2444 | though of course unknown POSIX names are subsequently rejected. Perl gives an |
2445 | "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE |
2446 | didn't consider this to be a POSIX class. Likewise for [:1234:]. |
2447 | |
2448 | The problem in trying to be exactly like Perl is in the handling of escapes. We |
2449 | have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX |
2450 | class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code |
2451 | below handles the special case of \], but does not try to do any other escape |
2452 | processing. This makes it different from Perl for cases such as [:l\ower:] |
2453 | where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize |
2454 | "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does, |
2455 | I think. |
2456 | |
2457 | A user pointed out that PCRE was rejecting [:a[:digit:]] whereas Perl was not. |
2458 | It seems that the appearance of a nested POSIX class supersedes an apparent |
2459 | external class. For example, [:a[:digit:]b:] matches "a", "b", ":", or |
2460 | a digit. |
2461 | |
2462 | In Perl, unescaped square brackets may also appear as part of class names. For |
2463 | example, [:a[:abc]b:] gives unknown POSIX class "[:abc]b:]". However, for |
2464 | [:a[:abc]b][b:] it gives unknown POSIX class "[:abc]b][b:]", which does not |
2465 | seem right at all. PCRE does not allow closing square brackets in POSIX class |
2466 | names. |
2467 | |
2468 | Arguments: |
2469 | ptr pointer to the initial [ |
2470 | endptr where to return the end pointer |
2471 | |
2472 | Returns: TRUE or FALSE |
2473 | */ |
2474 | |
2475 | static BOOL |
2476 | check_posix_syntax(const uschar *ptr, const uschar **endptr) |
2477 | { |
2478 | int terminator; /* Don't combine these lines; the Solaris cc */ |
2479 | terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */ |
2480 | for (++ptr; *ptr != 0; ptr++) |
2481 | { |
2482 | if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) |
2483 | ptr++; |
2484 | else if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE; |
2485 | else |
2486 | { |
2487 | if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) |
2488 | { |
2489 | *endptr = ptr; |
2490 | return TRUE; |
2491 | } |
2492 | if (*ptr == CHAR_LEFT_SQUARE_BRACKET && |
2493 | (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT || |
2494 | ptr[1] == CHAR_EQUALS_SIGN) && |
2495 | check_posix_syntax(ptr, endptr)) |
2496 | return FALSE; |
2497 | } |
2498 | } |
2499 | return FALSE; |
2500 | } |
2501 | |
2502 | |
2503 | |
2504 | |
2505 | /************************************************* |
2506 | * Check POSIX class name * |
2507 | *************************************************/ |
2508 | |
2509 | /* This function is called to check the name given in a POSIX-style class entry |
2510 | such as [:alnum:]. |
2511 | |
2512 | Arguments: |
2513 | ptr points to the first letter |
2514 | len the length of the name |
2515 | |
2516 | Returns: a value representing the name, or -1 if unknown |
2517 | */ |
2518 | |
2519 | static int |
2520 | check_posix_name(const uschar *ptr, int len) |
2521 | { |
2522 | const char *pn = posix_names; |
2523 | register int yield = 0; |
2524 | while (posix_name_lengths[yield] != 0) |
2525 | { |
2526 | if (len == posix_name_lengths[yield] && |
2527 | strncmp((const char *)ptr, pn, len) == 0) return yield; |
2528 | pn += posix_name_lengths[yield] + 1; |
2529 | yield++; |
2530 | } |
2531 | return -1; |
2532 | } |
2533 | |
2534 | |
2535 | /************************************************* |
2536 | * Adjust OP_RECURSE items in repeated group * |
2537 | *************************************************/ |
2538 | |
2539 | /* OP_RECURSE items contain an offset from the start of the regex to the group |
2540 | that is referenced. This means that groups can be replicated for fixed |
2541 | repetition simply by copying (because the recursion is allowed to refer to |
2542 | earlier groups that are outside the current group). However, when a group is |
2543 | optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is |
2544 | inserted before it, after it has been compiled. This means that any OP_RECURSE |
2545 | items within it that refer to the group itself or any contained groups have to |
2546 | have their offsets adjusted. That one of the jobs of this function. Before it |
2547 | is called, the partially compiled regex must be temporarily terminated with |
2548 | OP_END. |
2549 | |
2550 | This function has been extended with the possibility of forward references for |
2551 | recursions and subroutine calls. It must also check the list of such references |
2552 | for the group we are dealing with. If it finds that one of the recursions in |
2553 | the current group is on this list, it adjusts the offset in the list, not the |
2554 | value in the reference (which is a group number). |
2555 | |
2556 | Arguments: |
2557 | group points to the start of the group |
2558 | adjust the amount by which the group is to be moved |
2559 | utf8 TRUE in UTF-8 mode |
2560 | cd contains pointers to tables etc. |
2561 | save_hwm the hwm forward reference pointer at the start of the group |
2562 | |
2563 | Returns: nothing |
2564 | */ |
2565 | |
2566 | static void |
2567 | adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd, |
2568 | uschar *save_hwm) |
2569 | { |
2570 | uschar *ptr = group; |
2571 | |
2572 | while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL) |
2573 | { |
2574 | int offset; |
2575 | uschar *hc; |
2576 | |
2577 | /* See if this recursion is on the forward reference list. If so, adjust the |
2578 | reference. */ |
2579 | |
2580 | for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE) |
2581 | { |
2582 | offset = GET(hc, 0); |
2583 | if (cd->start_code + offset == ptr + 1) |
2584 | { |
2585 | PUT(hc, 0, offset + adjust); |
2586 | break; |
2587 | } |
2588 | } |
2589 | |
2590 | /* Otherwise, adjust the recursion offset if it's after the start of this |
2591 | group. */ |
2592 | |
2593 | if (hc >= cd->hwm) |
2594 | { |
2595 | offset = GET(ptr, 1); |
2596 | if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust); |
2597 | } |
2598 | |
2599 | ptr += 1 + LINK_SIZE; |
2600 | } |
2601 | } |
2602 | |
2603 | |
2604 | |
2605 | /************************************************* |
2606 | * Insert an automatic callout point * |
2607 | *************************************************/ |
2608 | |
2609 | /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert |
2610 | callout points before each pattern item. |
2611 | |
2612 | Arguments: |
2613 | code current code pointer |
2614 | ptr current pattern pointer |
2615 | cd pointers to tables etc |
2616 | |
2617 | Returns: new code pointer |
2618 | */ |
2619 | |
2620 | static uschar * |
2621 | auto_callout(uschar *code, const uschar *ptr, compile_data *cd) |
2622 | { |
2623 | *code++ = OP_CALLOUT; |
2624 | *code++ = 255; |
2625 | PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */ |
2626 | PUT(code, LINK_SIZE, 0); /* Default length */ |
2627 | return code + 2*LINK_SIZE; |
2628 | } |
2629 | |
2630 | |
2631 | |
2632 | /************************************************* |
2633 | * Complete a callout item * |
2634 | *************************************************/ |
2635 | |
2636 | /* A callout item contains the length of the next item in the pattern, which |
2637 | we can't fill in till after we have reached the relevant point. This is used |
2638 | for both automatic and manual callouts. |
2639 | |
2640 | Arguments: |
2641 | previous_callout points to previous callout item |
2642 | ptr current pattern pointer |
2643 | cd pointers to tables etc |
2644 | |
2645 | Returns: nothing |
2646 | */ |
2647 | |
2648 | static void |
2649 | complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd) |
2650 | { |
2651 | int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2)); |
2652 | PUT(previous_callout, 2 + LINK_SIZE, length); |
2653 | } |
2654 | |
2655 | |
2656 | |
2657 | #ifdef SUPPORT_UCP |
2658 | /************************************************* |
2659 | * Get othercase range * |
2660 | *************************************************/ |
2661 | |
2662 | /* This function is passed the start and end of a class range, in UTF-8 mode |
2663 | with UCP support. It searches up the characters, looking for internal ranges of |
2664 | characters in the "other" case. Each call returns the next one, updating the |
2665 | start address. |
2666 | |
2667 | Arguments: |
2668 | cptr points to starting character value; updated |
2669 | d end value |
2670 | ocptr where to put start of othercase range |
2671 | odptr where to put end of othercase range |
2672 | |
2673 | Yield: TRUE when range returned; FALSE when no more |
2674 | */ |
2675 | |
2676 | static BOOL |
2677 | get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr, |
2678 | unsigned int *odptr) |
2679 | { |
2680 | unsigned int c, othercase, next; |
2681 | |
2682 | for (c = *cptr; c <= d; c++) |
2683 | { if ((othercase = UCD_OTHERCASE(c)) != c) break; } |
2684 | |
2685 | if (c > d) return FALSE; |
2686 | |
2687 | *ocptr = othercase; |
2688 | next = othercase + 1; |
2689 | |
2690 | for (++c; c <= d; c++) |
2691 | { |
2692 | if (UCD_OTHERCASE(c) != next) break; |
2693 | next++; |
2694 | } |
2695 | |
2696 | *odptr = next - 1; |
2697 | *cptr = c; |
2698 | |
2699 | return TRUE; |
2700 | } |
2701 | |
2702 | |
2703 | |
2704 | /************************************************* |
2705 | * Check a character and a property * |
2706 | *************************************************/ |
2707 | |
2708 | /* This function is called by check_auto_possessive() when a property item |
2709 | is adjacent to a fixed character. |
2710 | |
2711 | Arguments: |
2712 | c the character |
2713 | ptype the property type |
2714 | pdata the data for the type |
2715 | negated TRUE if it's a negated property (\P or \p{^) |
2716 | |
2717 | Returns: TRUE if auto-possessifying is OK |
2718 | */ |
2719 | |
2720 | static BOOL |
2721 | check_char_prop(int c, int ptype, int pdata, BOOL negated) |
2722 | { |
2723 | const ucd_record *prop = GET_UCD(c); |
2724 | switch(ptype) |
2725 | { |
2726 | case PT_LAMP: |
2727 | return (prop->chartype == ucp_Lu || |
2728 | prop->chartype == ucp_Ll || |
2729 | prop->chartype == ucp_Lt) == negated; |
2730 | |
2731 | case PT_GC: |
2732 | return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated; |
2733 | |
2734 | case PT_PC: |
2735 | return (pdata == prop->chartype) == negated; |
2736 | |
2737 | case PT_SC: |
2738 | return (pdata == prop->script) == negated; |
2739 | |
2740 | /* These are specials */ |
2741 | |
2742 | case PT_ALNUM: |
2743 | return (_pcre_ucp_gentype[prop->chartype] == ucp_L || |
2744 | _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated; |
2745 | |
2746 | case PT_SPACE: /* Perl space */ |
2747 | return (_pcre_ucp_gentype[prop->chartype] == ucp_Z || |
2748 | c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR) |
2749 | == negated; |
2750 | |
2751 | case PT_PXSPACE: /* POSIX space */ |
2752 | return (_pcre_ucp_gentype[prop->chartype] == ucp_Z || |
2753 | c == CHAR_HT || c == CHAR_NL || c == CHAR_VT || |
2754 | c == CHAR_FF || c == CHAR_CR) |
2755 | == negated; |
2756 | |
2757 | case PT_WORD: |
2758 | return (_pcre_ucp_gentype[prop->chartype] == ucp_L || |
2759 | _pcre_ucp_gentype[prop->chartype] == ucp_N || |
2760 | c == CHAR_UNDERSCORE) == negated; |
2761 | } |
2762 | return FALSE; |
2763 | } |
2764 | #endif /* SUPPORT_UCP */ |
2765 | |
2766 | |
2767 | |
2768 | /************************************************* |
2769 | * Check if auto-possessifying is possible * |
2770 | *************************************************/ |
2771 | |
2772 | /* This function is called for unlimited repeats of certain items, to see |
2773 | whether the next thing could possibly match the repeated item. If not, it makes |
2774 | sense to automatically possessify the repeated item. |
2775 | |
2776 | Arguments: |
2777 | previous pointer to the repeated opcode |
2778 | utf8 TRUE in UTF-8 mode |
2779 | ptr next character in pattern |
2780 | options options bits |
2781 | cd contains pointers to tables etc. |
2782 | |
2783 | Returns: TRUE if possessifying is wanted |
2784 | */ |
2785 | |
2786 | static BOOL |
2787 | check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr, |
2788 | int options, compile_data *cd) |
2789 | { |
2790 | int c, next; |
2791 | int op_code = *previous++; |
2792 | |
2793 | /* Skip whitespace and comments in extended mode */ |
2794 | |
2795 | if ((options & PCRE_EXTENDED) != 0) |
2796 | { |
2797 | for (;;) |
2798 | { |
2799 | while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++; |
2800 | if (*ptr == CHAR_NUMBER_SIGN) |
2801 | { |
2802 | ptr++; |
2803 | while (*ptr != 0) |
2804 | { |
2805 | if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; } |
2806 | ptr++; |
2807 | #ifdef SUPPORT_UTF8 |
2808 | if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++; |
2809 | #endif |
2810 | } |
2811 | } |
2812 | else break; |
2813 | } |
2814 | } |
2815 | |
2816 | /* If the next item is one that we can handle, get its value. A non-negative |
2817 | value is a character, a negative value is an escape value. */ |
2818 | |
2819 | if (*ptr == CHAR_BACKSLASH) |
2820 | { |
2821 | int temperrorcode = 0; |
2822 | next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE); |
2823 | if (temperrorcode != 0) return FALSE; |
2824 | ptr++; /* Point after the escape sequence */ |
2825 | } |
2826 | |
2827 | else if ((cd->ctypes[*ptr] & ctype_meta) == 0) |
2828 | { |
2829 | #ifdef SUPPORT_UTF8 |
2830 | if (utf8) { GETCHARINC(next, ptr); } else |
2831 | #endif |
2832 | next = *ptr++; |
2833 | } |
2834 | |
2835 | else return FALSE; |
2836 | |
2837 | /* Skip whitespace and comments in extended mode */ |
2838 | |
2839 | if ((options & PCRE_EXTENDED) != 0) |
2840 | { |
2841 | for (;;) |
2842 | { |
2843 | while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++; |
2844 | if (*ptr == CHAR_NUMBER_SIGN) |
2845 | { |
2846 | ptr++; |
2847 | while (*ptr != 0) |
2848 | { |
2849 | if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; } |
2850 | ptr++; |
2851 | #ifdef SUPPORT_UTF8 |
2852 | if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++; |
2853 | #endif |
2854 | } |
2855 | } |
2856 | else break; |
2857 | } |
2858 | } |
2859 | |
2860 | /* If the next thing is itself optional, we have to give up. */ |
2861 | |
2862 | if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK || |
2863 | strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0) |
2864 | return FALSE; |
2865 | |
2866 | /* Now compare the next item with the previous opcode. First, handle cases when |
2867 | the next item is a character. */ |
2868 | |
2869 | if (next >= 0) switch(op_code) |
2870 | { |
2871 | case OP_CHAR: |
2872 | #ifdef SUPPORT_UTF8 |
2873 | GETCHARTEST(c, previous); |
2874 | #else |
2875 | c = *previous; |
2876 | #endif |
2877 | return c != next; |
2878 | |
2879 | /* For CHARI (caseless character) we must check the other case. If we have |
2880 | Unicode property support, we can use it to test the other case of |
2881 | high-valued characters. */ |
2882 | |
2883 | case OP_CHARI: |
2884 | #ifdef SUPPORT_UTF8 |
2885 | GETCHARTEST(c, previous); |
2886 | #else |
2887 | c = *previous; |
2888 | #endif |
2889 | if (c == next) return FALSE; |
2890 | #ifdef SUPPORT_UTF8 |
2891 | if (utf8) |
2892 | { |
2893 | unsigned int othercase; |
2894 | if (next < 128) othercase = cd->fcc[next]; else |
2895 | #ifdef SUPPORT_UCP |
2896 | othercase = UCD_OTHERCASE((unsigned int)next); |
2897 | #else |
2898 | othercase = NOTACHAR; |
2899 | #endif |
2900 | return (unsigned int)c != othercase; |
2901 | } |
2902 | else |
2903 | #endif /* SUPPORT_UTF8 */ |
2904 | return (c != cd->fcc[next]); /* Non-UTF-8 mode */ |
2905 | |
2906 | /* For OP_NOT and OP_NOTI, the data is always a single-byte character. These |
2907 | opcodes are not used for multi-byte characters, because they are coded using |
2908 | an XCLASS instead. */ |
2909 | |
2910 | case OP_NOT: |
2911 | return (c = *previous) == next; |
2912 | |
2913 | case OP_NOTI: |
2914 | if ((c = *previous) == next) return TRUE; |
2915 | #ifdef SUPPORT_UTF8 |
2916 | if (utf8) |
2917 | { |
2918 | unsigned int othercase; |
2919 | if (next < 128) othercase = cd->fcc[next]; else |
2920 | #ifdef SUPPORT_UCP |
2921 | othercase = UCD_OTHERCASE(next); |
2922 | #else |
2923 | othercase = NOTACHAR; |
2924 | #endif |
2925 | return (unsigned int)c == othercase; |
2926 | } |
2927 | else |
2928 | #endif /* SUPPORT_UTF8 */ |
2929 | return (c == cd->fcc[next]); /* Non-UTF-8 mode */ |
2930 | |
2931 | /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set. |
2932 | When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */ |
2933 | |
2934 | case OP_DIGIT: |
2935 | return next > 127 || (cd->ctypes[next] & ctype_digit) == 0; |
2936 | |
2937 | case OP_NOT_DIGIT: |
2938 | return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0; |
2939 | |
2940 | case OP_WHITESPACE: |
2941 | return next > 127 || (cd->ctypes[next] & ctype_space) == 0; |
2942 | |
2943 | case OP_NOT_WHITESPACE: |
2944 | return next <= 127 && (cd->ctypes[next] & ctype_space) != 0; |
2945 | |
2946 | case OP_WORDCHAR: |
2947 | return next > 127 || (cd->ctypes[next] & ctype_word) == 0; |
2948 | |
2949 | case OP_NOT_WORDCHAR: |
2950 | return next <= 127 && (cd->ctypes[next] & ctype_word) != 0; |
2951 | |
2952 | case OP_HSPACE: |
2953 | case OP_NOT_HSPACE: |
2954 | switch(next) |
2955 | { |
2956 | case 0x09: |
2957 | case 0x20: |
2958 | case 0xa0: |
2959 | case 0x1680: |
2960 | case 0x180e: |
2961 | case 0x2000: |
2962 | case 0x2001: |
2963 | case 0x2002: |
2964 | case 0x2003: |
2965 | case 0x2004: |
2966 | case 0x2005: |
2967 | case 0x2006: |
2968 | case 0x2007: |
2969 | case 0x2008: |
2970 | case 0x2009: |
2971 | case 0x200A: |
2972 | case 0x202f: |
2973 | case 0x205f: |
2974 | case 0x3000: |
2975 | return op_code == OP_NOT_HSPACE; |
2976 | default: |
2977 | return op_code != OP_NOT_HSPACE; |
2978 | } |
2979 | |
2980 | case OP_ANYNL: |
2981 | case OP_VSPACE: |
2982 | case OP_NOT_VSPACE: |
2983 | switch(next) |
2984 | { |
2985 | case 0x0a: |
2986 | case 0x0b: |
2987 | case 0x0c: |
2988 | case 0x0d: |
2989 | case 0x85: |
2990 | case 0x2028: |
2991 | case 0x2029: |
2992 | return op_code == OP_NOT_VSPACE; |
2993 | default: |
2994 | return op_code != OP_NOT_VSPACE; |
2995 | } |
2996 | |
2997 | #ifdef SUPPORT_UCP |
2998 | case OP_PROP: |
2999 | return check_char_prop(next, previous[0], previous[1], FALSE); |
3000 | |
3001 | case OP_NOTPROP: |
3002 | return check_char_prop(next, previous[0], previous[1], TRUE); |
3003 | #endif |
3004 | |
3005 | default: |
3006 | return FALSE; |
3007 | } |
3008 | |
3009 | |
3010 | /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP |
3011 | is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are |
3012 | generated only when PCRE_UCP is *not* set, that is, when only ASCII |
3013 | characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are |
3014 | replaced by OP_PROP codes when PCRE_UCP is set. */ |
3015 | |
3016 | switch(op_code) |
3017 | { |
3018 | case OP_CHAR: |
3019 | case OP_CHARI: |
3020 | #ifdef SUPPORT_UTF8 |
3021 | GETCHARTEST(c, previous); |
3022 | #else |
3023 | c = *previous; |
3024 | #endif |
3025 | switch(-next) |
3026 | { |
3027 | case ESC_d: |
3028 | return c > 127 || (cd->ctypes[c] & ctype_digit) == 0; |
3029 | |
3030 | case ESC_D: |
3031 | return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0; |
3032 | |
3033 | case ESC_s: |
3034 | return c > 127 || (cd->ctypes[c] & ctype_space) == 0; |
3035 | |
3036 | case ESC_S: |
3037 | return c <= 127 && (cd->ctypes[c] & ctype_space) != 0; |
3038 | |
3039 | case ESC_w: |
3040 | return c > 127 || (cd->ctypes[c] & ctype_word) == 0; |
3041 | |
3042 | case ESC_W: |
3043 | return c <= 127 && (cd->ctypes[c] & ctype_word) != 0; |
3044 | |
3045 | case ESC_h: |
3046 | case ESC_H: |
3047 | switch(c) |
3048 | { |
3049 | case 0x09: |
3050 | case 0x20: |
3051 | case 0xa0: |
3052 | case 0x1680: |
3053 | case 0x180e: |
3054 | case 0x2000: |
3055 | case 0x2001: |
3056 | case 0x2002: |
3057 | case 0x2003: |
3058 | case 0x2004: |
3059 | case 0x2005: |
3060 | case 0x2006: |
3061 | case 0x2007: |
3062 | case 0x2008: |
3063 | case 0x2009: |
3064 | case 0x200A: |
3065 | case 0x202f: |
3066 | case 0x205f: |
3067 | case 0x3000: |
3068 | return -next != ESC_h; |
3069 | default: |
3070 | return -next == ESC_h; |
3071 | } |
3072 | |
3073 | case ESC_v: |
3074 | case ESC_V: |
3075 | switch(c) |
3076 | { |
3077 | case 0x0a: |
3078 | case 0x0b: |
3079 | case 0x0c: |
3080 | case 0x0d: |
3081 | case 0x85: |
3082 | case 0x2028: |
3083 | case 0x2029: |
3084 | return -next != ESC_v; |
3085 | default: |
3086 | return -next == ESC_v; |
3087 | } |
3088 | |
3089 | /* When PCRE_UCP is set, these values get generated for \d etc. Find |
3090 | their substitutions and process them. The result will always be either |
3091 | -ESC_p or -ESC_P. Then fall through to process those values. */ |
3092 | |
3093 | #ifdef SUPPORT_UCP |
3094 | case ESC_du: |
3095 | case ESC_DU: |
3096 | case ESC_wu: |
3097 | case ESC_WU: |
3098 | case ESC_su: |
3099 | case ESC_SU: |
3100 | { |
3101 | int temperrorcode = 0; |
3102 | ptr = substitutes[-next - ESC_DU]; |
3103 | next = check_escape(&ptr, &temperrorcode, 0, options, FALSE); |
3104 | if (temperrorcode != 0) return FALSE; |
3105 | ptr++; /* For compatibility */ |
3106 | } |
3107 | /* Fall through */ |
3108 | |
3109 | case ESC_p: |
3110 | case ESC_P: |
3111 | { |
3112 | int ptype, pdata, errorcodeptr; |
3113 | BOOL negated; |
3114 | |
3115 | ptr--; /* Make ptr point at the p or P */ |
3116 | ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr); |
3117 | if (ptype < 0) return FALSE; |
3118 | ptr++; /* Point past the final curly ket */ |
3119 | |
3120 | /* If the property item is optional, we have to give up. (When generated |
3121 | from \d etc by PCRE_UCP, this test will have been applied much earlier, |
3122 | to the original \d etc. At this point, ptr will point to a zero byte. */ |
3123 | |
3124 | if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK || |
3125 | strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0) |
3126 | return FALSE; |
3127 | |
3128 | /* Do the property check. */ |
3129 | |
3130 | return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated); |
3131 | } |
3132 | #endif |
3133 | |
3134 | default: |
3135 | return FALSE; |
3136 | } |
3137 | |
3138 | /* In principle, support for Unicode properties should be integrated here as |
3139 | well. It means re-organizing the above code so as to get hold of the property |
3140 | values before switching on the op-code. However, I wonder how many patterns |
3141 | combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set, |
3142 | these op-codes are never generated.) */ |
3143 | |
3144 | case OP_DIGIT: |
3145 | return next == -ESC_D || next == -ESC_s || next == -ESC_W || |
3146 | next == -ESC_h || next == -ESC_v || next == -ESC_R; |
3147 | |
3148 | case OP_NOT_DIGIT: |
3149 | return next == -ESC_d; |
3150 | |
3151 | case OP_WHITESPACE: |
3152 | return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R; |
3153 | |
3154 | case OP_NOT_WHITESPACE: |
3155 | return next == -ESC_s || next == -ESC_h || next == -ESC_v; |
3156 | |
3157 | case OP_HSPACE: |
3158 | return next == -ESC_S || next == -ESC_H || next == -ESC_d || |
3159 | next == -ESC_w || next == -ESC_v || next == -ESC_R; |
3160 | |
3161 | case OP_NOT_HSPACE: |
3162 | return next == -ESC_h; |
3163 | |
3164 | /* Can't have \S in here because VT matches \S (Perl anomaly) */ |
3165 | case OP_ANYNL: |
3166 | case OP_VSPACE: |
3167 | return next == -ESC_V || next == -ESC_d || next == -ESC_w; |
3168 | |
3169 | case OP_NOT_VSPACE: |
3170 | return next == -ESC_v || next == -ESC_R; |
3171 | |
3172 | case OP_WORDCHAR: |
3173 | return next == -ESC_W || next == -ESC_s || next == -ESC_h || |
3174 | next == -ESC_v || next == -ESC_R; |
3175 | |
3176 | case OP_NOT_WORDCHAR: |
3177 | return next == -ESC_w || next == -ESC_d; |
3178 | |
3179 | default: |
3180 | return FALSE; |
3181 | } |
3182 | |
3183 | /* Control does not reach here */ |
3184 | } |
3185 | |
3186 | |
3187 | |
3188 | /************************************************* |
3189 | * Compile one branch * |
3190 | *************************************************/ |
3191 | |
3192 | /* Scan the pattern, compiling it into the a vector. If the options are |
3193 | changed during the branch, the pointer is used to change the external options |
3194 | bits. This function is used during the pre-compile phase when we are trying |
3195 | to find out the amount of memory needed, as well as during the real compile |
3196 | phase. The value of lengthptr distinguishes the two phases. |
3197 | |
3198 | Arguments: |
3199 | optionsptr pointer to the option bits |
3200 | codeptr points to the pointer to the current code point |
3201 | ptrptr points to the current pattern pointer |
3202 | errorcodeptr points to error code variable |
3203 | firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE) |
3204 | reqbyteptr set to the last literal character required, else < 0 |
3205 | bcptr points to current branch chain |
3206 | cond_depth conditional nesting depth |
3207 | cd contains pointers to tables etc. |
3208 | lengthptr NULL during the real compile phase |
3209 | points to length accumulator during pre-compile phase |
3210 | |
3211 | Returns: TRUE on success |
3212 | FALSE, with *errorcodeptr set non-zero on error |
3213 | */ |
3214 | |
3215 | static BOOL |
3216 | compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr, |
3217 | int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, |
3218 | int cond_depth, compile_data *cd, int *lengthptr) |
3219 | { |
3220 | int repeat_type, op_type; |
3221 | int repeat_min = 0, repeat_max = 0; /* To please picky compilers */ |
3222 | int bravalue = 0; |
3223 | int greedy_default, greedy_non_default; |
3224 | int firstbyte, reqbyte; |
3225 | int zeroreqbyte, zerofirstbyte; |
3226 | int req_caseopt, reqvary, tempreqvary; |
3227 | int options = *optionsptr; /* May change dynamically */ |
3228 | int after_manual_callout = 0; |
3229 | int length_prevgroup = 0; |
3230 | register int c; |
3231 | register uschar *code = *codeptr; |
3232 | uschar *last_code = code; |
3233 | uschar *orig_code = code; |
3234 | uschar *tempcode; |
3235 | BOOL inescq = FALSE; |
3236 | BOOL groupsetfirstbyte = FALSE; |
3237 | const uschar *ptr = *ptrptr; |
3238 | const uschar *tempptr; |
3239 | const uschar *nestptr = NULL; |
3240 | uschar *previous = NULL; |
3241 | uschar *previous_callout = NULL; |
3242 | uschar *save_hwm = NULL; |
3243 | uschar classbits[32]; |
3244 | |
3245 | /* We can fish out the UTF-8 setting once and for all into a BOOL, but we |
3246 | must not do this for other options (e.g. PCRE_EXTENDED) because they may change |
3247 | dynamically as we process the pattern. */ |
3248 | |
3249 | #ifdef SUPPORT_UTF8 |
3250 | BOOL class_utf8; |
3251 | BOOL utf8 = (options & PCRE_UTF8) != 0; |
3252 | uschar *class_utf8data; |
3253 | uschar *class_utf8data_base; |
3254 | uschar utf8_char[6]; |
3255 | #else |
3256 | BOOL utf8 = FALSE; |
3257 | #endif |
3258 | |
3259 | #ifdef PCRE_DEBUG |
3260 | if (lengthptr != NULL) DPRINTF((">> start branch\n")); |
3261 | #endif |
3262 | |
3263 | /* Set up the default and non-default settings for greediness */ |
3264 | |
3265 | greedy_default = ((options & PCRE_UNGREEDY) != 0); |
3266 | greedy_non_default = greedy_default ^ 1; |
3267 | |
3268 | /* Initialize no first byte, no required byte. REQ_UNSET means "no char |
3269 | matching encountered yet". It gets changed to REQ_NONE if we hit something that |
3270 | matches a non-fixed char first char; reqbyte just remains unset if we never |
3271 | find one. |
3272 | |
3273 | When we hit a repeat whose minimum is zero, we may have to adjust these values |
3274 | to take the zero repeat into account. This is implemented by setting them to |
3275 | zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual |
3276 | item types that can be repeated set these backoff variables appropriately. */ |
3277 | |
3278 | firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET; |
3279 | |
3280 | /* The variable req_caseopt contains either the REQ_CASELESS value or zero, |
3281 | according to the current setting of the caseless flag. REQ_CASELESS is a bit |
3282 | value > 255. It is added into the firstbyte or reqbyte variables to record the |
3283 | case status of the value. This is used only for ASCII characters. */ |
3284 | |
3285 | req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0; |
3286 | |
3287 | /* Switch on next character until the end of the branch */ |
3288 | |
3289 | for (;; ptr++) |
3290 | { |
3291 | BOOL negate_class; |
3292 | BOOL should_flip_negation; |
3293 | BOOL possessive_quantifier; |
3294 | BOOL is_quantifier; |
3295 | BOOL is_recurse; |
3296 | BOOL reset_bracount; |
3297 | int class_charcount; |
3298 | int class_lastchar; |
3299 | int newoptions; |
3300 | int recno; |
3301 | int refsign; |
3302 | int skipbytes; |
3303 | int subreqbyte; |
3304 | int subfirstbyte; |
3305 | int terminator; |
3306 | int mclength; |
3307 | int tempbracount; |
3308 | uschar mcbuffer[8]; |
3309 | |
3310 | /* Get next byte in the pattern */ |
3311 | |
3312 | c = *ptr; |
3313 | |
3314 | /* If we are at the end of a nested substitution, revert to the outer level |
3315 | string. Nesting only happens one level deep. */ |
3316 | |
3317 | if (c == 0 && nestptr != NULL) |
3318 | { |
3319 | ptr = nestptr; |
3320 | nestptr = NULL; |
3321 | c = *ptr; |
3322 | } |
3323 | |
3324 | /* If we are in the pre-compile phase, accumulate the length used for the |
3325 | previous cycle of this loop. */ |
3326 | |
3327 | if (lengthptr != NULL) |
3328 | { |
3329 | #ifdef PCRE_DEBUG |
3330 | if (code > cd->hwm) cd->hwm = code; /* High water info */ |
3331 | #endif |
3332 | if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */ |
3333 | { |
3334 | *errorcodeptr = ERR52; |
3335 | goto FAILED; |
3336 | } |
3337 | |
3338 | /* There is at least one situation where code goes backwards: this is the |
3339 | case of a zero quantifier after a class (e.g. [ab]{0}). At compile time, |
3340 | the class is simply eliminated. However, it is created first, so we have to |
3341 | allow memory for it. Therefore, don't ever reduce the length at this point. |
3342 | */ |
3343 | |
3344 | if (code < last_code) code = last_code; |
3345 | |
3346 | /* Paranoid check for integer overflow */ |
3347 | |
3348 | if (OFLOW_MAX - *lengthptr < code - last_code) |
3349 | { |
3350 | *errorcodeptr = ERR20; |
3351 | goto FAILED; |
3352 | } |
3353 | |
3354 | *lengthptr += (int)(code - last_code); |
3355 | DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c)); |
3356 | |
3357 | /* If "previous" is set and it is not at the start of the work space, move |
3358 | it back to there, in order to avoid filling up the work space. Otherwise, |
3359 | if "previous" is NULL, reset the current code pointer to the start. */ |
3360 | |
3361 | if (previous != NULL) |
3362 | { |
3363 | if (previous > orig_code) |
3364 | { |
3365 | memmove(orig_code, previous, code - previous); |
3366 | code -= previous - orig_code; |
3367 | previous = orig_code; |
3368 | } |
3369 | } |
3370 | else code = orig_code; |
3371 | |
3372 | /* Remember where this code item starts so we can pick up the length |
3373 | next time round. */ |
3374 | |
3375 | last_code = code; |
3376 | } |
3377 | |
3378 | /* In the real compile phase, just check the workspace used by the forward |
3379 | reference list. */ |
3380 | |
3381 | else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK) |
3382 | { |
3383 | *errorcodeptr = ERR52; |
3384 | goto FAILED; |
3385 | } |
3386 | |
3387 | /* If in \Q...\E, check for the end; if not, we have a literal */ |
3388 | |
3389 | if (inescq && c != 0) |
3390 | { |
3391 | if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) |
3392 | { |
3393 | inescq = FALSE; |
3394 | ptr++; |
3395 | continue; |
3396 | } |
3397 | else |
3398 | { |
3399 | if (previous_callout != NULL) |
3400 | { |
3401 | if (lengthptr == NULL) /* Don't attempt in pre-compile phase */ |
3402 | complete_callout(previous_callout, ptr, cd); |
3403 | previous_callout = NULL; |
3404 | } |
3405 | if ((options & PCRE_AUTO_CALLOUT) != 0) |
3406 | { |
3407 | previous_callout = code; |
3408 | code = auto_callout(code, ptr, cd); |
3409 | } |
3410 | goto NORMAL_CHAR; |
3411 | } |
3412 | } |
3413 | |
3414 | /* Fill in length of a previous callout, except when the next thing is |
3415 | a quantifier. */ |
3416 | |
3417 | is_quantifier = |
3418 | c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK || |
3419 | (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1)); |
3420 | |
3421 | if (!is_quantifier && previous_callout != NULL && |
3422 | after_manual_callout-- <= 0) |
3423 | { |
3424 | if (lengthptr == NULL) /* Don't attempt in pre-compile phase */ |
3425 | complete_callout(previous_callout, ptr, cd); |
3426 | previous_callout = NULL; |
3427 | } |
3428 | |
3429 | /* In extended mode, skip white space and comments. */ |
3430 | |
3431 | if ((options & PCRE_EXTENDED) != 0) |
3432 | { |
3433 | if ((cd->ctypes[c] & ctype_space) != 0) continue; |
3434 | if (c == CHAR_NUMBER_SIGN) |
3435 | { |
3436 | ptr++; |
3437 | while (*ptr != 0) |
3438 | { |
3439 | if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; } |
3440 | ptr++; |
3441 | #ifdef SUPPORT_UTF8 |
3442 | if (utf8) while ((*ptr & 0xc0) == 0x80) ptr++; |
3443 | #endif |
3444 | } |
3445 | if (*ptr != 0) continue; |
3446 | |
3447 | /* Else fall through to handle end of string */ |
3448 | c = 0; |
3449 | } |
3450 | } |
3451 | |
3452 | /* No auto callout for quantifiers. */ |
3453 | |
3454 | if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier) |
3455 | { |
3456 | previous_callout = code; |
3457 | code = auto_callout(code, ptr, cd); |
3458 | } |
3459 | |
3460 | switch(c) |
3461 | { |
3462 | /* ===================================================================*/ |
3463 | case 0: /* The branch terminates at string end */ |
3464 | case CHAR_VERTICAL_LINE: /* or | or ) */ |
3465 | case CHAR_RIGHT_PARENTHESIS: |
3466 | *firstbyteptr = firstbyte; |
3467 | *reqbyteptr = reqbyte; |
3468 | *codeptr = code; |
3469 | *ptrptr = ptr; |
3470 | if (lengthptr != NULL) |
3471 | { |
3472 | if (OFLOW_MAX - *lengthptr < code - last_code) |
3473 | { |
3474 | *errorcodeptr = ERR20; |
3475 | goto FAILED; |
3476 | } |
3477 | *lengthptr += (int)(code - last_code); /* To include callout length */ |
3478 | DPRINTF((">> end branch\n")); |
3479 | } |
3480 | return TRUE; |
3481 | |
3482 | |
3483 | /* ===================================================================*/ |
3484 | /* Handle single-character metacharacters. In multiline mode, ^ disables |
3485 | the setting of any following char as a first character. */ |
3486 | |
3487 | case CHAR_CIRCUMFLEX_ACCENT: |
3488 | previous = NULL; |
3489 | if ((options & PCRE_MULTILINE) != 0) |
3490 | { |
3491 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
3492 | *code++ = OP_CIRCM; |
3493 | } |
3494 | else *code++ = OP_CIRC; |
3495 | break; |
3496 | |
3497 | case CHAR_DOLLAR_SIGN: |
3498 | previous = NULL; |
3499 | *code++ = ((options & PCRE_MULTILINE) != 0)? OP_DOLLM : OP_DOLL; |
3500 | break; |
3501 | |
3502 | /* There can never be a first char if '.' is first, whatever happens about |
3503 | repeats. The value of reqbyte doesn't change either. */ |
3504 | |
3505 | case CHAR_DOT: |
3506 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
3507 | zerofirstbyte = firstbyte; |
3508 | zeroreqbyte = reqbyte; |
3509 | previous = code; |
3510 | *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY; |
3511 | break; |
3512 | |
3513 | |
3514 | /* ===================================================================*/ |
3515 | /* Character classes. If the included characters are all < 256, we build a |
3516 | 32-byte bitmap of the permitted characters, except in the special case |
3517 | where there is only one such character. For negated classes, we build the |
3518 | map as usual, then invert it at the end. However, we use a different opcode |
3519 | so that data characters > 255 can be handled correctly. |
3520 | |
3521 | If the class contains characters outside the 0-255 range, a different |
3522 | opcode is compiled. It may optionally have a bit map for characters < 256, |
3523 | but those above are are explicitly listed afterwards. A flag byte tells |
3524 | whether the bitmap is present, and whether this is a negated class or not. |
3525 | |
3526 | In JavaScript compatibility mode, an isolated ']' causes an error. In |
3527 | default (Perl) mode, it is treated as a data character. */ |
3528 | |
3529 | case CHAR_RIGHT_SQUARE_BRACKET: |
3530 | if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0) |
3531 | { |
3532 | *errorcodeptr = ERR64; |
3533 | goto FAILED; |
3534 | } |
3535 | goto NORMAL_CHAR; |
3536 | |
3537 | case CHAR_LEFT_SQUARE_BRACKET: |
3538 | previous = code; |
3539 | |
3540 | /* PCRE supports POSIX class stuff inside a class. Perl gives an error if |
3541 | they are encountered at the top level, so we'll do that too. */ |
3542 | |
3543 | if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT || |
3544 | ptr[1] == CHAR_EQUALS_SIGN) && |
3545 | check_posix_syntax(ptr, &tempptr)) |
3546 | { |
3547 | *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31; |
3548 | goto FAILED; |
3549 | } |
3550 | |
3551 | /* If the first character is '^', set the negation flag and skip it. Also, |
3552 | if the first few characters (either before or after ^) are \Q\E or \E we |
3553 | skip them too. This makes for compatibility with Perl. */ |
3554 | |
3555 | negate_class = FALSE; |
3556 | for (;;) |
3557 | { |
3558 | c = *(++ptr); |
3559 | if (c == CHAR_BACKSLASH) |
3560 | { |
3561 | if (ptr[1] == CHAR_E) |
3562 | ptr++; |
3563 | else if (strncmp((const char *)ptr+1, |
3564 | STR_Q STR_BACKSLASH STR_E, 3) == 0) |
3565 | ptr += 3; |
3566 | else |
3567 | break; |
3568 | } |
3569 | else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT) |
3570 | negate_class = TRUE; |
3571 | else break; |
3572 | } |
3573 | |
3574 | /* Empty classes are allowed in JavaScript compatibility mode. Otherwise, |
3575 | an initial ']' is taken as a data character -- the code below handles |
3576 | that. In JS mode, [] must always fail, so generate OP_FAIL, whereas |
3577 | [^] must match any character, so generate OP_ALLANY. */ |
3578 | |
3579 | if (c == CHAR_RIGHT_SQUARE_BRACKET && |
3580 | (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0) |
3581 | { |
3582 | *code++ = negate_class? OP_ALLANY : OP_FAIL; |
3583 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
3584 | zerofirstbyte = firstbyte; |
3585 | break; |
3586 | } |
3587 | |
3588 | /* If a class contains a negative special such as \S, we need to flip the |
3589 | negation flag at the end, so that support for characters > 255 works |
3590 | correctly (they are all included in the class). */ |
3591 | |
3592 | should_flip_negation = FALSE; |
3593 | |
3594 | /* Keep a count of chars with values < 256 so that we can optimize the case |
3595 | of just a single character (as long as it's < 256). However, For higher |
3596 | valued UTF-8 characters, we don't yet do any optimization. */ |
3597 | |
3598 | class_charcount = 0; |
3599 | class_lastchar = -1; |
3600 | |
3601 | /* Initialize the 32-char bit map to all zeros. We build the map in a |
3602 | temporary bit of memory, in case the class contains only 1 character (less |
3603 | than 256), because in that case the compiled code doesn't use the bit map. |
3604 | */ |
3605 | |
3606 | memset(classbits, 0, 32 * sizeof(uschar)); |
3607 | |
3608 | #ifdef SUPPORT_UTF8 |
3609 | class_utf8 = FALSE; /* No chars >= 256 */ |
3610 | class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */ |
3611 | class_utf8data_base = class_utf8data; /* For resetting in pass 1 */ |
3612 | #endif |
3613 | |
3614 | /* Process characters until ] is reached. By writing this as a "do" it |
3615 | means that an initial ] is taken as a data character. At the start of the |
3616 | loop, c contains the first byte of the character. */ |
3617 | |
3618 | if (c != 0) do |
3619 | { |
3620 | const uschar *oldptr; |
3621 | |
3622 | #ifdef SUPPORT_UTF8 |
3623 | if (utf8 && c > 127) |
3624 | { /* Braces are required because the */ |
3625 | GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */ |
3626 | } |
3627 | |
3628 | /* In the pre-compile phase, accumulate the length of any UTF-8 extra |
3629 | data and reset the pointer. This is so that very large classes that |
3630 | contain a zillion UTF-8 characters no longer overwrite the work space |
3631 | (which is on the stack). */ |
3632 | |
3633 | if (lengthptr != NULL) |
3634 | { |
3635 | *lengthptr += class_utf8data - class_utf8data_base; |
3636 | class_utf8data = class_utf8data_base; |
3637 | } |
3638 | |
3639 | #endif |
3640 | |
3641 | /* Inside \Q...\E everything is literal except \E */ |
3642 | |
3643 | if (inescq) |
3644 | { |
3645 | if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */ |
3646 | { |
3647 | inescq = FALSE; /* Reset literal state */ |
3648 | ptr++; /* Skip the 'E' */ |
3649 | continue; /* Carry on with next */ |
3650 | } |
3651 | goto CHECK_RANGE; /* Could be range if \E follows */ |
3652 | } |
3653 | |
3654 | /* Handle POSIX class names. Perl allows a negation extension of the |
3655 | form [:^name:]. A square bracket that doesn't match the syntax is |
3656 | treated as a literal. We also recognize the POSIX constructions |
3657 | [.ch.] and [=ch=] ("collating elements") and fault them, as Perl |
3658 | 5.6 and 5.8 do. */ |
3659 | |
3660 | if (c == CHAR_LEFT_SQUARE_BRACKET && |
3661 | (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT || |
3662 | ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr)) |
3663 | { |
3664 | BOOL local_negate = FALSE; |
3665 | int posix_class, taboffset, tabopt; |
3666 | register const uschar *cbits = cd->cbits; |
3667 | uschar pbits[32]; |
3668 | |
3669 | if (ptr[1] != CHAR_COLON) |
3670 | { |
3671 | *errorcodeptr = ERR31; |
3672 | goto FAILED; |
3673 | } |
3674 | |
3675 | ptr += 2; |
3676 | if (*ptr == CHAR_CIRCUMFLEX_ACCENT) |
3677 | { |
3678 | local_negate = TRUE; |
3679 | should_flip_negation = TRUE; /* Note negative special */ |
3680 | ptr++; |
3681 | } |
3682 | |
3683 | posix_class = check_posix_name(ptr, (int)(tempptr - ptr)); |
3684 | if (posix_class < 0) |
3685 | { |
3686 | *errorcodeptr = ERR30; |
3687 | goto FAILED; |
3688 | } |
3689 | |
3690 | /* If matching is caseless, upper and lower are converted to |
3691 | alpha. This relies on the fact that the class table starts with |
3692 | alpha, lower, upper as the first 3 entries. */ |
3693 | |
3694 | if ((options & PCRE_CASELESS) != 0 && posix_class <= 2) |
3695 | posix_class = 0; |
3696 | |
3697 | /* When PCRE_UCP is set, some of the POSIX classes are converted to |
3698 | different escape sequences that use Unicode properties. */ |
3699 | |
3700 | #ifdef SUPPORT_UCP |
3701 | if ((options & PCRE_UCP) != 0) |
3702 | { |
3703 | int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0); |
3704 | if (posix_substitutes[pc] != NULL) |
3705 | { |
3706 | nestptr = tempptr + 1; |
3707 | ptr = posix_substitutes[pc] - 1; |
3708 | continue; |
3709 | } |
3710 | } |
3711 | #endif |
3712 | /* In the non-UCP case, we build the bit map for the POSIX class in a |
3713 | chunk of local store because we may be adding and subtracting from it, |
3714 | and we don't want to subtract bits that may be in the main map already. |
3715 | At the end we or the result into the bit map that is being built. */ |
3716 | |
3717 | posix_class *= 3; |
3718 | |
3719 | /* Copy in the first table (always present) */ |
3720 | |
3721 | memcpy(pbits, cbits + posix_class_maps[posix_class], |
3722 | 32 * sizeof(uschar)); |
3723 | |
3724 | /* If there is a second table, add or remove it as required. */ |
3725 | |
3726 | taboffset = posix_class_maps[posix_class + 1]; |
3727 | tabopt = posix_class_maps[posix_class + 2]; |
3728 | |
3729 | if (taboffset >= 0) |
3730 | { |
3731 | if (tabopt >= 0) |
3732 | for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset]; |
3733 | else |
3734 | for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset]; |
3735 | } |
3736 | |
3737 | /* Not see if we need to remove any special characters. An option |
3738 | value of 1 removes vertical space and 2 removes underscore. */ |
3739 | |
3740 | if (tabopt < 0) tabopt = -tabopt; |
3741 | if (tabopt == 1) pbits[1] &= ~0x3c; |
3742 | else if (tabopt == 2) pbits[11] &= 0x7f; |
3743 | |
3744 | /* Add the POSIX table or its complement into the main table that is |
3745 | being built and we are done. */ |
3746 | |
3747 | if (local_negate) |
3748 | for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c]; |
3749 | else |
3750 | for (c = 0; c < 32; c++) classbits[c] |= pbits[c]; |
3751 | |
3752 | ptr = tempptr + 1; |
3753 | class_charcount = 10; /* Set > 1; assumes more than 1 per class */ |
3754 | continue; /* End of POSIX syntax handling */ |
3755 | } |
3756 | |
3757 | /* Backslash may introduce a single character, or it may introduce one |
3758 | of the specials, which just set a flag. The sequence \b is a special |
3759 | case. Inside a class (and only there) it is treated as backspace. We |
3760 | assume that other escapes have more than one character in them, so set |
3761 | class_charcount bigger than one. Unrecognized escapes fall through and |
3762 | are either treated as literal characters (by default), or are faulted if |
3763 | PCRE_EXTRA is set. */ |
3764 | |
3765 | if (c == CHAR_BACKSLASH) |
3766 | { |
3767 | c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE); |
3768 | if (*errorcodeptr != 0) goto FAILED; |
3769 | |
3770 | if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */ |
3771 | else if (-c == ESC_Q) /* Handle start of quoted string */ |
3772 | { |
3773 | if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E) |
3774 | { |
3775 | ptr += 2; /* avoid empty string */ |
3776 | } |
3777 | else inescq = TRUE; |
3778 | continue; |
3779 | } |
3780 | else if (-c == ESC_E) continue; /* Ignore orphan \E */ |
3781 | |
3782 | if (c < 0) |
3783 | { |
3784 | register const uschar *cbits = cd->cbits; |
3785 | class_charcount += 2; /* Greater than 1 is what matters */ |
3786 | |
3787 | switch (-c) |
3788 | { |
3789 | #ifdef SUPPORT_UCP |
3790 | case ESC_du: /* These are the values given for \d etc */ |
3791 | case ESC_DU: /* when PCRE_UCP is set. We replace the */ |
3792 | case ESC_wu: /* escape sequence with an appropriate \p */ |
3793 | case ESC_WU: /* or \P to test Unicode properties instead */ |
3794 | case ESC_su: /* of the default ASCII testing. */ |
3795 | case ESC_SU: |
3796 | nestptr = ptr; |
3797 | ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */ |
3798 | class_charcount -= 2; /* Undo! */ |
3799 | continue; |
3800 | #endif |
3801 | case ESC_d: |
3802 | for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit]; |
3803 | continue; |
3804 | |
3805 | case ESC_D: |
3806 | should_flip_negation = TRUE; |
3807 | for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit]; |
3808 | continue; |
3809 | |
3810 | case ESC_w: |
3811 | for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word]; |
3812 | continue; |
3813 | |
3814 | case ESC_W: |
3815 | should_flip_negation = TRUE; |
3816 | for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word]; |
3817 | continue; |
3818 | |
3819 | /* Perl 5.004 onwards omits VT from \s, but we must preserve it |
3820 | if it was previously set by something earlier in the character |
3821 | class. */ |
3822 | |
3823 | case ESC_s: |
3824 | classbits[0] |= cbits[cbit_space]; |
3825 | classbits[1] |= cbits[cbit_space+1] & ~0x08; |
3826 | for (c = 2; c < 32; c++) classbits[c] |= cbits[c+cbit_space]; |
3827 | continue; |
3828 | |
3829 | case ESC_S: |
3830 | should_flip_negation = TRUE; |
3831 | for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space]; |
3832 | classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */ |
3833 | continue; |
3834 | |
3835 | case ESC_h: |
3836 | SETBIT(classbits, 0x09); /* VT */ |
3837 | SETBIT(classbits, 0x20); /* SPACE */ |
3838 | SETBIT(classbits, 0xa0); /* NSBP */ |
3839 | #ifdef SUPPORT_UTF8 |
3840 | if (utf8) |
3841 | { |
3842 | class_utf8 = TRUE; |
3843 | *class_utf8data++ = XCL_SINGLE; |
3844 | class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data); |
3845 | *class_utf8data++ = XCL_SINGLE; |
3846 | class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data); |
3847 | *class_utf8data++ = XCL_RANGE; |
3848 | class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data); |
3849 | class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data); |
3850 | *class_utf8data++ = XCL_SINGLE; |
3851 | class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data); |
3852 | *class_utf8data++ = XCL_SINGLE; |
3853 | class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data); |
3854 | *class_utf8data++ = XCL_SINGLE; |
3855 | class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data); |
3856 | } |
3857 | #endif |
3858 | continue; |
3859 | |
3860 | case ESC_H: |
3861 | for (c = 0; c < 32; c++) |
3862 | { |
3863 | int x = 0xff; |
3864 | switch (c) |
3865 | { |
3866 | case 0x09/8: x ^= 1 << (0x09%8); break; |
3867 | case 0x20/8: x ^= 1 << (0x20%8); break; |
3868 | case 0xa0/8: x ^= 1 << (0xa0%8); break; |
3869 | default: break; |
3870 | } |
3871 | classbits[c] |= x; |
3872 | } |
3873 | |
3874 | #ifdef SUPPORT_UTF8 |
3875 | if (utf8) |
3876 | { |
3877 | class_utf8 = TRUE; |
3878 | *class_utf8data++ = XCL_RANGE; |
3879 | class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data); |
3880 | class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data); |
3881 | *class_utf8data++ = XCL_RANGE; |
3882 | class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data); |
3883 | class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data); |
3884 | *class_utf8data++ = XCL_RANGE; |
3885 | class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data); |
3886 | class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data); |
3887 | *class_utf8data++ = XCL_RANGE; |
3888 | class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data); |
3889 | class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data); |
3890 | *class_utf8data++ = XCL_RANGE; |
3891 | class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data); |
3892 | class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data); |
3893 | *class_utf8data++ = XCL_RANGE; |
3894 | class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data); |
3895 | class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data); |
3896 | *class_utf8data++ = XCL_RANGE; |
3897 | class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data); |
3898 | class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data); |
3899 | } |
3900 | #endif |
3901 | continue; |
3902 | |
3903 | case ESC_v: |
3904 | SETBIT(classbits, 0x0a); /* LF */ |
3905 | SETBIT(classbits, 0x0b); /* VT */ |
3906 | SETBIT(classbits, 0x0c); /* FF */ |
3907 | SETBIT(classbits, 0x0d); /* CR */ |
3908 | SETBIT(classbits, 0x85); /* NEL */ |
3909 | #ifdef SUPPORT_UTF8 |
3910 | if (utf8) |
3911 | { |
3912 | class_utf8 = TRUE; |
3913 | *class_utf8data++ = XCL_RANGE; |
3914 | class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data); |
3915 | class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data); |
3916 | } |
3917 | #endif |
3918 | continue; |
3919 | |
3920 | case ESC_V: |
3921 | for (c = 0; c < 32; c++) |
3922 | { |
3923 | int x = 0xff; |
3924 | switch (c) |
3925 | { |
3926 | case 0x0a/8: x ^= 1 << (0x0a%8); |
3927 | x ^= 1 << (0x0b%8); |
3928 | x ^= 1 << (0x0c%8); |
3929 | x ^= 1 << (0x0d%8); |
3930 | break; |
3931 | case 0x85/8: x ^= 1 << (0x85%8); break; |
3932 | default: break; |
3933 | } |
3934 | classbits[c] |= x; |
3935 | } |
3936 | |
3937 | #ifdef SUPPORT_UTF8 |
3938 | if (utf8) |
3939 | { |
3940 | class_utf8 = TRUE; |
3941 | *class_utf8data++ = XCL_RANGE; |
3942 | class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data); |
3943 | class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data); |
3944 | *class_utf8data++ = XCL_RANGE; |
3945 | class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data); |
3946 | class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data); |
3947 | } |
3948 | #endif |
3949 | continue; |
3950 | |
3951 | #ifdef SUPPORT_UCP |
3952 | case ESC_p: |
3953 | case ESC_P: |
3954 | { |
3955 | BOOL negated; |
3956 | int pdata; |
3957 | int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr); |
3958 | if (ptype < 0) goto FAILED; |
3959 | class_utf8 = TRUE; |
3960 | *class_utf8data++ = ((-c == ESC_p) != negated)? |
3961 | XCL_PROP : XCL_NOTPROP; |
3962 | *class_utf8data++ = ptype; |
3963 | *class_utf8data++ = pdata; |
3964 | class_charcount -= 2; /* Not a < 256 character */ |
3965 | continue; |
3966 | } |
3967 | #endif |
3968 | /* Unrecognized escapes are faulted if PCRE is running in its |
3969 | strict mode. By default, for compatibility with Perl, they are |
3970 | treated as literals. */ |
3971 | |
3972 | default: |
3973 | if ((options & PCRE_EXTRA) != 0) |
3974 | { |
3975 | *errorcodeptr = ERR7; |
3976 | goto FAILED; |
3977 | } |
3978 | class_charcount -= 2; /* Undo the default count from above */ |
3979 | c = *ptr; /* Get the final character and fall through */ |
3980 | break; |
3981 | } |
3982 | } |
3983 | |
3984 | /* Fall through if we have a single character (c >= 0). This may be |
3985 | greater than 256 in UTF-8 mode. */ |
3986 | |
3987 | } /* End of backslash handling */ |
3988 | |
3989 | /* A single character may be followed by '-' to form a range. However, |
3990 | Perl does not permit ']' to be the end of the range. A '-' character |
3991 | at the end is treated as a literal. Perl ignores orphaned \E sequences |
3992 | entirely. The code for handling \Q and \E is messy. */ |
3993 | |
3994 | CHECK_RANGE: |
3995 | while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E) |
3996 | { |
3997 | inescq = FALSE; |
3998 | ptr += 2; |
3999 | } |
4000 | |
4001 | oldptr = ptr; |
4002 | |
4003 | /* Remember \r or \n */ |
4004 | |
4005 | if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF; |
4006 | |
4007 | /* Check for range */ |
4008 | |
4009 | if (!inescq && ptr[1] == CHAR_MINUS) |
4010 | { |
4011 | int d; |
4012 | ptr += 2; |
4013 | while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2; |
4014 | |
4015 | /* If we hit \Q (not followed by \E) at this point, go into escaped |
4016 | mode. */ |
4017 | |
4018 | while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q) |
4019 | { |
4020 | ptr += 2; |
4021 | if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) |
4022 | { ptr += 2; continue; } |
4023 | inescq = TRUE; |
4024 | break; |
4025 | } |
4026 | |
4027 | if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET)) |
4028 | { |
4029 | ptr = oldptr; |
4030 | goto LONE_SINGLE_CHARACTER; |
4031 | } |
4032 | |
4033 | #ifdef SUPPORT_UTF8 |
4034 | if (utf8) |
4035 | { /* Braces are required because the */ |
4036 | GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */ |
4037 | } |
4038 | else |
4039 | #endif |
4040 | d = *ptr; /* Not UTF-8 mode */ |
4041 | |
4042 | /* The second part of a range can be a single-character escape, but |
4043 | not any of the other escapes. Perl 5.6 treats a hyphen as a literal |
4044 | in such circumstances. */ |
4045 | |
4046 | if (!inescq && d == CHAR_BACKSLASH) |
4047 | { |
4048 | d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE); |
4049 | if (*errorcodeptr != 0) goto FAILED; |
4050 | |
4051 | /* \b is backspace; any other special means the '-' was literal */ |
4052 | |
4053 | if (d < 0) |
4054 | { |
4055 | if (d == -ESC_b) d = CHAR_BS; else |
4056 | { |
4057 | ptr = oldptr; |
4058 | goto LONE_SINGLE_CHARACTER; /* A few lines below */ |
4059 | } |
4060 | } |
4061 | } |
4062 | |
4063 | /* Check that the two values are in the correct order. Optimize |
4064 | one-character ranges */ |
4065 | |
4066 | if (d < c) |
4067 | { |
4068 | *errorcodeptr = ERR8; |
4069 | goto FAILED; |
4070 | } |
4071 | |
4072 | if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */ |
4073 | |
4074 | /* Remember \r or \n */ |
4075 | |
4076 | if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF; |
4077 | |
4078 | /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless |
4079 | matching, we have to use an XCLASS with extra data items. Caseless |
4080 | matching for characters > 127 is available only if UCP support is |
4081 | available. */ |
4082 | |
4083 | #ifdef SUPPORT_UTF8 |
4084 | if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127))) |
4085 | { |
4086 | class_utf8 = TRUE; |
4087 | |
4088 | /* With UCP support, we can find the other case equivalents of |
4089 | the relevant characters. There may be several ranges. Optimize how |
4090 | they fit with the basic range. */ |
4091 | |
4092 | #ifdef SUPPORT_UCP |
4093 | if ((options & PCRE_CASELESS) != 0) |
4094 | { |
4095 | unsigned int occ, ocd; |
4096 | unsigned int cc = c; |
4097 | unsigned int origd = d; |
4098 | while (get_othercase_range(&cc, origd, &occ, &ocd)) |
4099 | { |
4100 | if (occ >= (unsigned int)c && |
4101 | ocd <= (unsigned int)d) |
4102 | continue; /* Skip embedded ranges */ |
4103 | |
4104 | if (occ < (unsigned int)c && |
4105 | ocd >= (unsigned int)c - 1) /* Extend the basic range */ |
4106 | { /* if there is overlap, */ |
4107 | c = occ; /* noting that if occ < c */ |
4108 | continue; /* we can't have ocd > d */ |
4109 | } /* because a subrange is */ |
4110 | if (ocd > (unsigned int)d && |
4111 | occ <= (unsigned int)d + 1) /* always shorter than */ |
4112 | { /* the basic range. */ |
4113 | d = ocd; |
4114 | continue; |
4115 | } |
4116 | |
4117 | if (occ == ocd) |
4118 | { |
4119 | *class_utf8data++ = XCL_SINGLE; |
4120 | } |
4121 | else |
4122 | { |
4123 | *class_utf8data++ = XCL_RANGE; |
4124 | class_utf8data += _pcre_ord2utf8(occ, class_utf8data); |
4125 | } |
4126 | class_utf8data += _pcre_ord2utf8(ocd, class_utf8data); |
4127 | } |
4128 | } |
4129 | #endif /* SUPPORT_UCP */ |
4130 | |
4131 | /* Now record the original range, possibly modified for UCP caseless |
4132 | overlapping ranges. */ |
4133 | |
4134 | *class_utf8data++ = XCL_RANGE; |
4135 | class_utf8data += _pcre_ord2utf8(c, class_utf8data); |
4136 | class_utf8data += _pcre_ord2utf8(d, class_utf8data); |
4137 | |
4138 | /* With UCP support, we are done. Without UCP support, there is no |
4139 | caseless matching for UTF-8 characters > 127; we can use the bit map |
4140 | for the smaller ones. */ |
4141 | |
4142 | #ifdef SUPPORT_UCP |
4143 | continue; /* With next character in the class */ |
4144 | #else |
4145 | if ((options & PCRE_CASELESS) == 0 || c > 127) continue; |
4146 | |
4147 | /* Adjust upper limit and fall through to set up the map */ |
4148 | |
4149 | d = 127; |
4150 | |
4151 | #endif /* SUPPORT_UCP */ |
4152 | } |
4153 | #endif /* SUPPORT_UTF8 */ |
4154 | |
4155 | /* We use the bit map for all cases when not in UTF-8 mode; else |
4156 | ranges that lie entirely within 0-127 when there is UCP support; else |
4157 | for partial ranges without UCP support. */ |
4158 | |
4159 | class_charcount += d - c + 1; |
4160 | class_lastchar = d; |
4161 | |
4162 | /* We can save a bit of time by skipping this in the pre-compile. */ |
4163 | |
4164 | if (lengthptr == NULL) for (; c <= d; c++) |
4165 | { |
4166 | classbits[c/8] |= (1 << (c&7)); |
4167 | if ((options & PCRE_CASELESS) != 0) |
4168 | { |
4169 | int uc = cd->fcc[c]; /* flip case */ |
4170 | classbits[uc/8] |= (1 << (uc&7)); |
4171 | } |
4172 | } |
4173 | |
4174 | continue; /* Go get the next char in the class */ |
4175 | } |
4176 | |
4177 | /* Handle a lone single character - we can get here for a normal |
4178 | non-escape char, or after \ that introduces a single character or for an |
4179 | apparent range that isn't. */ |
4180 | |
4181 | LONE_SINGLE_CHARACTER: |
4182 | |
4183 | /* Handle a character that cannot go in the bit map */ |
4184 | |
4185 | #ifdef SUPPORT_UTF8 |
4186 | if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127))) |
4187 | { |
4188 | class_utf8 = TRUE; |
4189 | *class_utf8data++ = XCL_SINGLE; |
4190 | class_utf8data += _pcre_ord2utf8(c, class_utf8data); |
4191 | |
4192 | #ifdef SUPPORT_UCP |
4193 | if ((options & PCRE_CASELESS) != 0) |
4194 | { |
4195 | unsigned int othercase; |
4196 | if ((othercase = UCD_OTHERCASE(c)) != c) |
4197 | { |
4198 | *class_utf8data++ = XCL_SINGLE; |
4199 | class_utf8data += _pcre_ord2utf8(othercase, class_utf8data); |
4200 | } |
4201 | } |
4202 | #endif /* SUPPORT_UCP */ |
4203 | |
4204 | } |
4205 | else |
4206 | #endif /* SUPPORT_UTF8 */ |
4207 | |
4208 | /* Handle a single-byte character */ |
4209 | { |
4210 | classbits[c/8] |= (1 << (c&7)); |
4211 | if ((options & PCRE_CASELESS) != 0) |
4212 | { |
4213 | c = cd->fcc[c]; /* flip case */ |
4214 | classbits[c/8] |= (1 << (c&7)); |
4215 | } |
4216 | class_charcount++; |
4217 | class_lastchar = c; |
4218 | } |
4219 | } |
4220 | |
4221 | /* Loop until ']' reached. This "while" is the end of the "do" far above. |
4222 | If we are at the end of an internal nested string, revert to the outer |
4223 | string. */ |
4224 | |
4225 | while (((c = *(++ptr)) != 0 || |
4226 | (nestptr != NULL && |
4227 | (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) && |
4228 | (c != CHAR_RIGHT_SQUARE_BRACKET || inescq)); |
4229 | |
4230 | /* Check for missing terminating ']' */ |
4231 | |
4232 | if (c == 0) |
4233 | { |
4234 | *errorcodeptr = ERR6; |
4235 | goto FAILED; |
4236 | } |
4237 | |
4238 | /* If class_charcount is 1, we saw precisely one character whose value is |
4239 | less than 256. As long as there were no characters >= 128 and there was no |
4240 | use of \p or \P, in other words, no use of any XCLASS features, we can |
4241 | optimize. |
4242 | |
4243 | In UTF-8 mode, we can optimize the negative case only if there were no |
4244 | characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR |
4245 | operate on single-bytes characters only. This is an historical hangover. |
4246 | Maybe one day we can tidy these opcodes to handle multi-byte characters. |
4247 | |
4248 | The optimization throws away the bit map. We turn the item into a |
4249 | 1-character OP_CHAR[I] if it's positive, or OP_NOT[I] if it's negative. |
4250 | Note that OP_NOT[I] does not support multibyte characters. In the positive |
4251 | case, it can cause firstbyte to be set. Otherwise, there can be no first |
4252 | char if this item is first, whatever repeat count may follow. In the case |
4253 | of reqbyte, save the previous value for reinstating. */ |
4254 | |
4255 | #ifdef SUPPORT_UTF8 |
4256 | if (class_charcount == 1 && !class_utf8 && |
4257 | (!utf8 || !negate_class || class_lastchar < 128)) |
4258 | #else |
4259 | if (class_charcount == 1) |
4260 | #endif |
4261 | { |
4262 | zeroreqbyte = reqbyte; |
4263 | |
4264 | /* The OP_NOT[I] opcodes work on one-byte characters only. */ |
4265 | |
4266 | if (negate_class) |
4267 | { |
4268 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
4269 | zerofirstbyte = firstbyte; |
4270 | *code++ = ((options & PCRE_CASELESS) != 0)? OP_NOTI: OP_NOT; |
4271 | *code++ = class_lastchar; |
4272 | break; |
4273 | } |
4274 | |
4275 | /* For a single, positive character, get the value into mcbuffer, and |
4276 | then we can handle this with the normal one-character code. */ |
4277 | |
4278 | #ifdef SUPPORT_UTF8 |
4279 | if (utf8 && class_lastchar > 127) |
4280 | mclength = _pcre_ord2utf8(class_lastchar, mcbuffer); |
4281 | else |
4282 | #endif |
4283 | { |
4284 | mcbuffer[0] = class_lastchar; |
4285 | mclength = 1; |
4286 | } |
4287 | goto ONE_CHAR; |
4288 | } /* End of 1-char optimization */ |
4289 | |
4290 | /* The general case - not the one-char optimization. If this is the first |
4291 | thing in the branch, there can be no first char setting, whatever the |
4292 | repeat count. Any reqbyte setting must remain unchanged after any kind of |
4293 | repeat. */ |
4294 | |
4295 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
4296 | zerofirstbyte = firstbyte; |
4297 | zeroreqbyte = reqbyte; |
4298 | |
4299 | /* If there are characters with values > 255, we have to compile an |
4300 | extended class, with its own opcode, unless there was a negated special |
4301 | such as \S in the class, and PCRE_UCP is not set, because in that case all |
4302 | characters > 255 are in the class, so any that were explicitly given as |
4303 | well can be ignored. If (when there are explicit characters > 255 that must |
4304 | be listed) there are no characters < 256, we can omit the bitmap in the |
4305 | actual compiled code. */ |
4306 | |
4307 | #ifdef SUPPORT_UTF8 |
4308 | if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0)) |
4309 | { |
4310 | *class_utf8data++ = XCL_END; /* Marks the end of extra data */ |
4311 | *code++ = OP_XCLASS; |
4312 | code += LINK_SIZE; |
4313 | *code = negate_class? XCL_NOT : 0; |
4314 | |
4315 | /* If the map is required, move up the extra data to make room for it; |
4316 | otherwise just move the code pointer to the end of the extra data. */ |
4317 | |
4318 | if (class_charcount > 0) |
4319 | { |
4320 | *code++ |= XCL_MAP; |
4321 | memmove(code + 32, code, class_utf8data - code); |
4322 | memcpy(code, classbits, 32); |
4323 | code = class_utf8data + 32; |
4324 | } |
4325 | else code = class_utf8data; |
4326 | |
4327 | /* Now fill in the complete length of the item */ |
4328 | |
4329 | PUT(previous, 1, code - previous); |
4330 | break; /* End of class handling */ |
4331 | } |
4332 | #endif |
4333 | |
4334 | /* If there are no characters > 255, or they are all to be included or |
4335 | excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the |
4336 | whole class was negated and whether there were negative specials such as \S |
4337 | (non-UCP) in the class. Then copy the 32-byte map into the code vector, |
4338 | negating it if necessary. */ |
4339 | |
4340 | *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS; |
4341 | if (negate_class) |
4342 | { |
4343 | if (lengthptr == NULL) /* Save time in the pre-compile phase */ |
4344 | for (c = 0; c < 32; c++) code[c] = ~classbits[c]; |
4345 | } |
4346 | else |
4347 | { |
4348 | memcpy(code, classbits, 32); |
4349 | } |
4350 | code += 32; |
4351 | break; |
4352 | |
4353 | |
4354 | /* ===================================================================*/ |
4355 | /* Various kinds of repeat; '{' is not necessarily a quantifier, but this |
4356 | has been tested above. */ |
4357 | |
4358 | case CHAR_LEFT_CURLY_BRACKET: |
4359 | if (!is_quantifier) goto NORMAL_CHAR; |
4360 | ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr); |
4361 | if (*errorcodeptr != 0) goto FAILED; |
4362 | goto REPEAT; |
4363 | |
4364 | case CHAR_ASTERISK: |
4365 | repeat_min = 0; |
4366 | repeat_max = -1; |
4367 | goto REPEAT; |
4368 | |
4369 | case CHAR_PLUS: |
4370 | repeat_min = 1; |
4371 | repeat_max = -1; |
4372 | goto REPEAT; |
4373 | |
4374 | case CHAR_QUESTION_MARK: |
4375 | repeat_min = 0; |
4376 | repeat_max = 1; |
4377 | |
4378 | REPEAT: |
4379 | if (previous == NULL) |
4380 | { |
4381 | *errorcodeptr = ERR9; |
4382 | goto FAILED; |
4383 | } |
4384 | |
4385 | if (repeat_min == 0) |
4386 | { |
4387 | firstbyte = zerofirstbyte; /* Adjust for zero repeat */ |
4388 | reqbyte = zeroreqbyte; /* Ditto */ |
4389 | } |
4390 | |
4391 | /* Remember whether this is a variable length repeat */ |
4392 | |
4393 | reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY; |
4394 | |
4395 | op_type = 0; /* Default single-char op codes */ |
4396 | possessive_quantifier = FALSE; /* Default not possessive quantifier */ |
4397 | |
4398 | /* Save start of previous item, in case we have to move it up in order to |
4399 | insert something before it. */ |
4400 | |
4401 | tempcode = previous; |
4402 | |
4403 | /* If the next character is '+', we have a possessive quantifier. This |
4404 | implies greediness, whatever the setting of the PCRE_UNGREEDY option. |
4405 | If the next character is '?' this is a minimizing repeat, by default, |
4406 | but if PCRE_UNGREEDY is set, it works the other way round. We change the |
4407 | repeat type to the non-default. */ |
4408 | |
4409 | if (ptr[1] == CHAR_PLUS) |
4410 | { |
4411 | repeat_type = 0; /* Force greedy */ |
4412 | possessive_quantifier = TRUE; |
4413 | ptr++; |
4414 | } |
4415 | else if (ptr[1] == CHAR_QUESTION_MARK) |
4416 | { |
4417 | repeat_type = greedy_non_default; |
4418 | ptr++; |
4419 | } |
4420 | else repeat_type = greedy_default; |
4421 | |
4422 | /* If previous was a recursion call, wrap it in atomic brackets so that |
4423 | previous becomes the atomic group. All recursions were so wrapped in the |
4424 | past, but it no longer happens for non-repeated recursions. In fact, the |
4425 | repeated ones could be re-implemented independently so as not to need this, |
4426 | but for the moment we rely on the code for repeating groups. */ |
4427 | |
4428 | if (*previous == OP_RECURSE) |
4429 | { |
4430 | memmove(previous + 1 + LINK_SIZE, previous, 1 + LINK_SIZE); |
4431 | *previous = OP_ONCE; |
4432 | PUT(previous, 1, 2 + 2*LINK_SIZE); |
4433 | previous[2 + 2*LINK_SIZE] = OP_KET; |
4434 | PUT(previous, 3 + 2*LINK_SIZE, 2 + 2*LINK_SIZE); |
4435 | code += 2 + 2 * LINK_SIZE; |
4436 | length_prevgroup = 3 + 3*LINK_SIZE; |
4437 | |
4438 | /* When actually compiling, we need to check whether this was a forward |
4439 | reference, and if so, adjust the offset. */ |
4440 | |
4441 | if (lengthptr == NULL && cd->hwm >= cd->start_workspace + LINK_SIZE) |
4442 | { |
4443 | int offset = GET(cd->hwm, -LINK_SIZE); |
4444 | if (offset == previous + 1 - cd->start_code) |
4445 | PUT(cd->hwm, -LINK_SIZE, offset + 1 + LINK_SIZE); |
4446 | } |
4447 | } |
4448 | |
4449 | /* Now handle repetition for the different types of item. */ |
4450 | |
4451 | /* If previous was a character match, abolish the item and generate a |
4452 | repeat item instead. If a char item has a minumum of more than one, ensure |
4453 | that it is set in reqbyte - it might not be if a sequence such as x{3} is |
4454 | the first thing in a branch because the x will have gone into firstbyte |
4455 | instead. */ |
4456 | |
4457 | if (*previous == OP_CHAR || *previous == OP_CHARI) |
4458 | { |
4459 | op_type = (*previous == OP_CHAR)? 0 : OP_STARI - OP_STAR; |
4460 | |
4461 | /* Deal with UTF-8 characters that take up more than one byte. It's |
4462 | easier to write this out separately than try to macrify it. Use c to |
4463 | hold the length of the character in bytes, plus 0x80 to flag that it's a |
4464 | length rather than a small character. */ |
4465 | |
4466 | #ifdef SUPPORT_UTF8 |
4467 | if (utf8 && (code[-1] & 0x80) != 0) |
4468 | { |
4469 | uschar *lastchar = code - 1; |
4470 | while((*lastchar & 0xc0) == 0x80) lastchar--; |
4471 | c = code - lastchar; /* Length of UTF-8 character */ |
4472 | memcpy(utf8_char, lastchar, c); /* Save the char */ |
4473 | c |= 0x80; /* Flag c as a length */ |
4474 | } |
4475 | else |
4476 | #endif |
4477 | |
4478 | /* Handle the case of a single byte - either with no UTF8 support, or |
4479 | with UTF-8 disabled, or for a UTF-8 character < 128. */ |
4480 | |
4481 | { |
4482 | c = code[-1]; |
4483 | if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt; |
4484 | } |
4485 | |
4486 | /* If the repetition is unlimited, it pays to see if the next thing on |
4487 | the line is something that cannot possibly match this character. If so, |
4488 | automatically possessifying this item gains some performance in the case |
4489 | where the match fails. */ |
4490 | |
4491 | if (!possessive_quantifier && |
4492 | repeat_max < 0 && |
4493 | check_auto_possessive(previous, utf8, ptr + 1, options, cd)) |
4494 | { |
4495 | repeat_type = 0; /* Force greedy */ |
4496 | possessive_quantifier = TRUE; |
4497 | } |
4498 | |
4499 | goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */ |
4500 | } |
4501 | |
4502 | /* If previous was a single negated character ([^a] or similar), we use |
4503 | one of the special opcodes, replacing it. The code is shared with single- |
4504 | character repeats by setting opt_type to add a suitable offset into |
4505 | repeat_type. We can also test for auto-possessification. OP_NOT and OP_NOTI |
4506 | are currently used only for single-byte chars. */ |
4507 | |
4508 | else if (*previous == OP_NOT || *previous == OP_NOTI) |
4509 | { |
4510 | op_type = ((*previous == OP_NOT)? OP_NOTSTAR : OP_NOTSTARI) - OP_STAR; |
4511 | c = previous[1]; |
4512 | if (!possessive_quantifier && |
4513 | repeat_max < 0 && |
4514 | check_auto_possessive(previous, utf8, ptr + 1, options, cd)) |
4515 | { |
4516 | repeat_type = 0; /* Force greedy */ |
4517 | possessive_quantifier = TRUE; |
4518 | } |
4519 | goto OUTPUT_SINGLE_REPEAT; |
4520 | } |
4521 | |
4522 | /* If previous was a character type match (\d or similar), abolish it and |
4523 | create a suitable repeat item. The code is shared with single-character |
4524 | repeats by setting op_type to add a suitable offset into repeat_type. Note |
4525 | the the Unicode property types will be present only when SUPPORT_UCP is |
4526 | defined, but we don't wrap the little bits of code here because it just |
4527 | makes it horribly messy. */ |
4528 | |
4529 | else if (*previous < OP_EODN) |
4530 | { |
4531 | uschar *oldcode; |
4532 | int prop_type, prop_value; |
4533 | op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */ |
4534 | c = *previous; |
4535 | |
4536 | if (!possessive_quantifier && |
4537 | repeat_max < 0 && |
4538 | check_auto_possessive(previous, utf8, ptr + 1, options, cd)) |
4539 | { |
4540 | repeat_type = 0; /* Force greedy */ |
4541 | possessive_quantifier = TRUE; |
4542 | } |
4543 | |
4544 | OUTPUT_SINGLE_REPEAT: |
4545 | if (*previous == OP_PROP || *previous == OP_NOTPROP) |
4546 | { |
4547 | prop_type = previous[1]; |
4548 | prop_value = previous[2]; |
4549 | } |
4550 | else prop_type = prop_value = -1; |
4551 | |
4552 | oldcode = code; |
4553 | code = previous; /* Usually overwrite previous item */ |
4554 | |
4555 | /* If the maximum is zero then the minimum must also be zero; Perl allows |
4556 | this case, so we do too - by simply omitting the item altogether. */ |
4557 | |
4558 | if (repeat_max == 0) goto END_REPEAT; |
4559 | |
4560 | /*--------------------------------------------------------------------*/ |
4561 | /* This code is obsolete from release 8.00; the restriction was finally |
4562 | removed: */ |
4563 | |
4564 | /* All real repeats make it impossible to handle partial matching (maybe |
4565 | one day we will be able to remove this restriction). */ |
4566 | |
4567 | /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */ |
4568 | /*--------------------------------------------------------------------*/ |
4569 | |
4570 | /* Combine the op_type with the repeat_type */ |
4571 | |
4572 | repeat_type += op_type; |
4573 | |
4574 | /* A minimum of zero is handled either as the special case * or ?, or as |
4575 | an UPTO, with the maximum given. */ |
4576 | |
4577 | if (repeat_min == 0) |
4578 | { |
4579 | if (repeat_max == -1) *code++ = OP_STAR + repeat_type; |
4580 | else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type; |
4581 | else |
4582 | { |
4583 | *code++ = OP_UPTO + repeat_type; |
4584 | PUT2INC(code, 0, repeat_max); |
4585 | } |
4586 | } |
4587 | |
4588 | /* A repeat minimum of 1 is optimized into some special cases. If the |
4589 | maximum is unlimited, we use OP_PLUS. Otherwise, the original item is |
4590 | left in place and, if the maximum is greater than 1, we use OP_UPTO with |
4591 | one less than the maximum. */ |
4592 | |
4593 | else if (repeat_min == 1) |
4594 | { |
4595 | if (repeat_max == -1) |
4596 | *code++ = OP_PLUS + repeat_type; |
4597 | else |
4598 | { |
4599 | code = oldcode; /* leave previous item in place */ |
4600 | if (repeat_max == 1) goto END_REPEAT; |
4601 | *code++ = OP_UPTO + repeat_type; |
4602 | PUT2INC(code, 0, repeat_max - 1); |
4603 | } |
4604 | } |
4605 | |
4606 | /* The case {n,n} is just an EXACT, while the general case {n,m} is |
4607 | handled as an EXACT followed by an UPTO. */ |
4608 | |
4609 | else |
4610 | { |
4611 | *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */ |
4612 | PUT2INC(code, 0, repeat_min); |
4613 | |
4614 | /* If the maximum is unlimited, insert an OP_STAR. Before doing so, |
4615 | we have to insert the character for the previous code. For a repeated |
4616 | Unicode property match, there are two extra bytes that define the |
4617 | required property. In UTF-8 mode, long characters have their length in |
4618 | c, with the 0x80 bit as a flag. */ |
4619 | |
4620 | if (repeat_max < 0) |
4621 | { |
4622 | #ifdef SUPPORT_UTF8 |
4623 | if (utf8 && c >= 128) |
4624 | { |
4625 | memcpy(code, utf8_char, c & 7); |
4626 | code += c & 7; |
4627 | } |
4628 | else |
4629 | #endif |
4630 | { |
4631 | *code++ = c; |
4632 | if (prop_type >= 0) |
4633 | { |
4634 | *code++ = prop_type; |
4635 | *code++ = prop_value; |
4636 | } |
4637 | } |
4638 | *code++ = OP_STAR + repeat_type; |
4639 | } |
4640 | |
4641 | /* Else insert an UPTO if the max is greater than the min, again |
4642 | preceded by the character, for the previously inserted code. If the |
4643 | UPTO is just for 1 instance, we can use QUERY instead. */ |
4644 | |
4645 | else if (repeat_max != repeat_min) |
4646 | { |
4647 | #ifdef SUPPORT_UTF8 |
4648 | if (utf8 && c >= 128) |
4649 | { |
4650 | memcpy(code, utf8_char, c & 7); |
4651 | code += c & 7; |
4652 | } |
4653 | else |
4654 | #endif |
4655 | *code++ = c; |
4656 | if (prop_type >= 0) |
4657 | { |
4658 | *code++ = prop_type; |
4659 | *code++ = prop_value; |
4660 | } |
4661 | repeat_max -= repeat_min; |
4662 | |
4663 | if (repeat_max == 1) |
4664 | { |
4665 | *code++ = OP_QUERY + repeat_type; |
4666 | } |
4667 | else |
4668 | { |
4669 | *code++ = OP_UPTO + repeat_type; |
4670 | PUT2INC(code, 0, repeat_max); |
4671 | } |
4672 | } |
4673 | } |
4674 | |
4675 | /* The character or character type itself comes last in all cases. */ |
4676 | |
4677 | #ifdef SUPPORT_UTF8 |
4678 | if (utf8 && c >= 128) |
4679 | { |
4680 | memcpy(code, utf8_char, c & 7); |
4681 | code += c & 7; |
4682 | } |
4683 | else |
4684 | #endif |
4685 | *code++ = c; |
4686 | |
4687 | /* For a repeated Unicode property match, there are two extra bytes that |
4688 | define the required property. */ |
4689 | |
4690 | #ifdef SUPPORT_UCP |
4691 | if (prop_type >= 0) |
4692 | { |
4693 | *code++ = prop_type; |
4694 | *code++ = prop_value; |
4695 | } |
4696 | #endif |
4697 | } |
4698 | |
4699 | /* If previous was a character class or a back reference, we put the repeat |
4700 | stuff after it, but just skip the item if the repeat was {0,0}. */ |
4701 | |
4702 | else if (*previous == OP_CLASS || |
4703 | *previous == OP_NCLASS || |
4704 | #ifdef SUPPORT_UTF8 |
4705 | *previous == OP_XCLASS || |
4706 | #endif |
4707 | *previous == OP_REF || |
4708 | *previous == OP_REFI) |
4709 | { |
4710 | if (repeat_max == 0) |
4711 | { |
4712 | code = previous; |
4713 | goto END_REPEAT; |
4714 | } |
4715 | |
4716 | /*--------------------------------------------------------------------*/ |
4717 | /* This code is obsolete from release 8.00; the restriction was finally |
4718 | removed: */ |
4719 | |
4720 | /* All real repeats make it impossible to handle partial matching (maybe |
4721 | one day we will be able to remove this restriction). */ |
4722 | |
4723 | /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */ |
4724 | /*--------------------------------------------------------------------*/ |
4725 | |
4726 | if (repeat_min == 0 && repeat_max == -1) |
4727 | *code++ = OP_CRSTAR + repeat_type; |
4728 | else if (repeat_min == 1 && repeat_max == -1) |
4729 | *code++ = OP_CRPLUS + repeat_type; |
4730 | else if (repeat_min == 0 && repeat_max == 1) |
4731 | *code++ = OP_CRQUERY + repeat_type; |
4732 | else |
4733 | { |
4734 | *code++ = OP_CRRANGE + repeat_type; |
4735 | PUT2INC(code, 0, repeat_min); |
4736 | if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */ |
4737 | PUT2INC(code, 0, repeat_max); |
4738 | } |
4739 | } |
4740 | |
4741 | /* If previous was a bracket group, we may have to replicate it in certain |
4742 | cases. Note that at this point we can encounter only the "basic" bracket |
4743 | opcodes such as BRA and CBRA, as this is the place where they get converted |
4744 | into the more special varieties such as BRAPOS and SBRA. A test for >= |
4745 | OP_ASSERT and <= OP_COND includes ASSERT, ASSERT_NOT, ASSERTBACK, |
4746 | ASSERTBACK_NOT, ONCE, BRA, CBRA, and COND. Originally, PCRE did not allow |
4747 | repetition of assertions, but now it does, for Perl compatibility. */ |
4748 | |
4749 | else if (*previous >= OP_ASSERT && *previous <= OP_COND) |
4750 | { |
4751 | register int i; |
4752 | int len = (int)(code - previous); |
4753 | uschar *bralink = NULL; |
4754 | uschar *brazeroptr = NULL; |
4755 | |
4756 | /* Repeating a DEFINE group is pointless, but Perl allows the syntax, so |
4757 | we just ignore the repeat. */ |
4758 | |
4759 | if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF) |
4760 | goto END_REPEAT; |
4761 | |
4762 | /* There is no sense in actually repeating assertions. The only potential |
4763 | use of repetition is in cases when the assertion is optional. Therefore, |
4764 | if the minimum is greater than zero, just ignore the repeat. If the |
4765 | maximum is not not zero or one, set it to 1. */ |
4766 | |
4767 | if (*previous < OP_ONCE) /* Assertion */ |
4768 | { |
4769 | if (repeat_min > 0) goto END_REPEAT; |
4770 | if (repeat_max < 0 || repeat_max > 1) repeat_max = 1; |
4771 | } |
4772 | |
4773 | /* The case of a zero minimum is special because of the need to stick |
4774 | OP_BRAZERO in front of it, and because the group appears once in the |
4775 | data, whereas in other cases it appears the minimum number of times. For |
4776 | this reason, it is simplest to treat this case separately, as otherwise |
4777 | the code gets far too messy. There are several special subcases when the |
4778 | minimum is zero. */ |
4779 | |
4780 | if (repeat_min == 0) |
4781 | { |
4782 | /* If the maximum is also zero, we used to just omit the group from the |
4783 | output altogether, like this: |
4784 | |
4785 | ** if (repeat_max == 0) |
4786 | ** { |
4787 | ** code = previous; |
4788 | ** goto END_REPEAT; |
4789 | ** } |
4790 | |
4791 | However, that fails when a group or a subgroup within it is referenced |
4792 | as a subroutine from elsewhere in the pattern, so now we stick in |
4793 | OP_SKIPZERO in front of it so that it is skipped on execution. As we |
4794 | don't have a list of which groups are referenced, we cannot do this |
4795 | selectively. |
4796 | |
4797 | If the maximum is 1 or unlimited, we just have to stick in the BRAZERO |
4798 | and do no more at this point. However, we do need to adjust any |
4799 | OP_RECURSE calls inside the group that refer to the group itself or any |
4800 | internal or forward referenced group, because the offset is from the |
4801 | start of the whole regex. Temporarily terminate the pattern while doing |
4802 | this. */ |
4803 | |
4804 | if (repeat_max <= 1) /* Covers 0, 1, and unlimited */ |
4805 | { |
4806 | *code = OP_END; |
4807 | adjust_recurse(previous, 1, utf8, cd, save_hwm); |
4808 | memmove(previous+1, previous, len); |
4809 | code++; |
4810 | if (repeat_max == 0) |
4811 | { |
4812 | *previous++ = OP_SKIPZERO; |
4813 | goto END_REPEAT; |
4814 | } |
4815 | brazeroptr = previous; /* Save for possessive optimizing */ |
4816 | *previous++ = OP_BRAZERO + repeat_type; |
4817 | } |
4818 | |
4819 | /* If the maximum is greater than 1 and limited, we have to replicate |
4820 | in a nested fashion, sticking OP_BRAZERO before each set of brackets. |
4821 | The first one has to be handled carefully because it's the original |
4822 | copy, which has to be moved up. The remainder can be handled by code |
4823 | that is common with the non-zero minimum case below. We have to |
4824 | adjust the value or repeat_max, since one less copy is required. Once |
4825 | again, we may have to adjust any OP_RECURSE calls inside the group. */ |
4826 | |
4827 | else |
4828 | { |
4829 | int offset; |
4830 | *code = OP_END; |
4831 | adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm); |
4832 | memmove(previous + 2 + LINK_SIZE, previous, len); |
4833 | code += 2 + LINK_SIZE; |
4834 | *previous++ = OP_BRAZERO + repeat_type; |
4835 | *previous++ = OP_BRA; |
4836 | |
4837 | /* We chain together the bracket offset fields that have to be |
4838 | filled in later when the ends of the brackets are reached. */ |
4839 | |
4840 | offset = (bralink == NULL)? 0 : (int)(previous - bralink); |
4841 | bralink = previous; |
4842 | PUTINC(previous, 0, offset); |
4843 | } |
4844 | |
4845 | repeat_max--; |
4846 | } |
4847 | |
4848 | /* If the minimum is greater than zero, replicate the group as many |
4849 | times as necessary, and adjust the maximum to the number of subsequent |
4850 | copies that we need. If we set a first char from the group, and didn't |
4851 | set a required char, copy the latter from the former. If there are any |
4852 | forward reference subroutine calls in the group, there will be entries on |
4853 | the workspace list; replicate these with an appropriate increment. */ |
4854 | |
4855 | else |
4856 | { |
4857 | if (repeat_min > 1) |
4858 | { |
4859 | /* In the pre-compile phase, we don't actually do the replication. We |
4860 | just adjust the length as if we had. Do some paranoid checks for |
4861 | potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit |
4862 | integer type when available, otherwise double. */ |
4863 | |
4864 | if (lengthptr != NULL) |
4865 | { |
4866 | int delta = (repeat_min - 1)*length_prevgroup; |
4867 | if ((INT64_OR_DOUBLE)(repeat_min - 1)* |
4868 | (INT64_OR_DOUBLE)length_prevgroup > |
4869 | (INT64_OR_DOUBLE)INT_MAX || |
4870 | OFLOW_MAX - *lengthptr < delta) |
4871 | { |
4872 | *errorcodeptr = ERR20; |
4873 | goto FAILED; |
4874 | } |
4875 | *lengthptr += delta; |
4876 | } |
4877 | |
4878 | /* This is compiling for real */ |
4879 | |
4880 | else |
4881 | { |
4882 | if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte; |
4883 | for (i = 1; i < repeat_min; i++) |
4884 | { |
4885 | uschar *hc; |
4886 | uschar *this_hwm = cd->hwm; |
4887 | memcpy(code, previous, len); |
4888 | for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE) |
4889 | { |
4890 | PUT(cd->hwm, 0, GET(hc, 0) + len); |
4891 | cd->hwm += LINK_SIZE; |
4892 | } |
4893 | save_hwm = this_hwm; |
4894 | code += len; |
4895 | } |
4896 | } |
4897 | } |
4898 | |
4899 | if (repeat_max > 0) repeat_max -= repeat_min; |
4900 | } |
4901 | |
4902 | /* This code is common to both the zero and non-zero minimum cases. If |
4903 | the maximum is limited, it replicates the group in a nested fashion, |
4904 | remembering the bracket starts on a stack. In the case of a zero minimum, |
4905 | the first one was set up above. In all cases the repeat_max now specifies |
4906 | the number of additional copies needed. Again, we must remember to |
4907 | replicate entries on the forward reference list. */ |
4908 | |
4909 | if (repeat_max >= 0) |
4910 | { |
4911 | /* In the pre-compile phase, we don't actually do the replication. We |
4912 | just adjust the length as if we had. For each repetition we must add 1 |
4913 | to the length for BRAZERO and for all but the last repetition we must |
4914 | add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some |
4915 | paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is |
4916 | a 64-bit integer type when available, otherwise double. */ |
4917 | |
4918 | if (lengthptr != NULL && repeat_max > 0) |
4919 | { |
4920 | int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) - |
4921 | 2 - 2*LINK_SIZE; /* Last one doesn't nest */ |
4922 | if ((INT64_OR_DOUBLE)repeat_max * |
4923 | (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE) |
4924 | > (INT64_OR_DOUBLE)INT_MAX || |
4925 | OFLOW_MAX - *lengthptr < delta) |
4926 | { |
4927 | *errorcodeptr = ERR20; |
4928 | goto FAILED; |
4929 | } |
4930 | *lengthptr += delta; |
4931 | } |
4932 | |
4933 | /* This is compiling for real */ |
4934 | |
4935 | else for (i = repeat_max - 1; i >= 0; i--) |
4936 | { |
4937 | uschar *hc; |
4938 | uschar *this_hwm = cd->hwm; |
4939 | |
4940 | *code++ = OP_BRAZERO + repeat_type; |
4941 | |
4942 | /* All but the final copy start a new nesting, maintaining the |
4943 | chain of brackets outstanding. */ |
4944 | |
4945 | if (i != 0) |
4946 | { |
4947 | int offset; |
4948 | *code++ = OP_BRA; |
4949 | offset = (bralink == NULL)? 0 : (int)(code - bralink); |
4950 | bralink = code; |
4951 | PUTINC(code, 0, offset); |
4952 | } |
4953 | |
4954 | memcpy(code, previous, len); |
4955 | for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE) |
4956 | { |
4957 | PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1)); |
4958 | cd->hwm += LINK_SIZE; |
4959 | } |
4960 | save_hwm = this_hwm; |
4961 | code += len; |
4962 | } |
4963 | |
4964 | /* Now chain through the pending brackets, and fill in their length |
4965 | fields (which are holding the chain links pro tem). */ |
4966 | |
4967 | while (bralink != NULL) |
4968 | { |
4969 | int oldlinkoffset; |
4970 | int offset = (int)(code - bralink + 1); |
4971 | uschar *bra = code - offset; |
4972 | oldlinkoffset = GET(bra, 1); |
4973 | bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset; |
4974 | *code++ = OP_KET; |
4975 | PUTINC(code, 0, offset); |
4976 | PUT(bra, 1, offset); |
4977 | } |
4978 | } |
4979 | |
4980 | /* If the maximum is unlimited, set a repeater in the final copy. For |
4981 | ONCE brackets, that's all we need to do. However, possessively repeated |
4982 | ONCE brackets can be converted into non-capturing brackets, as the |
4983 | behaviour of (?:xx)++ is the same as (?>xx)++ and this saves having to |
4984 | deal with possessive ONCEs specially. |
4985 | |
4986 | Otherwise, when we are doing the actual compile phase, check to see |
4987 | whether this group is one that could match an empty string. If so, |
4988 | convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so |
4989 | that runtime checking can be done. [This check is also applied to ONCE |
4990 | groups at runtime, but in a different way.] |
4991 | |
4992 | Then, if the quantifier was possessive and the bracket is not a |
4993 | conditional, we convert the BRA code to the POS form, and the KET code to |
4994 | KETRPOS. (It turns out to be convenient at runtime to detect this kind of |
4995 | subpattern at both the start and at the end.) The use of special opcodes |
4996 | makes it possible to reduce greatly the stack usage in pcre_exec(). If |
4997 | the group is preceded by OP_BRAZERO, convert this to OP_BRAPOSZERO. |
4998 | |
4999 | Then, if the minimum number of matches is 1 or 0, cancel the possessive |
5000 | flag so that the default action below, of wrapping everything inside |
5001 | atomic brackets, does not happen. When the minimum is greater than 1, |
5002 | there will be earlier copies of the group, and so we still have to wrap |
5003 | the whole thing. */ |
5004 | |
5005 | else |
5006 | { |
5007 | uschar *ketcode = code - 1 - LINK_SIZE; |
5008 | uschar *bracode = ketcode - GET(ketcode, 1); |
5009 | |
5010 | /* Convert possessive ONCE brackets to non-capturing */ |
5011 | |
5012 | if ((*bracode == OP_ONCE || *bracode == OP_ONCE_NC) && |
5013 | possessive_quantifier) *bracode = OP_BRA; |
5014 | |
5015 | /* For non-possessive ONCE brackets, all we need to do is to |
5016 | set the KET. */ |
5017 | |
5018 | if (*bracode == OP_ONCE || *bracode == OP_ONCE_NC) |
5019 | *ketcode = OP_KETRMAX + repeat_type; |
5020 | |
5021 | /* Handle non-ONCE brackets and possessive ONCEs (which have been |
5022 | converted to non-capturing above). */ |
5023 | |
5024 | else |
5025 | { |
5026 | /* In the compile phase, check for empty string matching. */ |
5027 | |
5028 | if (lengthptr == NULL) |
5029 | { |
5030 | uschar *scode = bracode; |
5031 | do |
5032 | { |
5033 | if (could_be_empty_branch(scode, ketcode, utf8, cd)) |
5034 | { |
5035 | *bracode += OP_SBRA - OP_BRA; |
5036 | break; |
5037 | } |
5038 | scode += GET(scode, 1); |
5039 | } |
5040 | while (*scode == OP_ALT); |
5041 | } |
5042 | |
5043 | /* Handle possessive quantifiers. */ |
5044 | |
5045 | if (possessive_quantifier) |
5046 | { |
5047 | /* For COND brackets, we wrap the whole thing in a possessively |
5048 | repeated non-capturing bracket, because we have not invented POS |
5049 | versions of the COND opcodes. Because we are moving code along, we |
5050 | must ensure that any pending recursive references are updated. */ |
5051 | |
5052 | if (*bracode == OP_COND || *bracode == OP_SCOND) |
5053 | { |
5054 | int nlen = (int)(code - bracode); |
5055 | *code = OP_END; |
5056 | adjust_recurse(bracode, 1 + LINK_SIZE, utf8, cd, save_hwm); |
5057 | memmove(bracode + 1+LINK_SIZE, bracode, nlen); |
5058 | code += 1 + LINK_SIZE; |
5059 | nlen += 1 + LINK_SIZE; |
5060 | *bracode = OP_BRAPOS; |
5061 | *code++ = OP_KETRPOS; |
5062 | PUTINC(code, 0, nlen); |
5063 | PUT(bracode, 1, nlen); |
5064 | } |
5065 | |
5066 | /* For non-COND brackets, we modify the BRA code and use KETRPOS. */ |
5067 | |
5068 | else |
5069 | { |
5070 | *bracode += 1; /* Switch to xxxPOS opcodes */ |
5071 | *ketcode = OP_KETRPOS; |
5072 | } |
5073 | |
5074 | /* If the minimum is zero, mark it as possessive, then unset the |
5075 | possessive flag when the minimum is 0 or 1. */ |
5076 | |
5077 | if (brazeroptr != NULL) *brazeroptr = OP_BRAPOSZERO; |
5078 | if (repeat_min < 2) possessive_quantifier = FALSE; |
5079 | } |
5080 | |
5081 | /* Non-possessive quantifier */ |
5082 | |
5083 | else *ketcode = OP_KETRMAX + repeat_type; |
5084 | } |
5085 | } |
5086 | } |
5087 | |
5088 | /* If previous is OP_FAIL, it was generated by an empty class [] in |
5089 | JavaScript mode. The other ways in which OP_FAIL can be generated, that is |
5090 | by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat" |
5091 | error above. We can just ignore the repeat in JS case. */ |
5092 | |
5093 | else if (*previous == OP_FAIL) goto END_REPEAT; |
5094 | |
5095 | /* Else there's some kind of shambles */ |
5096 | |
5097 | else |
5098 | { |
5099 | *errorcodeptr = ERR11; |
5100 | goto FAILED; |
5101 | } |
5102 | |
5103 | /* If the character following a repeat is '+', or if certain optimization |
5104 | tests above succeeded, possessive_quantifier is TRUE. For some opcodes, |
5105 | there are special alternative opcodes for this case. For anything else, we |
5106 | wrap the entire repeated item inside OP_ONCE brackets. Logically, the '+' |
5107 | notation is just syntactic sugar, taken from Sun's Java package, but the |
5108 | special opcodes can optimize it. |
5109 | |
5110 | Some (but not all) possessively repeated subpatterns have already been |
5111 | completely handled in the code just above. For them, possessive_quantifier |
5112 | is always FALSE at this stage. |
5113 | |
5114 | Note that the repeated item starts at tempcode, not at previous, which |
5115 | might be the first part of a string whose (former) last char we repeated. |
5116 | |
5117 | Possessifying an 'exact' quantifier has no effect, so we can ignore it. But |
5118 | an 'upto' may follow. We skip over an 'exact' item, and then test the |
5119 | length of what remains before proceeding. */ |
5120 | |
5121 | if (possessive_quantifier) |
5122 | { |
5123 | int len; |
5124 | |
5125 | if (*tempcode == OP_TYPEEXACT) |
5126 | tempcode += _pcre_OP_lengths[*tempcode] + |
5127 | ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0); |
5128 | |
5129 | else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT) |
5130 | { |
5131 | tempcode += _pcre_OP_lengths[*tempcode]; |
5132 | #ifdef SUPPORT_UTF8 |
5133 | if (utf8 && tempcode[-1] >= 0xc0) |
5134 | tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f]; |
5135 | #endif |
5136 | } |
5137 | |
5138 | len = (int)(code - tempcode); |
5139 | if (len > 0) switch (*tempcode) |
5140 | { |
5141 | case OP_STAR: *tempcode = OP_POSSTAR; break; |
5142 | case OP_PLUS: *tempcode = OP_POSPLUS; break; |
5143 | case OP_QUERY: *tempcode = OP_POSQUERY; break; |
5144 | case OP_UPTO: *tempcode = OP_POSUPTO; break; |
5145 | |
5146 | case OP_STARI: *tempcode = OP_POSSTARI; break; |
5147 | case OP_PLUSI: *tempcode = OP_POSPLUSI; break; |
5148 | case OP_QUERYI: *tempcode = OP_POSQUERYI; break; |
5149 | case OP_UPTOI: *tempcode = OP_POSUPTOI; break; |
5150 | |
5151 | case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break; |
5152 | case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break; |
5153 | case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break; |
5154 | case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break; |
5155 | |
5156 | case OP_NOTSTARI: *tempcode = OP_NOTPOSSTARI; break; |
5157 | case OP_NOTPLUSI: *tempcode = OP_NOTPOSPLUSI; break; |
5158 | case OP_NOTQUERYI: *tempcode = OP_NOTPOSQUERYI; break; |
5159 | case OP_NOTUPTOI: *tempcode = OP_NOTPOSUPTOI; break; |
5160 | |
5161 | case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break; |
5162 | case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break; |
5163 | case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break; |
5164 | case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break; |
5165 | |
5166 | /* Because we are moving code along, we must ensure that any |
5167 | pending recursive references are updated. */ |
5168 | |
5169 | default: |
5170 | *code = OP_END; |
5171 | adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm); |
5172 | memmove(tempcode + 1+LINK_SIZE, tempcode, len); |
5173 | code += 1 + LINK_SIZE; |
5174 | len += 1 + LINK_SIZE; |
5175 | tempcode[0] = OP_ONCE; |
5176 | *code++ = OP_KET; |
5177 | PUTINC(code, 0, len); |
5178 | PUT(tempcode, 1, len); |
5179 | break; |
5180 | } |
5181 | } |
5182 | |
5183 | /* In all case we no longer have a previous item. We also set the |
5184 | "follows varying string" flag for subsequently encountered reqbytes if |
5185 | it isn't already set and we have just passed a varying length item. */ |
5186 | |
5187 | END_REPEAT: |
5188 | previous = NULL; |
5189 | cd->req_varyopt |= reqvary; |
5190 | break; |
5191 | |
5192 | |
5193 | /* ===================================================================*/ |
5194 | /* Start of nested parenthesized sub-expression, or comment or lookahead or |
5195 | lookbehind or option setting or condition or all the other extended |
5196 | parenthesis forms. */ |
5197 | |
5198 | case CHAR_LEFT_PARENTHESIS: |
5199 | newoptions = options; |
5200 | skipbytes = 0; |
5201 | bravalue = OP_CBRA; |
5202 | save_hwm = cd->hwm; |
5203 | reset_bracount = FALSE; |
5204 | |
5205 | /* First deal with various "verbs" that can be introduced by '*'. */ |
5206 | |
5207 | if (*(++ptr) == CHAR_ASTERISK && |
5208 | ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':')) |
5209 | { |
5210 | int i, namelen; |
5211 | int arglen = 0; |
5212 | const char *vn = verbnames; |
5213 | const uschar *name = ptr + 1; |
5214 | const uschar *arg = NULL; |
5215 | previous = NULL; |
5216 | while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {}; |
5217 | namelen = (int)(ptr - name); |
5218 | |
5219 | /* It appears that Perl allows any characters whatsoever, other than |
5220 | a closing parenthesis, to appear in arguments, so we no longer insist on |
5221 | letters, digits, and underscores. */ |
5222 | |
5223 | if (*ptr == CHAR_COLON) |
5224 | { |
5225 | arg = ++ptr; |
5226 | while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; |
5227 | arglen = (int)(ptr - arg); |
5228 | } |
5229 | |
5230 | if (*ptr != CHAR_RIGHT_PARENTHESIS) |
5231 | { |
5232 | *errorcodeptr = ERR60; |
5233 | goto FAILED; |
5234 | } |
5235 | |
5236 | /* Scan the table of verb names */ |
5237 | |
5238 | for (i = 0; i < verbcount; i++) |
5239 | { |
5240 | if (namelen == verbs[i].len && |
5241 | strncmp((char *)name, vn, namelen) == 0) |
5242 | { |
5243 | /* Check for open captures before ACCEPT and convert it to |
5244 | ASSERT_ACCEPT if in an assertion. */ |
5245 | |
5246 | if (verbs[i].op == OP_ACCEPT) |
5247 | { |
5248 | open_capitem *oc; |
5249 | if (arglen != 0) |
5250 | { |
5251 | *errorcodeptr = ERR59; |
5252 | goto FAILED; |
5253 | } |
5254 | cd->had_accept = TRUE; |
5255 | for (oc = cd->open_caps; oc != NULL; oc = oc->next) |
5256 | { |
5257 | *code++ = OP_CLOSE; |
5258 | PUT2INC(code, 0, oc->number); |
5259 | } |
5260 | *code++ = (cd->assert_depth > 0)? OP_ASSERT_ACCEPT : OP_ACCEPT; |
5261 | |
5262 | /* Do not set firstbyte after *ACCEPT */ |
5263 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
5264 | } |
5265 | |
5266 | /* Handle other cases with/without an argument */ |
5267 | |
5268 | else if (arglen == 0) |
5269 | { |
5270 | if (verbs[i].op < 0) /* Argument is mandatory */ |
5271 | { |
5272 | *errorcodeptr = ERR66; |
5273 | goto FAILED; |
5274 | } |
5275 | *code = verbs[i].op; |
5276 | if (*code++ == OP_THEN) cd->external_flags |= PCRE_HASTHEN; |
5277 | } |
5278 | |
5279 | else |
5280 | { |
5281 | if (verbs[i].op_arg < 0) /* Argument is forbidden */ |
5282 | { |
5283 | *errorcodeptr = ERR59; |
5284 | goto FAILED; |
5285 | } |
5286 | *code = verbs[i].op_arg; |
5287 | if (*code++ == OP_THEN_ARG) cd->external_flags |= PCRE_HASTHEN; |
5288 | *code++ = arglen; |
5289 | memcpy(code, arg, arglen); |
5290 | code += arglen; |
5291 | *code++ = 0; |
5292 | } |
5293 | |
5294 | break; /* Found verb, exit loop */ |
5295 | } |
5296 | |
5297 | vn += verbs[i].len + 1; |
5298 | } |
5299 | |
5300 | if (i < verbcount) continue; /* Successfully handled a verb */ |
5301 | *errorcodeptr = ERR60; /* Verb not recognized */ |
5302 | goto FAILED; |
5303 | } |
5304 | |
5305 | /* Deal with the extended parentheses; all are introduced by '?', and the |
5306 | appearance of any of them means that this is not a capturing group. */ |
5307 | |
5308 | else if (*ptr == CHAR_QUESTION_MARK) |
5309 | { |
5310 | int i, set, unset, namelen; |
5311 | int *optset; |
5312 | const uschar *name; |
5313 | uschar *slot; |
5314 | |
5315 | switch (*(++ptr)) |
5316 | { |
5317 | case CHAR_NUMBER_SIGN: /* Comment; skip to ket */ |
5318 | ptr++; |
5319 | while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++; |
5320 | if (*ptr == 0) |
5321 | { |
5322 | *errorcodeptr = ERR18; |
5323 | goto FAILED; |
5324 | } |
5325 | continue; |
5326 | |
5327 | |
5328 | /* ------------------------------------------------------------ */ |
5329 | case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */ |
5330 | reset_bracount = TRUE; |
5331 | /* Fall through */ |
5332 | |
5333 | /* ------------------------------------------------------------ */ |
5334 | case CHAR_COLON: /* Non-capturing bracket */ |
5335 | bravalue = OP_BRA; |
5336 | ptr++; |
5337 | break; |
5338 | |
5339 | |
5340 | /* ------------------------------------------------------------ */ |
5341 | case CHAR_LEFT_PARENTHESIS: |
5342 | bravalue = OP_COND; /* Conditional group */ |
5343 | |
5344 | /* A condition can be an assertion, a number (referring to a numbered |
5345 | group), a name (referring to a named group), or 'R', referring to |
5346 | recursion. R<digits> and R&name are also permitted for recursion tests. |
5347 | |
5348 | There are several syntaxes for testing a named group: (?(name)) is used |
5349 | by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')). |
5350 | |
5351 | There are two unfortunate ambiguities, caused by history. (a) 'R' can |
5352 | be the recursive thing or the name 'R' (and similarly for 'R' followed |
5353 | by digits), and (b) a number could be a name that consists of digits. |
5354 | In both cases, we look for a name first; if not found, we try the other |
5355 | cases. */ |
5356 | |
5357 | /* For conditions that are assertions, check the syntax, and then exit |
5358 | the switch. This will take control down to where bracketed groups, |
5359 | including assertions, are processed. */ |
5360 | |
5361 | if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN || |
5362 | ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN)) |
5363 | break; |
5364 | |
5365 | /* Most other conditions use OP_CREF (a couple change to OP_RREF |
5366 | below), and all need to skip 3 bytes at the start of the group. */ |
5367 | |
5368 | code[1+LINK_SIZE] = OP_CREF; |
5369 | skipbytes = 3; |
5370 | refsign = -1; |
5371 | |
5372 | /* Check for a test for recursion in a named group. */ |
5373 | |
5374 | if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND) |
5375 | { |
5376 | terminator = -1; |
5377 | ptr += 2; |
5378 | code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */ |
5379 | } |
5380 | |
5381 | /* Check for a test for a named group's having been set, using the Perl |
5382 | syntax (?(<name>) or (?('name') */ |
5383 | |
5384 | else if (ptr[1] == CHAR_LESS_THAN_SIGN) |
5385 | { |
5386 | terminator = CHAR_GREATER_THAN_SIGN; |
5387 | ptr++; |
5388 | } |
5389 | else if (ptr[1] == CHAR_APOSTROPHE) |
5390 | { |
5391 | terminator = CHAR_APOSTROPHE; |
5392 | ptr++; |
5393 | } |
5394 | else |
5395 | { |
5396 | terminator = 0; |
5397 | if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr); |
5398 | } |
5399 | |
5400 | /* We now expect to read a name; any thing else is an error */ |
5401 | |
5402 | if ((cd->ctypes[ptr[1]] & ctype_word) == 0) |
5403 | { |
5404 | ptr += 1; /* To get the right offset */ |
5405 | *errorcodeptr = ERR28; |
5406 | goto FAILED; |
5407 | } |
5408 | |
5409 | /* Read the name, but also get it as a number if it's all digits */ |
5410 | |
5411 | recno = 0; |
5412 | name = ++ptr; |
5413 | while ((cd->ctypes[*ptr] & ctype_word) != 0) |
5414 | { |
5415 | if (recno >= 0) |
5416 | recno = ((digitab[*ptr] & ctype_digit) != 0)? |
5417 | recno * 10 + *ptr - CHAR_0 : -1; |
5418 | ptr++; |
5419 | } |
5420 | namelen = (int)(ptr - name); |
5421 | |
5422 | if ((terminator > 0 && *ptr++ != terminator) || |
5423 | *ptr++ != CHAR_RIGHT_PARENTHESIS) |
5424 | { |
5425 | ptr--; /* Error offset */ |
5426 | *errorcodeptr = ERR26; |
5427 | goto FAILED; |
5428 | } |
5429 | |
5430 | /* Do no further checking in the pre-compile phase. */ |
5431 | |
5432 | if (lengthptr != NULL) break; |
5433 | |
5434 | /* In the real compile we do the work of looking for the actual |
5435 | reference. If the string started with "+" or "-" we require the rest to |
5436 | be digits, in which case recno will be set. */ |
5437 | |
5438 | if (refsign > 0) |
5439 | { |
5440 | if (recno <= 0) |
5441 | { |
5442 | *errorcodeptr = ERR58; |
5443 | goto FAILED; |
5444 | } |
5445 | recno = (refsign == CHAR_MINUS)? |
5446 | cd->bracount - recno + 1 : recno +cd->bracount; |
5447 | if (recno <= 0 || recno > cd->final_bracount) |
5448 | { |
5449 | *errorcodeptr = ERR15; |
5450 | goto FAILED; |
5451 | } |
5452 | PUT2(code, 2+LINK_SIZE, recno); |
5453 | break; |
5454 | } |
5455 | |
5456 | /* Otherwise (did not start with "+" or "-"), start by looking for the |
5457 | name. If we find a name, add one to the opcode to change OP_CREF or |
5458 | OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same, |
5459 | except they record that the reference was originally to a name. The |
5460 | information is used to check duplicate names. */ |
5461 | |
5462 | slot = cd->name_table; |
5463 | for (i = 0; i < cd->names_found; i++) |
5464 | { |
5465 | if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break; |
5466 | slot += cd->name_entry_size; |
5467 | } |
5468 | |
5469 | /* Found a previous named subpattern */ |
5470 | |
5471 | if (i < cd->names_found) |
5472 | { |
5473 | recno = GET2(slot, 0); |
5474 | PUT2(code, 2+LINK_SIZE, recno); |
5475 | code[1+LINK_SIZE]++; |
5476 | } |
5477 | |
5478 | /* Search the pattern for a forward reference */ |
5479 | |
5480 | else if ((i = find_parens(cd, name, namelen, |
5481 | (options & PCRE_EXTENDED) != 0, utf8)) > 0) |
5482 | { |
5483 | PUT2(code, 2+LINK_SIZE, i); |
5484 | code[1+LINK_SIZE]++; |
5485 | } |
5486 | |
5487 | /* If terminator == 0 it means that the name followed directly after |
5488 | the opening parenthesis [e.g. (?(abc)...] and in this case there are |
5489 | some further alternatives to try. For the cases where terminator != 0 |
5490 | [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have |
5491 | now checked all the possibilities, so give an error. */ |
5492 | |
5493 | else if (terminator != 0) |
5494 | { |
5495 | *errorcodeptr = ERR15; |
5496 | goto FAILED; |
5497 | } |
5498 | |
5499 | /* Check for (?(R) for recursion. Allow digits after R to specify a |
5500 | specific group number. */ |
5501 | |
5502 | else if (*name == CHAR_R) |
5503 | { |
5504 | recno = 0; |
5505 | for (i = 1; i < namelen; i++) |
5506 | { |
5507 | if ((digitab[name[i]] & ctype_digit) == 0) |
5508 | { |
5509 | *errorcodeptr = ERR15; |
5510 | goto FAILED; |
5511 | } |
5512 | recno = recno * 10 + name[i] - CHAR_0; |
5513 | } |
5514 | if (recno == 0) recno = RREF_ANY; |
5515 | code[1+LINK_SIZE] = OP_RREF; /* Change test type */ |
5516 | PUT2(code, 2+LINK_SIZE, recno); |
5517 | } |
5518 | |
5519 | /* Similarly, check for the (?(DEFINE) "condition", which is always |
5520 | false. */ |
5521 | |
5522 | else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0) |
5523 | { |
5524 | code[1+LINK_SIZE] = OP_DEF; |
5525 | skipbytes = 1; |
5526 | } |
5527 | |
5528 | /* Check for the "name" actually being a subpattern number. We are |
5529 | in the second pass here, so final_bracount is set. */ |
5530 | |
5531 | else if (recno > 0 && recno <= cd->final_bracount) |
5532 | { |
5533 | PUT2(code, 2+LINK_SIZE, recno); |
5534 | } |
5535 | |
5536 | /* Either an unidentified subpattern, or a reference to (?(0) */ |
5537 | |
5538 | else |
5539 | { |
5540 | *errorcodeptr = (recno == 0)? ERR35: ERR15; |
5541 | goto FAILED; |
5542 | } |
5543 | break; |
5544 | |
5545 | |
5546 | /* ------------------------------------------------------------ */ |
5547 | case CHAR_EQUALS_SIGN: /* Positive lookahead */ |
5548 | bravalue = OP_ASSERT; |
5549 | cd->assert_depth += 1; |
5550 | ptr++; |
5551 | break; |
5552 | |
5553 | |
5554 | /* ------------------------------------------------------------ */ |
5555 | case CHAR_EXCLAMATION_MARK: /* Negative lookahead */ |
5556 | ptr++; |
5557 | if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */ |
5558 | { |
5559 | *code++ = OP_FAIL; |
5560 | previous = NULL; |
5561 | continue; |
5562 | } |
5563 | bravalue = OP_ASSERT_NOT; |
5564 | cd->assert_depth += 1; |
5565 | break; |
5566 | |
5567 | |
5568 | /* ------------------------------------------------------------ */ |
5569 | case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */ |
5570 | switch (ptr[1]) |
5571 | { |
5572 | case CHAR_EQUALS_SIGN: /* Positive lookbehind */ |
5573 | bravalue = OP_ASSERTBACK; |
5574 | cd->assert_depth += 1; |
5575 | ptr += 2; |
5576 | break; |
5577 | |
5578 | case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */ |
5579 | bravalue = OP_ASSERTBACK_NOT; |
5580 | cd->assert_depth += 1; |
5581 | ptr += 2; |
5582 | break; |
5583 | |
5584 | default: /* Could be name define, else bad */ |
5585 | if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME; |
5586 | ptr++; /* Correct offset for error */ |
5587 | *errorcodeptr = ERR24; |
5588 | goto FAILED; |
5589 | } |
5590 | break; |
5591 | |
5592 | |
5593 | /* ------------------------------------------------------------ */ |
5594 | case CHAR_GREATER_THAN_SIGN: /* One-time brackets */ |
5595 | bravalue = OP_ONCE; |
5596 | ptr++; |
5597 | break; |
5598 | |
5599 | |
5600 | /* ------------------------------------------------------------ */ |
5601 | case CHAR_C: /* Callout - may be followed by digits; */ |
5602 | previous_callout = code; /* Save for later completion */ |
5603 | after_manual_callout = 1; /* Skip one item before completing */ |
5604 | *code++ = OP_CALLOUT; |
5605 | { |
5606 | int n = 0; |
5607 | while ((digitab[*(++ptr)] & ctype_digit) != 0) |
5608 | n = n * 10 + *ptr - CHAR_0; |
5609 | if (*ptr != CHAR_RIGHT_PARENTHESIS) |
5610 | { |
5611 | *errorcodeptr = ERR39; |
5612 | goto FAILED; |
5613 | } |
5614 | if (n > 255) |
5615 | { |
5616 | *errorcodeptr = ERR38; |
5617 | goto FAILED; |
5618 | } |
5619 | *code++ = n; |
5620 | PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */ |
5621 | PUT(code, LINK_SIZE, 0); /* Default length */ |
5622 | code += 2 * LINK_SIZE; |
5623 | } |
5624 | previous = NULL; |
5625 | continue; |
5626 | |
5627 | |
5628 | /* ------------------------------------------------------------ */ |
5629 | case CHAR_P: /* Python-style named subpattern handling */ |
5630 | if (*(++ptr) == CHAR_EQUALS_SIGN || |
5631 | *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */ |
5632 | { |
5633 | is_recurse = *ptr == CHAR_GREATER_THAN_SIGN; |
5634 | terminator = CHAR_RIGHT_PARENTHESIS; |
5635 | goto NAMED_REF_OR_RECURSE; |
5636 | } |
5637 | else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */ |
5638 | { |
5639 | *errorcodeptr = ERR41; |
5640 | goto FAILED; |
5641 | } |
5642 | /* Fall through to handle (?P< as (?< is handled */ |
5643 | |
5644 | |
5645 | /* ------------------------------------------------------------ */ |
5646 | DEFINE_NAME: /* Come here from (?< handling */ |
5647 | case CHAR_APOSTROPHE: |
5648 | { |
5649 | terminator = (*ptr == CHAR_LESS_THAN_SIGN)? |
5650 | CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE; |
5651 | name = ++ptr; |
5652 | |
5653 | while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++; |
5654 | namelen = (int)(ptr - name); |
5655 | |
5656 | /* In the pre-compile phase, just do a syntax check. */ |
5657 | |
5658 | if (lengthptr != NULL) |
5659 | { |
5660 | if (*ptr != terminator) |
5661 | { |
5662 | *errorcodeptr = ERR42; |
5663 | goto FAILED; |
5664 | } |
5665 | if (cd->names_found >= MAX_NAME_COUNT) |
5666 | { |
5667 | *errorcodeptr = ERR49; |
5668 | goto FAILED; |
5669 | } |
5670 | if (namelen + 3 > cd->name_entry_size) |
5671 | { |
5672 | cd->name_entry_size = namelen + 3; |
5673 | if (namelen > MAX_NAME_SIZE) |
5674 | { |
5675 | *errorcodeptr = ERR48; |
5676 | goto FAILED; |
5677 | } |
5678 | } |
5679 | } |
5680 | |
5681 | /* In the real compile, create the entry in the table, maintaining |
5682 | alphabetical order. Duplicate names for different numbers are |
5683 | permitted only if PCRE_DUPNAMES is set. Duplicate names for the same |
5684 | number are always OK. (An existing number can be re-used if (?| |
5685 | appears in the pattern.) In either event, a duplicate name results in |
5686 | a duplicate entry in the table, even if the number is the same. This |
5687 | is because the number of names, and hence the table size, is computed |
5688 | in the pre-compile, and it affects various numbers and pointers which |
5689 | would all have to be modified, and the compiled code moved down, if |
5690 | duplicates with the same number were omitted from the table. This |
5691 | doesn't seem worth the hassle. However, *different* names for the |
5692 | same number are not permitted. */ |
5693 | |
5694 | else |
5695 | { |
5696 | BOOL dupname = FALSE; |
5697 | slot = cd->name_table; |
5698 | |
5699 | for (i = 0; i < cd->names_found; i++) |
5700 | { |
5701 | int crc = memcmp(name, slot+2, namelen); |
5702 | if (crc == 0) |
5703 | { |
5704 | if (slot[2+namelen] == 0) |
5705 | { |
5706 | if (GET2(slot, 0) != cd->bracount + 1 && |
5707 | (options & PCRE_DUPNAMES) == 0) |
5708 | { |
5709 | *errorcodeptr = ERR43; |
5710 | goto FAILED; |
5711 | } |
5712 | else dupname = TRUE; |
5713 | } |
5714 | else crc = -1; /* Current name is a substring */ |
5715 | } |
5716 | |
5717 | /* Make space in the table and break the loop for an earlier |
5718 | name. For a duplicate or later name, carry on. We do this for |
5719 | duplicates so that in the simple case (when ?(| is not used) they |
5720 | are in order of their numbers. */ |
5721 | |
5722 | if (crc < 0) |
5723 | { |
5724 | memmove(slot + cd->name_entry_size, slot, |
5725 | (cd->names_found - i) * cd->name_entry_size); |
5726 | break; |
5727 | } |
5728 | |
5729 | /* Continue the loop for a later or duplicate name */ |
5730 | |
5731 | slot += cd->name_entry_size; |
5732 | } |
5733 | |
5734 | /* For non-duplicate names, check for a duplicate number before |
5735 | adding the new name. */ |
5736 | |
5737 | if (!dupname) |
5738 | { |
5739 | uschar *cslot = cd->name_table; |
5740 | for (i = 0; i < cd->names_found; i++) |
5741 | { |
5742 | if (cslot != slot) |
5743 | { |
5744 | if (GET2(cslot, 0) == cd->bracount + 1) |
5745 | { |
5746 | *errorcodeptr = ERR65; |
5747 | goto FAILED; |
5748 | } |
5749 | } |
5750 | else i--; |
5751 | cslot += cd->name_entry_size; |
5752 | } |
5753 | } |
5754 | |
5755 | PUT2(slot, 0, cd->bracount + 1); |
5756 | memcpy(slot + 2, name, namelen); |
5757 | slot[2+namelen] = 0; |
5758 | } |
5759 | } |
5760 | |
5761 | /* In both pre-compile and compile, count the number of names we've |
5762 | encountered. */ |
5763 | |
5764 | cd->names_found++; |
5765 | ptr++; /* Move past > or ' */ |
5766 | goto NUMBERED_GROUP; |
5767 | |
5768 | |
5769 | /* ------------------------------------------------------------ */ |
5770 | case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */ |
5771 | terminator = CHAR_RIGHT_PARENTHESIS; |
5772 | is_recurse = TRUE; |
5773 | /* Fall through */ |
5774 | |
5775 | /* We come here from the Python syntax above that handles both |
5776 | references (?P=name) and recursion (?P>name), as well as falling |
5777 | through from the Perl recursion syntax (?&name). We also come here from |
5778 | the Perl \k<name> or \k'name' back reference syntax and the \k{name} |
5779 | .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */ |
5780 | |
5781 | NAMED_REF_OR_RECURSE: |
5782 | name = ++ptr; |
5783 | while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++; |
5784 | namelen = (int)(ptr - name); |
5785 | |
5786 | /* In the pre-compile phase, do a syntax check. We used to just set |
5787 | a dummy reference number, because it was not used in the first pass. |
5788 | However, with the change of recursive back references to be atomic, |
5789 | we have to look for the number so that this state can be identified, as |
5790 | otherwise the incorrect length is computed. If it's not a backwards |
5791 | reference, the dummy number will do. */ |
5792 | |
5793 | if (lengthptr != NULL) |
5794 | { |
5795 | const uschar *temp; |
5796 | |
5797 | if (namelen == 0) |
5798 | { |
5799 | *errorcodeptr = ERR62; |
5800 | goto FAILED; |
5801 | } |
5802 | if (*ptr != terminator) |
5803 | { |
5804 | *errorcodeptr = ERR42; |
5805 | goto FAILED; |
5806 | } |
5807 | if (namelen > MAX_NAME_SIZE) |
5808 | { |
5809 | *errorcodeptr = ERR48; |
5810 | goto FAILED; |
5811 | } |
5812 | |
5813 | /* The name table does not exist in the first pass, so we cannot |
5814 | do a simple search as in the code below. Instead, we have to scan the |
5815 | pattern to find the number. It is important that we scan it only as |
5816 | far as we have got because the syntax of named subpatterns has not |
5817 | been checked for the rest of the pattern, and find_parens() assumes |
5818 | correct syntax. In any case, it's a waste of resources to scan |
5819 | further. We stop the scan at the current point by temporarily |
5820 | adjusting the value of cd->endpattern. */ |
5821 | |
5822 | temp = cd->end_pattern; |
5823 | cd->end_pattern = ptr; |
5824 | recno = find_parens(cd, name, namelen, |
5825 | (options & PCRE_EXTENDED) != 0, utf8); |
5826 | cd->end_pattern = temp; |
5827 | if (recno < 0) recno = 0; /* Forward ref; set dummy number */ |
5828 | } |
5829 | |
5830 | /* In the real compile, seek the name in the table. We check the name |
5831 | first, and then check that we have reached the end of the name in the |
5832 | table. That way, if the name that is longer than any in the table, |
5833 | the comparison will fail without reading beyond the table entry. */ |
5834 | |
5835 | else |
5836 | { |
5837 | slot = cd->name_table; |
5838 | for (i = 0; i < cd->names_found; i++) |
5839 | { |
5840 | if (strncmp((char *)name, (char *)slot+2, namelen) == 0 && |
5841 | slot[2+namelen] == 0) |
5842 | break; |
5843 | slot += cd->name_entry_size; |
5844 | } |
5845 | |
5846 | if (i < cd->names_found) /* Back reference */ |
5847 | { |
5848 | recno = GET2(slot, 0); |
5849 | } |
5850 | else if ((recno = /* Forward back reference */ |
5851 | find_parens(cd, name, namelen, |
5852 | (options & PCRE_EXTENDED) != 0, utf8)) <= 0) |
5853 | { |
5854 | *errorcodeptr = ERR15; |
5855 | goto FAILED; |
5856 | } |
5857 | } |
5858 | |
5859 | /* In both phases, we can now go to the code than handles numerical |
5860 | recursion or backreferences. */ |
5861 | |
5862 | if (is_recurse) goto HANDLE_RECURSION; |
5863 | else goto HANDLE_REFERENCE; |
5864 | |
5865 | |
5866 | /* ------------------------------------------------------------ */ |
5867 | case CHAR_R: /* Recursion */ |
5868 | ptr++; /* Same as (?0) */ |
5869 | /* Fall through */ |
5870 | |
5871 | |
5872 | /* ------------------------------------------------------------ */ |
5873 | case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */ |
5874 | case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: |
5875 | case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9: |
5876 | { |
5877 | const uschar *called; |
5878 | terminator = CHAR_RIGHT_PARENTHESIS; |
5879 | |
5880 | /* Come here from the \g<...> and \g'...' code (Oniguruma |
5881 | compatibility). However, the syntax has been checked to ensure that |
5882 | the ... are a (signed) number, so that neither ERR63 nor ERR29 will |
5883 | be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY |
5884 | ever be taken. */ |
5885 | |
5886 | HANDLE_NUMERICAL_RECURSION: |
5887 | |
5888 | if ((refsign = *ptr) == CHAR_PLUS) |
5889 | { |
5890 | ptr++; |
5891 | if ((digitab[*ptr] & ctype_digit) == 0) |
5892 | { |
5893 | *errorcodeptr = ERR63; |
5894 | goto FAILED; |
5895 | } |
5896 | } |
5897 | else if (refsign == CHAR_MINUS) |
5898 | { |
5899 | if ((digitab[ptr[1]] & ctype_digit) == 0) |
5900 | goto OTHER_CHAR_AFTER_QUERY; |
5901 | ptr++; |
5902 | } |
5903 | |
5904 | recno = 0; |
5905 | while((digitab[*ptr] & ctype_digit) != 0) |
5906 | recno = recno * 10 + *ptr++ - CHAR_0; |
5907 | |
5908 | if (*ptr != terminator) |
5909 | { |
5910 | *errorcodeptr = ERR29; |
5911 | goto FAILED; |
5912 | } |
5913 | |
5914 | if (refsign == CHAR_MINUS) |
5915 | { |
5916 | if (recno == 0) |
5917 | { |
5918 | *errorcodeptr = ERR58; |
5919 | goto FAILED; |
5920 | } |
5921 | recno = cd->bracount - recno + 1; |
5922 | if (recno <= 0) |
5923 | { |
5924 | *errorcodeptr = ERR15; |
5925 | goto FAILED; |
5926 | } |
5927 | } |
5928 | else if (refsign == CHAR_PLUS) |
5929 | { |
5930 | if (recno == 0) |
5931 | { |
5932 | *errorcodeptr = ERR58; |
5933 | goto FAILED; |
5934 | } |
5935 | recno += cd->bracount; |
5936 | } |
5937 | |
5938 | /* Come here from code above that handles a named recursion */ |
5939 | |
5940 | HANDLE_RECURSION: |
5941 | |
5942 | previous = code; |
5943 | called = cd->start_code; |
5944 | |
5945 | /* When we are actually compiling, find the bracket that is being |
5946 | referenced. Temporarily end the regex in case it doesn't exist before |
5947 | this point. If we end up with a forward reference, first check that |
5948 | the bracket does occur later so we can give the error (and position) |
5949 | now. Then remember this forward reference in the workspace so it can |
5950 | be filled in at the end. */ |
5951 | |
5952 | if (lengthptr == NULL) |
5953 | { |
5954 | *code = OP_END; |
5955 | if (recno != 0) |
5956 | called = _pcre_find_bracket(cd->start_code, utf8, recno); |
5957 | |
5958 | /* Forward reference */ |
5959 | |
5960 | if (called == NULL) |
5961 | { |
5962 | if (find_parens(cd, NULL, recno, |
5963 | (options & PCRE_EXTENDED) != 0, utf8) < 0) |
5964 | { |
5965 | *errorcodeptr = ERR15; |
5966 | goto FAILED; |
5967 | } |
5968 | |
5969 | /* Fudge the value of "called" so that when it is inserted as an |
5970 | offset below, what it actually inserted is the reference number |
5971 | of the group. Then remember the forward reference. */ |
5972 | |
5973 | called = cd->start_code + recno; |
5974 | PUTINC(cd->hwm, 0, (int)(code + 1 - cd->start_code)); |
5975 | } |
5976 | |
5977 | /* If not a forward reference, and the subpattern is still open, |
5978 | this is a recursive call. We check to see if this is a left |
5979 | recursion that could loop for ever, and diagnose that case. We |
5980 | must not, however, do this check if we are in a conditional |
5981 | subpattern because the condition might be testing for recursion in |
5982 | a pattern such as /(?(R)a+|(?R)b)/, which is perfectly valid. |
5983 | Forever loops are also detected at runtime, so those that occur in |
5984 | conditional subpatterns will be picked up then. */ |
5985 | |
5986 | else if (GET(called, 1) == 0 && cond_depth <= 0 && |
5987 | could_be_empty(called, code, bcptr, utf8, cd)) |
5988 | { |
5989 | *errorcodeptr = ERR40; |
5990 | goto FAILED; |
5991 | } |
5992 | } |
5993 | |
5994 | /* Insert the recursion/subroutine item. */ |
5995 | |
5996 | *code = OP_RECURSE; |
5997 | PUT(code, 1, (int)(called - cd->start_code)); |
5998 | code += 1 + LINK_SIZE; |
5999 | } |
6000 | |
6001 | /* Can't determine a first byte now */ |
6002 | |
6003 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
6004 | continue; |
6005 | |
6006 | |
6007 | /* ------------------------------------------------------------ */ |
6008 | default: /* Other characters: check option setting */ |
6009 | OTHER_CHAR_AFTER_QUERY: |
6010 | set = unset = 0; |
6011 | optset = &set; |
6012 | |
6013 | while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON) |
6014 | { |
6015 | switch (*ptr++) |
6016 | { |
6017 | case CHAR_MINUS: optset = &unset; break; |
6018 | |
6019 | case CHAR_J: /* Record that it changed in the external options */ |
6020 | *optset |= PCRE_DUPNAMES; |
6021 | cd->external_flags |= PCRE_JCHANGED; |
6022 | break; |
6023 | |
6024 | case CHAR_i: *optset |= PCRE_CASELESS; break; |
6025 | case CHAR_m: *optset |= PCRE_MULTILINE; break; |
6026 | case CHAR_s: *optset |= PCRE_DOTALL; break; |
6027 | case CHAR_x: *optset |= PCRE_EXTENDED; break; |
6028 | case CHAR_U: *optset |= PCRE_UNGREEDY; break; |
6029 | case CHAR_X: *optset |= PCRE_EXTRA; break; |
6030 | |
6031 | default: *errorcodeptr = ERR12; |
6032 | ptr--; /* Correct the offset */ |
6033 | goto FAILED; |
6034 | } |
6035 | } |
6036 | |
6037 | /* Set up the changed option bits, but don't change anything yet. */ |
6038 | |
6039 | newoptions = (options | set) & (~unset); |
6040 | |
6041 | /* If the options ended with ')' this is not the start of a nested |
6042 | group with option changes, so the options change at this level. If this |
6043 | item is right at the start of the pattern, the options can be |
6044 | abstracted and made external in the pre-compile phase, and ignored in |
6045 | the compile phase. This can be helpful when matching -- for instance in |
6046 | caseless checking of required bytes. |
6047 | |
6048 | If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are |
6049 | definitely *not* at the start of the pattern because something has been |
6050 | compiled. In the pre-compile phase, however, the code pointer can have |
6051 | that value after the start, because it gets reset as code is discarded |
6052 | during the pre-compile. However, this can happen only at top level - if |
6053 | we are within parentheses, the starting BRA will still be present. At |
6054 | any parenthesis level, the length value can be used to test if anything |
6055 | has been compiled at that level. Thus, a test for both these conditions |
6056 | is necessary to ensure we correctly detect the start of the pattern in |
6057 | both phases. |
6058 | |
6059 | If we are not at the pattern start, reset the greedy defaults and the |
6060 | case value for firstbyte and reqbyte. */ |
6061 | |
6062 | if (*ptr == CHAR_RIGHT_PARENTHESIS) |
6063 | { |
6064 | if (code == cd->start_code + 1 + LINK_SIZE && |
6065 | (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE)) |
6066 | { |
6067 | cd->external_options = newoptions; |
6068 | } |
6069 | else |
6070 | { |
6071 | greedy_default = ((newoptions & PCRE_UNGREEDY) != 0); |
6072 | greedy_non_default = greedy_default ^ 1; |
6073 | req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0; |
6074 | } |
6075 | |
6076 | /* Change options at this level, and pass them back for use |
6077 | in subsequent branches. */ |
6078 | |
6079 | *optionsptr = options = newoptions; |
6080 | previous = NULL; /* This item can't be repeated */ |
6081 | continue; /* It is complete */ |
6082 | } |
6083 | |
6084 | /* If the options ended with ':' we are heading into a nested group |
6085 | with possible change of options. Such groups are non-capturing and are |
6086 | not assertions of any kind. All we need to do is skip over the ':'; |
6087 | the newoptions value is handled below. */ |
6088 | |
6089 | bravalue = OP_BRA; |
6090 | ptr++; |
6091 | } /* End of switch for character following (? */ |
6092 | } /* End of (? handling */ |
6093 | |
6094 | /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE |
6095 | is set, all unadorned brackets become non-capturing and behave like (?:...) |
6096 | brackets. */ |
6097 | |
6098 | else if ((options & PCRE_NO_AUTO_CAPTURE) != 0) |
6099 | { |
6100 | bravalue = OP_BRA; |
6101 | } |
6102 | |
6103 | /* Else we have a capturing group. */ |
6104 | |
6105 | else |
6106 | { |
6107 | NUMBERED_GROUP: |
6108 | cd->bracount += 1; |
6109 | PUT2(code, 1+LINK_SIZE, cd->bracount); |
6110 | skipbytes = 2; |
6111 | } |
6112 | |
6113 | /* Process nested bracketed regex. Assertions used not to be repeatable, |
6114 | but this was changed for Perl compatibility, so all kinds can now be |
6115 | repeated. We copy code into a non-register variable (tempcode) in order to |
6116 | be able to pass its address because some compilers complain otherwise. */ |
6117 | |
6118 | previous = code; /* For handling repetition */ |
6119 | *code = bravalue; |
6120 | tempcode = code; |
6121 | tempreqvary = cd->req_varyopt; /* Save value before bracket */ |
6122 | tempbracount = cd->bracount; /* Save value before bracket */ |
6123 | length_prevgroup = 0; /* Initialize for pre-compile phase */ |
6124 | |
6125 | if (!compile_regex( |
6126 | newoptions, /* The complete new option state */ |
6127 | &tempcode, /* Where to put code (updated) */ |
6128 | &ptr, /* Input pointer (updated) */ |
6129 | errorcodeptr, /* Where to put an error message */ |
6130 | (bravalue == OP_ASSERTBACK || |
6131 | bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */ |
6132 | reset_bracount, /* True if (?| group */ |
6133 | skipbytes, /* Skip over bracket number */ |
6134 | cond_depth + |
6135 | ((bravalue == OP_COND)?1:0), /* Depth of condition subpatterns */ |
6136 | &subfirstbyte, /* For possible first char */ |
6137 | &subreqbyte, /* For possible last char */ |
6138 | bcptr, /* Current branch chain */ |
6139 | cd, /* Tables block */ |
6140 | (lengthptr == NULL)? NULL : /* Actual compile phase */ |
6141 | &length_prevgroup /* Pre-compile phase */ |
6142 | )) |
6143 | goto FAILED; |
6144 | |
6145 | /* If this was an atomic group and there are no capturing groups within it, |
6146 | generate OP_ONCE_NC instead of OP_ONCE. */ |
6147 | |
6148 | if (bravalue == OP_ONCE && cd->bracount <= tempbracount) |
6149 | *code = OP_ONCE_NC; |
6150 | |
6151 | if (bravalue >= OP_ASSERT && bravalue <= OP_ASSERTBACK_NOT) |
6152 | cd->assert_depth -= 1; |
6153 | |
6154 | /* At the end of compiling, code is still pointing to the start of the |
6155 | group, while tempcode has been updated to point past the end of the group. |
6156 | The pattern pointer (ptr) is on the bracket. |
6157 | |
6158 | If this is a conditional bracket, check that there are no more than |
6159 | two branches in the group, or just one if it's a DEFINE group. We do this |
6160 | in the real compile phase, not in the pre-pass, where the whole group may |
6161 | not be available. */ |
6162 | |
6163 | if (bravalue == OP_COND && lengthptr == NULL) |
6164 | { |
6165 | uschar *tc = code; |
6166 | int condcount = 0; |
6167 | |
6168 | do { |
6169 | condcount++; |
6170 | tc += GET(tc,1); |
6171 | } |
6172 | while (*tc != OP_KET); |
6173 | |
6174 | /* A DEFINE group is never obeyed inline (the "condition" is always |
6175 | false). It must have only one branch. */ |
6176 | |
6177 | if (code[LINK_SIZE+1] == OP_DEF) |
6178 | { |
6179 | if (condcount > 1) |
6180 | { |
6181 | *errorcodeptr = ERR54; |
6182 | goto FAILED; |
6183 | } |
6184 | bravalue = OP_DEF; /* Just a flag to suppress char handling below */ |
6185 | } |
6186 | |
6187 | /* A "normal" conditional group. If there is just one branch, we must not |
6188 | make use of its firstbyte or reqbyte, because this is equivalent to an |
6189 | empty second branch. */ |
6190 | |
6191 | else |
6192 | { |
6193 | if (condcount > 2) |
6194 | { |
6195 | *errorcodeptr = ERR27; |
6196 | goto FAILED; |
6197 | } |
6198 | if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE; |
6199 | } |
6200 | } |
6201 | |
6202 | /* Error if hit end of pattern */ |
6203 | |
6204 | if (*ptr != CHAR_RIGHT_PARENTHESIS) |
6205 | { |
6206 | *errorcodeptr = ERR14; |
6207 | goto FAILED; |
6208 | } |
6209 | |
6210 | /* In the pre-compile phase, update the length by the length of the group, |
6211 | less the brackets at either end. Then reduce the compiled code to just a |
6212 | set of non-capturing brackets so that it doesn't use much memory if it is |
6213 | duplicated by a quantifier.*/ |
6214 | |
6215 | if (lengthptr != NULL) |
6216 | { |
6217 | if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE) |
6218 | { |
6219 | *errorcodeptr = ERR20; |
6220 | goto FAILED; |
6221 | } |
6222 | *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE; |
6223 | code++; /* This already contains bravalue */ |
6224 | PUTINC(code, 0, 1 + LINK_SIZE); |
6225 | *code++ = OP_KET; |
6226 | PUTINC(code, 0, 1 + LINK_SIZE); |
6227 | break; /* No need to waste time with special character handling */ |
6228 | } |
6229 | |
6230 | /* Otherwise update the main code pointer to the end of the group. */ |
6231 | |
6232 | code = tempcode; |
6233 | |
6234 | /* For a DEFINE group, required and first character settings are not |
6235 | relevant. */ |
6236 | |
6237 | if (bravalue == OP_DEF) break; |
6238 | |
6239 | /* Handle updating of the required and first characters for other types of |
6240 | group. Update for normal brackets of all kinds, and conditions with two |
6241 | branches (see code above). If the bracket is followed by a quantifier with |
6242 | zero repeat, we have to back off. Hence the definition of zeroreqbyte and |
6243 | zerofirstbyte outside the main loop so that they can be accessed for the |
6244 | back off. */ |
6245 | |
6246 | zeroreqbyte = reqbyte; |
6247 | zerofirstbyte = firstbyte; |
6248 | groupsetfirstbyte = FALSE; |
6249 | |
6250 | if (bravalue >= OP_ONCE) |
6251 | { |
6252 | /* If we have not yet set a firstbyte in this branch, take it from the |
6253 | subpattern, remembering that it was set here so that a repeat of more |
6254 | than one can replicate it as reqbyte if necessary. If the subpattern has |
6255 | no firstbyte, set "none" for the whole branch. In both cases, a zero |
6256 | repeat forces firstbyte to "none". */ |
6257 | |
6258 | if (firstbyte == REQ_UNSET) |
6259 | { |
6260 | if (subfirstbyte >= 0) |
6261 | { |
6262 | firstbyte = subfirstbyte; |
6263 | groupsetfirstbyte = TRUE; |
6264 | } |
6265 | else firstbyte = REQ_NONE; |
6266 | zerofirstbyte = REQ_NONE; |
6267 | } |
6268 | |
6269 | /* If firstbyte was previously set, convert the subpattern's firstbyte |
6270 | into reqbyte if there wasn't one, using the vary flag that was in |
6271 | existence beforehand. */ |
6272 | |
6273 | else if (subfirstbyte >= 0 && subreqbyte < 0) |
6274 | subreqbyte = subfirstbyte | tempreqvary; |
6275 | |
6276 | /* If the subpattern set a required byte (or set a first byte that isn't |
6277 | really the first byte - see above), set it. */ |
6278 | |
6279 | if (subreqbyte >= 0) reqbyte = subreqbyte; |
6280 | } |
6281 | |
6282 | /* For a forward assertion, we take the reqbyte, if set. This can be |
6283 | helpful if the pattern that follows the assertion doesn't set a different |
6284 | char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte |
6285 | for an assertion, however because it leads to incorrect effect for patterns |
6286 | such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead |
6287 | of a firstbyte. This is overcome by a scan at the end if there's no |
6288 | firstbyte, looking for an asserted first char. */ |
6289 | |
6290 | else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte; |
6291 | break; /* End of processing '(' */ |
6292 | |
6293 | |
6294 | /* ===================================================================*/ |
6295 | /* Handle metasequences introduced by \. For ones like \d, the ESC_ values |
6296 | are arranged to be the negation of the corresponding OP_values in the |
6297 | default case when PCRE_UCP is not set. For the back references, the values |
6298 | are ESC_REF plus the reference number. Only back references and those types |
6299 | that consume a character may be repeated. We can test for values between |
6300 | ESC_b and ESC_Z for the latter; this may have to change if any new ones are |
6301 | ever created. */ |
6302 | |
6303 | case CHAR_BACKSLASH: |
6304 | tempptr = ptr; |
6305 | c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE); |
6306 | if (*errorcodeptr != 0) goto FAILED; |
6307 | |
6308 | if (c < 0) |
6309 | { |
6310 | if (-c == ESC_Q) /* Handle start of quoted string */ |
6311 | { |
6312 | if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E) |
6313 | ptr += 2; /* avoid empty string */ |
6314 | else inescq = TRUE; |
6315 | continue; |
6316 | } |
6317 | |
6318 | if (-c == ESC_E) continue; /* Perl ignores an orphan \E */ |
6319 | |
6320 | /* For metasequences that actually match a character, we disable the |
6321 | setting of a first character if it hasn't already been set. */ |
6322 | |
6323 | if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z) |
6324 | firstbyte = REQ_NONE; |
6325 | |
6326 | /* Set values to reset to if this is followed by a zero repeat. */ |
6327 | |
6328 | zerofirstbyte = firstbyte; |
6329 | zeroreqbyte = reqbyte; |
6330 | |
6331 | /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n' |
6332 | is a subroutine call by number (Oniguruma syntax). In fact, the value |
6333 | -ESC_g is returned only for these cases. So we don't need to check for < |
6334 | or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is |
6335 | -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as |
6336 | that is a synonym for a named back reference). */ |
6337 | |
6338 | if (-c == ESC_g) |
6339 | { |
6340 | const uschar *p; |
6341 | save_hwm = cd->hwm; /* Normally this is set when '(' is read */ |
6342 | terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)? |
6343 | CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE; |
6344 | |
6345 | /* These two statements stop the compiler for warning about possibly |
6346 | unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In |
6347 | fact, because we actually check for a number below, the paths that |
6348 | would actually be in error are never taken. */ |
6349 | |
6350 | skipbytes = 0; |
6351 | reset_bracount = FALSE; |
6352 | |
6353 | /* Test for a name */ |
6354 | |
6355 | if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS) |
6356 | { |
6357 | BOOL isnumber = TRUE; |
6358 | for (p = ptr + 1; *p != 0 && *p != terminator; p++) |
6359 | { |
6360 | if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE; |
6361 | if ((cd->ctypes[*p] & ctype_word) == 0) break; |
6362 | } |
6363 | if (*p != terminator) |
6364 | { |
6365 | *errorcodeptr = ERR57; |
6366 | break; |
6367 | } |
6368 | if (isnumber) |
6369 | { |
6370 | ptr++; |
6371 | goto HANDLE_NUMERICAL_RECURSION; |
6372 | } |
6373 | is_recurse = TRUE; |
6374 | goto NAMED_REF_OR_RECURSE; |
6375 | } |
6376 | |
6377 | /* Test a signed number in angle brackets or quotes. */ |
6378 | |
6379 | p = ptr + 2; |
6380 | while ((digitab[*p] & ctype_digit) != 0) p++; |
6381 | if (*p != terminator) |
6382 | { |
6383 | *errorcodeptr = ERR57; |
6384 | break; |
6385 | } |
6386 | ptr++; |
6387 | goto HANDLE_NUMERICAL_RECURSION; |
6388 | } |
6389 | |
6390 | /* \k<name> or \k'name' is a back reference by name (Perl syntax). |
6391 | We also support \k{name} (.NET syntax). */ |
6392 | |
6393 | if (-c == ESC_k) |
6394 | { |
6395 | if ((ptr[1] != CHAR_LESS_THAN_SIGN && |
6396 | ptr[1] != CHAR_APOSTROPHE && ptr[1] != CHAR_LEFT_CURLY_BRACKET)) |
6397 | { |
6398 | *errorcodeptr = ERR69; |
6399 | break; |
6400 | } |
6401 | is_recurse = FALSE; |
6402 | terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)? |
6403 | CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)? |
6404 | CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET; |
6405 | goto NAMED_REF_OR_RECURSE; |
6406 | } |
6407 | |
6408 | /* Back references are handled specially; must disable firstbyte if |
6409 | not set to cope with cases like (?=(\w+))\1: which would otherwise set |
6410 | ':' later. */ |
6411 | |
6412 | if (-c >= ESC_REF) |
6413 | { |
6414 | open_capitem *oc; |
6415 | recno = -c - ESC_REF; |
6416 | |
6417 | HANDLE_REFERENCE: /* Come here from named backref handling */ |
6418 | if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE; |
6419 | previous = code; |
6420 | *code++ = ((options & PCRE_CASELESS) != 0)? OP_REFI : OP_REF; |
6421 | PUT2INC(code, 0, recno); |
6422 | cd->backref_map |= (recno < 32)? (1 << recno) : 1; |
6423 | if (recno > cd->top_backref) cd->top_backref = recno; |
6424 | |
6425 | /* Check to see if this back reference is recursive, that it, it |
6426 | is inside the group that it references. A flag is set so that the |
6427 | group can be made atomic. */ |
6428 | |
6429 | for (oc = cd->open_caps; oc != NULL; oc = oc->next) |
6430 | { |
6431 | if (oc->number == recno) |
6432 | { |
6433 | oc->flag = TRUE; |
6434 | break; |
6435 | } |
6436 | } |
6437 | } |
6438 | |
6439 | /* So are Unicode property matches, if supported. */ |
6440 | |
6441 | #ifdef SUPPORT_UCP |
6442 | else if (-c == ESC_P || -c == ESC_p) |
6443 | { |
6444 | BOOL negated; |
6445 | int pdata; |
6446 | int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr); |
6447 | if (ptype < 0) goto FAILED; |
6448 | previous = code; |
6449 | *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP; |
6450 | *code++ = ptype; |
6451 | *code++ = pdata; |
6452 | } |
6453 | #else |
6454 | |
6455 | /* If Unicode properties are not supported, \X, \P, and \p are not |
6456 | allowed. */ |
6457 | |
6458 | else if (-c == ESC_X || -c == ESC_P || -c == ESC_p) |
6459 | { |
6460 | *errorcodeptr = ERR45; |
6461 | goto FAILED; |
6462 | } |
6463 | #endif |
6464 | |
6465 | /* For the rest (including \X when Unicode properties are supported), we |
6466 | can obtain the OP value by negating the escape value in the default |
6467 | situation when PCRE_UCP is not set. When it *is* set, we substitute |
6468 | Unicode property tests. */ |
6469 | |
6470 | else |
6471 | { |
6472 | #ifdef SUPPORT_UCP |
6473 | if (-c >= ESC_DU && -c <= ESC_wu) |
6474 | { |
6475 | nestptr = ptr + 1; /* Where to resume */ |
6476 | ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */ |
6477 | } |
6478 | else |
6479 | #endif |
6480 | { |
6481 | previous = (-c > ESC_b && -c < ESC_Z)? code : NULL; |
6482 | *code++ = -c; |
6483 | } |
6484 | } |
6485 | continue; |
6486 | } |
6487 | |
6488 | /* We have a data character whose value is in c. In UTF-8 mode it may have |
6489 | a value > 127. We set its representation in the length/buffer, and then |
6490 | handle it as a data character. */ |
6491 | |
6492 | #ifdef SUPPORT_UTF8 |
6493 | if (utf8 && c > 127) |
6494 | mclength = _pcre_ord2utf8(c, mcbuffer); |
6495 | else |
6496 | #endif |
6497 | |
6498 | { |
6499 | mcbuffer[0] = c; |
6500 | mclength = 1; |
6501 | } |
6502 | goto ONE_CHAR; |
6503 | |
6504 | |
6505 | /* ===================================================================*/ |
6506 | /* Handle a literal character. It is guaranteed not to be whitespace or # |
6507 | when the extended flag is set. If we are in UTF-8 mode, it may be a |
6508 | multi-byte literal character. */ |
6509 | |
6510 | default: |
6511 | NORMAL_CHAR: |
6512 | mclength = 1; |
6513 | mcbuffer[0] = c; |
6514 | |
6515 | #ifdef SUPPORT_UTF8 |
6516 | if (utf8 && c >= 0xc0) |
6517 | { |
6518 | while ((ptr[1] & 0xc0) == 0x80) |
6519 | mcbuffer[mclength++] = *(++ptr); |
6520 | } |
6521 | #endif |
6522 | |
6523 | /* At this point we have the character's bytes in mcbuffer, and the length |
6524 | in mclength. When not in UTF-8 mode, the length is always 1. */ |
6525 | |
6526 | ONE_CHAR: |
6527 | previous = code; |
6528 | *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARI : OP_CHAR; |
6529 | for (c = 0; c < mclength; c++) *code++ = mcbuffer[c]; |
6530 | |
6531 | /* Remember if \r or \n were seen */ |
6532 | |
6533 | if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL) |
6534 | cd->external_flags |= PCRE_HASCRORLF; |
6535 | |
6536 | /* Set the first and required bytes appropriately. If no previous first |
6537 | byte, set it from this character, but revert to none on a zero repeat. |
6538 | Otherwise, leave the firstbyte value alone, and don't change it on a zero |
6539 | repeat. */ |
6540 | |
6541 | if (firstbyte == REQ_UNSET) |
6542 | { |
6543 | zerofirstbyte = REQ_NONE; |
6544 | zeroreqbyte = reqbyte; |
6545 | |
6546 | /* If the character is more than one byte long, we can set firstbyte |
6547 | only if it is not to be matched caselessly. */ |
6548 | |
6549 | if (mclength == 1 || req_caseopt == 0) |
6550 | { |
6551 | firstbyte = mcbuffer[0] | req_caseopt; |
6552 | if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt; |
6553 | } |
6554 | else firstbyte = reqbyte = REQ_NONE; |
6555 | } |
6556 | |
6557 | /* firstbyte was previously set; we can set reqbyte only if the length is |
6558 | 1 or the matching is caseful. */ |
6559 | |
6560 | else |
6561 | { |
6562 | zerofirstbyte = firstbyte; |
6563 | zeroreqbyte = reqbyte; |
6564 | if (mclength == 1 || req_caseopt == 0) |
6565 | reqbyte = code[-1] | req_caseopt | cd->req_varyopt; |
6566 | } |
6567 | |
6568 | break; /* End of literal character handling */ |
6569 | } |
6570 | } /* end of big loop */ |
6571 | |
6572 | |
6573 | /* Control never reaches here by falling through, only by a goto for all the |
6574 | error states. Pass back the position in the pattern so that it can be displayed |
6575 | to the user for diagnosing the error. */ |
6576 | |
6577 | FAILED: |
6578 | *ptrptr = ptr; |
6579 | return FALSE; |
6580 | } |
6581 | |
6582 | |
6583 | |
6584 | |
6585 | /************************************************* |
6586 | * Compile sequence of alternatives * |
6587 | *************************************************/ |
6588 | |
6589 | /* On entry, ptr is pointing past the bracket character, but on return it |
6590 | points to the closing bracket, or vertical bar, or end of string. The code |
6591 | variable is pointing at the byte into which the BRA operator has been stored. |
6592 | This function is used during the pre-compile phase when we are trying to find |
6593 | out the amount of memory needed, as well as during the real compile phase. The |
6594 | value of lengthptr distinguishes the two phases. |
6595 | |
6596 | Arguments: |
6597 | options option bits, including any changes for this subpattern |
6598 | codeptr -> the address of the current code pointer |
6599 | ptrptr -> the address of the current pattern pointer |
6600 | errorcodeptr -> pointer to error code variable |
6601 | lookbehind TRUE if this is a lookbehind assertion |
6602 | reset_bracount TRUE to reset the count for each branch |
6603 | skipbytes skip this many bytes at start (for brackets and OP_COND) |
6604 | cond_depth depth of nesting for conditional subpatterns |
6605 | firstbyteptr place to put the first required character, or a negative number |
6606 | reqbyteptr place to put the last required character, or a negative number |
6607 | bcptr pointer to the chain of currently open branches |
6608 | cd points to the data block with tables pointers etc. |
6609 | lengthptr NULL during the real compile phase |
6610 | points to length accumulator during pre-compile phase |
6611 | |
6612 | Returns: TRUE on success |
6613 | */ |
6614 | |
6615 | static BOOL |
6616 | compile_regex(int options, uschar **codeptr, const uschar **ptrptr, |
6617 | int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes, |
6618 | int cond_depth, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, |
6619 | compile_data *cd, int *lengthptr) |
6620 | { |
6621 | const uschar *ptr = *ptrptr; |
6622 | uschar *code = *codeptr; |
6623 | uschar *last_branch = code; |
6624 | uschar *start_bracket = code; |
6625 | uschar *reverse_count = NULL; |
6626 | open_capitem capitem; |
6627 | int capnumber = 0; |
6628 | int firstbyte, reqbyte; |
6629 | int branchfirstbyte, branchreqbyte; |
6630 | int length; |
6631 | int orig_bracount; |
6632 | int max_bracount; |
6633 | branch_chain bc; |
6634 | |
6635 | bc.outer = bcptr; |
6636 | bc.current_branch = code; |
6637 | |
6638 | firstbyte = reqbyte = REQ_UNSET; |
6639 | |
6640 | /* Accumulate the length for use in the pre-compile phase. Start with the |
6641 | length of the BRA and KET and any extra bytes that are required at the |
6642 | beginning. We accumulate in a local variable to save frequent testing of |
6643 | lenthptr for NULL. We cannot do this by looking at the value of code at the |
6644 | start and end of each alternative, because compiled items are discarded during |
6645 | the pre-compile phase so that the work space is not exceeded. */ |
6646 | |
6647 | length = 2 + 2*LINK_SIZE + skipbytes; |
6648 | |
6649 |