Annotation of win32/pcre/pcre_compile.c, revision 1.4
1.1 misha 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
1.4 ! misha 9: Copyright (c) 1997-2010 University of Cambridge
1.1 misha 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:
1.4 ! misha 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. */
1.1 misha 59:
1.4 ! misha 60: #ifdef PCRE_DEBUG
1.1 misha 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:
1.4 ! misha 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:
1.1 misha 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:
1.3 misha 106: #ifndef EBCDIC
107:
108: /* This is the "normal" table for ASCII systems or for EBCDIC systems running
109: in UTF-8 mode. */
110:
1.1 misha 111: static const short int escapes[] = {
1.3 misha 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,
1.4 ! misha 127: -ESC_N, 0,
1.3 misha 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
1.1 misha 150: };
151:
1.3 misha 152: #else
153:
154: /* This is the "abnormal" table for EBCDIC systems without UTF-8 support. */
155:
1.1 misha 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,
1.4 ! misha 174: /* D0 */ '}', 0, -ESC_K, 0, 0,-ESC_N, 0, -ESC_P,
1.1 misha 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
1.3 misha 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. */
1.1 misha 189:
190: typedef struct verbitem {
1.4 ! misha 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 */
1.1 misha 194: } verbitem;
195:
196: static const char verbnames[] =
1.4 ! misha 197: "\0" /* Empty name is a shorthand for MARK */
! 198: STRING_MARK0
1.3 misha 199: STRING_ACCEPT0
200: STRING_COMMIT0
201: STRING_F0
202: STRING_FAIL0
203: STRING_PRUNE0
204: STRING_SKIP0
205: STRING_THEN;
1.1 misha 206:
207: static const verbitem verbs[] = {
1.4 ! misha 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 }
1.1 misha 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[] =
1.3 misha 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;
1.1 misha 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:
1.4 ! misha 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
1.1 misha 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
1.4 ! misha 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. */
1.1 misha 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"
1.4 ! misha 374: "PCRE does not support \\L, \\l, \\N{name}, \\U, or \\u\0"
1.1 misha 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"
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"
1.4 ! misha 400: "an argument is not allowed for (*ACCEPT), (*FAIL), or (*COMMIT)\0"
1.1 misha 401: /* 60 */
402: "(*VERB) not recognized\0"
403: "number is too big\0"
404: "subpattern name expected\0"
405: "digit expected after (?+\0"
1.4 ! misha 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: ;
1.1 misha 412:
413: /* Table to identify digits and hex digits. This is used when compiling
414: patterns. Note that the tables in chartables are dependent on the locale, and
415: may mark arbitrary characters as digits - but the PCRE compiling code expects
416: to handle only 0-9, a-z, and A-Z as digits when compiling. That is why we have
417: a private table here. It costs 256 bytes, but it is a lot faster than doing
418: character value tests (at least in some simple cases I timed), and in some
419: applications one wants PCRE to compile efficiently as well as match
420: efficiently.
421:
422: For convenience, we use the same bit definitions as in chartables:
423:
424: 0x04 decimal digit
425: 0x08 hexadecimal digit
426:
427: Then we can use ctype_digit and ctype_xdigit in the code. */
428:
1.3 misha 429: #ifndef EBCDIC
430:
431: /* This is the "normal" case, for ASCII systems, and EBCDIC systems running in
432: UTF-8 mode. */
433:
1.1 misha 434: static const unsigned char digitab[] =
435: {
436: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 */
437: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
438: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 */
439: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
440: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - ' */
441: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ( - / */
442: 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 */
443: 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00, /* 8 - ? */
444: 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* @ - G */
445: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H - O */
446: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* P - W */
447: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* X - _ */
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 -127 */
452: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 128-135 */
453: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 136-143 */
454: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144-151 */
455: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 152-159 */
456: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160-167 */
457: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 168-175 */
458: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 176-183 */
459: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
460: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 192-199 */
461: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 200-207 */
462: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 208-215 */
463: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 216-223 */
464: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 224-231 */
465: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 232-239 */
466: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 240-247 */
467: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};/* 248-255 */
468:
1.3 misha 469: #else
470:
471: /* This is the "abnormal" case, for EBCDIC systems not running in UTF-8 mode. */
472:
1.1 misha 473: static const unsigned char digitab[] =
474: {
475: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 0- 7 0 */
476: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 8- 15 */
477: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 16- 23 10 */
478: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
479: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 32- 39 20 */
480: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
481: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 30 */
482: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
483: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 40 */
484: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 72- | */
485: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 50 */
486: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 88- 95 */
487: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 60 */
488: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 104- ? */
489: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 70 */
490: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
491: 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* 128- g 80 */
492: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
493: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 144- p 90 */
494: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
495: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 160- x A0 */
496: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
497: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 B0 */
498: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
499: 0x00,0x08,0x08,0x08,0x08,0x08,0x08,0x00, /* { - G C0 */
500: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
501: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* } - P D0 */
502: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
503: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* \ - X E0 */
504: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
505: 0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c,0x0c, /* 0 - 7 F0 */
506: 0x0c,0x0c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
507:
508: static const unsigned char ebcdic_chartab[] = { /* chartable partial dup */
509: 0x80,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 0- 7 */
510: 0x00,0x00,0x00,0x00,0x01,0x01,0x00,0x00, /* 8- 15 */
511: 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 16- 23 */
512: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 24- 31 */
513: 0x00,0x00,0x00,0x00,0x00,0x01,0x00,0x00, /* 32- 39 */
514: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 40- 47 */
515: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 48- 55 */
516: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 56- 63 */
517: 0x01,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - 71 */
518: 0x00,0x00,0x00,0x80,0x00,0x80,0x80,0x80, /* 72- | */
519: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* & - 87 */
520: 0x00,0x00,0x00,0x80,0x80,0x80,0x00,0x00, /* 88- 95 */
521: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* - -103 */
522: 0x00,0x00,0x00,0x00,0x00,0x10,0x00,0x80, /* 104- ? */
523: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 112-119 */
524: 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* 120- " */
525: 0x00,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* 128- g */
526: 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* h -143 */
527: 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* 144- p */
528: 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* q -159 */
529: 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* 160- x */
530: 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* y -175 */
531: 0x80,0x00,0x00,0x00,0x00,0x00,0x00,0x00, /* ^ -183 */
532: 0x00,0x00,0x80,0x00,0x00,0x00,0x00,0x00, /* 184-191 */
533: 0x80,0x1a,0x1a,0x1a,0x1a,0x1a,0x1a,0x12, /* { - G */
534: 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* H -207 */
535: 0x00,0x12,0x12,0x12,0x12,0x12,0x12,0x12, /* } - P */
536: 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Q -223 */
537: 0x00,0x00,0x12,0x12,0x12,0x12,0x12,0x12, /* \ - X */
538: 0x12,0x12,0x00,0x00,0x00,0x00,0x00,0x00, /* Y -239 */
539: 0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c,0x1c, /* 0 - 7 */
540: 0x1c,0x1c,0x00,0x00,0x00,0x00,0x00,0x00};/* 8 -255 */
541: #endif
542:
543:
544: /* Definition to allow mutual recursion */
545:
546: static BOOL
547: compile_regex(int, int, uschar **, const uschar **, int *, BOOL, BOOL, int,
548: int *, int *, branch_chain *, compile_data *, int *);
549:
550:
551:
552: /*************************************************
553: * Find an error text *
554: *************************************************/
555:
556: /* The error texts are now all in one long string, to save on relocations. As
557: some of the text is of unknown length, we can't use a table of offsets.
558: Instead, just count through the strings. This is not a performance issue
559: because it happens only when there has been a compilation error.
560:
561: Argument: the error number
562: Returns: pointer to the error string
563: */
564:
565: static const char *
566: find_error_text(int n)
567: {
568: const char *s = error_texts;
1.4 ! misha 569: for (; n > 0; n--)
! 570: {
! 571: while (*s++ != 0) {};
! 572: if (*s == 0) return "Error text not found (please report)";
! 573: }
1.1 misha 574: return s;
575: }
576:
577:
578: /*************************************************
579: * Handle escapes *
580: *************************************************/
581:
582: /* This function is called when a \ has been encountered. It either returns a
583: positive value for a simple escape such as \n, or a negative value which
584: encodes one of the more complicated things such as \d. A backreference to group
585: n is returned as -(ESC_REF + n); ESC_REF is the highest ESC_xxx macro. When
586: UTF-8 is enabled, a positive value greater than 255 may be returned. On entry,
587: ptr is pointing at the \. On exit, it is on the final character of the escape
588: sequence.
589:
590: Arguments:
591: ptrptr points to the pattern position pointer
592: errorcodeptr points to the errorcode variable
593: bracount number of previous extracting brackets
594: options the options bits
595: isclass TRUE if inside a character class
596:
597: Returns: zero or positive => a data character
598: negative => a special escape sequence
599: on error, errorcodeptr is set
600: */
601:
602: static int
603: check_escape(const uschar **ptrptr, int *errorcodeptr, int bracount,
604: int options, BOOL isclass)
605: {
606: BOOL utf8 = (options & PCRE_UTF8) != 0;
607: const uschar *ptr = *ptrptr + 1;
608: int c, i;
609:
610: GETCHARINCTEST(c, ptr); /* Get character value, increment pointer */
611: ptr--; /* Set pointer back to the last byte */
612:
613: /* If backslash is at the end of the pattern, it's an error. */
614:
615: if (c == 0) *errorcodeptr = ERR1;
616:
617: /* Non-alphanumerics are literals. For digits or letters, do an initial lookup
618: in a table. A non-zero result is something that can be returned immediately.
619: Otherwise further processing may be required. */
620:
1.3 misha 621: #ifndef EBCDIC /* ASCII/UTF-8 coding */
622: else if (c < CHAR_0 || c > CHAR_z) {} /* Not alphanumeric */
623: else if ((i = escapes[c - CHAR_0]) != 0) c = i;
1.1 misha 624:
625: #else /* EBCDIC coding */
626: else if (c < 'a' || (ebcdic_chartab[c] & 0x0E) == 0) {} /* Not alphanumeric */
627: else if ((i = escapes[c - 0x48]) != 0) c = i;
628: #endif
629:
630: /* Escapes that need further processing, or are illegal. */
631:
632: else
633: {
634: const uschar *oldptr;
635: BOOL braced, negated;
636:
637: switch (c)
638: {
639: /* A number of Perl escapes are not handled by PCRE. We give an explicit
640: error. */
641:
1.3 misha 642: case CHAR_l:
643: case CHAR_L:
644: case CHAR_u:
645: case CHAR_U:
1.1 misha 646: *errorcodeptr = ERR37;
647: break;
648:
649: /* \g must be followed by one of a number of specific things:
650:
651: (1) A number, either plain or braced. If positive, it is an absolute
652: backreference. If negative, it is a relative backreference. This is a Perl
653: 5.10 feature.
654:
655: (2) Perl 5.10 also supports \g{name} as a reference to a named group. This
656: is part of Perl's movement towards a unified syntax for back references. As
657: this is synonymous with \k{name}, we fudge it up by pretending it really
658: was \k.
659:
660: (3) For Oniguruma compatibility we also support \g followed by a name or a
661: number either in angle brackets or in single quotes. However, these are
662: (possibly recursive) subroutine calls, _not_ backreferences. Just return
663: the -ESC_g code (cf \k). */
664:
1.3 misha 665: case CHAR_g:
666: if (ptr[1] == CHAR_LESS_THAN_SIGN || ptr[1] == CHAR_APOSTROPHE)
1.1 misha 667: {
668: c = -ESC_g;
669: break;
670: }
671:
672: /* Handle the Perl-compatible cases */
673:
1.3 misha 674: if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1.1 misha 675: {
676: const uschar *p;
1.3 misha 677: for (p = ptr+2; *p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET; p++)
678: if (*p != CHAR_MINUS && (digitab[*p] & ctype_digit) == 0) break;
679: if (*p != 0 && *p != CHAR_RIGHT_CURLY_BRACKET)
1.1 misha 680: {
681: c = -ESC_k;
682: break;
683: }
684: braced = TRUE;
685: ptr++;
686: }
687: else braced = FALSE;
688:
1.3 misha 689: if (ptr[1] == CHAR_MINUS)
1.1 misha 690: {
691: negated = TRUE;
692: ptr++;
693: }
694: else negated = FALSE;
695:
696: c = 0;
697: while ((digitab[ptr[1]] & ctype_digit) != 0)
1.3 misha 698: c = c * 10 + *(++ptr) - CHAR_0;
1.1 misha 699:
700: if (c < 0) /* Integer overflow */
701: {
702: *errorcodeptr = ERR61;
703: break;
704: }
705:
1.3 misha 706: if (braced && *(++ptr) != CHAR_RIGHT_CURLY_BRACKET)
1.1 misha 707: {
708: *errorcodeptr = ERR57;
709: break;
710: }
711:
712: if (c == 0)
713: {
714: *errorcodeptr = ERR58;
715: break;
716: }
717:
718: if (negated)
719: {
720: if (c > bracount)
721: {
722: *errorcodeptr = ERR15;
723: break;
724: }
725: c = bracount - (c - 1);
726: }
727:
728: c = -(ESC_REF + c);
729: break;
730:
731: /* The handling of escape sequences consisting of a string of digits
732: starting with one that is not zero is not straightforward. By experiment,
733: the way Perl works seems to be as follows:
734:
735: Outside a character class, the digits are read as a decimal number. If the
736: number is less than 10, or if there are that many previous extracting
737: left brackets, then it is a back reference. Otherwise, up to three octal
738: digits are read to form an escaped byte. Thus \123 is likely to be octal
739: 123 (cf \0123, which is octal 012 followed by the literal 3). If the octal
740: value is greater than 377, the least significant 8 bits are taken. Inside a
741: character class, \ followed by a digit is always an octal number. */
742:
1.3 misha 743: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4: case CHAR_5:
744: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
1.1 misha 745:
746: if (!isclass)
747: {
748: oldptr = ptr;
1.3 misha 749: c -= CHAR_0;
1.1 misha 750: while ((digitab[ptr[1]] & ctype_digit) != 0)
1.3 misha 751: c = c * 10 + *(++ptr) - CHAR_0;
1.1 misha 752: if (c < 0) /* Integer overflow */
753: {
754: *errorcodeptr = ERR61;
755: break;
756: }
757: if (c < 10 || c <= bracount)
758: {
759: c = -(ESC_REF + c);
760: break;
761: }
762: ptr = oldptr; /* Put the pointer back and fall through */
763: }
764:
765: /* Handle an octal number following \. If the first digit is 8 or 9, Perl
766: generates a binary zero byte and treats the digit as a following literal.
767: Thus we have to pull back the pointer by one. */
768:
1.3 misha 769: if ((c = *ptr) >= CHAR_8)
1.1 misha 770: {
771: ptr--;
772: c = 0;
773: break;
774: }
775:
776: /* \0 always starts an octal number, but we may drop through to here with a
777: larger first octal digit. The original code used just to take the least
778: significant 8 bits of octal numbers (I think this is what early Perls used
779: to do). Nowadays we allow for larger numbers in UTF-8 mode, but no more
780: than 3 octal digits. */
781:
1.3 misha 782: case CHAR_0:
783: c -= CHAR_0;
784: while(i++ < 2 && ptr[1] >= CHAR_0 && ptr[1] <= CHAR_7)
785: c = c * 8 + *(++ptr) - CHAR_0;
1.1 misha 786: if (!utf8 && c > 255) *errorcodeptr = ERR51;
787: break;
788:
789: /* \x is complicated. \x{ddd} is a character number which can be greater
790: than 0xff in utf8 mode, but only if the ddd are hex digits. If not, { is
791: treated as a data character. */
792:
1.3 misha 793: case CHAR_x:
794: if (ptr[1] == CHAR_LEFT_CURLY_BRACKET)
1.1 misha 795: {
796: const uschar *pt = ptr + 2;
797: int count = 0;
798:
799: c = 0;
800: while ((digitab[*pt] & ctype_xdigit) != 0)
801: {
802: register int cc = *pt++;
1.3 misha 803: if (c == 0 && cc == CHAR_0) continue; /* Leading zeroes */
1.1 misha 804: count++;
805:
1.3 misha 806: #ifndef EBCDIC /* ASCII/UTF-8 coding */
807: if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
808: c = (c << 4) + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1.1 misha 809: #else /* EBCDIC coding */
1.3 misha 810: if (cc >= CHAR_a && cc <= CHAR_z) cc += 64; /* Convert to upper case */
811: c = (c << 4) + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1.1 misha 812: #endif
813: }
814:
1.3 misha 815: if (*pt == CHAR_RIGHT_CURLY_BRACKET)
1.1 misha 816: {
817: if (c < 0 || count > (utf8? 8 : 2)) *errorcodeptr = ERR34;
818: ptr = pt;
819: break;
820: }
821:
822: /* If the sequence of hex digits does not end with '}', then we don't
823: recognize this construct; fall through to the normal \x handling. */
824: }
825:
826: /* Read just a single-byte hex-defined char */
827:
828: c = 0;
829: while (i++ < 2 && (digitab[ptr[1]] & ctype_xdigit) != 0)
830: {
1.3 misha 831: int cc; /* Some compilers don't like */
832: cc = *(++ptr); /* ++ in initializers */
833: #ifndef EBCDIC /* ASCII/UTF-8 coding */
834: if (cc >= CHAR_a) cc -= 32; /* Convert to upper case */
835: c = c * 16 + cc - ((cc < CHAR_A)? CHAR_0 : (CHAR_A - 10));
1.1 misha 836: #else /* EBCDIC coding */
1.3 misha 837: if (cc <= CHAR_z) cc += 64; /* Convert to upper case */
838: c = c * 16 + cc - ((cc >= CHAR_0)? CHAR_0 : (CHAR_A - 10));
1.1 misha 839: #endif
840: }
841: break;
842:
843: /* For \c, a following letter is upper-cased; then the 0x40 bit is flipped.
844: This coding is ASCII-specific, but then the whole concept of \cx is
845: ASCII-specific. (However, an EBCDIC equivalent has now been added.) */
846:
1.3 misha 847: case CHAR_c:
1.1 misha 848: c = *(++ptr);
849: if (c == 0)
850: {
851: *errorcodeptr = ERR2;
852: break;
853: }
854:
1.3 misha 855: #ifndef EBCDIC /* ASCII/UTF-8 coding */
856: if (c >= CHAR_a && c <= CHAR_z) c -= 32;
1.1 misha 857: c ^= 0x40;
858: #else /* EBCDIC coding */
1.3 misha 859: if (c >= CHAR_a && c <= CHAR_z) c += 64;
1.1 misha 860: c ^= 0xC0;
861: #endif
862: break;
863:
864: /* PCRE_EXTRA enables extensions to Perl in the matter of escapes. Any
865: other alphanumeric following \ is an error if PCRE_EXTRA was set;
866: otherwise, for Perl compatibility, it is a literal. This code looks a bit
867: odd, but there used to be some cases other than the default, and there may
868: be again in future, so I haven't "optimized" it. */
869:
870: default:
871: if ((options & PCRE_EXTRA) != 0) switch(c)
872: {
873: default:
874: *errorcodeptr = ERR3;
875: break;
876: }
877: break;
878: }
879: }
880:
1.4 ! misha 881: /* Perl supports \N{name} for character names, as well as plain \N for "not
! 882: newline". PCRE does not support \N{name}. */
! 883:
! 884: if (c == -ESC_N && ptr[1] == CHAR_LEFT_CURLY_BRACKET)
! 885: *errorcodeptr = ERR37;
! 886:
! 887: /* If PCRE_UCP is set, we change the values for \d etc. */
! 888:
! 889: if ((options & PCRE_UCP) != 0 && c <= -ESC_D && c >= -ESC_w)
! 890: c -= (ESC_DU - ESC_D);
! 891:
! 892: /* Set the pointer to the final character before returning. */
! 893:
1.1 misha 894: *ptrptr = ptr;
895: return c;
896: }
897:
898:
899:
900: #ifdef SUPPORT_UCP
901: /*************************************************
902: * Handle \P and \p *
903: *************************************************/
904:
905: /* This function is called after \P or \p has been encountered, provided that
906: PCRE is compiled with support for Unicode properties. On entry, ptrptr is
907: pointing at the P or p. On exit, it is pointing at the final character of the
908: escape sequence.
909:
910: Argument:
911: ptrptr points to the pattern position pointer
912: negptr points to a boolean that is set TRUE for negation else FALSE
913: dptr points to an int that is set to the detailed property value
914: errorcodeptr points to the error code variable
915:
916: Returns: type value from ucp_type_table, or -1 for an invalid type
917: */
918:
919: static int
920: get_ucp(const uschar **ptrptr, BOOL *negptr, int *dptr, int *errorcodeptr)
921: {
922: int c, i, bot, top;
923: const uschar *ptr = *ptrptr;
924: char name[32];
925:
926: c = *(++ptr);
927: if (c == 0) goto ERROR_RETURN;
928:
929: *negptr = FALSE;
930:
931: /* \P or \p can be followed by a name in {}, optionally preceded by ^ for
932: negation. */
933:
1.3 misha 934: if (c == CHAR_LEFT_CURLY_BRACKET)
1.1 misha 935: {
1.3 misha 936: if (ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
1.1 misha 937: {
938: *negptr = TRUE;
939: ptr++;
940: }
941: for (i = 0; i < (int)sizeof(name) - 1; i++)
942: {
943: c = *(++ptr);
944: if (c == 0) goto ERROR_RETURN;
1.3 misha 945: if (c == CHAR_RIGHT_CURLY_BRACKET) break;
1.1 misha 946: name[i] = c;
947: }
1.3 misha 948: if (c != CHAR_RIGHT_CURLY_BRACKET) goto ERROR_RETURN;
1.1 misha 949: name[i] = 0;
950: }
951:
952: /* Otherwise there is just one following character */
953:
954: else
955: {
956: name[0] = c;
957: name[1] = 0;
958: }
959:
960: *ptrptr = ptr;
961:
962: /* Search for a recognized property name using binary chop */
963:
964: bot = 0;
965: top = _pcre_utt_size;
966:
967: while (bot < top)
968: {
969: i = (bot + top) >> 1;
970: c = strcmp(name, _pcre_utt_names + _pcre_utt[i].name_offset);
971: if (c == 0)
972: {
973: *dptr = _pcre_utt[i].value;
974: return _pcre_utt[i].type;
975: }
976: if (c > 0) bot = i + 1; else top = i;
977: }
978:
979: *errorcodeptr = ERR47;
980: *ptrptr = ptr;
981: return -1;
982:
983: ERROR_RETURN:
984: *errorcodeptr = ERR46;
985: *ptrptr = ptr;
986: return -1;
987: }
988: #endif
989:
990:
991:
992:
993: /*************************************************
994: * Check for counted repeat *
995: *************************************************/
996:
997: /* This function is called when a '{' is encountered in a place where it might
998: start a quantifier. It looks ahead to see if it really is a quantifier or not.
999: It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd}
1000: where the ddds are digits.
1001:
1002: Arguments:
1003: p pointer to the first char after '{'
1004:
1005: Returns: TRUE or FALSE
1006: */
1007:
1008: static BOOL
1009: is_counted_repeat(const uschar *p)
1010: {
1011: if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1012: while ((digitab[*p] & ctype_digit) != 0) p++;
1.3 misha 1013: if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1.1 misha 1014:
1.3 misha 1015: if (*p++ != CHAR_COMMA) return FALSE;
1016: if (*p == CHAR_RIGHT_CURLY_BRACKET) return TRUE;
1.1 misha 1017:
1018: if ((digitab[*p++] & ctype_digit) == 0) return FALSE;
1019: while ((digitab[*p] & ctype_digit) != 0) p++;
1020:
1.3 misha 1021: return (*p == CHAR_RIGHT_CURLY_BRACKET);
1.1 misha 1022: }
1023:
1024:
1025:
1026: /*************************************************
1027: * Read repeat counts *
1028: *************************************************/
1029:
1030: /* Read an item of the form {n,m} and return the values. This is called only
1031: after is_counted_repeat() has confirmed that a repeat-count quantifier exists,
1032: so the syntax is guaranteed to be correct, but we need to check the values.
1033:
1034: Arguments:
1035: p pointer to first char after '{'
1036: minp pointer to int for min
1037: maxp pointer to int for max
1038: returned as -1 if no max
1039: errorcodeptr points to error code variable
1040:
1041: Returns: pointer to '}' on success;
1042: current ptr on error, with errorcodeptr set non-zero
1043: */
1044:
1045: static const uschar *
1046: read_repeat_counts(const uschar *p, int *minp, int *maxp, int *errorcodeptr)
1047: {
1048: int min = 0;
1049: int max = -1;
1050:
1051: /* Read the minimum value and do a paranoid check: a negative value indicates
1052: an integer overflow. */
1053:
1.3 misha 1054: while ((digitab[*p] & ctype_digit) != 0) min = min * 10 + *p++ - CHAR_0;
1.1 misha 1055: if (min < 0 || min > 65535)
1056: {
1057: *errorcodeptr = ERR5;
1058: return p;
1059: }
1060:
1061: /* Read the maximum value if there is one, and again do a paranoid on its size.
1062: Also, max must not be less than min. */
1063:
1.3 misha 1064: if (*p == CHAR_RIGHT_CURLY_BRACKET) max = min; else
1.1 misha 1065: {
1.3 misha 1066: if (*(++p) != CHAR_RIGHT_CURLY_BRACKET)
1.1 misha 1067: {
1068: max = 0;
1.3 misha 1069: while((digitab[*p] & ctype_digit) != 0) max = max * 10 + *p++ - CHAR_0;
1.1 misha 1070: if (max < 0 || max > 65535)
1071: {
1072: *errorcodeptr = ERR5;
1073: return p;
1074: }
1075: if (max < min)
1076: {
1077: *errorcodeptr = ERR4;
1078: return p;
1079: }
1080: }
1081: }
1082:
1083: /* Fill in the required variables, and pass back the pointer to the terminating
1084: '}'. */
1085:
1086: *minp = min;
1087: *maxp = max;
1088: return p;
1089: }
1090:
1091:
1092:
1093: /*************************************************
1.3 misha 1094: * Subroutine for finding forward reference *
1.1 misha 1095: *************************************************/
1096:
1.3 misha 1097: /* This recursive function is called only from find_parens() below. The
1098: top-level call starts at the beginning of the pattern. All other calls must
1099: start at a parenthesis. It scans along a pattern's text looking for capturing
1.1 misha 1100: subpatterns, and counting them. If it finds a named pattern that matches the
1101: name it is given, it returns its number. Alternatively, if the name is NULL, it
1.3 misha 1102: returns when it reaches a given numbered subpattern. We know that if (?P< is
1103: encountered, the name will be terminated by '>' because that is checked in the
1104: first pass. Recursion is used to keep track of subpatterns that reset the
1105: capturing group numbers - the (?| feature.
1.1 misha 1106:
1107: Arguments:
1.3 misha 1108: ptrptr address of the current character pointer (updated)
1.1 misha 1109: cd compile background data
1110: name name to seek, or NULL if seeking a numbered subpattern
1111: lorn name length, or subpattern number if name is NULL
1112: xmode TRUE if we are in /x mode
1.3 misha 1113: count pointer to the current capturing subpattern number (updated)
1.1 misha 1114:
1115: Returns: the number of the named subpattern, or -1 if not found
1116: */
1117:
1118: static int
1.3 misha 1119: find_parens_sub(uschar **ptrptr, compile_data *cd, const uschar *name, int lorn,
1120: BOOL xmode, int *count)
1.1 misha 1121: {
1.3 misha 1122: uschar *ptr = *ptrptr;
1123: int start_count = *count;
1124: int hwm_count = start_count;
1125: BOOL dup_parens = FALSE;
1126:
1127: /* If the first character is a parenthesis, check on the type of group we are
1128: dealing with. The very first call may not start with a parenthesis. */
1129:
1130: if (ptr[0] == CHAR_LEFT_PARENTHESIS)
1131: {
1.4 ! misha 1132: /* Handle specials such as (*SKIP) or (*UTF8) etc. */
! 1133:
! 1134: if (ptr[1] == CHAR_ASTERISK) ptr += 2;
! 1135:
! 1136: /* Handle a normal, unnamed capturing parenthesis. */
! 1137:
! 1138: else if (ptr[1] != CHAR_QUESTION_MARK)
! 1139: {
! 1140: *count += 1;
! 1141: if (name == NULL && *count == lorn) return *count;
! 1142: ptr++;
! 1143: }
! 1144:
! 1145: /* All cases now have (? at the start. Remember when we are in a group
! 1146: where the parenthesis numbers are duplicated. */
! 1147:
! 1148: else if (ptr[2] == CHAR_VERTICAL_LINE)
1.3 misha 1149: {
1150: ptr += 3;
1151: dup_parens = TRUE;
1152: }
1153:
1.4 ! misha 1154: /* Handle comments; all characters are allowed until a ket is reached. */
1.3 misha 1155:
1.4 ! misha 1156: else if (ptr[2] == CHAR_NUMBER_SIGN)
1.3 misha 1157: {
1.4 ! misha 1158: for (ptr += 3; *ptr != 0; ptr++) if (*ptr == CHAR_RIGHT_PARENTHESIS) break;
! 1159: goto FAIL_EXIT;
1.3 misha 1160: }
1161:
1162: /* Handle a condition. If it is an assertion, just carry on so that it
1163: is processed as normal. If not, skip to the closing parenthesis of the
1.4 ! misha 1164: condition (there can't be any nested parens). */
1.3 misha 1165:
1166: else if (ptr[2] == CHAR_LEFT_PARENTHESIS)
1167: {
1168: ptr += 2;
1169: if (ptr[1] != CHAR_QUESTION_MARK)
1170: {
1171: while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1172: if (*ptr != 0) ptr++;
1173: }
1174: }
1175:
1.4 ! misha 1176: /* Start with (? but not a condition. */
1.3 misha 1177:
1178: else
1179: {
1180: ptr += 2;
1181: if (*ptr == CHAR_P) ptr++; /* Allow optional P */
1182:
1183: /* We have to disambiguate (?<! and (?<= from (?<name> for named groups */
1184:
1185: if ((*ptr == CHAR_LESS_THAN_SIGN && ptr[1] != CHAR_EXCLAMATION_MARK &&
1186: ptr[1] != CHAR_EQUALS_SIGN) || *ptr == CHAR_APOSTROPHE)
1187: {
1188: int term;
1189: const uschar *thisname;
1190: *count += 1;
1191: if (name == NULL && *count == lorn) return *count;
1192: term = *ptr++;
1193: if (term == CHAR_LESS_THAN_SIGN) term = CHAR_GREATER_THAN_SIGN;
1194: thisname = ptr;
1195: while (*ptr != term) ptr++;
1196: if (name != NULL && lorn == ptr - thisname &&
1197: strncmp((const char *)name, (const char *)thisname, lorn) == 0)
1198: return *count;
1.4 ! misha 1199: term++;
1.3 misha 1200: }
1201: }
1202: }
1203:
1204: /* Past any initial parenthesis handling, scan for parentheses or vertical
1205: bars. */
1.1 misha 1206:
1207: for (; *ptr != 0; ptr++)
1208: {
1209: /* Skip over backslashed characters and also entire \Q...\E */
1210:
1.3 misha 1211: if (*ptr == CHAR_BACKSLASH)
1.1 misha 1212: {
1.3 misha 1213: if (*(++ptr) == 0) goto FAIL_EXIT;
1214: if (*ptr == CHAR_Q) for (;;)
1.1 misha 1215: {
1.3 misha 1216: while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1217: if (*ptr == 0) goto FAIL_EXIT;
1218: if (*(++ptr) == CHAR_E) break;
1.1 misha 1219: }
1220: continue;
1221: }
1222:
1223: /* Skip over character classes; this logic must be similar to the way they
1224: are handled for real. If the first character is '^', skip it. Also, if the
1225: first few characters (either before or after ^) are \Q\E or \E we skip them
1.3 misha 1226: too. This makes for compatibility with Perl. Note the use of STR macros to
1227: encode "Q\\E" so that it works in UTF-8 on EBCDIC platforms. */
1.1 misha 1228:
1.3 misha 1229: if (*ptr == CHAR_LEFT_SQUARE_BRACKET)
1.1 misha 1230: {
1231: BOOL negate_class = FALSE;
1232: for (;;)
1233: {
1.4 ! misha 1234: if (ptr[1] == CHAR_BACKSLASH)
1.1 misha 1235: {
1.4 ! misha 1236: if (ptr[2] == CHAR_E)
! 1237: ptr+= 2;
! 1238: else if (strncmp((const char *)ptr+2,
1.3 misha 1239: STR_Q STR_BACKSLASH STR_E, 3) == 0)
1.4 ! misha 1240: ptr += 4;
1.3 misha 1241: else
1242: break;
1.1 misha 1243: }
1.4 ! misha 1244: else if (!negate_class && ptr[1] == CHAR_CIRCUMFLEX_ACCENT)
! 1245: {
1.1 misha 1246: negate_class = TRUE;
1.4 ! misha 1247: ptr++;
! 1248: }
1.1 misha 1249: else break;
1250: }
1251:
1252: /* If the next character is ']', it is a data character that must be
1253: skipped, except in JavaScript compatibility mode. */
1254:
1.3 misha 1255: if (ptr[1] == CHAR_RIGHT_SQUARE_BRACKET &&
1256: (cd->external_options & PCRE_JAVASCRIPT_COMPAT) == 0)
1.1 misha 1257: ptr++;
1258:
1.3 misha 1259: while (*(++ptr) != CHAR_RIGHT_SQUARE_BRACKET)
1.1 misha 1260: {
1261: if (*ptr == 0) return -1;
1.3 misha 1262: if (*ptr == CHAR_BACKSLASH)
1.1 misha 1263: {
1.3 misha 1264: if (*(++ptr) == 0) goto FAIL_EXIT;
1265: if (*ptr == CHAR_Q) for (;;)
1.1 misha 1266: {
1.3 misha 1267: while (*(++ptr) != 0 && *ptr != CHAR_BACKSLASH) {};
1268: if (*ptr == 0) goto FAIL_EXIT;
1269: if (*(++ptr) == CHAR_E) break;
1.1 misha 1270: }
1271: continue;
1272: }
1273: }
1274: continue;
1275: }
1276:
1277: /* Skip comments in /x mode */
1278:
1.3 misha 1279: if (xmode && *ptr == CHAR_NUMBER_SIGN)
1.1 misha 1280: {
1.3 misha 1281: while (*(++ptr) != 0 && *ptr != CHAR_NL) {};
1282: if (*ptr == 0) goto FAIL_EXIT;
1.1 misha 1283: continue;
1284: }
1285:
1.3 misha 1286: /* Check for the special metacharacters */
1287:
1288: if (*ptr == CHAR_LEFT_PARENTHESIS)
1289: {
1290: int rc = find_parens_sub(&ptr, cd, name, lorn, xmode, count);
1291: if (rc > 0) return rc;
1292: if (*ptr == 0) goto FAIL_EXIT;
1293: }
1294:
1295: else if (*ptr == CHAR_RIGHT_PARENTHESIS)
1296: {
1297: if (dup_parens && *count < hwm_count) *count = hwm_count;
1.4 ! misha 1298: goto FAIL_EXIT;
1.3 misha 1299: }
1.1 misha 1300:
1.3 misha 1301: else if (*ptr == CHAR_VERTICAL_LINE && dup_parens)
1.1 misha 1302: {
1.3 misha 1303: if (*count > hwm_count) hwm_count = *count;
1304: *count = start_count;
1.1 misha 1305: }
1.3 misha 1306: }
1307:
1308: FAIL_EXIT:
1309: *ptrptr = ptr;
1310: return -1;
1311: }
1.1 misha 1312:
1313:
1314:
1315:
1.3 misha 1316: /*************************************************
1317: * Find forward referenced subpattern *
1318: *************************************************/
1319:
1320: /* This function scans along a pattern's text looking for capturing
1321: subpatterns, and counting them. If it finds a named pattern that matches the
1322: name it is given, it returns its number. Alternatively, if the name is NULL, it
1323: returns when it reaches a given numbered subpattern. This is used for forward
1324: references to subpatterns. We used to be able to start this scan from the
1325: current compiling point, using the current count value from cd->bracount, and
1326: do it all in a single loop, but the addition of the possibility of duplicate
1327: subpattern numbers means that we have to scan from the very start, in order to
1328: take account of such duplicates, and to use a recursive function to keep track
1329: of the different types of group.
1330:
1331: Arguments:
1332: cd compile background data
1333: name name to seek, or NULL if seeking a numbered subpattern
1334: lorn name length, or subpattern number if name is NULL
1335: xmode TRUE if we are in /x mode
1336:
1337: Returns: the number of the found subpattern, or -1 if not found
1338: */
1339:
1340: static int
1341: find_parens(compile_data *cd, const uschar *name, int lorn, BOOL xmode)
1342: {
1343: uschar *ptr = (uschar *)cd->start_pattern;
1344: int count = 0;
1345: int rc;
1346:
1347: /* If the pattern does not start with an opening parenthesis, the first call
1348: to find_parens_sub() will scan right to the end (if necessary). However, if it
1349: does start with a parenthesis, find_parens_sub() will return when it hits the
1350: matching closing parens. That is why we have to have a loop. */
1.1 misha 1351:
1.3 misha 1352: for (;;)
1353: {
1354: rc = find_parens_sub(&ptr, cd, name, lorn, xmode, &count);
1355: if (rc > 0 || *ptr++ == 0) break;
1.1 misha 1356: }
1357:
1.3 misha 1358: return rc;
1.1 misha 1359: }
1360:
1361:
1362:
1.3 misha 1363:
1.1 misha 1364: /*************************************************
1365: * Find first significant op code *
1366: *************************************************/
1367:
1368: /* This is called by several functions that scan a compiled expression looking
1369: for a fixed first character, or an anchoring op code etc. It skips over things
1370: that do not influence this. For some calls, a change of option is important.
1371: For some calls, it makes sense to skip negative forward and all backward
1372: assertions, and also the \b assertion; for others it does not.
1373:
1374: Arguments:
1375: code pointer to the start of the group
1376: options pointer to external options
1377: optbit the option bit whose changing is significant, or
1378: zero if none are
1379: skipassert TRUE if certain assertions are to be skipped
1380:
1381: Returns: pointer to the first significant opcode
1382: */
1383:
1384: static const uschar*
1385: first_significant_code(const uschar *code, int *options, int optbit,
1386: BOOL skipassert)
1387: {
1388: for (;;)
1389: {
1390: switch ((int)*code)
1391: {
1392: case OP_OPT:
1393: if (optbit > 0 && ((int)code[1] & optbit) != (*options & optbit))
1394: *options = (int)code[1];
1395: code += 2;
1396: break;
1397:
1398: case OP_ASSERT_NOT:
1399: case OP_ASSERTBACK:
1400: case OP_ASSERTBACK_NOT:
1401: if (!skipassert) return code;
1402: do code += GET(code, 1); while (*code == OP_ALT);
1403: code += _pcre_OP_lengths[*code];
1404: break;
1405:
1406: case OP_WORD_BOUNDARY:
1407: case OP_NOT_WORD_BOUNDARY:
1408: if (!skipassert) return code;
1409: /* Fall through */
1410:
1411: case OP_CALLOUT:
1412: case OP_CREF:
1.4 ! misha 1413: case OP_NCREF:
1.1 misha 1414: case OP_RREF:
1.4 ! misha 1415: case OP_NRREF:
1.1 misha 1416: case OP_DEF:
1417: code += _pcre_OP_lengths[*code];
1418: break;
1419:
1420: default:
1421: return code;
1422: }
1423: }
1424: /* Control never reaches here */
1425: }
1426:
1427:
1428:
1429:
1430: /*************************************************
1.4 ! misha 1431: * Find the fixed length of a branch *
1.1 misha 1432: *************************************************/
1433:
1.4 ! misha 1434: /* Scan a branch and compute the fixed length of subject that will match it,
1.1 misha 1435: if the length is fixed. This is needed for dealing with backward assertions.
1.4 ! misha 1436: In UTF8 mode, the result is in characters rather than bytes. The branch is
! 1437: temporarily terminated with OP_END when this function is called.
! 1438:
! 1439: This function is called when a backward assertion is encountered, so that if it
! 1440: fails, the error message can point to the correct place in the pattern.
! 1441: However, we cannot do this when the assertion contains subroutine calls,
! 1442: because they can be forward references. We solve this by remembering this case
! 1443: and doing the check at the end; a flag specifies which mode we are running in.
1.1 misha 1444:
1445: Arguments:
1446: code points to the start of the pattern (the bracket)
1447: options the compiling options
1.4 ! misha 1448: atend TRUE if called when the pattern is complete
! 1449: cd the "compile data" structure
1.1 misha 1450:
1.4 ! misha 1451: Returns: the fixed length,
! 1452: or -1 if there is no fixed length,
1.1 misha 1453: or -2 if \C was encountered
1.4 ! misha 1454: or -3 if an OP_RECURSE item was encountered and atend is FALSE
1.1 misha 1455: */
1456:
1457: static int
1.4 ! misha 1458: find_fixedlength(uschar *code, int options, BOOL atend, compile_data *cd)
1.1 misha 1459: {
1460: int length = -1;
1461:
1462: register int branchlength = 0;
1463: register uschar *cc = code + 1 + LINK_SIZE;
1464:
1465: /* Scan along the opcodes for this branch. If we get to the end of the
1466: branch, check the length against that of the other branches. */
1467:
1468: for (;;)
1469: {
1470: int d;
1.4 ! misha 1471: uschar *ce, *cs;
1.1 misha 1472: register int op = *cc;
1473: switch (op)
1474: {
1475: case OP_CBRA:
1476: case OP_BRA:
1477: case OP_ONCE:
1478: case OP_COND:
1.4 ! misha 1479: d = find_fixedlength(cc + ((op == OP_CBRA)? 2:0), options, atend, cd);
1.1 misha 1480: if (d < 0) return d;
1481: branchlength += d;
1482: do cc += GET(cc, 1); while (*cc == OP_ALT);
1483: cc += 1 + LINK_SIZE;
1484: break;
1485:
1486: /* Reached end of a branch; if it's a ket it is the end of a nested
1487: call. If it's ALT it is an alternation in a nested call. If it is
1488: END it's the end of the outer call. All can be handled by the same code. */
1489:
1490: case OP_ALT:
1491: case OP_KET:
1492: case OP_KETRMAX:
1493: case OP_KETRMIN:
1494: case OP_END:
1495: if (length < 0) length = branchlength;
1496: else if (length != branchlength) return -1;
1497: if (*cc != OP_ALT) return length;
1498: cc += 1 + LINK_SIZE;
1499: branchlength = 0;
1500: break;
1501:
1.4 ! misha 1502: /* A true recursion implies not fixed length, but a subroutine call may
! 1503: be OK. If the subroutine is a forward reference, we can't deal with
! 1504: it until the end of the pattern, so return -3. */
! 1505:
! 1506: case OP_RECURSE:
! 1507: if (!atend) return -3;
! 1508: cs = ce = (uschar *)cd->start_code + GET(cc, 1); /* Start subpattern */
! 1509: do ce += GET(ce, 1); while (*ce == OP_ALT); /* End subpattern */
! 1510: if (cc > cs && cc < ce) return -1; /* Recursion */
! 1511: d = find_fixedlength(cs + 2, options, atend, cd);
! 1512: if (d < 0) return d;
! 1513: branchlength += d;
! 1514: cc += 1 + LINK_SIZE;
! 1515: break;
! 1516:
1.1 misha 1517: /* Skip over assertive subpatterns */
1518:
1519: case OP_ASSERT:
1520: case OP_ASSERT_NOT:
1521: case OP_ASSERTBACK:
1522: case OP_ASSERTBACK_NOT:
1523: do cc += GET(cc, 1); while (*cc == OP_ALT);
1524: /* Fall through */
1525:
1526: /* Skip over things that don't match chars */
1527:
1528: case OP_REVERSE:
1529: case OP_CREF:
1.4 ! misha 1530: case OP_NCREF:
1.1 misha 1531: case OP_RREF:
1.4 ! misha 1532: case OP_NRREF:
1.1 misha 1533: case OP_DEF:
1534: case OP_OPT:
1535: case OP_CALLOUT:
1536: case OP_SOD:
1537: case OP_SOM:
1.4 ! misha 1538: case OP_SET_SOM:
1.1 misha 1539: case OP_EOD:
1540: case OP_EODN:
1541: case OP_CIRC:
1542: case OP_DOLL:
1543: case OP_NOT_WORD_BOUNDARY:
1544: case OP_WORD_BOUNDARY:
1545: cc += _pcre_OP_lengths[*cc];
1546: break;
1547:
1548: /* Handle literal characters */
1549:
1550: case OP_CHAR:
1551: case OP_CHARNC:
1552: case OP_NOT:
1553: branchlength++;
1554: cc += 2;
1555: #ifdef SUPPORT_UTF8
1.4 ! misha 1556: if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
! 1557: cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1.1 misha 1558: #endif
1559: break;
1560:
1561: /* Handle exact repetitions. The count is already in characters, but we
1562: need to skip over a multibyte character in UTF8 mode. */
1563:
1564: case OP_EXACT:
1565: branchlength += GET2(cc,1);
1566: cc += 4;
1567: #ifdef SUPPORT_UTF8
1.4 ! misha 1568: if ((options & PCRE_UTF8) != 0 && cc[-1] >= 0xc0)
! 1569: cc += _pcre_utf8_table4[cc[-1] & 0x3f];
1.1 misha 1570: #endif
1571: break;
1572:
1573: case OP_TYPEEXACT:
1574: branchlength += GET2(cc,1);
1575: if (cc[3] == OP_PROP || cc[3] == OP_NOTPROP) cc += 2;
1576: cc += 4;
1577: break;
1578:
1579: /* Handle single-char matchers */
1580:
1581: case OP_PROP:
1582: case OP_NOTPROP:
1583: cc += 2;
1584: /* Fall through */
1585:
1586: case OP_NOT_DIGIT:
1587: case OP_DIGIT:
1588: case OP_NOT_WHITESPACE:
1589: case OP_WHITESPACE:
1590: case OP_NOT_WORDCHAR:
1591: case OP_WORDCHAR:
1592: case OP_ANY:
1593: case OP_ALLANY:
1594: branchlength++;
1595: cc++;
1596: break;
1597:
1598: /* The single-byte matcher isn't allowed */
1599:
1600: case OP_ANYBYTE:
1601: return -2;
1602:
1603: /* Check a class for variable quantification */
1604:
1605: #ifdef SUPPORT_UTF8
1606: case OP_XCLASS:
1607: cc += GET(cc, 1) - 33;
1608: /* Fall through */
1609: #endif
1610:
1611: case OP_CLASS:
1612: case OP_NCLASS:
1613: cc += 33;
1614:
1615: switch (*cc)
1616: {
1617: case OP_CRSTAR:
1618: case OP_CRMINSTAR:
1619: case OP_CRQUERY:
1620: case OP_CRMINQUERY:
1621: return -1;
1622:
1623: case OP_CRRANGE:
1624: case OP_CRMINRANGE:
1625: if (GET2(cc,1) != GET2(cc,3)) return -1;
1626: branchlength += GET2(cc,1);
1627: cc += 5;
1628: break;
1629:
1630: default:
1631: branchlength++;
1632: }
1633: break;
1634:
1635: /* Anything else is variable length */
1636:
1637: default:
1638: return -1;
1639: }
1640: }
1641: /* Control never gets here */
1642: }
1643:
1644:
1645:
1646:
1647: /*************************************************
1.4 ! misha 1648: * Scan compiled regex for specific bracket *
1.1 misha 1649: *************************************************/
1650:
1651: /* This little function scans through a compiled pattern until it finds a
1.4 ! misha 1652: capturing bracket with the given number, or, if the number is negative, an
! 1653: instance of OP_REVERSE for a lookbehind. The function is global in the C sense
! 1654: so that it can be called from pcre_study() when finding the minimum matching
! 1655: length.
1.1 misha 1656:
1657: Arguments:
1658: code points to start of expression
1659: utf8 TRUE in UTF-8 mode
1.4 ! misha 1660: number the required bracket number or negative to find a lookbehind
1.1 misha 1661:
1662: Returns: pointer to the opcode for the bracket, or NULL if not found
1663: */
1664:
1.4 ! misha 1665: const uschar *
! 1666: _pcre_find_bracket(const uschar *code, BOOL utf8, int number)
1.1 misha 1667: {
1668: for (;;)
1669: {
1670: register int c = *code;
1671: if (c == OP_END) return NULL;
1672:
1673: /* XCLASS is used for classes that cannot be represented just by a bit
1674: map. This includes negated single high-valued characters. The length in
1675: the table is zero; the actual length is stored in the compiled code. */
1676:
1677: if (c == OP_XCLASS) code += GET(code, 1);
1678:
1.4 ! misha 1679: /* Handle recursion */
! 1680:
! 1681: else if (c == OP_REVERSE)
! 1682: {
! 1683: if (number < 0) return (uschar *)code;
! 1684: code += _pcre_OP_lengths[c];
! 1685: }
! 1686:
1.1 misha 1687: /* Handle capturing bracket */
1688:
1689: else if (c == OP_CBRA)
1690: {
1691: int n = GET2(code, 1+LINK_SIZE);
1692: if (n == number) return (uschar *)code;
1693: code += _pcre_OP_lengths[c];
1694: }
1695:
1696: /* Otherwise, we can get the item's length from the table, except that for
1697: repeated character types, we have to test for \p and \P, which have an extra
1.4 ! misha 1698: two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
! 1699: must add in its length. */
1.1 misha 1700:
1701: else
1702: {
1703: switch(c)
1704: {
1705: case OP_TYPESTAR:
1706: case OP_TYPEMINSTAR:
1707: case OP_TYPEPLUS:
1708: case OP_TYPEMINPLUS:
1709: case OP_TYPEQUERY:
1710: case OP_TYPEMINQUERY:
1711: case OP_TYPEPOSSTAR:
1712: case OP_TYPEPOSPLUS:
1713: case OP_TYPEPOSQUERY:
1714: if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1715: break;
1716:
1717: case OP_TYPEUPTO:
1718: case OP_TYPEMINUPTO:
1719: case OP_TYPEEXACT:
1720: case OP_TYPEPOSUPTO:
1721: if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1722: break;
1.4 ! misha 1723:
! 1724: case OP_MARK:
! 1725: case OP_PRUNE_ARG:
! 1726: case OP_SKIP_ARG:
! 1727: case OP_THEN_ARG:
! 1728: code += code[1];
! 1729: break;
1.1 misha 1730: }
1731:
1732: /* Add in the fixed length from the table */
1733:
1734: code += _pcre_OP_lengths[c];
1735:
1736: /* In UTF-8 mode, opcodes that are followed by a character may be followed by
1737: a multi-byte character. The length in the table is a minimum, so we have to
1738: arrange to skip the extra bytes. */
1739:
1740: #ifdef SUPPORT_UTF8
1741: if (utf8) switch(c)
1742: {
1743: case OP_CHAR:
1744: case OP_CHARNC:
1745: case OP_EXACT:
1746: case OP_UPTO:
1747: case OP_MINUPTO:
1748: case OP_POSUPTO:
1749: case OP_STAR:
1750: case OP_MINSTAR:
1751: case OP_POSSTAR:
1752: case OP_PLUS:
1753: case OP_MINPLUS:
1754: case OP_POSPLUS:
1755: case OP_QUERY:
1756: case OP_MINQUERY:
1757: case OP_POSQUERY:
1758: if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1759: break;
1760: }
1.2 misha 1761: #else
1762: (void)(utf8); /* Keep compiler happy by referencing function argument */
1.1 misha 1763: #endif
1764: }
1765: }
1766: }
1767:
1768:
1769:
1770: /*************************************************
1771: * Scan compiled regex for recursion reference *
1772: *************************************************/
1773:
1774: /* This little function scans through a compiled pattern until it finds an
1775: instance of OP_RECURSE.
1776:
1777: Arguments:
1778: code points to start of expression
1779: utf8 TRUE in UTF-8 mode
1780:
1781: Returns: pointer to the opcode for OP_RECURSE, or NULL if not found
1782: */
1783:
1784: static const uschar *
1785: find_recurse(const uschar *code, BOOL utf8)
1786: {
1787: for (;;)
1788: {
1789: register int c = *code;
1790: if (c == OP_END) return NULL;
1791: if (c == OP_RECURSE) return code;
1792:
1793: /* XCLASS is used for classes that cannot be represented just by a bit
1794: map. This includes negated single high-valued characters. The length in
1795: the table is zero; the actual length is stored in the compiled code. */
1796:
1797: if (c == OP_XCLASS) code += GET(code, 1);
1798:
1799: /* Otherwise, we can get the item's length from the table, except that for
1800: repeated character types, we have to test for \p and \P, which have an extra
1.4 ! misha 1801: two bytes of parameters, and for MARK/PRUNE/SKIP/THEN with an argument, we
! 1802: must add in its length. */
1.1 misha 1803:
1804: else
1805: {
1806: switch(c)
1807: {
1808: case OP_TYPESTAR:
1809: case OP_TYPEMINSTAR:
1810: case OP_TYPEPLUS:
1811: case OP_TYPEMINPLUS:
1812: case OP_TYPEQUERY:
1813: case OP_TYPEMINQUERY:
1814: case OP_TYPEPOSSTAR:
1815: case OP_TYPEPOSPLUS:
1816: case OP_TYPEPOSQUERY:
1817: if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
1818: break;
1819:
1820: case OP_TYPEPOSUPTO:
1821: case OP_TYPEUPTO:
1822: case OP_TYPEMINUPTO:
1823: case OP_TYPEEXACT:
1824: if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
1825: break;
1.4 ! misha 1826:
! 1827: case OP_MARK:
! 1828: case OP_PRUNE_ARG:
! 1829: case OP_SKIP_ARG:
! 1830: case OP_THEN_ARG:
! 1831: code += code[1];
! 1832: break;
1.1 misha 1833: }
1834:
1835: /* Add in the fixed length from the table */
1836:
1837: code += _pcre_OP_lengths[c];
1838:
1839: /* In UTF-8 mode, opcodes that are followed by a character may be followed
1840: by a multi-byte character. The length in the table is a minimum, so we have
1841: to arrange to skip the extra bytes. */
1842:
1843: #ifdef SUPPORT_UTF8
1844: if (utf8) switch(c)
1845: {
1846: case OP_CHAR:
1847: case OP_CHARNC:
1848: case OP_EXACT:
1849: case OP_UPTO:
1850: case OP_MINUPTO:
1851: case OP_POSUPTO:
1852: case OP_STAR:
1853: case OP_MINSTAR:
1854: case OP_POSSTAR:
1855: case OP_PLUS:
1856: case OP_MINPLUS:
1857: case OP_POSPLUS:
1858: case OP_QUERY:
1859: case OP_MINQUERY:
1860: case OP_POSQUERY:
1861: if (code[-1] >= 0xc0) code += _pcre_utf8_table4[code[-1] & 0x3f];
1862: break;
1863: }
1.2 misha 1864: #else
1865: (void)(utf8); /* Keep compiler happy by referencing function argument */
1.1 misha 1866: #endif
1867: }
1868: }
1869: }
1870:
1871:
1872:
1873: /*************************************************
1874: * Scan compiled branch for non-emptiness *
1875: *************************************************/
1876:
1877: /* This function scans through a branch of a compiled pattern to see whether it
1878: can match the empty string or not. It is called from could_be_empty()
1879: below and from compile_branch() when checking for an unlimited repeat of a
1880: group that can match nothing. Note that first_significant_code() skips over
1881: backward and negative forward assertions when its final argument is TRUE. If we
1882: hit an unclosed bracket, we return "empty" - this means we've struck an inner
1883: bracket whose current branch will already have been scanned.
1884:
1885: Arguments:
1886: code points to start of search
1887: endcode points to where to stop
1888: utf8 TRUE if in UTF8 mode
1.4 ! misha 1889: cd contains pointers to tables etc.
1.1 misha 1890:
1891: Returns: TRUE if what is matched could be empty
1892: */
1893:
1894: static BOOL
1.4 ! misha 1895: could_be_empty_branch(const uschar *code, const uschar *endcode, BOOL utf8,
! 1896: compile_data *cd)
1.1 misha 1897: {
1898: register int c;
1899: for (code = first_significant_code(code + _pcre_OP_lengths[*code], NULL, 0, TRUE);
1900: code < endcode;
1901: code = first_significant_code(code + _pcre_OP_lengths[c], NULL, 0, TRUE))
1902: {
1903: const uschar *ccode;
1904:
1905: c = *code;
1906:
1907: /* Skip over forward assertions; the other assertions are skipped by
1908: first_significant_code() with a TRUE final argument. */
1909:
1910: if (c == OP_ASSERT)
1911: {
1912: do code += GET(code, 1); while (*code == OP_ALT);
1913: c = *code;
1914: continue;
1915: }
1916:
1917: /* Groups with zero repeats can of course be empty; skip them. */
1918:
1919: if (c == OP_BRAZERO || c == OP_BRAMINZERO || c == OP_SKIPZERO)
1920: {
1921: code += _pcre_OP_lengths[c];
1922: do code += GET(code, 1); while (*code == OP_ALT);
1923: c = *code;
1924: continue;
1925: }
1926:
1.4 ! misha 1927: /* For a recursion/subroutine call, if its end has been reached, which
! 1928: implies a subroutine call, we can scan it. */
! 1929:
! 1930: if (c == OP_RECURSE)
! 1931: {
! 1932: BOOL empty_branch = FALSE;
! 1933: const uschar *scode = cd->start_code + GET(code, 1);
! 1934: if (GET(scode, 1) == 0) return TRUE; /* Unclosed */
! 1935: do
! 1936: {
! 1937: if (could_be_empty_branch(scode, endcode, utf8, cd))
! 1938: {
! 1939: empty_branch = TRUE;
! 1940: break;
! 1941: }
! 1942: scode += GET(scode, 1);
! 1943: }
! 1944: while (*scode == OP_ALT);
! 1945: if (!empty_branch) return FALSE; /* All branches are non-empty */
! 1946: continue;
! 1947: }
! 1948:
1.1 misha 1949: /* For other groups, scan the branches. */
1950:
1951: if (c == OP_BRA || c == OP_CBRA || c == OP_ONCE || c == OP_COND)
1952: {
1953: BOOL empty_branch;
1954: if (GET(code, 1) == 0) return TRUE; /* Hit unclosed bracket */
1955:
1.3 misha 1956: /* If a conditional group has only one branch, there is a second, implied,
1957: empty branch, so just skip over the conditional, because it could be empty.
1958: Otherwise, scan the individual branches of the group. */
1.1 misha 1959:
1.3 misha 1960: if (c == OP_COND && code[GET(code, 1)] != OP_ALT)
1961: code += GET(code, 1);
1962: else
1.1 misha 1963: {
1.3 misha 1964: empty_branch = FALSE;
1965: do
1966: {
1.4 ! misha 1967: if (!empty_branch && could_be_empty_branch(code, endcode, utf8, cd))
1.3 misha 1968: empty_branch = TRUE;
1969: code += GET(code, 1);
1970: }
1971: while (*code == OP_ALT);
1972: if (!empty_branch) return FALSE; /* All branches are non-empty */
1.1 misha 1973: }
1.3 misha 1974:
1.1 misha 1975: c = *code;
1976: continue;
1977: }
1978:
1979: /* Handle the other opcodes */
1980:
1981: switch (c)
1982: {
1983: /* Check for quantifiers after a class. XCLASS is used for classes that
1984: cannot be represented just by a bit map. This includes negated single
1985: high-valued characters. The length in _pcre_OP_lengths[] is zero; the
1986: actual length is stored in the compiled code, so we must update "code"
1987: here. */
1988:
1989: #ifdef SUPPORT_UTF8
1990: case OP_XCLASS:
1991: ccode = code += GET(code, 1);
1992: goto CHECK_CLASS_REPEAT;
1993: #endif
1994:
1995: case OP_CLASS:
1996: case OP_NCLASS:
1997: ccode = code + 33;
1998:
1999: #ifdef SUPPORT_UTF8
2000: CHECK_CLASS_REPEAT:
2001: #endif
2002:
2003: switch (*ccode)
2004: {
2005: case OP_CRSTAR: /* These could be empty; continue */
2006: case OP_CRMINSTAR:
2007: case OP_CRQUERY:
2008: case OP_CRMINQUERY:
2009: break;
2010:
2011: default: /* Non-repeat => class must match */
2012: case OP_CRPLUS: /* These repeats aren't empty */
2013: case OP_CRMINPLUS:
2014: return FALSE;
2015:
2016: case OP_CRRANGE:
2017: case OP_CRMINRANGE:
2018: if (GET2(ccode, 1) > 0) return FALSE; /* Minimum > 0 */
2019: break;
2020: }
2021: break;
2022:
2023: /* Opcodes that must match a character */
2024:
2025: case OP_PROP:
2026: case OP_NOTPROP:
2027: case OP_EXTUNI:
2028: case OP_NOT_DIGIT:
2029: case OP_DIGIT:
2030: case OP_NOT_WHITESPACE:
2031: case OP_WHITESPACE:
2032: case OP_NOT_WORDCHAR:
2033: case OP_WORDCHAR:
2034: case OP_ANY:
2035: case OP_ALLANY:
2036: case OP_ANYBYTE:
2037: case OP_CHAR:
2038: case OP_CHARNC:
2039: case OP_NOT:
2040: case OP_PLUS:
2041: case OP_MINPLUS:
2042: case OP_POSPLUS:
2043: case OP_EXACT:
2044: case OP_NOTPLUS:
2045: case OP_NOTMINPLUS:
2046: case OP_NOTPOSPLUS:
2047: case OP_NOTEXACT:
2048: case OP_TYPEPLUS:
2049: case OP_TYPEMINPLUS:
2050: case OP_TYPEPOSPLUS:
2051: case OP_TYPEEXACT:
2052: return FALSE;
2053:
2054: /* These are going to continue, as they may be empty, but we have to
2055: fudge the length for the \p and \P cases. */
2056:
2057: case OP_TYPESTAR:
2058: case OP_TYPEMINSTAR:
2059: case OP_TYPEPOSSTAR:
2060: case OP_TYPEQUERY:
2061: case OP_TYPEMINQUERY:
2062: case OP_TYPEPOSQUERY:
2063: if (code[1] == OP_PROP || code[1] == OP_NOTPROP) code += 2;
2064: break;
2065:
2066: /* Same for these */
2067:
2068: case OP_TYPEUPTO:
2069: case OP_TYPEMINUPTO:
2070: case OP_TYPEPOSUPTO:
2071: if (code[3] == OP_PROP || code[3] == OP_NOTPROP) code += 2;
2072: break;
2073:
2074: /* End of branch */
2075:
2076: case OP_KET:
2077: case OP_KETRMAX:
2078: case OP_KETRMIN:
2079: case OP_ALT:
2080: return TRUE;
2081:
2082: /* In UTF-8 mode, STAR, MINSTAR, POSSTAR, QUERY, MINQUERY, POSQUERY, UPTO,
2083: MINUPTO, and POSUPTO may be followed by a multibyte character */
2084:
2085: #ifdef SUPPORT_UTF8
2086: case OP_STAR:
2087: case OP_MINSTAR:
2088: case OP_POSSTAR:
2089: case OP_QUERY:
2090: case OP_MINQUERY:
2091: case OP_POSQUERY:
1.4 ! misha 2092: if (utf8 && code[1] >= 0xc0) code += _pcre_utf8_table4[code[1] & 0x3f];
! 2093: break;
! 2094:
1.1 misha 2095: case OP_UPTO:
2096: case OP_MINUPTO:
2097: case OP_POSUPTO:
1.4 ! misha 2098: if (utf8 && code[3] >= 0xc0) code += _pcre_utf8_table4[code[3] & 0x3f];
1.1 misha 2099: break;
2100: #endif
1.4 ! misha 2101:
! 2102: /* MARK, and PRUNE/SKIP/THEN with an argument must skip over the argument
! 2103: string. */
! 2104:
! 2105: case OP_MARK:
! 2106: case OP_PRUNE_ARG:
! 2107: case OP_SKIP_ARG:
! 2108: case OP_THEN_ARG:
! 2109: code += code[1];
! 2110: break;
! 2111:
! 2112: /* None of the remaining opcodes are required to match a character. */
! 2113:
! 2114: default:
! 2115: break;
1.1 misha 2116: }
2117: }
2118:
2119: return TRUE;
2120: }
2121:
2122:
2123:
2124: /*************************************************
2125: * Scan compiled regex for non-emptiness *
2126: *************************************************/
2127:
2128: /* This function is called to check for left recursive calls. We want to check
2129: the current branch of the current pattern to see if it could match the empty
2130: string. If it could, we must look outwards for branches at other levels,
2131: stopping when we pass beyond the bracket which is the subject of the recursion.
2132:
2133: Arguments:
2134: code points to start of the recursion
2135: endcode points to where to stop (current RECURSE item)
2136: bcptr points to the chain of current (unclosed) branch starts
2137: utf8 TRUE if in UTF-8 mode
1.4 ! misha 2138: cd pointers to tables etc
1.1 misha 2139:
2140: Returns: TRUE if what is matched could be empty
2141: */
2142:
2143: static BOOL
2144: could_be_empty(const uschar *code, const uschar *endcode, branch_chain *bcptr,
1.4 ! misha 2145: BOOL utf8, compile_data *cd)
1.1 misha 2146: {
1.4 ! misha 2147: while (bcptr != NULL && bcptr->current_branch >= code)
1.1 misha 2148: {
1.4 ! misha 2149: if (!could_be_empty_branch(bcptr->current_branch, endcode, utf8, cd))
! 2150: return FALSE;
1.1 misha 2151: bcptr = bcptr->outer;
2152: }
2153: return TRUE;
2154: }
2155:
2156:
2157:
2158: /*************************************************
2159: * Check for POSIX class syntax *
2160: *************************************************/
2161:
2162: /* This function is called when the sequence "[:" or "[." or "[=" is
2163: encountered in a character class. It checks whether this is followed by a
2164: sequence of characters terminated by a matching ":]" or ".]" or "=]". If we
2165: reach an unescaped ']' without the special preceding character, return FALSE.
2166:
2167: Originally, this function only recognized a sequence of letters between the
2168: terminators, but it seems that Perl recognizes any sequence of characters,
2169: though of course unknown POSIX names are subsequently rejected. Perl gives an
2170: "Unknown POSIX class" error for [:f\oo:] for example, where previously PCRE
2171: didn't consider this to be a POSIX class. Likewise for [:1234:].
2172:
2173: The problem in trying to be exactly like Perl is in the handling of escapes. We
2174: have to be sure that [abc[:x\]pqr] is *not* treated as containing a POSIX
2175: class, but [abc[:x\]pqr:]] is (so that an error can be generated). The code
2176: below handles the special case of \], but does not try to do any other escape
2177: processing. This makes it different from Perl for cases such as [:l\ower:]
2178: where Perl recognizes it as the POSIX class "lower" but PCRE does not recognize
2179: "l\ower". This is a lesser evil that not diagnosing bad classes when Perl does,
2180: I think.
2181:
2182: Arguments:
2183: ptr pointer to the initial [
2184: endptr where to return the end pointer
2185:
2186: Returns: TRUE or FALSE
2187: */
2188:
2189: static BOOL
2190: check_posix_syntax(const uschar *ptr, const uschar **endptr)
2191: {
2192: int terminator; /* Don't combine these lines; the Solaris cc */
2193: terminator = *(++ptr); /* compiler warns about "non-constant" initializer. */
2194: for (++ptr; *ptr != 0; ptr++)
2195: {
1.3 misha 2196: if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET) ptr++; else
1.1 misha 2197: {
1.3 misha 2198: if (*ptr == CHAR_RIGHT_SQUARE_BRACKET) return FALSE;
2199: if (*ptr == terminator && ptr[1] == CHAR_RIGHT_SQUARE_BRACKET)
1.1 misha 2200: {
2201: *endptr = ptr;
2202: return TRUE;
2203: }
2204: }
2205: }
2206: return FALSE;
2207: }
2208:
2209:
2210:
2211:
2212: /*************************************************
2213: * Check POSIX class name *
2214: *************************************************/
2215:
2216: /* This function is called to check the name given in a POSIX-style class entry
2217: such as [:alnum:].
2218:
2219: Arguments:
2220: ptr points to the first letter
2221: len the length of the name
2222:
2223: Returns: a value representing the name, or -1 if unknown
2224: */
2225:
2226: static int
2227: check_posix_name(const uschar *ptr, int len)
2228: {
2229: const char *pn = posix_names;
2230: register int yield = 0;
2231: while (posix_name_lengths[yield] != 0)
2232: {
2233: if (len == posix_name_lengths[yield] &&
2234: strncmp((const char *)ptr, pn, len) == 0) return yield;
2235: pn += posix_name_lengths[yield] + 1;
2236: yield++;
2237: }
2238: return -1;
2239: }
2240:
2241:
2242: /*************************************************
2243: * Adjust OP_RECURSE items in repeated group *
2244: *************************************************/
2245:
2246: /* OP_RECURSE items contain an offset from the start of the regex to the group
2247: that is referenced. This means that groups can be replicated for fixed
2248: repetition simply by copying (because the recursion is allowed to refer to
2249: earlier groups that are outside the current group). However, when a group is
2250: optional (i.e. the minimum quantifier is zero), OP_BRAZERO or OP_SKIPZERO is
2251: inserted before it, after it has been compiled. This means that any OP_RECURSE
2252: items within it that refer to the group itself or any contained groups have to
2253: have their offsets adjusted. That one of the jobs of this function. Before it
2254: is called, the partially compiled regex must be temporarily terminated with
2255: OP_END.
2256:
2257: This function has been extended with the possibility of forward references for
2258: recursions and subroutine calls. It must also check the list of such references
2259: for the group we are dealing with. If it finds that one of the recursions in
2260: the current group is on this list, it adjusts the offset in the list, not the
2261: value in the reference (which is a group number).
2262:
2263: Arguments:
2264: group points to the start of the group
2265: adjust the amount by which the group is to be moved
2266: utf8 TRUE in UTF-8 mode
2267: cd contains pointers to tables etc.
2268: save_hwm the hwm forward reference pointer at the start of the group
2269:
2270: Returns: nothing
2271: */
2272:
2273: static void
2274: adjust_recurse(uschar *group, int adjust, BOOL utf8, compile_data *cd,
2275: uschar *save_hwm)
2276: {
2277: uschar *ptr = group;
2278:
2279: while ((ptr = (uschar *)find_recurse(ptr, utf8)) != NULL)
2280: {
2281: int offset;
2282: uschar *hc;
2283:
2284: /* See if this recursion is on the forward reference list. If so, adjust the
2285: reference. */
2286:
2287: for (hc = save_hwm; hc < cd->hwm; hc += LINK_SIZE)
2288: {
2289: offset = GET(hc, 0);
2290: if (cd->start_code + offset == ptr + 1)
2291: {
2292: PUT(hc, 0, offset + adjust);
2293: break;
2294: }
2295: }
2296:
2297: /* Otherwise, adjust the recursion offset if it's after the start of this
2298: group. */
2299:
2300: if (hc >= cd->hwm)
2301: {
2302: offset = GET(ptr, 1);
2303: if (cd->start_code + offset >= group) PUT(ptr, 1, offset + adjust);
2304: }
2305:
2306: ptr += 1 + LINK_SIZE;
2307: }
2308: }
2309:
2310:
2311:
2312: /*************************************************
2313: * Insert an automatic callout point *
2314: *************************************************/
2315:
2316: /* This function is called when the PCRE_AUTO_CALLOUT option is set, to insert
2317: callout points before each pattern item.
2318:
2319: Arguments:
2320: code current code pointer
2321: ptr current pattern pointer
2322: cd pointers to tables etc
2323:
2324: Returns: new code pointer
2325: */
2326:
2327: static uschar *
2328: auto_callout(uschar *code, const uschar *ptr, compile_data *cd)
2329: {
2330: *code++ = OP_CALLOUT;
2331: *code++ = 255;
1.4 ! misha 2332: PUT(code, 0, (int)(ptr - cd->start_pattern)); /* Pattern offset */
! 2333: PUT(code, LINK_SIZE, 0); /* Default length */
1.1 misha 2334: return code + 2*LINK_SIZE;
2335: }
2336:
2337:
2338:
2339: /*************************************************
2340: * Complete a callout item *
2341: *************************************************/
2342:
2343: /* A callout item contains the length of the next item in the pattern, which
2344: we can't fill in till after we have reached the relevant point. This is used
2345: for both automatic and manual callouts.
2346:
2347: Arguments:
2348: previous_callout points to previous callout item
2349: ptr current pattern pointer
2350: cd pointers to tables etc
2351:
2352: Returns: nothing
2353: */
2354:
2355: static void
2356: complete_callout(uschar *previous_callout, const uschar *ptr, compile_data *cd)
2357: {
1.4 ! misha 2358: int length = (int)(ptr - cd->start_pattern - GET(previous_callout, 2));
1.1 misha 2359: PUT(previous_callout, 2 + LINK_SIZE, length);
2360: }
2361:
2362:
2363:
2364: #ifdef SUPPORT_UCP
2365: /*************************************************
2366: * Get othercase range *
2367: *************************************************/
2368:
2369: /* This function is passed the start and end of a class range, in UTF-8 mode
2370: with UCP support. It searches up the characters, looking for internal ranges of
2371: characters in the "other" case. Each call returns the next one, updating the
2372: start address.
2373:
2374: Arguments:
2375: cptr points to starting character value; updated
2376: d end value
2377: ocptr where to put start of othercase range
2378: odptr where to put end of othercase range
2379:
2380: Yield: TRUE when range returned; FALSE when no more
2381: */
2382:
2383: static BOOL
2384: get_othercase_range(unsigned int *cptr, unsigned int d, unsigned int *ocptr,
2385: unsigned int *odptr)
2386: {
2387: unsigned int c, othercase, next;
2388:
2389: for (c = *cptr; c <= d; c++)
1.2 misha 2390: { if ((othercase = UCD_OTHERCASE(c)) != c) break; }
1.1 misha 2391:
2392: if (c > d) return FALSE;
2393:
2394: *ocptr = othercase;
2395: next = othercase + 1;
2396:
2397: for (++c; c <= d; c++)
2398: {
1.2 misha 2399: if (UCD_OTHERCASE(c) != next) break;
1.1 misha 2400: next++;
2401: }
2402:
2403: *odptr = next - 1;
2404: *cptr = c;
2405:
2406: return TRUE;
2407: }
1.4 ! misha 2408:
! 2409:
! 2410:
! 2411: /*************************************************
! 2412: * Check a character and a property *
! 2413: *************************************************/
! 2414:
! 2415: /* This function is called by check_auto_possessive() when a property item
! 2416: is adjacent to a fixed character.
! 2417:
! 2418: Arguments:
! 2419: c the character
! 2420: ptype the property type
! 2421: pdata the data for the type
! 2422: negated TRUE if it's a negated property (\P or \p{^)
! 2423:
! 2424: Returns: TRUE if auto-possessifying is OK
! 2425: */
! 2426:
! 2427: static BOOL
! 2428: check_char_prop(int c, int ptype, int pdata, BOOL negated)
! 2429: {
! 2430: const ucd_record *prop = GET_UCD(c);
! 2431: switch(ptype)
! 2432: {
! 2433: case PT_LAMP:
! 2434: return (prop->chartype == ucp_Lu ||
! 2435: prop->chartype == ucp_Ll ||
! 2436: prop->chartype == ucp_Lt) == negated;
! 2437:
! 2438: case PT_GC:
! 2439: return (pdata == _pcre_ucp_gentype[prop->chartype]) == negated;
! 2440:
! 2441: case PT_PC:
! 2442: return (pdata == prop->chartype) == negated;
! 2443:
! 2444: case PT_SC:
! 2445: return (pdata == prop->script) == negated;
! 2446:
! 2447: /* These are specials */
! 2448:
! 2449: case PT_ALNUM:
! 2450: return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
! 2451: _pcre_ucp_gentype[prop->chartype] == ucp_N) == negated;
! 2452:
! 2453: case PT_SPACE: /* Perl space */
! 2454: return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
! 2455: c == CHAR_HT || c == CHAR_NL || c == CHAR_FF || c == CHAR_CR)
! 2456: == negated;
! 2457:
! 2458: case PT_PXSPACE: /* POSIX space */
! 2459: return (_pcre_ucp_gentype[prop->chartype] == ucp_Z ||
! 2460: c == CHAR_HT || c == CHAR_NL || c == CHAR_VT ||
! 2461: c == CHAR_FF || c == CHAR_CR)
! 2462: == negated;
! 2463:
! 2464: case PT_WORD:
! 2465: return (_pcre_ucp_gentype[prop->chartype] == ucp_L ||
! 2466: _pcre_ucp_gentype[prop->chartype] == ucp_N ||
! 2467: c == CHAR_UNDERSCORE) == negated;
! 2468: }
! 2469: return FALSE;
! 2470: }
1.1 misha 2471: #endif /* SUPPORT_UCP */
2472:
2473:
2474:
2475: /*************************************************
2476: * Check if auto-possessifying is possible *
2477: *************************************************/
2478:
2479: /* This function is called for unlimited repeats of certain items, to see
2480: whether the next thing could possibly match the repeated item. If not, it makes
2481: sense to automatically possessify the repeated item.
2482:
2483: Arguments:
1.4 ! misha 2484: previous pointer to the repeated opcode
1.1 misha 2485: utf8 TRUE in UTF-8 mode
2486: ptr next character in pattern
2487: options options bits
2488: cd contains pointers to tables etc.
2489:
2490: Returns: TRUE if possessifying is wanted
2491: */
2492:
2493: static BOOL
1.4 ! misha 2494: check_auto_possessive(const uschar *previous, BOOL utf8, const uschar *ptr,
! 2495: int options, compile_data *cd)
1.1 misha 2496: {
1.4 ! misha 2497: int c, next;
! 2498: int op_code = *previous++;
1.1 misha 2499:
2500: /* Skip whitespace and comments in extended mode */
2501:
2502: if ((options & PCRE_EXTENDED) != 0)
2503: {
2504: for (;;)
2505: {
2506: while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
1.3 misha 2507: if (*ptr == CHAR_NUMBER_SIGN)
1.1 misha 2508: {
2509: while (*(++ptr) != 0)
2510: if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2511: }
2512: else break;
2513: }
2514: }
2515:
2516: /* If the next item is one that we can handle, get its value. A non-negative
2517: value is a character, a negative value is an escape value. */
2518:
1.3 misha 2519: if (*ptr == CHAR_BACKSLASH)
1.1 misha 2520: {
2521: int temperrorcode = 0;
2522: next = check_escape(&ptr, &temperrorcode, cd->bracount, options, FALSE);
2523: if (temperrorcode != 0) return FALSE;
2524: ptr++; /* Point after the escape sequence */
2525: }
2526:
2527: else if ((cd->ctypes[*ptr] & ctype_meta) == 0)
2528: {
2529: #ifdef SUPPORT_UTF8
2530: if (utf8) { GETCHARINC(next, ptr); } else
2531: #endif
2532: next = *ptr++;
2533: }
2534:
2535: else return FALSE;
2536:
2537: /* Skip whitespace and comments in extended mode */
2538:
2539: if ((options & PCRE_EXTENDED) != 0)
2540: {
2541: for (;;)
2542: {
2543: while ((cd->ctypes[*ptr] & ctype_space) != 0) ptr++;
1.3 misha 2544: if (*ptr == CHAR_NUMBER_SIGN)
1.1 misha 2545: {
2546: while (*(++ptr) != 0)
2547: if (IS_NEWLINE(ptr)) { ptr += cd->nllen; break; }
2548: }
2549: else break;
2550: }
2551: }
2552:
2553: /* If the next thing is itself optional, we have to give up. */
2554:
1.3 misha 2555: if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
2556: strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
2557: return FALSE;
1.1 misha 2558:
1.4 ! misha 2559: /* Now compare the next item with the previous opcode. First, handle cases when
! 2560: the next item is a character. */
1.1 misha 2561:
2562: if (next >= 0) switch(op_code)
2563: {
2564: case OP_CHAR:
2565: #ifdef SUPPORT_UTF8
1.4 ! misha 2566: GETCHARTEST(c, previous);
1.2 misha 2567: #else
1.4 ! misha 2568: c = *previous;
1.1 misha 2569: #endif
1.4 ! misha 2570: return c != next;
1.1 misha 2571:
2572: /* For CHARNC (caseless character) we must check the other case. If we have
2573: Unicode property support, we can use it to test the other case of
2574: high-valued characters. */
2575:
2576: case OP_CHARNC:
2577: #ifdef SUPPORT_UTF8
1.4 ! misha 2578: GETCHARTEST(c, previous);
! 2579: #else
! 2580: c = *previous;
1.1 misha 2581: #endif
1.4 ! misha 2582: if (c == next) return FALSE;
1.1 misha 2583: #ifdef SUPPORT_UTF8
2584: if (utf8)
2585: {
2586: unsigned int othercase;
2587: if (next < 128) othercase = cd->fcc[next]; else
2588: #ifdef SUPPORT_UCP
1.2 misha 2589: othercase = UCD_OTHERCASE((unsigned int)next);
1.1 misha 2590: #else
2591: othercase = NOTACHAR;
2592: #endif
1.4 ! misha 2593: return (unsigned int)c != othercase;
1.1 misha 2594: }
2595: else
2596: #endif /* SUPPORT_UTF8 */
1.4 ! misha 2597: return (c != cd->fcc[next]); /* Non-UTF-8 mode */
1.1 misha 2598:
1.4 ! misha 2599: /* For OP_NOT, its data is always a single-byte character. */
1.1 misha 2600:
2601: case OP_NOT:
1.4 ! misha 2602: if ((c = *previous) == next) return TRUE;
1.1 misha 2603: if ((options & PCRE_CASELESS) == 0) return FALSE;
2604: #ifdef SUPPORT_UTF8
2605: if (utf8)
2606: {
2607: unsigned int othercase;
2608: if (next < 128) othercase = cd->fcc[next]; else
2609: #ifdef SUPPORT_UCP
1.2 misha 2610: othercase = UCD_OTHERCASE(next);
1.1 misha 2611: #else
2612: othercase = NOTACHAR;
2613: #endif
1.4 ! misha 2614: return (unsigned int)c == othercase;
1.1 misha 2615: }
2616: else
2617: #endif /* SUPPORT_UTF8 */
1.4 ! misha 2618: return (c == cd->fcc[next]); /* Non-UTF-8 mode */
! 2619:
! 2620: /* Note that OP_DIGIT etc. are generated only when PCRE_UCP is *not* set.
! 2621: When it is set, \d etc. are converted into OP_(NOT_)PROP codes. */
1.1 misha 2622:
2623: case OP_DIGIT:
2624: return next > 127 || (cd->ctypes[next] & ctype_digit) == 0;
2625:
2626: case OP_NOT_DIGIT:
2627: return next <= 127 && (cd->ctypes[next] & ctype_digit) != 0;
2628:
2629: case OP_WHITESPACE:
2630: return next > 127 || (cd->ctypes[next] & ctype_space) == 0;
2631:
2632: case OP_NOT_WHITESPACE:
2633: return next <= 127 && (cd->ctypes[next] & ctype_space) != 0;
2634:
2635: case OP_WORDCHAR:
2636: return next > 127 || (cd->ctypes[next] & ctype_word) == 0;
2637:
2638: case OP_NOT_WORDCHAR:
2639: return next <= 127 && (cd->ctypes[next] & ctype_word) != 0;
2640:
2641: case OP_HSPACE:
2642: case OP_NOT_HSPACE:
2643: switch(next)
2644: {
2645: case 0x09:
2646: case 0x20:
2647: case 0xa0:
2648: case 0x1680:
2649: case 0x180e:
2650: case 0x2000:
2651: case 0x2001:
2652: case 0x2002:
2653: case 0x2003:
2654: case 0x2004:
2655: case 0x2005:
2656: case 0x2006:
2657: case 0x2007:
2658: case 0x2008:
2659: case 0x2009:
2660: case 0x200A:
2661: case 0x202f:
2662: case 0x205f:
2663: case 0x3000:
1.4 ! misha 2664: return op_code == OP_NOT_HSPACE;
1.1 misha 2665: default:
1.4 ! misha 2666: return op_code != OP_NOT_HSPACE;
1.1 misha 2667: }
2668:
1.4 ! misha 2669: case OP_ANYNL:
1.1 misha 2670: case OP_VSPACE:
2671: case OP_NOT_VSPACE:
2672: switch(next)
2673: {
2674: case 0x0a:
2675: case 0x0b:
2676: case 0x0c:
2677: case 0x0d:
2678: case 0x85:
2679: case 0x2028:
2680: case 0x2029:
1.4 ! misha 2681: return op_code == OP_NOT_VSPACE;
1.1 misha 2682: default:
1.4 ! misha 2683: return op_code != OP_NOT_VSPACE;
1.1 misha 2684: }
2685:
1.4 ! misha 2686: #ifdef SUPPORT_UCP
! 2687: case OP_PROP:
! 2688: return check_char_prop(next, previous[0], previous[1], FALSE);
! 2689:
! 2690: case OP_NOTPROP:
! 2691: return check_char_prop(next, previous[0], previous[1], TRUE);
! 2692: #endif
! 2693:
1.1 misha 2694: default:
2695: return FALSE;
2696: }
2697:
2698:
1.4 ! misha 2699: /* Handle the case when the next item is \d, \s, etc. Note that when PCRE_UCP
! 2700: is set, \d turns into ESC_du rather than ESC_d, etc., so ESC_d etc. are
! 2701: generated only when PCRE_UCP is *not* set, that is, when only ASCII
! 2702: characteristics are recognized. Similarly, the opcodes OP_DIGIT etc. are
! 2703: replaced by OP_PROP codes when PCRE_UCP is set. */
1.1 misha 2704:
2705: switch(op_code)
2706: {
2707: case OP_CHAR:
2708: case OP_CHARNC:
2709: #ifdef SUPPORT_UTF8
1.4 ! misha 2710: GETCHARTEST(c, previous);
! 2711: #else
! 2712: c = *previous;
1.1 misha 2713: #endif
2714: switch(-next)
2715: {
2716: case ESC_d:
1.4 ! misha 2717: return c > 127 || (cd->ctypes[c] & ctype_digit) == 0;
1.1 misha 2718:
2719: case ESC_D:
1.4 ! misha 2720: return c <= 127 && (cd->ctypes[c] & ctype_digit) != 0;
1.1 misha 2721:
2722: case ESC_s:
1.4 ! misha 2723: return c > 127 || (cd->ctypes[c] & ctype_space) == 0;
1.1 misha 2724:
2725: case ESC_S:
1.4 ! misha 2726: return c <= 127 && (cd->ctypes[c] & ctype_space) != 0;
1.1 misha 2727:
2728: case ESC_w:
1.4 ! misha 2729: return c > 127 || (cd->ctypes[c] & ctype_word) == 0;
1.1 misha 2730:
2731: case ESC_W:
1.4 ! misha 2732: return c <= 127 && (cd->ctypes[c] & ctype_word) != 0;
1.1 misha 2733:
2734: case ESC_h:
2735: case ESC_H:
1.4 ! misha 2736: switch(c)
1.1 misha 2737: {
2738: case 0x09:
2739: case 0x20:
2740: case 0xa0:
2741: case 0x1680:
2742: case 0x180e:
2743: case 0x2000:
2744: case 0x2001:
2745: case 0x2002:
2746: case 0x2003:
2747: case 0x2004:
2748: case 0x2005:
2749: case 0x2006:
2750: case 0x2007:
2751: case 0x2008:
2752: case 0x2009:
2753: case 0x200A:
2754: case 0x202f:
2755: case 0x205f:
2756: case 0x3000:
2757: return -next != ESC_h;
2758: default:
2759: return -next == ESC_h;
2760: }
2761:
2762: case ESC_v:
2763: case ESC_V:
1.4 ! misha 2764: switch(c)
1.1 misha 2765: {
2766: case 0x0a:
2767: case 0x0b:
2768: case 0x0c:
2769: case 0x0d:
2770: case 0x85:
2771: case 0x2028:
2772: case 0x2029:
2773: return -next != ESC_v;
2774: default:
2775: return -next == ESC_v;
2776: }
2777:
1.4 ! misha 2778: /* When PCRE_UCP is set, these values get generated for \d etc. Find
! 2779: their substitutions and process them. The result will always be either
! 2780: -ESC_p or -ESC_P. Then fall through to process those values. */
! 2781:
! 2782: #ifdef SUPPORT_UCP
! 2783: case ESC_du:
! 2784: case ESC_DU:
! 2785: case ESC_wu:
! 2786: case ESC_WU:
! 2787: case ESC_su:
! 2788: case ESC_SU:
! 2789: {
! 2790: int temperrorcode = 0;
! 2791: ptr = substitutes[-next - ESC_DU];
! 2792: next = check_escape(&ptr, &temperrorcode, 0, options, FALSE);
! 2793: if (temperrorcode != 0) return FALSE;
! 2794: ptr++; /* For compatibility */
! 2795: }
! 2796: /* Fall through */
! 2797:
! 2798: case ESC_p:
! 2799: case ESC_P:
! 2800: {
! 2801: int ptype, pdata, errorcodeptr;
! 2802: BOOL negated;
! 2803:
! 2804: ptr--; /* Make ptr point at the p or P */
! 2805: ptype = get_ucp(&ptr, &negated, &pdata, &errorcodeptr);
! 2806: if (ptype < 0) return FALSE;
! 2807: ptr++; /* Point past the final curly ket */
! 2808:
! 2809: /* If the property item is optional, we have to give up. (When generated
! 2810: from \d etc by PCRE_UCP, this test will have been applied much earlier,
! 2811: to the original \d etc. At this point, ptr will point to a zero byte. */
! 2812:
! 2813: if (*ptr == CHAR_ASTERISK || *ptr == CHAR_QUESTION_MARK ||
! 2814: strncmp((char *)ptr, STR_LEFT_CURLY_BRACKET STR_0 STR_COMMA, 3) == 0)
! 2815: return FALSE;
! 2816:
! 2817: /* Do the property check. */
! 2818:
! 2819: return check_char_prop(c, ptype, pdata, (next == -ESC_P) != negated);
! 2820: }
! 2821: #endif
! 2822:
1.1 misha 2823: default:
2824: return FALSE;
2825: }
2826:
1.4 ! misha 2827: /* In principle, support for Unicode properties should be integrated here as
! 2828: well. It means re-organizing the above code so as to get hold of the property
! 2829: values before switching on the op-code. However, I wonder how many patterns
! 2830: combine ASCII \d etc with Unicode properties? (Note that if PCRE_UCP is set,
! 2831: these op-codes are never generated.) */
! 2832:
1.1 misha 2833: case OP_DIGIT:
2834: return next == -ESC_D || next == -ESC_s || next == -ESC_W ||
1.4 ! misha 2835: next == -ESC_h || next == -ESC_v || next == -ESC_R;
1.1 misha 2836:
2837: case OP_NOT_DIGIT:
2838: return next == -ESC_d;
2839:
2840: case OP_WHITESPACE:
1.4 ! misha 2841: return next == -ESC_S || next == -ESC_d || next == -ESC_w || next == -ESC_R;
1.1 misha 2842:
2843: case OP_NOT_WHITESPACE:
2844: return next == -ESC_s || next == -ESC_h || next == -ESC_v;
2845:
2846: case OP_HSPACE:
1.4 ! misha 2847: return next == -ESC_S || next == -ESC_H || next == -ESC_d ||
! 2848: next == -ESC_w || next == -ESC_v || next == -ESC_R;
1.1 misha 2849:
2850: case OP_NOT_HSPACE:
2851: return next == -ESC_h;
2852:
2853: /* Can't have \S in here because VT matches \S (Perl anomaly) */
1.4 ! misha 2854: case OP_ANYNL:
1.1 misha 2855: case OP_VSPACE:
2856: return next == -ESC_V || next == -ESC_d || next == -ESC_w;
2857:
2858: case OP_NOT_VSPACE:
1.4 ! misha 2859: return next == -ESC_v || next == -ESC_R;
1.1 misha 2860:
2861: case OP_WORDCHAR:
1.4 ! misha 2862: return next == -ESC_W || next == -ESC_s || next == -ESC_h ||
! 2863: next == -ESC_v || next == -ESC_R;
1.1 misha 2864:
2865: case OP_NOT_WORDCHAR:
2866: return next == -ESC_w || next == -ESC_d;
2867:
2868: default:
2869: return FALSE;
2870: }
2871:
2872: /* Control does not reach here */
2873: }
2874:
2875:
2876:
2877: /*************************************************
2878: * Compile one branch *
2879: *************************************************/
2880:
2881: /* Scan the pattern, compiling it into the a vector. If the options are
2882: changed during the branch, the pointer is used to change the external options
2883: bits. This function is used during the pre-compile phase when we are trying
2884: to find out the amount of memory needed, as well as during the real compile
2885: phase. The value of lengthptr distinguishes the two phases.
2886:
2887: Arguments:
2888: optionsptr pointer to the option bits
2889: codeptr points to the pointer to the current code point
2890: ptrptr points to the current pattern pointer
2891: errorcodeptr points to error code variable
2892: firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE)
2893: reqbyteptr set to the last literal character required, else < 0
2894: bcptr points to current branch chain
2895: cd contains pointers to tables etc.
2896: lengthptr NULL during the real compile phase
2897: points to length accumulator during pre-compile phase
2898:
2899: Returns: TRUE on success
2900: FALSE, with *errorcodeptr set non-zero on error
2901: */
2902:
2903: static BOOL
2904: compile_branch(int *optionsptr, uschar **codeptr, const uschar **ptrptr,
2905: int *errorcodeptr, int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr,
2906: compile_data *cd, int *lengthptr)
2907: {
2908: int repeat_type, op_type;
2909: int repeat_min = 0, repeat_max = 0; /* To please picky compilers */
2910: int bravalue = 0;
2911: int greedy_default, greedy_non_default;
2912: int firstbyte, reqbyte;
2913: int zeroreqbyte, zerofirstbyte;
2914: int req_caseopt, reqvary, tempreqvary;
2915: int options = *optionsptr;
2916: int after_manual_callout = 0;
2917: int length_prevgroup = 0;
2918: register int c;
2919: register uschar *code = *codeptr;
2920: uschar *last_code = code;
2921: uschar *orig_code = code;
2922: uschar *tempcode;
2923: BOOL inescq = FALSE;
2924: BOOL groupsetfirstbyte = FALSE;
2925: const uschar *ptr = *ptrptr;
2926: const uschar *tempptr;
1.4 ! misha 2927: const uschar *nestptr = NULL;
1.1 misha 2928: uschar *previous = NULL;
2929: uschar *previous_callout = NULL;
2930: uschar *save_hwm = NULL;
2931: uschar classbits[32];
2932:
2933: #ifdef SUPPORT_UTF8
2934: BOOL class_utf8;
2935: BOOL utf8 = (options & PCRE_UTF8) != 0;
2936: uschar *class_utf8data;
2937: uschar *class_utf8data_base;
2938: uschar utf8_char[6];
2939: #else
2940: BOOL utf8 = FALSE;
2941: uschar *utf8_char = NULL;
2942: #endif
2943:
1.4 ! misha 2944: #ifdef PCRE_DEBUG
1.1 misha 2945: if (lengthptr != NULL) DPRINTF((">> start branch\n"));
2946: #endif
2947:
2948: /* Set up the default and non-default settings for greediness */
2949:
2950: greedy_default = ((options & PCRE_UNGREEDY) != 0);
2951: greedy_non_default = greedy_default ^ 1;
2952:
2953: /* Initialize no first byte, no required byte. REQ_UNSET means "no char
2954: matching encountered yet". It gets changed to REQ_NONE if we hit something that
2955: matches a non-fixed char first char; reqbyte just remains unset if we never
2956: find one.
2957:
2958: When we hit a repeat whose minimum is zero, we may have to adjust these values
2959: to take the zero repeat into account. This is implemented by setting them to
2960: zerofirstbyte and zeroreqbyte when such a repeat is encountered. The individual
2961: item types that can be repeated set these backoff variables appropriately. */
2962:
2963: firstbyte = reqbyte = zerofirstbyte = zeroreqbyte = REQ_UNSET;
2964:
2965: /* The variable req_caseopt contains either the REQ_CASELESS value or zero,
2966: according to the current setting of the caseless flag. REQ_CASELESS is a bit
2967: value > 255. It is added into the firstbyte or reqbyte variables to record the
2968: case status of the value. This is used only for ASCII characters. */
2969:
2970: req_caseopt = ((options & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
2971:
2972: /* Switch on next character until the end of the branch */
2973:
2974: for (;; ptr++)
2975: {
2976: BOOL negate_class;
2977: BOOL should_flip_negation;
2978: BOOL possessive_quantifier;
2979: BOOL is_quantifier;
2980: BOOL is_recurse;
2981: BOOL reset_bracount;
2982: int class_charcount;
2983: int class_lastchar;
2984: int newoptions;
2985: int recno;
2986: int refsign;
2987: int skipbytes;
2988: int subreqbyte;
2989: int subfirstbyte;
2990: int terminator;
2991: int mclength;
2992: uschar mcbuffer[8];
2993:
2994: /* Get next byte in the pattern */
2995:
2996: c = *ptr;
2997:
1.4 ! misha 2998: /* If we are at the end of a nested substitution, revert to the outer level
! 2999: string. Nesting only happens one level deep. */
! 3000:
! 3001: if (c == 0 && nestptr != NULL)
! 3002: {
! 3003: ptr = nestptr;
! 3004: nestptr = NULL;
! 3005: c = *ptr;
! 3006: }
! 3007:
1.1 misha 3008: /* If we are in the pre-compile phase, accumulate the length used for the
3009: previous cycle of this loop. */
3010:
3011: if (lengthptr != NULL)
3012: {
1.4 ! misha 3013: #ifdef PCRE_DEBUG
1.1 misha 3014: if (code > cd->hwm) cd->hwm = code; /* High water info */
3015: #endif
1.4 ! misha 3016: if (code > cd->start_workspace + WORK_SIZE_CHECK) /* Check for overrun */
1.1 misha 3017: {
3018: *errorcodeptr = ERR52;
3019: goto FAILED;
3020: }
3021:
3022: /* There is at least one situation where code goes backwards: this is the
3023: case of a zero quantifier after a class (e.g. [ab]{0}). At compile time,
3024: the class is simply eliminated. However, it is created first, so we have to
3025: allow memory for it. Therefore, don't ever reduce the length at this point.
3026: */
3027:
3028: if (code < last_code) code = last_code;
3029:
3030: /* Paranoid check for integer overflow */
3031:
3032: if (OFLOW_MAX - *lengthptr < code - last_code)
3033: {
3034: *errorcodeptr = ERR20;
3035: goto FAILED;
3036: }
3037:
1.4 ! misha 3038: *lengthptr += (int)(code - last_code);
1.1 misha 3039: DPRINTF(("length=%d added %d c=%c\n", *lengthptr, code - last_code, c));
3040:
3041: /* If "previous" is set and it is not at the start of the work space, move
3042: it back to there, in order to avoid filling up the work space. Otherwise,
3043: if "previous" is NULL, reset the current code pointer to the start. */
3044:
3045: if (previous != NULL)
3046: {
3047: if (previous > orig_code)
3048: {
3049: memmove(orig_code, previous, code - previous);
3050: code -= previous - orig_code;
3051: previous = orig_code;
3052: }
3053: }
3054: else code = orig_code;
3055:
3056: /* Remember where this code item starts so we can pick up the length
3057: next time round. */
3058:
3059: last_code = code;
3060: }
3061:
3062: /* In the real compile phase, just check the workspace used by the forward
3063: reference list. */
3064:
1.4 ! misha 3065: else if (cd->hwm > cd->start_workspace + WORK_SIZE_CHECK)
1.1 misha 3066: {
3067: *errorcodeptr = ERR52;
3068: goto FAILED;
3069: }
3070:
3071: /* If in \Q...\E, check for the end; if not, we have a literal */
3072:
3073: if (inescq && c != 0)
3074: {
1.3 misha 3075: if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E)
1.1 misha 3076: {
3077: inescq = FALSE;
3078: ptr++;
3079: continue;
3080: }
3081: else
3082: {
3083: if (previous_callout != NULL)
3084: {
3085: if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3086: complete_callout(previous_callout, ptr, cd);
3087: previous_callout = NULL;
3088: }
3089: if ((options & PCRE_AUTO_CALLOUT) != 0)
3090: {
3091: previous_callout = code;
3092: code = auto_callout(code, ptr, cd);
3093: }
3094: goto NORMAL_CHAR;
3095: }
3096: }
3097:
3098: /* Fill in length of a previous callout, except when the next thing is
3099: a quantifier. */
3100:
1.3 misha 3101: is_quantifier =
3102: c == CHAR_ASTERISK || c == CHAR_PLUS || c == CHAR_QUESTION_MARK ||
3103: (c == CHAR_LEFT_CURLY_BRACKET && is_counted_repeat(ptr+1));
1.1 misha 3104:
3105: if (!is_quantifier && previous_callout != NULL &&
3106: after_manual_callout-- <= 0)
3107: {
3108: if (lengthptr == NULL) /* Don't attempt in pre-compile phase */
3109: complete_callout(previous_callout, ptr, cd);
3110: previous_callout = NULL;
3111: }
3112:
3113: /* In extended mode, skip white space and comments */
3114:
3115: if ((options & PCRE_EXTENDED) != 0)
3116: {
3117: if ((cd->ctypes[c] & ctype_space) != 0) continue;
1.3 misha 3118: if (c == CHAR_NUMBER_SIGN)
1.1 misha 3119: {
3120: while (*(++ptr) != 0)
3121: {
3122: if (IS_NEWLINE(ptr)) { ptr += cd->nllen - 1; break; }
3123: }
3124: if (*ptr != 0) continue;
3125:
3126: /* Else fall through to handle end of string */
3127: c = 0;
3128: }
3129: }
3130:
3131: /* No auto callout for quantifiers. */
3132:
3133: if ((options & PCRE_AUTO_CALLOUT) != 0 && !is_quantifier)
3134: {
3135: previous_callout = code;
3136: code = auto_callout(code, ptr, cd);
3137: }
3138:
3139: switch(c)
3140: {
3141: /* ===================================================================*/
3142: case 0: /* The branch terminates at string end */
1.3 misha 3143: case CHAR_VERTICAL_LINE: /* or | or ) */
3144: case CHAR_RIGHT_PARENTHESIS:
1.1 misha 3145: *firstbyteptr = firstbyte;
3146: *reqbyteptr = reqbyte;
3147: *codeptr = code;
3148: *ptrptr = ptr;
3149: if (lengthptr != NULL)
3150: {
3151: if (OFLOW_MAX - *lengthptr < code - last_code)
3152: {
3153: *errorcodeptr = ERR20;
3154: goto FAILED;
3155: }
1.4 ! misha 3156: *lengthptr += (int)(code - last_code); /* To include callout length */
1.1 misha 3157: DPRINTF((">> end branch\n"));
3158: }
3159: return TRUE;
3160:
3161:
3162: /* ===================================================================*/
3163: /* Handle single-character metacharacters. In multiline mode, ^ disables
3164: the setting of any following char as a first character. */
3165:
1.3 misha 3166: case CHAR_CIRCUMFLEX_ACCENT:
1.1 misha 3167: if ((options & PCRE_MULTILINE) != 0)
3168: {
3169: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3170: }
3171: previous = NULL;
3172: *code++ = OP_CIRC;
3173: break;
3174:
1.3 misha 3175: case CHAR_DOLLAR_SIGN:
1.1 misha 3176: previous = NULL;
3177: *code++ = OP_DOLL;
3178: break;
3179:
3180: /* There can never be a first char if '.' is first, whatever happens about
3181: repeats. The value of reqbyte doesn't change either. */
3182:
1.3 misha 3183: case CHAR_DOT:
1.1 misha 3184: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3185: zerofirstbyte = firstbyte;
3186: zeroreqbyte = reqbyte;
3187: previous = code;
3188: *code++ = ((options & PCRE_DOTALL) != 0)? OP_ALLANY: OP_ANY;
3189: break;
3190:
3191:
3192: /* ===================================================================*/
3193: /* Character classes. If the included characters are all < 256, we build a
3194: 32-byte bitmap of the permitted characters, except in the special case
3195: where there is only one such character. For negated classes, we build the
3196: map as usual, then invert it at the end. However, we use a different opcode
3197: so that data characters > 255 can be handled correctly.
3198:
3199: If the class contains characters outside the 0-255 range, a different
3200: opcode is compiled. It may optionally have a bit map for characters < 256,
3201: but those above are are explicitly listed afterwards. A flag byte tells
3202: whether the bitmap is present, and whether this is a negated class or not.
3203:
3204: In JavaScript compatibility mode, an isolated ']' causes an error. In
3205: default (Perl) mode, it is treated as a data character. */
3206:
1.3 misha 3207: case CHAR_RIGHT_SQUARE_BRACKET:
1.1 misha 3208: if ((cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
3209: {
3210: *errorcodeptr = ERR64;
3211: goto FAILED;
3212: }
3213: goto NORMAL_CHAR;
3214:
1.3 misha 3215: case CHAR_LEFT_SQUARE_BRACKET:
1.1 misha 3216: previous = code;
3217:
3218: /* PCRE supports POSIX class stuff inside a class. Perl gives an error if
3219: they are encountered at the top level, so we'll do that too. */
3220:
1.3 misha 3221: if ((ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3222: ptr[1] == CHAR_EQUALS_SIGN) &&
1.1 misha 3223: check_posix_syntax(ptr, &tempptr))
3224: {
1.3 misha 3225: *errorcodeptr = (ptr[1] == CHAR_COLON)? ERR13 : ERR31;
1.1 misha 3226: goto FAILED;
3227: }
3228:
3229: /* If the first character is '^', set the negation flag and skip it. Also,
3230: if the first few characters (either before or after ^) are \Q\E or \E we
3231: skip them too. This makes for compatibility with Perl. */
3232:
3233: negate_class = FALSE;
3234: for (;;)
3235: {
3236: c = *(++ptr);
1.3 misha 3237: if (c == CHAR_BACKSLASH)
1.1 misha 3238: {
1.3 misha 3239: if (ptr[1] == CHAR_E)
3240: ptr++;
3241: else if (strncmp((const char *)ptr+1,
3242: STR_Q STR_BACKSLASH STR_E, 3) == 0)
3243: ptr += 3;
3244: else
3245: break;
1.1 misha 3246: }
1.3 misha 3247: else if (!negate_class && c == CHAR_CIRCUMFLEX_ACCENT)
1.1 misha 3248: negate_class = TRUE;
3249: else break;
3250: }
3251:
3252: /* Empty classes are allowed in JavaScript compatibility mode. Otherwise,
3253: an initial ']' is taken as a data character -- the code below handles
3254: that. In JS mode, [] must always fail, so generate OP_FAIL, whereas
3255: [^] must match any character, so generate OP_ALLANY. */
3256:
1.3 misha 3257: if (c == CHAR_RIGHT_SQUARE_BRACKET &&
3258: (cd->external_options & PCRE_JAVASCRIPT_COMPAT) != 0)
1.1 misha 3259: {
3260: *code++ = negate_class? OP_ALLANY : OP_FAIL;
3261: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3262: zerofirstbyte = firstbyte;
3263: break;
3264: }
3265:
3266: /* If a class contains a negative special such as \S, we need to flip the
3267: negation flag at the end, so that support for characters > 255 works
3268: correctly (they are all included in the class). */
3269:
3270: should_flip_negation = FALSE;
3271:
3272: /* Keep a count of chars with values < 256 so that we can optimize the case
3273: of just a single character (as long as it's < 256). However, For higher
3274: valued UTF-8 characters, we don't yet do any optimization. */
3275:
3276: class_charcount = 0;
3277: class_lastchar = -1;
3278:
3279: /* Initialize the 32-char bit map to all zeros. We build the map in a
3280: temporary bit of memory, in case the class contains only 1 character (less
3281: than 256), because in that case the compiled code doesn't use the bit map.
3282: */
3283:
3284: memset(classbits, 0, 32 * sizeof(uschar));
3285:
3286: #ifdef SUPPORT_UTF8
3287: class_utf8 = FALSE; /* No chars >= 256 */
3288: class_utf8data = code + LINK_SIZE + 2; /* For UTF-8 items */
3289: class_utf8data_base = class_utf8data; /* For resetting in pass 1 */
3290: #endif
3291:
3292: /* Process characters until ] is reached. By writing this as a "do" it
3293: means that an initial ] is taken as a data character. At the start of the
3294: loop, c contains the first byte of the character. */
3295:
3296: if (c != 0) do
3297: {
3298: const uschar *oldptr;
3299:
3300: #ifdef SUPPORT_UTF8
3301: if (utf8 && c > 127)
3302: { /* Braces are required because the */
3303: GETCHARLEN(c, ptr, ptr); /* macro generates multiple statements */
3304: }
3305:
3306: /* In the pre-compile phase, accumulate the length of any UTF-8 extra
3307: data and reset the pointer. This is so that very large classes that
3308: contain a zillion UTF-8 characters no longer overwrite the work space
3309: (which is on the stack). */
3310:
3311: if (lengthptr != NULL)
3312: {
3313: *lengthptr += class_utf8data - class_utf8data_base;
3314: class_utf8data = class_utf8data_base;
3315: }
3316:
3317: #endif
3318:
3319: /* Inside \Q...\E everything is literal except \E */
3320:
3321: if (inescq)
3322: {
1.3 misha 3323: if (c == CHAR_BACKSLASH && ptr[1] == CHAR_E) /* If we are at \E */
1.1 misha 3324: {
3325: inescq = FALSE; /* Reset literal state */
3326: ptr++; /* Skip the 'E' */
3327: continue; /* Carry on with next */
3328: }
3329: goto CHECK_RANGE; /* Could be range if \E follows */
3330: }
3331:
3332: /* Handle POSIX class names. Perl allows a negation extension of the
3333: form [:^name:]. A square bracket that doesn't match the syntax is
3334: treated as a literal. We also recognize the POSIX constructions
3335: [.ch.] and [=ch=] ("collating elements") and fault them, as Perl
3336: 5.6 and 5.8 do. */
3337:
1.3 misha 3338: if (c == CHAR_LEFT_SQUARE_BRACKET &&
3339: (ptr[1] == CHAR_COLON || ptr[1] == CHAR_DOT ||
3340: ptr[1] == CHAR_EQUALS_SIGN) && check_posix_syntax(ptr, &tempptr))
1.1 misha 3341: {
3342: BOOL local_negate = FALSE;
3343: int posix_class, taboffset, tabopt;
3344: register const uschar *cbits = cd->cbits;
3345: uschar pbits[32];
3346:
1.3 misha 3347: if (ptr[1] != CHAR_COLON)
1.1 misha 3348: {
3349: *errorcodeptr = ERR31;
3350: goto FAILED;
3351: }
3352:
3353: ptr += 2;
1.3 misha 3354: if (*ptr == CHAR_CIRCUMFLEX_ACCENT)
1.1 misha 3355: {
3356: local_negate = TRUE;
3357: should_flip_negation = TRUE; /* Note negative special */
3358: ptr++;
3359: }
3360:
1.4 ! misha 3361: posix_class = check_posix_name(ptr, (int)(tempptr - ptr));
1.1 misha 3362: if (posix_class < 0)
3363: {
3364: *errorcodeptr = ERR30;
3365: goto FAILED;
3366: }
3367:
3368: /* If matching is caseless, upper and lower are converted to
3369: alpha. This relies on the fact that the class table starts with
3370: alpha, lower, upper as the first 3 entries. */
3371:
3372: if ((options & PCRE_CASELESS) != 0 && posix_class <= 2)
3373: posix_class = 0;
3374:
1.4 ! misha 3375: /* When PCRE_UCP is set, some of the POSIX classes are converted to
! 3376: different escape sequences that use Unicode properties. */
! 3377:
! 3378: #ifdef SUPPORT_UCP
! 3379: if ((options & PCRE_UCP) != 0)
! 3380: {
! 3381: int pc = posix_class + ((local_negate)? POSIX_SUBSIZE/2 : 0);
! 3382: if (posix_substitutes[pc] != NULL)
! 3383: {
! 3384: nestptr = tempptr + 1;
! 3385: ptr = posix_substitutes[pc] - 1;
! 3386: continue;
! 3387: }
! 3388: }
! 3389: #endif
! 3390: /* In the non-UCP case, we build the bit map for the POSIX class in a
! 3391: chunk of local store because we may be adding and subtracting from it,
! 3392: and we don't want to subtract bits that may be in the main map already.
! 3393: At the end we or the result into the bit map that is being built. */
1.1 misha 3394:
3395: posix_class *= 3;
3396:
3397: /* Copy in the first table (always present) */
3398:
3399: memcpy(pbits, cbits + posix_class_maps[posix_class],
3400: 32 * sizeof(uschar));
3401:
3402: /* If there is a second table, add or remove it as required. */
3403:
3404: taboffset = posix_class_maps[posix_class + 1];
3405: tabopt = posix_class_maps[posix_class + 2];
3406:
3407: if (taboffset >= 0)
3408: {
3409: if (tabopt >= 0)
3410: for (c = 0; c < 32; c++) pbits[c] |= cbits[c + taboffset];
3411: else
3412: for (c = 0; c < 32; c++) pbits[c] &= ~cbits[c + taboffset];
3413: }
3414:
3415: /* Not see if we need to remove any special characters. An option
3416: value of 1 removes vertical space and 2 removes underscore. */
3417:
3418: if (tabopt < 0) tabopt = -tabopt;
3419: if (tabopt == 1) pbits[1] &= ~0x3c;
3420: else if (tabopt == 2) pbits[11] &= 0x7f;
3421:
3422: /* Add the POSIX table or its complement into the main table that is
3423: being built and we are done. */
3424:
3425: if (local_negate)
3426: for (c = 0; c < 32; c++) classbits[c] |= ~pbits[c];
3427: else
3428: for (c = 0; c < 32; c++) classbits[c] |= pbits[c];
3429:
3430: ptr = tempptr + 1;
3431: class_charcount = 10; /* Set > 1; assumes more than 1 per class */
3432: continue; /* End of POSIX syntax handling */
3433: }
3434:
3435: /* Backslash may introduce a single character, or it may introduce one
3436: of the specials, which just set a flag. The sequence \b is a special
1.4 ! misha 3437: case. Inside a class (and only there) it is treated as backspace. We
! 3438: assume that other escapes have more than one character in them, so set
! 3439: class_charcount bigger than one. Unrecognized escapes fall through and
! 3440: are either treated as literal characters (by default), or are faulted if
! 3441: PCRE_EXTRA is set. */
1.1 misha 3442:
1.3 misha 3443: if (c == CHAR_BACKSLASH)
1.1 misha 3444: {
3445: c = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3446: if (*errorcodeptr != 0) goto FAILED;
3447:
1.4 ! misha 3448: if (-c == ESC_b) c = CHAR_BS; /* \b is backspace in a class */
1.1 misha 3449: else if (-c == ESC_Q) /* Handle start of quoted string */
3450: {
1.3 misha 3451: if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
1.1 misha 3452: {
3453: ptr += 2; /* avoid empty string */
3454: }
3455: else inescq = TRUE;
3456: continue;
3457: }
3458: else if (-c == ESC_E) continue; /* Ignore orphan \E */
3459:
3460: if (c < 0)
3461: {
3462: register const uschar *cbits = cd->cbits;
3463: class_charcount += 2; /* Greater than 1 is what matters */
3464:
1.4 ! misha 3465: switch (-c)
1.1 misha 3466: {
1.4 ! misha 3467: #ifdef SUPPORT_UCP
! 3468: case ESC_du: /* These are the values given for \d etc */
! 3469: case ESC_DU: /* when PCRE_UCP is set. We replace the */
! 3470: case ESC_wu: /* escape sequence with an appropriate \p */
! 3471: case ESC_WU: /* or \P to test Unicode properties instead */
! 3472: case ESC_su: /* of the default ASCII testing. */
! 3473: case ESC_SU:
! 3474: nestptr = ptr;
! 3475: ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
! 3476: class_charcount -= 2; /* Undo! */
! 3477: continue;
! 3478: #endif
1.1 misha 3479: case ESC_d:
3480: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_digit];
3481: continue;
3482:
3483: case ESC_D:
3484: should_flip_negation = TRUE;
3485: for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_digit];
3486: continue;
3487:
3488: case ESC_w:
3489: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_word];
3490: continue;
3491:
3492: case ESC_W:
3493: should_flip_negation = TRUE;
3494: for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_word];
3495: continue;
3496:
3497: case ESC_s:
3498: for (c = 0; c < 32; c++) classbits[c] |= cbits[c+cbit_space];
3499: classbits[1] &= ~0x08; /* Perl 5.004 onwards omits VT from \s */
3500: continue;
3501:
3502: case ESC_S:
3503: should_flip_negation = TRUE;
3504: for (c = 0; c < 32; c++) classbits[c] |= ~cbits[c+cbit_space];
3505: classbits[1] |= 0x08; /* Perl 5.004 onwards omits VT from \s */
3506: continue;
3507:
1.4 ! misha 3508: case ESC_h:
1.1 misha 3509: SETBIT(classbits, 0x09); /* VT */
3510: SETBIT(classbits, 0x20); /* SPACE */
3511: SETBIT(classbits, 0xa0); /* NSBP */
3512: #ifdef SUPPORT_UTF8
3513: if (utf8)
3514: {
3515: class_utf8 = TRUE;
3516: *class_utf8data++ = XCL_SINGLE;
3517: class_utf8data += _pcre_ord2utf8(0x1680, class_utf8data);
3518: *class_utf8data++ = XCL_SINGLE;
3519: class_utf8data += _pcre_ord2utf8(0x180e, class_utf8data);
3520: *class_utf8data++ = XCL_RANGE;
3521: class_utf8data += _pcre_ord2utf8(0x2000, class_utf8data);
3522: class_utf8data += _pcre_ord2utf8(0x200A, class_utf8data);
3523: *class_utf8data++ = XCL_SINGLE;
3524: class_utf8data += _pcre_ord2utf8(0x202f, class_utf8data);
3525: *class_utf8data++ = XCL_SINGLE;
3526: class_utf8data += _pcre_ord2utf8(0x205f, class_utf8data);
3527: *class_utf8data++ = XCL_SINGLE;
3528: class_utf8data += _pcre_ord2utf8(0x3000, class_utf8data);
3529: }
3530: #endif
3531: continue;
3532:
1.4 ! misha 3533: case ESC_H:
1.1 misha 3534: for (c = 0; c < 32; c++)
3535: {
3536: int x = 0xff;
3537: switch (c)
3538: {
3539: case 0x09/8: x ^= 1 << (0x09%8); break;
3540: case 0x20/8: x ^= 1 << (0x20%8); break;
3541: case 0xa0/8: x ^= 1 << (0xa0%8); break;
3542: default: break;
3543: }
3544: classbits[c] |= x;
3545: }
3546:
3547: #ifdef SUPPORT_UTF8
3548: if (utf8)
3549: {
3550: class_utf8 = TRUE;
3551: *class_utf8data++ = XCL_RANGE;
3552: class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3553: class_utf8data += _pcre_ord2utf8(0x167f, class_utf8data);
3554: *class_utf8data++ = XCL_RANGE;
3555: class_utf8data += _pcre_ord2utf8(0x1681, class_utf8data);
3556: class_utf8data += _pcre_ord2utf8(0x180d, class_utf8data);
3557: *class_utf8data++ = XCL_RANGE;
3558: class_utf8data += _pcre_ord2utf8(0x180f, class_utf8data);
3559: class_utf8data += _pcre_ord2utf8(0x1fff, class_utf8data);
3560: *class_utf8data++ = XCL_RANGE;
3561: class_utf8data += _pcre_ord2utf8(0x200B, class_utf8data);
3562: class_utf8data += _pcre_ord2utf8(0x202e, class_utf8data);
3563: *class_utf8data++ = XCL_RANGE;
3564: class_utf8data += _pcre_ord2utf8(0x2030, class_utf8data);
3565: class_utf8data += _pcre_ord2utf8(0x205e, class_utf8data);
3566: *class_utf8data++ = XCL_RANGE;
3567: class_utf8data += _pcre_ord2utf8(0x2060, class_utf8data);
3568: class_utf8data += _pcre_ord2utf8(0x2fff, class_utf8data);
3569: *class_utf8data++ = XCL_RANGE;
3570: class_utf8data += _pcre_ord2utf8(0x3001, class_utf8data);
3571: class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3572: }
3573: #endif
3574: continue;
3575:
1.4 ! misha 3576: case ESC_v:
1.1 misha 3577: SETBIT(classbits, 0x0a); /* LF */
3578: SETBIT(classbits, 0x0b); /* VT */
3579: SETBIT(classbits, 0x0c); /* FF */
3580: SETBIT(classbits, 0x0d); /* CR */
3581: SETBIT(classbits, 0x85); /* NEL */
3582: #ifdef SUPPORT_UTF8
3583: if (utf8)
3584: {
3585: class_utf8 = TRUE;
3586: *class_utf8data++ = XCL_RANGE;
3587: class_utf8data += _pcre_ord2utf8(0x2028, class_utf8data);
3588: class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3589: }
3590: #endif
3591: continue;
3592:
1.4 ! misha 3593: case ESC_V:
1.1 misha 3594: for (c = 0; c < 32; c++)
3595: {
3596: int x = 0xff;
3597: switch (c)
3598: {
3599: case 0x0a/8: x ^= 1 << (0x0a%8);
3600: x ^= 1 << (0x0b%8);
3601: x ^= 1 << (0x0c%8);
3602: x ^= 1 << (0x0d%8);
3603: break;
3604: case 0x85/8: x ^= 1 << (0x85%8); break;
3605: default: break;
3606: }
3607: classbits[c] |= x;
3608: }
3609:
3610: #ifdef SUPPORT_UTF8
3611: if (utf8)
3612: {
3613: class_utf8 = TRUE;
3614: *class_utf8data++ = XCL_RANGE;
3615: class_utf8data += _pcre_ord2utf8(0x0100, class_utf8data);
3616: class_utf8data += _pcre_ord2utf8(0x2027, class_utf8data);
3617: *class_utf8data++ = XCL_RANGE;
3618: class_utf8data += _pcre_ord2utf8(0x2029, class_utf8data);
3619: class_utf8data += _pcre_ord2utf8(0x7fffffff, class_utf8data);
3620: }
3621: #endif
3622: continue;
3623:
3624: #ifdef SUPPORT_UCP
1.4 ! misha 3625: case ESC_p:
! 3626: case ESC_P:
! 3627: {
! 3628: BOOL negated;
! 3629: int pdata;
! 3630: int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
! 3631: if (ptype < 0) goto FAILED;
! 3632: class_utf8 = TRUE;
! 3633: *class_utf8data++ = ((-c == ESC_p) != negated)?
! 3634: XCL_PROP : XCL_NOTPROP;
! 3635: *class_utf8data++ = ptype;
! 3636: *class_utf8data++ = pdata;
! 3637: class_charcount -= 2; /* Not a < 256 character */
! 3638: continue;
! 3639: }
1.1 misha 3640: #endif
1.4 ! misha 3641: /* Unrecognized escapes are faulted if PCRE is running in its
! 3642: strict mode. By default, for compatibility with Perl, they are
! 3643: treated as literals. */
1.1 misha 3644:
1.4 ! misha 3645: default:
! 3646: if ((options & PCRE_EXTRA) != 0)
! 3647: {
! 3648: *errorcodeptr = ERR7;
! 3649: goto FAILED;
! 3650: }
! 3651: class_charcount -= 2; /* Undo the default count from above */
! 3652: c = *ptr; /* Get the final character and fall through */
! 3653: break;
1.1 misha 3654: }
3655: }
3656:
3657: /* Fall through if we have a single character (c >= 0). This may be
3658: greater than 256 in UTF-8 mode. */
3659:
3660: } /* End of backslash handling */
3661:
3662: /* A single character may be followed by '-' to form a range. However,
3663: Perl does not permit ']' to be the end of the range. A '-' character
3664: at the end is treated as a literal. Perl ignores orphaned \E sequences
3665: entirely. The code for handling \Q and \E is messy. */
3666:
3667: CHECK_RANGE:
1.3 misha 3668: while (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
1.1 misha 3669: {
3670: inescq = FALSE;
3671: ptr += 2;
3672: }
3673:
3674: oldptr = ptr;
3675:
3676: /* Remember \r or \n */
3677:
1.3 misha 3678: if (c == CHAR_CR || c == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
1.1 misha 3679:
3680: /* Check for range */
3681:
1.3 misha 3682: if (!inescq && ptr[1] == CHAR_MINUS)
1.1 misha 3683: {
3684: int d;
3685: ptr += 2;
1.3 misha 3686: while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E) ptr += 2;
1.1 misha 3687:
3688: /* If we hit \Q (not followed by \E) at this point, go into escaped
3689: mode. */
3690:
1.3 misha 3691: while (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_Q)
1.1 misha 3692: {
3693: ptr += 2;
1.3 misha 3694: if (*ptr == CHAR_BACKSLASH && ptr[1] == CHAR_E)
3695: { ptr += 2; continue; }
1.1 misha 3696: inescq = TRUE;
3697: break;
3698: }
3699:
1.3 misha 3700: if (*ptr == 0 || (!inescq && *ptr == CHAR_RIGHT_SQUARE_BRACKET))
1.1 misha 3701: {
3702: ptr = oldptr;
3703: goto LONE_SINGLE_CHARACTER;
3704: }
3705:
3706: #ifdef SUPPORT_UTF8
3707: if (utf8)
3708: { /* Braces are required because the */
3709: GETCHARLEN(d, ptr, ptr); /* macro generates multiple statements */
3710: }
3711: else
3712: #endif
3713: d = *ptr; /* Not UTF-8 mode */
3714:
3715: /* The second part of a range can be a single-character escape, but
3716: not any of the other escapes. Perl 5.6 treats a hyphen as a literal
3717: in such circumstances. */
3718:
1.3 misha 3719: if (!inescq && d == CHAR_BACKSLASH)
1.1 misha 3720: {
3721: d = check_escape(&ptr, errorcodeptr, cd->bracount, options, TRUE);
3722: if (*errorcodeptr != 0) goto FAILED;
3723:
1.4 ! misha 3724: /* \b is backspace; any other special means the '-' was literal */
1.1 misha 3725:
3726: if (d < 0)
3727: {
1.4 ! misha 3728: if (d == -ESC_b) d = CHAR_BS; else
1.1 misha 3729: {
3730: ptr = oldptr;
3731: goto LONE_SINGLE_CHARACTER; /* A few lines below */
3732: }
3733: }
3734: }
3735:
3736: /* Check that the two values are in the correct order. Optimize
3737: one-character ranges */
3738:
3739: if (d < c)
3740: {
3741: *errorcodeptr = ERR8;
3742: goto FAILED;
3743: }
3744:
3745: if (d == c) goto LONE_SINGLE_CHARACTER; /* A few lines below */
3746:
3747: /* Remember \r or \n */
3748:
1.3 misha 3749: if (d == CHAR_CR || d == CHAR_NL) cd->external_flags |= PCRE_HASCRORLF;
1.1 misha 3750:
3751: /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless
3752: matching, we have to use an XCLASS with extra data items. Caseless
3753: matching for characters > 127 is available only if UCP support is
3754: available. */
3755:
3756: #ifdef SUPPORT_UTF8
3757: if (utf8 && (d > 255 || ((options & PCRE_CASELESS) != 0 && d > 127)))
3758: {
3759: class_utf8 = TRUE;
3760:
3761: /* With UCP support, we can find the other case equivalents of
3762: the relevant characters. There may be several ranges. Optimize how
3763: they fit with the basic range. */
3764:
3765: #ifdef SUPPORT_UCP
3766: if ((options & PCRE_CASELESS) != 0)
3767: {
3768: unsigned int occ, ocd;
3769: unsigned int cc = c;
3770: unsigned int origd = d;
3771: while (get_othercase_range(&cc, origd, &occ, &ocd))
3772: {
3773: if (occ >= (unsigned int)c &&
3774: ocd <= (unsigned int)d)
3775: continue; /* Skip embedded ranges */
3776:
3777: if (occ < (unsigned int)c &&
3778: ocd >= (unsigned int)c - 1) /* Extend the basic range */
3779: { /* if there is overlap, */
3780: c = occ; /* noting that if occ < c */
3781: continue; /* we can't have ocd > d */
3782: } /* because a subrange is */
3783: if (ocd > (unsigned int)d &&
3784: occ <= (unsigned int)d + 1) /* always shorter than */
3785: { /* the basic range. */
3786: d = ocd;
3787: continue;
3788: }
3789:
3790: if (occ == ocd)
3791: {
3792: *class_utf8data++ = XCL_SINGLE;
3793: }
3794: else
3795: {
3796: *class_utf8data++ = XCL_RANGE;
3797: class_utf8data += _pcre_ord2utf8(occ, class_utf8data);
3798: }
3799: class_utf8data += _pcre_ord2utf8(ocd, class_utf8data);
3800: }
3801: }
3802: #endif /* SUPPORT_UCP */
3803:
3804: /* Now record the original range, possibly modified for UCP caseless
3805: overlapping ranges. */
3806:
3807: *class_utf8data++ = XCL_RANGE;
3808: class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3809: class_utf8data += _pcre_ord2utf8(d, class_utf8data);
3810:
3811: /* With UCP support, we are done. Without UCP support, there is no
3812: caseless matching for UTF-8 characters > 127; we can use the bit map
3813: for the smaller ones. */
3814:
3815: #ifdef SUPPORT_UCP
3816: continue; /* With next character in the class */
3817: #else
3818: if ((options & PCRE_CASELESS) == 0 || c > 127) continue;
3819:
3820: /* Adjust upper limit and fall through to set up the map */
3821:
3822: d = 127;
3823:
3824: #endif /* SUPPORT_UCP */
3825: }
3826: #endif /* SUPPORT_UTF8 */
3827:
3828: /* We use the bit map for all cases when not in UTF-8 mode; else
3829: ranges that lie entirely within 0-127 when there is UCP support; else
3830: for partial ranges without UCP support. */
3831:
3832: class_charcount += d - c + 1;
3833: class_lastchar = d;
3834:
3835: /* We can save a bit of time by skipping this in the pre-compile. */
3836:
3837: if (lengthptr == NULL) for (; c <= d; c++)
3838: {
3839: classbits[c/8] |= (1 << (c&7));
3840: if ((options & PCRE_CASELESS) != 0)
3841: {
3842: int uc = cd->fcc[c]; /* flip case */
3843: classbits[uc/8] |= (1 << (uc&7));
3844: }
3845: }
3846:
3847: continue; /* Go get the next char in the class */
3848: }
3849:
3850: /* Handle a lone single character - we can get here for a normal
3851: non-escape char, or after \ that introduces a single character or for an
3852: apparent range that isn't. */
3853:
3854: LONE_SINGLE_CHARACTER:
3855:
3856: /* Handle a character that cannot go in the bit map */
3857:
3858: #ifdef SUPPORT_UTF8
3859: if (utf8 && (c > 255 || ((options & PCRE_CASELESS) != 0 && c > 127)))
3860: {
3861: class_utf8 = TRUE;
3862: *class_utf8data++ = XCL_SINGLE;
3863: class_utf8data += _pcre_ord2utf8(c, class_utf8data);
3864:
3865: #ifdef SUPPORT_UCP
3866: if ((options & PCRE_CASELESS) != 0)
3867: {
3868: unsigned int othercase;
1.2 misha 3869: if ((othercase = UCD_OTHERCASE(c)) != c)
1.1 misha 3870: {
3871: *class_utf8data++ = XCL_SINGLE;
3872: class_utf8data += _pcre_ord2utf8(othercase, class_utf8data);
3873: }
3874: }
3875: #endif /* SUPPORT_UCP */
3876:
3877: }
3878: else
3879: #endif /* SUPPORT_UTF8 */
3880:
3881: /* Handle a single-byte character */
3882: {
3883: classbits[c/8] |= (1 << (c&7));
3884: if ((options & PCRE_CASELESS) != 0)
3885: {
3886: c = cd->fcc[c]; /* flip case */
3887: classbits[c/8] |= (1 << (c&7));
3888: }
3889: class_charcount++;
3890: class_lastchar = c;
3891: }
3892: }
3893:
1.4 ! misha 3894: /* Loop until ']' reached. This "while" is the end of the "do" far above.
! 3895: If we are at the end of an internal nested string, revert to the outer
! 3896: string. */
! 3897:
! 3898: while (((c = *(++ptr)) != 0 ||
! 3899: (nestptr != NULL &&
! 3900: (ptr = nestptr, nestptr = NULL, c = *(++ptr)) != 0)) &&
! 3901: (c != CHAR_RIGHT_SQUARE_BRACKET || inescq));
1.1 misha 3902:
1.4 ! misha 3903: /* Check for missing terminating ']' */
1.1 misha 3904:
1.4 ! misha 3905: if (c == 0)
1.1 misha 3906: {
3907: *errorcodeptr = ERR6;
3908: goto FAILED;
3909: }
3910:
3911: /* If class_charcount is 1, we saw precisely one character whose value is
3912: less than 256. As long as there were no characters >= 128 and there was no
3913: use of \p or \P, in other words, no use of any XCLASS features, we can
3914: optimize.
3915:
3916: In UTF-8 mode, we can optimize the negative case only if there were no
3917: characters >= 128 because OP_NOT and the related opcodes like OP_NOTSTAR
3918: operate on single-bytes only. This is an historical hangover. Maybe one day
3919: we can tidy these opcodes to handle multi-byte characters.
3920:
3921: The optimization throws away the bit map. We turn the item into a
3922: 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note
3923: that OP_NOT does not support multibyte characters. In the positive case, it
3924: can cause firstbyte to be set. Otherwise, there can be no first char if
3925: this item is first, whatever repeat count may follow. In the case of
3926: reqbyte, save the previous value for reinstating. */
3927:
3928: #ifdef SUPPORT_UTF8
3929: if (class_charcount == 1 && !class_utf8 &&
3930: (!utf8 || !negate_class || class_lastchar < 128))
3931: #else
3932: if (class_charcount == 1)
3933: #endif
3934: {
3935: zeroreqbyte = reqbyte;
3936:
3937: /* The OP_NOT opcode works on one-byte characters only. */
3938:
3939: if (negate_class)
3940: {
3941: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3942: zerofirstbyte = firstbyte;
3943: *code++ = OP_NOT;
3944: *code++ = class_lastchar;
3945: break;
3946: }
3947:
3948: /* For a single, positive character, get the value into mcbuffer, and
3949: then we can handle this with the normal one-character code. */
3950:
3951: #ifdef SUPPORT_UTF8
3952: if (utf8 && class_lastchar > 127)
3953: mclength = _pcre_ord2utf8(class_lastchar, mcbuffer);
3954: else
3955: #endif
3956: {
3957: mcbuffer[0] = class_lastchar;
3958: mclength = 1;
3959: }
3960: goto ONE_CHAR;
3961: } /* End of 1-char optimization */
3962:
3963: /* The general case - not the one-char optimization. If this is the first
3964: thing in the branch, there can be no first char setting, whatever the
3965: repeat count. Any reqbyte setting must remain unchanged after any kind of
3966: repeat. */
3967:
3968: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
3969: zerofirstbyte = firstbyte;
3970: zeroreqbyte = reqbyte;
3971:
3972: /* If there are characters with values > 255, we have to compile an
3973: extended class, with its own opcode, unless there was a negated special
1.4 ! misha 3974: such as \S in the class, and PCRE_UCP is not set, because in that case all
! 3975: characters > 255 are in the class, so any that were explicitly given as
! 3976: well can be ignored. If (when there are explicit characters > 255 that must
! 3977: be listed) there are no characters < 256, we can omit the bitmap in the
! 3978: actual compiled code. */
1.1 misha 3979:
3980: #ifdef SUPPORT_UTF8
1.4 ! misha 3981: if (class_utf8 && (!should_flip_negation || (options & PCRE_UCP) != 0))
1.1 misha 3982: {
3983: *class_utf8data++ = XCL_END; /* Marks the end of extra data */
3984: *code++ = OP_XCLASS;
3985: code += LINK_SIZE;
3986: *code = negate_class? XCL_NOT : 0;
3987:
3988: /* If the map is required, move up the extra data to make room for it;
3989: otherwise just move the code pointer to the end of the extra data. */
3990:
3991: if (class_charcount > 0)
3992: {
3993: *code++ |= XCL_MAP;
3994: memmove(code + 32, code, class_utf8data - code);
3995: memcpy(code, classbits, 32);
3996: code = class_utf8data + 32;
3997: }
3998: else code = class_utf8data;
3999:
4000: /* Now fill in the complete length of the item */
4001:
4002: PUT(previous, 1, code - previous);
4003: break; /* End of class handling */
4004: }
4005: #endif
4006:
1.4 ! misha 4007: /* If there are no characters > 255, or they are all to be included or
! 4008: excluded, set the opcode to OP_CLASS or OP_NCLASS, depending on whether the
! 4009: whole class was negated and whether there were negative specials such as \S
! 4010: (non-UCP) in the class. Then copy the 32-byte map into the code vector,
! 4011: negating it if necessary. */
1.1 misha 4012:
4013: *code++ = (negate_class == should_flip_negation) ? OP_CLASS : OP_NCLASS;
4014: if (negate_class)
4015: {
4016: if (lengthptr == NULL) /* Save time in the pre-compile phase */
4017: for (c = 0; c < 32; c++) code[c] = ~classbits[c];
4018: }
4019: else
4020: {
4021: memcpy(code, classbits, 32);
4022: }
4023: code += 32;
4024: break;
4025:
4026:
4027: /* ===================================================================*/
4028: /* Various kinds of repeat; '{' is not necessarily a quantifier, but this
4029: has been tested above. */
4030:
1.3 misha 4031: case CHAR_LEFT_CURLY_BRACKET:
1.1 misha 4032: if (!is_quantifier) goto NORMAL_CHAR;
4033: ptr = read_repeat_counts(ptr+1, &repeat_min, &repeat_max, errorcodeptr);
4034: if (*errorcodeptr != 0) goto FAILED;
4035: goto REPEAT;
4036:
1.3 misha 4037: case CHAR_ASTERISK:
1.1 misha 4038: repeat_min = 0;
4039: repeat_max = -1;
4040: goto REPEAT;
4041:
1.3 misha 4042: case CHAR_PLUS:
1.1 misha 4043: repeat_min = 1;
4044: repeat_max = -1;
4045: goto REPEAT;
4046:
1.3 misha 4047: case CHAR_QUESTION_MARK:
1.1 misha 4048: repeat_min = 0;
4049: repeat_max = 1;
4050:
4051: REPEAT:
4052: if (previous == NULL)
4053: {
4054: *errorcodeptr = ERR9;
4055: goto FAILED;
4056: }
4057:
4058: if (repeat_min == 0)
4059: {
4060: firstbyte = zerofirstbyte; /* Adjust for zero repeat */
4061: reqbyte = zeroreqbyte; /* Ditto */
4062: }
4063:
4064: /* Remember whether this is a variable length repeat */
4065:
4066: reqvary = (repeat_min == repeat_max)? 0 : REQ_VARY;
4067:
4068: op_type = 0; /* Default single-char op codes */
4069: possessive_quantifier = FALSE; /* Default not possessive quantifier */
4070:
4071: /* Save start of previous item, in case we have to move it up to make space
4072: for an inserted OP_ONCE for the additional '+' extension. */
4073:
4074: tempcode = previous;
4075:
4076: /* If the next character is '+', we have a possessive quantifier. This
4077: implies greediness, whatever the setting of the PCRE_UNGREEDY option.
4078: If the next character is '?' this is a minimizing repeat, by default,
4079: but if PCRE_UNGREEDY is set, it works the other way round. We change the
4080: repeat type to the non-default. */
4081:
1.3 misha 4082: if (ptr[1] == CHAR_PLUS)
1.1 misha 4083: {
4084: repeat_type = 0; /* Force greedy */
4085: possessive_quantifier = TRUE;
4086: ptr++;
4087: }
1.3 misha 4088: else if (ptr[1] == CHAR_QUESTION_MARK)
1.1 misha 4089: {
4090: repeat_type = greedy_non_default;
4091: ptr++;
4092: }
4093: else repeat_type = greedy_default;
4094:
4095: /* If previous was a character match, abolish the item and generate a
4096: repeat item instead. If a char item has a minumum of more than one, ensure
4097: that it is set in reqbyte - it might not be if a sequence such as x{3} is
4098: the first thing in a branch because the x will have gone into firstbyte
4099: instead. */
4100:
4101: if (*previous == OP_CHAR || *previous == OP_CHARNC)
4102: {
4103: /* Deal with UTF-8 characters that take up more than one byte. It's
4104: easier to write this out separately than try to macrify it. Use c to
4105: hold the length of the character in bytes, plus 0x80 to flag that it's a
4106: length rather than a small character. */
4107:
4108: #ifdef SUPPORT_UTF8
4109: if (utf8 && (code[-1] & 0x80) != 0)
4110: {
4111: uschar *lastchar = code - 1;
4112: while((*lastchar & 0xc0) == 0x80) lastchar--;
4113: c = code - lastchar; /* Length of UTF-8 character */
4114: memcpy(utf8_char, lastchar, c); /* Save the char */
4115: c |= 0x80; /* Flag c as a length */
4116: }
4117: else
4118: #endif
4119:
4120: /* Handle the case of a single byte - either with no UTF8 support, or
4121: with UTF-8 disabled, or for a UTF-8 character < 128. */
4122:
4123: {
4124: c = code[-1];
4125: if (repeat_min > 1) reqbyte = c | req_caseopt | cd->req_varyopt;
4126: }
4127:
4128: /* If the repetition is unlimited, it pays to see if the next thing on
4129: the line is something that cannot possibly match this character. If so,
4130: automatically possessifying this item gains some performance in the case
4131: where the match fails. */
4132:
4133: if (!possessive_quantifier &&
4134: repeat_max < 0 &&
1.4 ! misha 4135: check_auto_possessive(previous, utf8, ptr + 1, options, cd))
1.1 misha 4136: {
4137: repeat_type = 0; /* Force greedy */
4138: possessive_quantifier = TRUE;
4139: }
4140:
4141: goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */
4142: }
4143:
4144: /* If previous was a single negated character ([^a] or similar), we use
4145: one of the special opcodes, replacing it. The code is shared with single-
4146: character repeats by setting opt_type to add a suitable offset into
4147: repeat_type. We can also test for auto-possessification. OP_NOT is
4148: currently used only for single-byte chars. */
4149:
4150: else if (*previous == OP_NOT)
4151: {
4152: op_type = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */
4153: c = previous[1];
4154: if (!possessive_quantifier &&
4155: repeat_max < 0 &&
1.4 ! misha 4156: check_auto_possessive(previous, utf8, ptr + 1, options, cd))
1.1 misha 4157: {
4158: repeat_type = 0; /* Force greedy */
4159: possessive_quantifier = TRUE;
4160: }
4161: goto OUTPUT_SINGLE_REPEAT;
4162: }
4163:
4164: /* If previous was a character type match (\d or similar), abolish it and
4165: create a suitable repeat item. The code is shared with single-character
4166: repeats by setting op_type to add a suitable offset into repeat_type. Note
4167: the the Unicode property types will be present only when SUPPORT_UCP is
4168: defined, but we don't wrap the little bits of code here because it just
4169: makes it horribly messy. */
4170:
4171: else if (*previous < OP_EODN)
4172: {
4173: uschar *oldcode;
4174: int prop_type, prop_value;
4175: op_type = OP_TYPESTAR - OP_STAR; /* Use type opcodes */
4176: c = *previous;
4177:
4178: if (!possessive_quantifier &&
4179: repeat_max < 0 &&
1.4 ! misha 4180: check_auto_possessive(previous, utf8, ptr + 1, options, cd))
1.1 misha 4181: {
4182: repeat_type = 0; /* Force greedy */
4183: possessive_quantifier = TRUE;
4184: }
4185:
4186: OUTPUT_SINGLE_REPEAT:
4187: if (*previous == OP_PROP || *previous == OP_NOTPROP)
4188: {
4189: prop_type = previous[1];
4190: prop_value = previous[2];
4191: }
4192: else prop_type = prop_value = -1;
4193:
4194: oldcode = code;
4195: code = previous; /* Usually overwrite previous item */
4196:
4197: /* If the maximum is zero then the minimum must also be zero; Perl allows
4198: this case, so we do too - by simply omitting the item altogether. */
4199:
4200: if (repeat_max == 0) goto END_REPEAT;
4201:
1.4 ! misha 4202: /*--------------------------------------------------------------------*/
! 4203: /* This code is obsolete from release 8.00; the restriction was finally
! 4204: removed: */
! 4205:
1.1 misha 4206: /* All real repeats make it impossible to handle partial matching (maybe
4207: one day we will be able to remove this restriction). */
4208:
1.4 ! misha 4209: /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
! 4210: /*--------------------------------------------------------------------*/
1.1 misha 4211:
4212: /* Combine the op_type with the repeat_type */
4213:
4214: repeat_type += op_type;
4215:
4216: /* A minimum of zero is handled either as the special case * or ?, or as
4217: an UPTO, with the maximum given. */
4218:
4219: if (repeat_min == 0)
4220: {
4221: if (repeat_max == -1) *code++ = OP_STAR + repeat_type;
4222: else if (repeat_max == 1) *code++ = OP_QUERY + repeat_type;
4223: else
4224: {
4225: *code++ = OP_UPTO + repeat_type;
4226: PUT2INC(code, 0, repeat_max);
4227: }
4228: }
4229:
4230: /* A repeat minimum of 1 is optimized into some special cases. If the
4231: maximum is unlimited, we use OP_PLUS. Otherwise, the original item is
4232: left in place and, if the maximum is greater than 1, we use OP_UPTO with
4233: one less than the maximum. */
4234:
4235: else if (repeat_min == 1)
4236: {
4237: if (repeat_max == -1)
4238: *code++ = OP_PLUS + repeat_type;
4239: else
4240: {
4241: code = oldcode; /* leave previous item in place */
4242: if (repeat_max == 1) goto END_REPEAT;
4243: *code++ = OP_UPTO + repeat_type;
4244: PUT2INC(code, 0, repeat_max - 1);
4245: }
4246: }
4247:
4248: /* The case {n,n} is just an EXACT, while the general case {n,m} is
4249: handled as an EXACT followed by an UPTO. */
4250:
4251: else
4252: {
4253: *code++ = OP_EXACT + op_type; /* NB EXACT doesn't have repeat_type */
4254: PUT2INC(code, 0, repeat_min);
4255:
4256: /* If the maximum is unlimited, insert an OP_STAR. Before doing so,
4257: we have to insert the character for the previous code. For a repeated
4258: Unicode property match, there are two extra bytes that define the
4259: required property. In UTF-8 mode, long characters have their length in
4260: c, with the 0x80 bit as a flag. */
4261:
4262: if (repeat_max < 0)
4263: {
4264: #ifdef SUPPORT_UTF8
4265: if (utf8 && c >= 128)
4266: {
4267: memcpy(code, utf8_char, c & 7);
4268: code += c & 7;
4269: }
4270: else
4271: #endif
4272: {
4273: *code++ = c;
4274: if (prop_type >= 0)
4275: {
4276: *code++ = prop_type;
4277: *code++ = prop_value;
4278: }
4279: }
4280: *code++ = OP_STAR + repeat_type;
4281: }
4282:
4283: /* Else insert an UPTO if the max is greater than the min, again
4284: preceded by the character, for the previously inserted code. If the
4285: UPTO is just for 1 instance, we can use QUERY instead. */
4286:
4287: else if (repeat_max != repeat_min)
4288: {
4289: #ifdef SUPPORT_UTF8
4290: if (utf8 && c >= 128)
4291: {
4292: memcpy(code, utf8_char, c & 7);
4293: code += c & 7;
4294: }
4295: else
4296: #endif
4297: *code++ = c;
4298: if (prop_type >= 0)
4299: {
4300: *code++ = prop_type;
4301: *code++ = prop_value;
4302: }
4303: repeat_max -= repeat_min;
4304:
4305: if (repeat_max == 1)
4306: {
4307: *code++ = OP_QUERY + repeat_type;
4308: }
4309: else
4310: {
4311: *code++ = OP_UPTO + repeat_type;
4312: PUT2INC(code, 0, repeat_max);
4313: }
4314: }
4315: }
4316:
4317: /* The character or character type itself comes last in all cases. */
4318:
4319: #ifdef SUPPORT_UTF8
4320: if (utf8 && c >= 128)
4321: {
4322: memcpy(code, utf8_char, c & 7);
4323: code += c & 7;
4324: }
4325: else
4326: #endif
4327: *code++ = c;
4328:
4329: /* For a repeated Unicode property match, there are two extra bytes that
4330: define the required property. */
4331:
4332: #ifdef SUPPORT_UCP
4333: if (prop_type >= 0)
4334: {
4335: *code++ = prop_type;
4336: *code++ = prop_value;
4337: }
4338: #endif
4339: }
4340:
4341: /* If previous was a character class or a back reference, we put the repeat
4342: stuff after it, but just skip the item if the repeat was {0,0}. */
4343:
4344: else if (*previous == OP_CLASS ||
4345: *previous == OP_NCLASS ||
4346: #ifdef SUPPORT_UTF8
4347: *previous == OP_XCLASS ||
4348: #endif
4349: *previous == OP_REF)
4350: {
4351: if (repeat_max == 0)
4352: {
4353: code = previous;
4354: goto END_REPEAT;
4355: }
4356:
1.4 ! misha 4357: /*--------------------------------------------------------------------*/
! 4358: /* This code is obsolete from release 8.00; the restriction was finally
! 4359: removed: */
! 4360:
1.1 misha 4361: /* All real repeats make it impossible to handle partial matching (maybe
4362: one day we will be able to remove this restriction). */
4363:
1.4 ! misha 4364: /* if (repeat_max != 1) cd->external_flags |= PCRE_NOPARTIAL; */
! 4365: /*--------------------------------------------------------------------*/
1.1 misha 4366:
4367: if (repeat_min == 0 && repeat_max == -1)
4368: *code++ = OP_CRSTAR + repeat_type;
4369: else if (repeat_min == 1 && repeat_max == -1)
4370: *code++ = OP_CRPLUS + repeat_type;
4371: else if (repeat_min == 0 && repeat_max == 1)
4372: *code++ = OP_CRQUERY + repeat_type;
4373: else
4374: {
4375: *code++ = OP_CRRANGE + repeat_type;
4376: PUT2INC(code, 0, repeat_min);
4377: if (repeat_max == -1) repeat_max = 0; /* 2-byte encoding for max */
4378: PUT2INC(code, 0, repeat_max);
4379: }
4380: }
4381:
4382: /* If previous was a bracket group, we may have to replicate it in certain
4383: cases. */
4384:
4385: else if (*previous == OP_BRA || *previous == OP_CBRA ||
4386: *previous == OP_ONCE || *previous == OP_COND)
4387: {
4388: register int i;
4389: int ketoffset = 0;
1.4 ! misha 4390: int len = (int)(code - previous);
1.1 misha 4391: uschar *bralink = NULL;
4392:
4393: /* Repeating a DEFINE group is pointless */
4394:
4395: if (*previous == OP_COND && previous[LINK_SIZE+1] == OP_DEF)
4396: {
4397: *errorcodeptr = ERR55;
4398: goto FAILED;
4399: }
4400:
4401: /* If the maximum repeat count is unlimited, find the end of the bracket
4402: by scanning through from the start, and compute the offset back to it
4403: from the current code pointer. There may be an OP_OPT setting following
4404: the final KET, so we can't find the end just by going back from the code
4405: pointer. */
4406:
4407: if (repeat_max == -1)
4408: {
4409: register uschar *ket = previous;
4410: do ket += GET(ket, 1); while (*ket != OP_KET);
1.4 ! misha 4411: ketoffset = (int)(code - ket);
1.1 misha 4412: }
4413:
4414: /* The case of a zero minimum is special because of the need to stick
4415: OP_BRAZERO in front of it, and because the group appears once in the
4416: data, whereas in other cases it appears the minimum number of times. For
4417: this reason, it is simplest to treat this case separately, as otherwise
4418: the code gets far too messy. There are several special subcases when the
4419: minimum is zero. */
4420:
4421: if (repeat_min == 0)
4422: {
4423: /* If the maximum is also zero, we used to just omit the group from the
4424: output altogether, like this:
4425:
4426: ** if (repeat_max == 0)
4427: ** {
4428: ** code = previous;
4429: ** goto END_REPEAT;
4430: ** }
4431:
4432: However, that fails when a group is referenced as a subroutine from
4433: elsewhere in the pattern, so now we stick in OP_SKIPZERO in front of it
4434: so that it is skipped on execution. As we don't have a list of which
4435: groups are referenced, we cannot do this selectively.
4436:
4437: If the maximum is 1 or unlimited, we just have to stick in the BRAZERO
4438: and do no more at this point. However, we do need to adjust any
4439: OP_RECURSE calls inside the group that refer to the group itself or any
4440: internal or forward referenced group, because the offset is from the
4441: start of the whole regex. Temporarily terminate the pattern while doing
4442: this. */
4443:
4444: if (repeat_max <= 1) /* Covers 0, 1, and unlimited */
4445: {
4446: *code = OP_END;
4447: adjust_recurse(previous, 1, utf8, cd, save_hwm);
4448: memmove(previous+1, previous, len);
4449: code++;
4450: if (repeat_max == 0)
4451: {
4452: *previous++ = OP_SKIPZERO;
4453: goto END_REPEAT;
4454: }
4455: *previous++ = OP_BRAZERO + repeat_type;
4456: }
4457:
4458: /* If the maximum is greater than 1 and limited, we have to replicate
4459: in a nested fashion, sticking OP_BRAZERO before each set of brackets.
4460: The first one has to be handled carefully because it's the original
4461: copy, which has to be moved up. The remainder can be handled by code
4462: that is common with the non-zero minimum case below. We have to
4463: adjust the value or repeat_max, since one less copy is required. Once
4464: again, we may have to adjust any OP_RECURSE calls inside the group. */
4465:
4466: else
4467: {
4468: int offset;
4469: *code = OP_END;
4470: adjust_recurse(previous, 2 + LINK_SIZE, utf8, cd, save_hwm);
4471: memmove(previous + 2 + LINK_SIZE, previous, len);
4472: code += 2 + LINK_SIZE;
4473: *previous++ = OP_BRAZERO + repeat_type;
4474: *previous++ = OP_BRA;
4475:
4476: /* We chain together the bracket offset fields that have to be
4477: filled in later when the ends of the brackets are reached. */
4478:
1.4 ! misha 4479: offset = (bralink == NULL)? 0 : (int)(previous - bralink);
1.1 misha 4480: bralink = previous;
4481: PUTINC(previous, 0, offset);
4482: }
4483:
4484: repeat_max--;
4485: }
4486:
4487: /* If the minimum is greater than zero, replicate the group as many
4488: times as necessary, and adjust the maximum to the number of subsequent
4489: copies that we need. If we set a first char from the group, and didn't
4490: set a required char, copy the latter from the former. If there are any
4491: forward reference subroutine calls in the group, there will be entries on
4492: the workspace list; replicate these with an appropriate increment. */
4493:
4494: else
4495: {
4496: if (repeat_min > 1)
4497: {
4498: /* In the pre-compile phase, we don't actually do the replication. We
4499: just adjust the length as if we had. Do some paranoid checks for
1.4 ! misha 4500: potential integer overflow. The INT64_OR_DOUBLE type is a 64-bit
! 4501: integer type when available, otherwise double. */
1.1 misha 4502:
4503: if (lengthptr != NULL)
4504: {
4505: int delta = (repeat_min - 1)*length_prevgroup;
1.4 ! misha 4506: if ((INT64_OR_DOUBLE)(repeat_min - 1)*
! 4507: (INT64_OR_DOUBLE)length_prevgroup >
! 4508: (INT64_OR_DOUBLE)INT_MAX ||
1.1 misha 4509: OFLOW_MAX - *lengthptr < delta)
4510: {
4511: *errorcodeptr = ERR20;
4512: goto FAILED;
4513: }
4514: *lengthptr += delta;
4515: }
4516:
4517: /* This is compiling for real */
4518:
4519: else
4520: {
4521: if (groupsetfirstbyte && reqbyte < 0) reqbyte = firstbyte;
4522: for (i = 1; i < repeat_min; i++)
4523: {
4524: uschar *hc;
4525: uschar *this_hwm = cd->hwm;
4526: memcpy(code, previous, len);
4527: for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4528: {
4529: PUT(cd->hwm, 0, GET(hc, 0) + len);
4530: cd->hwm += LINK_SIZE;
4531: }
4532: save_hwm = this_hwm;
4533: code += len;
4534: }
4535: }
4536: }
4537:
4538: if (repeat_max > 0) repeat_max -= repeat_min;
4539: }
4540:
4541: /* This code is common to both the zero and non-zero minimum cases. If
4542: the maximum is limited, it replicates the group in a nested fashion,
4543: remembering the bracket starts on a stack. In the case of a zero minimum,
4544: the first one was set up above. In all cases the repeat_max now specifies
4545: the number of additional copies needed. Again, we must remember to
4546: replicate entries on the forward reference list. */
4547:
4548: if (repeat_max >= 0)
4549: {
4550: /* In the pre-compile phase, we don't actually do the replication. We
4551: just adjust the length as if we had. For each repetition we must add 1
4552: to the length for BRAZERO and for all but the last repetition we must
4553: add 2 + 2*LINKSIZE to allow for the nesting that occurs. Do some
1.4 ! misha 4554: paranoid checks to avoid integer overflow. The INT64_OR_DOUBLE type is
! 4555: a 64-bit integer type when available, otherwise double. */
1.1 misha 4556:
4557: if (lengthptr != NULL && repeat_max > 0)
4558: {
4559: int delta = repeat_max * (length_prevgroup + 1 + 2 + 2*LINK_SIZE) -
4560: 2 - 2*LINK_SIZE; /* Last one doesn't nest */
1.4 ! misha 4561: if ((INT64_OR_DOUBLE)repeat_max *
! 4562: (INT64_OR_DOUBLE)(length_prevgroup + 1 + 2 + 2*LINK_SIZE)
! 4563: > (INT64_OR_DOUBLE)INT_MAX ||
1.1 misha 4564: OFLOW_MAX - *lengthptr < delta)
4565: {
4566: *errorcodeptr = ERR20;
4567: goto FAILED;
4568: }
4569: *lengthptr += delta;
4570: }
4571:
4572: /* This is compiling for real */
4573:
4574: else for (i = repeat_max - 1; i >= 0; i--)
4575: {
4576: uschar *hc;
4577: uschar *this_hwm = cd->hwm;
4578:
4579: *code++ = OP_BRAZERO + repeat_type;
4580:
4581: /* All but the final copy start a new nesting, maintaining the
4582: chain of brackets outstanding. */
4583:
4584: if (i != 0)
4585: {
4586: int offset;
4587: *code++ = OP_BRA;
1.4 ! misha 4588: offset = (bralink == NULL)? 0 : (int)(code - bralink);
1.1 misha 4589: bralink = code;
4590: PUTINC(code, 0, offset);
4591: }
4592:
4593: memcpy(code, previous, len);
4594: for (hc = save_hwm; hc < this_hwm; hc += LINK_SIZE)
4595: {
4596: PUT(cd->hwm, 0, GET(hc, 0) + len + ((i != 0)? 2+LINK_SIZE : 1));
4597: cd->hwm += LINK_SIZE;
4598: }
4599: save_hwm = this_hwm;
4600: code += len;
4601: }
4602:
4603: /* Now chain through the pending brackets, and fill in their length
4604: fields (which are holding the chain links pro tem). */
4605:
4606: while (bralink != NULL)
4607: {
4608: int oldlinkoffset;
1.4 ! misha 4609: int offset = (int)(code - bralink + 1);
1.1 misha 4610: uschar *bra = code - offset;
4611: oldlinkoffset = GET(bra, 1);
4612: bralink = (oldlinkoffset == 0)? NULL : bralink - oldlinkoffset;
4613: *code++ = OP_KET;
4614: PUTINC(code, 0, offset);
4615: PUT(bra, 1, offset);
4616: }
4617: }
4618:
4619: /* If the maximum is unlimited, set a repeater in the final copy. We
4620: can't just offset backwards from the current code point, because we
4621: don't know if there's been an options resetting after the ket. The
4622: correct offset was computed above.
4623:
4624: Then, when we are doing the actual compile phase, check to see whether
4625: this group is a non-atomic one that could match an empty string. If so,
4626: convert the initial operator to the S form (e.g. OP_BRA -> OP_SBRA) so
4627: that runtime checking can be done. [This check is also applied to
4628: atomic groups at runtime, but in a different way.] */
4629:
4630: else
4631: {
4632: uschar *ketcode = code - ketoffset;
4633: uschar *bracode = ketcode - GET(ketcode, 1);
4634: *ketcode = OP_KETRMAX + repeat_type;
4635: if (lengthptr == NULL && *bracode != OP_ONCE)
4636: {
4637: uschar *scode = bracode;
4638: do
4639: {
1.4 ! misha 4640: if (could_be_empty_branch(scode, ketcode, utf8, cd))
1.1 misha 4641: {
4642: *bracode += OP_SBRA - OP_BRA;
4643: break;
4644: }
4645: scode += GET(scode, 1);
4646: }
4647: while (*scode == OP_ALT);
4648: }
4649: }
4650: }
4651:
4652: /* If previous is OP_FAIL, it was generated by an empty class [] in
4653: JavaScript mode. The other ways in which OP_FAIL can be generated, that is
4654: by (*FAIL) or (?!) set previous to NULL, which gives a "nothing to repeat"
4655: error above. We can just ignore the repeat in JS case. */
4656:
4657: else if (*previous == OP_FAIL) goto END_REPEAT;
4658:
4659: /* Else there's some kind of shambles */
4660:
4661: else
4662: {
4663: *errorcodeptr = ERR11;
4664: goto FAILED;
4665: }
4666:
4667: /* If the character following a repeat is '+', or if certain optimization
4668: tests above succeeded, possessive_quantifier is TRUE. For some of the
4669: simpler opcodes, there is an special alternative opcode for this. For
4670: anything else, we wrap the entire repeated item inside OP_ONCE brackets.
4671: The '+' notation is just syntactic sugar, taken from Sun's Java package,
4672: but the special opcodes can optimize it a bit. The repeated item starts at
4673: tempcode, not at previous, which might be the first part of a string whose
4674: (former) last char we repeated.
4675:
4676: Possessifying an 'exact' quantifier has no effect, so we can ignore it. But
4677: an 'upto' may follow. We skip over an 'exact' item, and then test the
4678: length of what remains before proceeding. */
4679:
4680: if (possessive_quantifier)
4681: {
4682: int len;
1.4 ! misha 4683:
! 4684: if (*tempcode == OP_TYPEEXACT)
1.1 misha 4685: tempcode += _pcre_OP_lengths[*tempcode] +
1.4 ! misha 4686: ((tempcode[3] == OP_PROP || tempcode[3] == OP_NOTPROP)? 2 : 0);
! 4687:
! 4688: else if (*tempcode == OP_EXACT || *tempcode == OP_NOTEXACT)
! 4689: {
! 4690: tempcode += _pcre_OP_lengths[*tempcode];
! 4691: #ifdef SUPPORT_UTF8
! 4692: if (utf8 && tempcode[-1] >= 0xc0)
! 4693: tempcode += _pcre_utf8_table4[tempcode[-1] & 0x3f];
! 4694: #endif
! 4695: }
! 4696:
! 4697: len = (int)(code - tempcode);
1.1 misha 4698: if (len > 0) switch (*tempcode)
4699: {
4700: case OP_STAR: *tempcode = OP_POSSTAR; break;
4701: case OP_PLUS: *tempcode = OP_POSPLUS; break;
4702: case OP_QUERY: *tempcode = OP_POSQUERY; break;
4703: case OP_UPTO: *tempcode = OP_POSUPTO; break;
4704:
4705: case OP_TYPESTAR: *tempcode = OP_TYPEPOSSTAR; break;
4706: case OP_TYPEPLUS: *tempcode = OP_TYPEPOSPLUS; break;
4707: case OP_TYPEQUERY: *tempcode = OP_TYPEPOSQUERY; break;
4708: case OP_TYPEUPTO: *tempcode = OP_TYPEPOSUPTO; break;
4709:
4710: case OP_NOTSTAR: *tempcode = OP_NOTPOSSTAR; break;
4711: case OP_NOTPLUS: *tempcode = OP_NOTPOSPLUS; break;
4712: case OP_NOTQUERY: *tempcode = OP_NOTPOSQUERY; break;
4713: case OP_NOTUPTO: *tempcode = OP_NOTPOSUPTO; break;
4714:
1.4 ! misha 4715: /* Because we are moving code along, we must ensure that any
! 4716: pending recursive references are updated. */
! 4717:
1.1 misha 4718: default:
1.4 ! misha 4719: *code = OP_END;
! 4720: adjust_recurse(tempcode, 1 + LINK_SIZE, utf8, cd, save_hwm);
1.1 misha 4721: memmove(tempcode + 1+LINK_SIZE, tempcode, len);
4722: code += 1 + LINK_SIZE;
4723: len += 1 + LINK_SIZE;
4724: tempcode[0] = OP_ONCE;
4725: *code++ = OP_KET;
4726: PUTINC(code, 0, len);
4727: PUT(tempcode, 1, len);
4728: break;
4729: }
4730: }
4731:
4732: /* In all case we no longer have a previous item. We also set the
4733: "follows varying string" flag for subsequently encountered reqbytes if
4734: it isn't already set and we have just passed a varying length item. */
4735:
4736: END_REPEAT:
4737: previous = NULL;
4738: cd->req_varyopt |= reqvary;
4739: break;
4740:
4741:
4742: /* ===================================================================*/
4743: /* Start of nested parenthesized sub-expression, or comment or lookahead or
4744: lookbehind or option setting or condition or all the other extended
4745: parenthesis forms. */
4746:
1.3 misha 4747: case CHAR_LEFT_PARENTHESIS:
1.1 misha 4748: newoptions = options;
4749: skipbytes = 0;
4750: bravalue = OP_CBRA;
4751: save_hwm = cd->hwm;
4752: reset_bracount = FALSE;
4753:
4754: /* First deal with various "verbs" that can be introduced by '*'. */
4755:
1.4 ! misha 4756: if (*(++ptr) == CHAR_ASTERISK &&
! 4757: ((cd->ctypes[ptr[1]] & ctype_letter) != 0 || ptr[1] == ':'))
1.1 misha 4758: {
4759: int i, namelen;
1.4 ! misha 4760: int arglen = 0;
1.1 misha 4761: const char *vn = verbnames;
1.4 ! misha 4762: const uschar *name = ptr + 1;
! 4763: const uschar *arg = NULL;
1.1 misha 4764: previous = NULL;
1.2 misha 4765: while ((cd->ctypes[*++ptr] & ctype_letter) != 0) {};
1.4 ! misha 4766: namelen = (int)(ptr - name);
! 4767:
1.3 misha 4768: if (*ptr == CHAR_COLON)
1.1 misha 4769: {
1.4 ! misha 4770: arg = ++ptr;
! 4771: while ((cd->ctypes[*ptr] & (ctype_letter|ctype_digit)) != 0
! 4772: || *ptr == '_') ptr++;
! 4773: arglen = (int)(ptr - arg);
1.1 misha 4774: }
1.4 ! misha 4775:
1.3 misha 4776: if (*ptr != CHAR_RIGHT_PARENTHESIS)
1.1 misha 4777: {
4778: *errorcodeptr = ERR60;
4779: goto FAILED;
4780: }
1.4 ! misha 4781:
! 4782: /* Scan the table of verb names */
! 4783:
1.1 misha 4784: for (i = 0; i < verbcount; i++)
4785: {
4786: if (namelen == verbs[i].len &&
4787: strncmp((char *)name, vn, namelen) == 0)
4788: {
1.4 ! misha 4789: /* Check for open captures before ACCEPT */
! 4790:
! 4791: if (verbs[i].op == OP_ACCEPT)
! 4792: {
! 4793: open_capitem *oc;
! 4794: cd->had_accept = TRUE;
! 4795: for (oc = cd->open_caps; oc != NULL; oc = oc->next)
! 4796: {
! 4797: *code++ = OP_CLOSE;
! 4798: PUT2INC(code, 0, oc->number);
! 4799: }
! 4800: }
! 4801:
! 4802: /* Handle the cases with/without an argument */
! 4803:
! 4804: if (arglen == 0)
! 4805: {
! 4806: if (verbs[i].op < 0) /* Argument is mandatory */
! 4807: {
! 4808: *errorcodeptr = ERR66;
! 4809: goto FAILED;
! 4810: }
! 4811: *code++ = verbs[i].op;
! 4812: }
! 4813:
! 4814: else
! 4815: {
! 4816: if (verbs[i].op_arg < 0) /* Argument is forbidden */
! 4817: {
! 4818: *errorcodeptr = ERR59;
! 4819: goto FAILED;
! 4820: }
! 4821: *code++ = verbs[i].op_arg;
! 4822: *code++ = arglen;
! 4823: memcpy(code, arg, arglen);
! 4824: code += arglen;
! 4825: *code++ = 0;
! 4826: }
! 4827:
! 4828: break; /* Found verb, exit loop */
1.1 misha 4829: }
1.4 ! misha 4830:
1.1 misha 4831: vn += verbs[i].len + 1;
4832: }
1.4 ! misha 4833:
! 4834: if (i < verbcount) continue; /* Successfully handled a verb */
! 4835: *errorcodeptr = ERR60; /* Verb not recognized */
1.1 misha 4836: goto FAILED;
4837: }
4838:
4839: /* Deal with the extended parentheses; all are introduced by '?', and the
4840: appearance of any of them means that this is not a capturing group. */
4841:
1.3 misha 4842: else if (*ptr == CHAR_QUESTION_MARK)
1.1 misha 4843: {
4844: int i, set, unset, namelen;
4845: int *optset;
4846: const uschar *name;
4847: uschar *slot;
4848:
4849: switch (*(++ptr))
4850: {
1.3 misha 4851: case CHAR_NUMBER_SIGN: /* Comment; skip to ket */
1.1 misha 4852: ptr++;
1.3 misha 4853: while (*ptr != 0 && *ptr != CHAR_RIGHT_PARENTHESIS) ptr++;
1.1 misha 4854: if (*ptr == 0)
4855: {
4856: *errorcodeptr = ERR18;
4857: goto FAILED;
4858: }
4859: continue;
4860:
4861:
4862: /* ------------------------------------------------------------ */
1.3 misha 4863: case CHAR_VERTICAL_LINE: /* Reset capture count for each branch */
1.1 misha 4864: reset_bracount = TRUE;
4865: /* Fall through */
4866:
4867: /* ------------------------------------------------------------ */
1.3 misha 4868: case CHAR_COLON: /* Non-capturing bracket */
1.1 misha 4869: bravalue = OP_BRA;
4870: ptr++;
4871: break;
4872:
4873:
4874: /* ------------------------------------------------------------ */
1.3 misha 4875: case CHAR_LEFT_PARENTHESIS:
1.1 misha 4876: bravalue = OP_COND; /* Conditional group */
4877:
4878: /* A condition can be an assertion, a number (referring to a numbered
4879: group), a name (referring to a named group), or 'R', referring to
4880: recursion. R<digits> and R&name are also permitted for recursion tests.
4881:
4882: There are several syntaxes for testing a named group: (?(name)) is used
4883: by Python; Perl 5.10 onwards uses (?(<name>) or (?('name')).
4884:
4885: There are two unfortunate ambiguities, caused by history. (a) 'R' can
4886: be the recursive thing or the name 'R' (and similarly for 'R' followed
4887: by digits), and (b) a number could be a name that consists of digits.
4888: In both cases, we look for a name first; if not found, we try the other
4889: cases. */
4890:
4891: /* For conditions that are assertions, check the syntax, and then exit
4892: the switch. This will take control down to where bracketed groups,
4893: including assertions, are processed. */
4894:
1.3 misha 4895: if (ptr[1] == CHAR_QUESTION_MARK && (ptr[2] == CHAR_EQUALS_SIGN ||
4896: ptr[2] == CHAR_EXCLAMATION_MARK || ptr[2] == CHAR_LESS_THAN_SIGN))
1.1 misha 4897: break;
4898:
4899: /* Most other conditions use OP_CREF (a couple change to OP_RREF
4900: below), and all need to skip 3 bytes at the start of the group. */
4901:
4902: code[1+LINK_SIZE] = OP_CREF;
4903: skipbytes = 3;
4904: refsign = -1;
4905:
4906: /* Check for a test for recursion in a named group. */
4907:
1.3 misha 4908: if (ptr[1] == CHAR_R && ptr[2] == CHAR_AMPERSAND)
1.1 misha 4909: {
4910: terminator = -1;
4911: ptr += 2;
4912: code[1+LINK_SIZE] = OP_RREF; /* Change the type of test */
4913: }
4914:
4915: /* Check for a test for a named group's having been set, using the Perl
4916: syntax (?(<name>) or (?('name') */
4917:
1.3 misha 4918: else if (ptr[1] == CHAR_LESS_THAN_SIGN)
1.1 misha 4919: {
1.3 misha 4920: terminator = CHAR_GREATER_THAN_SIGN;
1.1 misha 4921: ptr++;
4922: }
1.3 misha 4923: else if (ptr[1] == CHAR_APOSTROPHE)
1.1 misha 4924: {
1.3 misha 4925: terminator = CHAR_APOSTROPHE;
1.1 misha 4926: ptr++;
4927: }
4928: else
4929: {
4930: terminator = 0;
1.3 misha 4931: if (ptr[1] == CHAR_MINUS || ptr[1] == CHAR_PLUS) refsign = *(++ptr);
1.1 misha 4932: }
4933:
4934: /* We now expect to read a name; any thing else is an error */
4935:
4936: if ((cd->ctypes[ptr[1]] & ctype_word) == 0)
4937: {
4938: ptr += 1; /* To get the right offset */
4939: *errorcodeptr = ERR28;
4940: goto FAILED;
4941: }
4942:
4943: /* Read the name, but also get it as a number if it's all digits */
4944:
4945: recno = 0;
4946: name = ++ptr;
4947: while ((cd->ctypes[*ptr] & ctype_word) != 0)
4948: {
4949: if (recno >= 0)
4950: recno = ((digitab[*ptr] & ctype_digit) != 0)?
1.3 misha 4951: recno * 10 + *ptr - CHAR_0 : -1;
1.1 misha 4952: ptr++;
4953: }
1.4 ! misha 4954: namelen = (int)(ptr - name);
1.1 misha 4955:
1.3 misha 4956: if ((terminator > 0 && *ptr++ != terminator) ||
4957: *ptr++ != CHAR_RIGHT_PARENTHESIS)
1.1 misha 4958: {
4959: ptr--; /* Error offset */
4960: *errorcodeptr = ERR26;
4961: goto FAILED;
4962: }
4963:
4964: /* Do no further checking in the pre-compile phase. */
4965:
4966: if (lengthptr != NULL) break;
4967:
4968: /* In the real compile we do the work of looking for the actual
4969: reference. If the string started with "+" or "-" we require the rest to
4970: be digits, in which case recno will be set. */
4971:
4972: if (refsign > 0)
4973: {
4974: if (recno <= 0)
4975: {
4976: *errorcodeptr = ERR58;
4977: goto FAILED;
4978: }
1.3 misha 4979: recno = (refsign == CHAR_MINUS)?
1.1 misha 4980: cd->bracount - recno + 1 : recno +cd->bracount;
4981: if (recno <= 0 || recno > cd->final_bracount)
4982: {
4983: *errorcodeptr = ERR15;
4984: goto FAILED;
4985: }
4986: PUT2(code, 2+LINK_SIZE, recno);
4987: break;
4988: }
4989:
4990: /* Otherwise (did not start with "+" or "-"), start by looking for the
1.4 ! misha 4991: name. If we find a name, add one to the opcode to change OP_CREF or
! 4992: OP_RREF into OP_NCREF or OP_NRREF. These behave exactly the same,
! 4993: except they record that the reference was originally to a name. The
! 4994: information is used to check duplicate names. */
1.1 misha 4995:
4996: slot = cd->name_table;
4997: for (i = 0; i < cd->names_found; i++)
4998: {
4999: if (strncmp((char *)name, (char *)slot+2, namelen) == 0) break;
5000: slot += cd->name_entry_size;
5001: }
5002:
5003: /* Found a previous named subpattern */
5004:
5005: if (i < cd->names_found)
5006: {
5007: recno = GET2(slot, 0);
5008: PUT2(code, 2+LINK_SIZE, recno);
1.4 ! misha 5009: code[1+LINK_SIZE]++;
1.1 misha 5010: }
5011:
5012: /* Search the pattern for a forward reference */
5013:
1.3 misha 5014: else if ((i = find_parens(cd, name, namelen,
1.1 misha 5015: (options & PCRE_EXTENDED) != 0)) > 0)
5016: {
5017: PUT2(code, 2+LINK_SIZE, i);
1.4 ! misha 5018: code[1+LINK_SIZE]++;
1.1 misha 5019: }
5020:
5021: /* If terminator == 0 it means that the name followed directly after
5022: the opening parenthesis [e.g. (?(abc)...] and in this case there are
5023: some further alternatives to try. For the cases where terminator != 0
5024: [things like (?(<name>... or (?('name')... or (?(R&name)... ] we have
5025: now checked all the possibilities, so give an error. */
5026:
5027: else if (terminator != 0)
5028: {
5029: *errorcodeptr = ERR15;
5030: goto FAILED;
5031: }
5032:
5033: /* Check for (?(R) for recursion. Allow digits after R to specify a
5034: specific group number. */
5035:
1.3 misha 5036: else if (*name == CHAR_R)
1.1 misha 5037: {
5038: recno = 0;
5039: for (i = 1; i < namelen; i++)
5040: {
5041: if ((digitab[name[i]] & ctype_digit) == 0)
5042: {
5043: *errorcodeptr = ERR15;
5044: goto FAILED;
5045: }
1.3 misha 5046: recno = recno * 10 + name[i] - CHAR_0;
1.1 misha 5047: }
5048: if (recno == 0) recno = RREF_ANY;
5049: code[1+LINK_SIZE] = OP_RREF; /* Change test type */
5050: PUT2(code, 2+LINK_SIZE, recno);
5051: }
5052:
5053: /* Similarly, check for the (?(DEFINE) "condition", which is always
5054: false. */
5055:
1.3 misha 5056: else if (namelen == 6 && strncmp((char *)name, STRING_DEFINE, 6) == 0)
1.1 misha 5057: {
5058: code[1+LINK_SIZE] = OP_DEF;
5059: skipbytes = 1;
5060: }
5061:
5062: /* Check for the "name" actually being a subpattern number. We are
5063: in the second pass here, so final_bracount is set. */
5064:
5065: else if (recno > 0 && recno <= cd->final_bracount)
5066: {
5067: PUT2(code, 2+LINK_SIZE, recno);
5068: }
5069:
5070: /* Either an unidentified subpattern, or a reference to (?(0) */
5071:
5072: else
5073: {
5074: *errorcodeptr = (recno == 0)? ERR35: ERR15;
5075: goto FAILED;
5076: }
5077: break;
5078:
5079:
5080: /* ------------------------------------------------------------ */
1.3 misha 5081: case CHAR_EQUALS_SIGN: /* Positive lookahead */
1.1 misha 5082: bravalue = OP_ASSERT;
5083: ptr++;
5084: break;
5085:
5086:
5087: /* ------------------------------------------------------------ */
1.3 misha 5088: case CHAR_EXCLAMATION_MARK: /* Negative lookahead */
1.1 misha 5089: ptr++;
1.3 misha 5090: if (*ptr == CHAR_RIGHT_PARENTHESIS) /* Optimize (?!) */
1.1 misha 5091: {
5092: *code++ = OP_FAIL;
5093: previous = NULL;
5094: continue;
5095: }
5096: bravalue = OP_ASSERT_NOT;
5097: break;
5098:
5099:
5100: /* ------------------------------------------------------------ */
1.3 misha 5101: case CHAR_LESS_THAN_SIGN: /* Lookbehind or named define */
1.1 misha 5102: switch (ptr[1])
5103: {
1.3 misha 5104: case CHAR_EQUALS_SIGN: /* Positive lookbehind */
1.1 misha 5105: bravalue = OP_ASSERTBACK;
5106: ptr += 2;
5107: break;
5108:
1.3 misha 5109: case CHAR_EXCLAMATION_MARK: /* Negative lookbehind */
1.1 misha 5110: bravalue = OP_ASSERTBACK_NOT;
5111: ptr += 2;
5112: break;
5113:
5114: default: /* Could be name define, else bad */
5115: if ((cd->ctypes[ptr[1]] & ctype_word) != 0) goto DEFINE_NAME;
5116: ptr++; /* Correct offset for error */
5117: *errorcodeptr = ERR24;
5118: goto FAILED;
5119: }
5120: break;
5121:
5122:
5123: /* ------------------------------------------------------------ */
1.3 misha 5124: case CHAR_GREATER_THAN_SIGN: /* One-time brackets */
1.1 misha 5125: bravalue = OP_ONCE;
5126: ptr++;
5127: break;
5128:
5129:
5130: /* ------------------------------------------------------------ */
1.3 misha 5131: case CHAR_C: /* Callout - may be followed by digits; */
1.1 misha 5132: previous_callout = code; /* Save for later completion */
5133: after_manual_callout = 1; /* Skip one item before completing */
5134: *code++ = OP_CALLOUT;
5135: {
5136: int n = 0;
5137: while ((digitab[*(++ptr)] & ctype_digit) != 0)
1.3 misha 5138: n = n * 10 + *ptr - CHAR_0;
5139: if (*ptr != CHAR_RIGHT_PARENTHESIS)
1.1 misha 5140: {
5141: *errorcodeptr = ERR39;
5142: goto FAILED;
5143: }
5144: if (n > 255)
5145: {
5146: *errorcodeptr = ERR38;
5147: goto FAILED;
5148: }
5149: *code++ = n;
1.4 ! misha 5150: PUT(code, 0, (int)(ptr - cd->start_pattern + 1)); /* Pattern offset */
! 5151: PUT(code, LINK_SIZE, 0); /* Default length */
1.1 misha 5152: code += 2 * LINK_SIZE;
5153: }
5154: previous = NULL;
5155: continue;
5156:
5157:
5158: /* ------------------------------------------------------------ */
1.3 misha 5159: case CHAR_P: /* Python-style named subpattern handling */
5160: if (*(++ptr) == CHAR_EQUALS_SIGN ||
5161: *ptr == CHAR_GREATER_THAN_SIGN) /* Reference or recursion */
1.1 misha 5162: {
1.3 misha 5163: is_recurse = *ptr == CHAR_GREATER_THAN_SIGN;
5164: terminator = CHAR_RIGHT_PARENTHESIS;
1.1 misha 5165: goto NAMED_REF_OR_RECURSE;
5166: }
1.3 misha 5167: else if (*ptr != CHAR_LESS_THAN_SIGN) /* Test for Python-style defn */
1.1 misha 5168: {
5169: *errorcodeptr = ERR41;
5170: goto FAILED;
5171: }
5172: /* Fall through to handle (?P< as (?< is handled */
5173:
5174:
5175: /* ------------------------------------------------------------ */
5176: DEFINE_NAME: /* Come here from (?< handling */
1.3 misha 5177: case CHAR_APOSTROPHE:
1.1 misha 5178: {
1.3 misha 5179: terminator = (*ptr == CHAR_LESS_THAN_SIGN)?
5180: CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
1.1 misha 5181: name = ++ptr;
5182:
5183: while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
1.4 ! misha 5184: namelen = (int)(ptr - name);
1.1 misha 5185:
5186: /* In the pre-compile phase, just do a syntax check. */
5187:
5188: if (lengthptr != NULL)
5189: {
5190: if (*ptr != terminator)
5191: {
5192: *errorcodeptr = ERR42;
5193: goto FAILED;
5194: }
5195: if (cd->names_found >= MAX_NAME_COUNT)
5196: {
5197: *errorcodeptr = ERR49;
5198: goto FAILED;
5199: }
5200: if (namelen + 3 > cd->name_entry_size)
5201: {
5202: cd->name_entry_size = namelen + 3;
5203: if (namelen > MAX_NAME_SIZE)
5204: {
5205: *errorcodeptr = ERR48;
5206: goto FAILED;
5207: }
5208: }
5209: }
5210:
1.4 ! misha 5211: /* In the real compile, create the entry in the table, maintaining
! 5212: alphabetical order. Duplicate names for different numbers are
! 5213: permitted only if PCRE_DUPNAMES is set. Duplicate names for the same
! 5214: number are always OK. (An existing number can be re-used if (?|
! 5215: appears in the pattern.) In either event, a duplicate name results in
! 5216: a duplicate entry in the table, even if the number is the same. This
! 5217: is because the number of names, and hence the table size, is computed
! 5218: in the pre-compile, and it affects various numbers and pointers which
! 5219: would all have to be modified, and the compiled code moved down, if
! 5220: duplicates with the same number were omitted from the table. This
! 5221: doesn't seem worth the hassle. However, *different* names for the
! 5222: same number are not permitted. */
1.1 misha 5223:
5224: else
5225: {
1.4 ! misha 5226: BOOL dupname = FALSE;
1.1 misha 5227: slot = cd->name_table;
1.4 ! misha 5228:
1.1 misha 5229: for (i = 0; i < cd->names_found; i++)
5230: {
5231: int crc = memcmp(name, slot+2, namelen);
5232: if (crc == 0)
5233: {
5234: if (slot[2+namelen] == 0)
5235: {
1.4 ! misha 5236: if (GET2(slot, 0) != cd->bracount + 1 &&
! 5237: (options & PCRE_DUPNAMES) == 0)
1.1 misha 5238: {
5239: *errorcodeptr = ERR43;
5240: goto FAILED;
5241: }
1.4 ! misha 5242: else dupname = TRUE;
1.1 misha 5243: }
1.4 ! misha 5244: else crc = -1; /* Current name is a substring */
1.1 misha 5245: }
1.4 ! misha 5246:
! 5247: /* Make space in the table and break the loop for an earlier
! 5248: name. For a duplicate or later name, carry on. We do this for
! 5249: duplicates so that in the simple case (when ?(| is not used) they
! 5250: are in order of their numbers. */
! 5251:
1.1 misha 5252: if (crc < 0)
5253: {
5254: memmove(slot + cd->name_entry_size, slot,
5255: (cd->names_found - i) * cd->name_entry_size);
5256: break;
5257: }
1.4 ! misha 5258:
! 5259: /* Continue the loop for a later or duplicate name */
! 5260:
1.1 misha 5261: slot += cd->name_entry_size;
5262: }
5263:
1.4 ! misha 5264: /* For non-duplicate names, check for a duplicate number before
! 5265: adding the new name. */
! 5266:
! 5267: if (!dupname)
! 5268: {
! 5269: uschar *cslot = cd->name_table;
! 5270: for (i = 0; i < cd->names_found; i++)
! 5271: {
! 5272: if (cslot != slot)
! 5273: {
! 5274: if (GET2(cslot, 0) == cd->bracount + 1)
! 5275: {
! 5276: *errorcodeptr = ERR65;
! 5277: goto FAILED;
! 5278: }
! 5279: }
! 5280: else i--;
! 5281: cslot += cd->name_entry_size;
! 5282: }
! 5283: }
! 5284:
1.1 misha 5285: PUT2(slot, 0, cd->bracount + 1);
5286: memcpy(slot + 2, name, namelen);
5287: slot[2+namelen] = 0;
5288: }
5289: }
5290:
1.4 ! misha 5291: /* In both pre-compile and compile, count the number of names we've
! 5292: encountered. */
1.1 misha 5293:
1.4 ! misha 5294: cd->names_found++;
1.1 misha 5295: ptr++; /* Move past > or ' */
5296: goto NUMBERED_GROUP;
5297:
5298:
5299: /* ------------------------------------------------------------ */
1.3 misha 5300: case CHAR_AMPERSAND: /* Perl recursion/subroutine syntax */
5301: terminator = CHAR_RIGHT_PARENTHESIS;
1.1 misha 5302: is_recurse = TRUE;
5303: /* Fall through */
5304:
5305: /* We come here from the Python syntax above that handles both
5306: references (?P=name) and recursion (?P>name), as well as falling
5307: through from the Perl recursion syntax (?&name). We also come here from
5308: the Perl \k<name> or \k'name' back reference syntax and the \k{name}
5309: .NET syntax, and the Oniguruma \g<...> and \g'...' subroutine syntax. */
5310:
5311: NAMED_REF_OR_RECURSE:
5312: name = ++ptr;
5313: while ((cd->ctypes[*ptr] & ctype_word) != 0) ptr++;
1.4 ! misha 5314: namelen = (int)(ptr - name);
1.1 misha 5315:
5316: /* In the pre-compile phase, do a syntax check and set a dummy
5317: reference number. */
5318:
5319: if (lengthptr != NULL)
5320: {
5321: if (namelen == 0)
5322: {
5323: *errorcodeptr = ERR62;
5324: goto FAILED;
5325: }
5326: if (*ptr != terminator)
5327: {
5328: *errorcodeptr = ERR42;
5329: goto FAILED;
5330: }
5331: if (namelen > MAX_NAME_SIZE)
5332: {
5333: *errorcodeptr = ERR48;
5334: goto FAILED;
5335: }
5336: recno = 0;
5337: }
5338:
5339: /* In the real compile, seek the name in the table. We check the name
5340: first, and then check that we have reached the end of the name in the
5341: table. That way, if the name that is longer than any in the table,
5342: the comparison will fail without reading beyond the table entry. */
5343:
5344: else
5345: {
5346: slot = cd->name_table;
5347: for (i = 0; i < cd->names_found; i++)
5348: {
5349: if (strncmp((char *)name, (char *)slot+2, namelen) == 0 &&
5350: slot[2+namelen] == 0)
5351: break;
5352: slot += cd->name_entry_size;
5353: }
5354:
5355: if (i < cd->names_found) /* Back reference */
5356: {
5357: recno = GET2(slot, 0);
5358: }
5359: else if ((recno = /* Forward back reference */
1.3 misha 5360: find_parens(cd, name, namelen,
1.1 misha 5361: (options & PCRE_EXTENDED) != 0)) <= 0)
5362: {
5363: *errorcodeptr = ERR15;
5364: goto FAILED;
5365: }
5366: }
5367:
5368: /* In both phases, we can now go to the code than handles numerical
5369: recursion or backreferences. */
5370:
5371: if (is_recurse) goto HANDLE_RECURSION;
5372: else goto HANDLE_REFERENCE;
5373:
5374:
5375: /* ------------------------------------------------------------ */
1.3 misha 5376: case CHAR_R: /* Recursion */
1.1 misha 5377: ptr++; /* Same as (?0) */
5378: /* Fall through */
5379:
5380:
5381: /* ------------------------------------------------------------ */
1.3 misha 5382: case CHAR_MINUS: case CHAR_PLUS: /* Recursion or subroutine */
5383: case CHAR_0: case CHAR_1: case CHAR_2: case CHAR_3: case CHAR_4:
5384: case CHAR_5: case CHAR_6: case CHAR_7: case CHAR_8: case CHAR_9:
1.1 misha 5385: {
5386: const uschar *called;
1.3 misha 5387: terminator = CHAR_RIGHT_PARENTHESIS;
1.1 misha 5388:
5389: /* Come here from the \g<...> and \g'...' code (Oniguruma
5390: compatibility). However, the syntax has been checked to ensure that
5391: the ... are a (signed) number, so that neither ERR63 nor ERR29 will
5392: be called on this path, nor with the jump to OTHER_CHAR_AFTER_QUERY
5393: ever be taken. */
5394:
5395: HANDLE_NUMERICAL_RECURSION:
5396:
1.3 misha 5397: if ((refsign = *ptr) == CHAR_PLUS)
1.1 misha 5398: {
5399: ptr++;
5400: if ((digitab[*ptr] & ctype_digit) == 0)
5401: {
5402: *errorcodeptr = ERR63;
5403: goto FAILED;
5404: }
5405: }
1.3 misha 5406: else if (refsign == CHAR_MINUS)
1.1 misha 5407: {
5408: if ((digitab[ptr[1]] & ctype_digit) == 0)
5409: goto OTHER_CHAR_AFTER_QUERY;
5410: ptr++;
5411: }
5412:
5413: recno = 0;
5414: while((digitab[*ptr] & ctype_digit) != 0)
1.3 misha 5415: recno = recno * 10 + *ptr++ - CHAR_0;
1.1 misha 5416:
5417: if (*ptr != terminator)
5418: {
5419: *errorcodeptr = ERR29;
5420: goto FAILED;
5421: }
5422:
1.3 misha 5423: if (refsign == CHAR_MINUS)
1.1 misha 5424: {
5425: if (recno == 0)
5426: {
5427: *errorcodeptr = ERR58;
5428: goto FAILED;
5429: }
5430: recno = cd->bracount - recno + 1;
5431: if (recno <= 0)
5432: {
5433: *errorcodeptr = ERR15;
5434: goto FAILED;
5435: }
5436: }
1.3 misha 5437: else if (refsign == CHAR_PLUS)
1.1 misha 5438: {
5439: if (recno == 0)
5440: {
5441: *errorcodeptr = ERR58;
5442: goto FAILED;
5443: }
5444: recno += cd->bracount;
5445: }
5446:
5447: /* Come here from code above that handles a named recursion */
5448:
5449: HANDLE_RECURSION:
5450:
5451: previous = code;
5452: called = cd->start_code;
5453:
5454: /* When we are actually compiling, find the bracket that is being
5455: referenced. Temporarily end the regex in case it doesn't exist before
5456: this point. If we end up with a forward reference, first check that
5457: the bracket does occur later so we can give the error (and position)
5458: now. Then remember this forward reference in the workspace so it can
5459: be filled in at the end. */
5460:
5461: if (lengthptr == NULL)
5462: {
5463: *code = OP_END;
1.4 ! misha 5464: if (recno != 0)
! 5465: called = _pcre_find_bracket(cd->start_code, utf8, recno);
1.1 misha 5466:
5467: /* Forward reference */
5468:
5469: if (called == NULL)
5470: {
1.3 misha 5471: if (find_parens(cd, NULL, recno,
1.1 misha 5472: (options & PCRE_EXTENDED) != 0) < 0)
5473: {
5474: *errorcodeptr = ERR15;
5475: goto FAILED;
5476: }
1.4 ! misha 5477:
! 5478: /* Fudge the value of "called" so that when it is inserted as an
! 5479: offset below, what it actually inserted is the reference number
! 5480: of the group. */
! 5481:
1.1 misha 5482: called = cd->start_code + recno;
1.4 ! misha 5483: PUTINC(cd->hwm, 0, (int)(code + 2 + LINK_SIZE - cd->start_code));
1.1 misha 5484: }
5485:
5486: /* If not a forward reference, and the subpattern is still open,
5487: this is a recursive call. We check to see if this is a left
5488: recursion that could loop for ever, and diagnose that case. */
5489:
5490: else if (GET(called, 1) == 0 &&
1.4 ! misha 5491: could_be_empty(called, code, bcptr, utf8, cd))
1.1 misha 5492: {
5493: *errorcodeptr = ERR40;
5494: goto FAILED;
5495: }
5496: }
5497:
5498: /* Insert the recursion/subroutine item, automatically wrapped inside
5499: "once" brackets. Set up a "previous group" length so that a
5500: subsequent quantifier will work. */
5501:
5502: *code = OP_ONCE;
5503: PUT(code, 1, 2 + 2*LINK_SIZE);
5504: code += 1 + LINK_SIZE;
5505:
5506: *code = OP_RECURSE;
1.4 ! misha 5507: PUT(code, 1, (int)(called - cd->start_code));
1.1 misha 5508: code += 1 + LINK_SIZE;
5509:
5510: *code = OP_KET;
5511: PUT(code, 1, 2 + 2*LINK_SIZE);
5512: code += 1 + LINK_SIZE;
5513:
5514: length_prevgroup = 3 + 3*LINK_SIZE;
5515: }
5516:
5517: /* Can't determine a first byte now */
5518:
5519: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5520: continue;
5521:
5522:
5523: /* ------------------------------------------------------------ */
5524: default: /* Other characters: check option setting */
5525: OTHER_CHAR_AFTER_QUERY:
5526: set = unset = 0;
5527: optset = &set;
5528:
1.3 misha 5529: while (*ptr != CHAR_RIGHT_PARENTHESIS && *ptr != CHAR_COLON)
1.1 misha 5530: {
5531: switch (*ptr++)
5532: {
1.3 misha 5533: case CHAR_MINUS: optset = &unset; break;
1.1 misha 5534:
1.3 misha 5535: case CHAR_J: /* Record that it changed in the external options */
1.1 misha 5536: *optset |= PCRE_DUPNAMES;
5537: cd->external_flags |= PCRE_JCHANGED;
5538: break;
5539:
1.3 misha 5540: case CHAR_i: *optset |= PCRE_CASELESS; break;
5541: case CHAR_m: *optset |= PCRE_MULTILINE; break;
5542: case CHAR_s: *optset |= PCRE_DOTALL; break;
5543: case CHAR_x: *optset |= PCRE_EXTENDED; break;
5544: case CHAR_U: *optset |= PCRE_UNGREEDY; break;
5545: case CHAR_X: *optset |= PCRE_EXTRA; break;
1.1 misha 5546:
5547: default: *errorcodeptr = ERR12;
5548: ptr--; /* Correct the offset */
5549: goto FAILED;
5550: }
5551: }
5552:
5553: /* Set up the changed option bits, but don't change anything yet. */
5554:
5555: newoptions = (options | set) & (~unset);
5556:
5557: /* If the options ended with ')' this is not the start of a nested
5558: group with option changes, so the options change at this level. If this
5559: item is right at the start of the pattern, the options can be
5560: abstracted and made external in the pre-compile phase, and ignored in
5561: the compile phase. This can be helpful when matching -- for instance in
5562: caseless checking of required bytes.
5563:
5564: If the code pointer is not (cd->start_code + 1 + LINK_SIZE), we are
5565: definitely *not* at the start of the pattern because something has been
5566: compiled. In the pre-compile phase, however, the code pointer can have
5567: that value after the start, because it gets reset as code is discarded
5568: during the pre-compile. However, this can happen only at top level - if
5569: we are within parentheses, the starting BRA will still be present. At
5570: any parenthesis level, the length value can be used to test if anything
5571: has been compiled at that level. Thus, a test for both these conditions
5572: is necessary to ensure we correctly detect the start of the pattern in
5573: both phases.
5574:
5575: If we are not at the pattern start, compile code to change the ims
1.2 misha 5576: options if this setting actually changes any of them, and reset the
5577: greedy defaults and the case value for firstbyte and reqbyte. */
1.1 misha 5578:
1.3 misha 5579: if (*ptr == CHAR_RIGHT_PARENTHESIS)
1.1 misha 5580: {
5581: if (code == cd->start_code + 1 + LINK_SIZE &&
5582: (lengthptr == NULL || *lengthptr == 2 + 2*LINK_SIZE))
5583: {
5584: cd->external_options = newoptions;
5585: }
1.4 ! misha 5586: else
1.1 misha 5587: {
5588: if ((options & PCRE_IMS) != (newoptions & PCRE_IMS))
5589: {
5590: *code++ = OP_OPT;
5591: *code++ = newoptions & PCRE_IMS;
5592: }
5593: greedy_default = ((newoptions & PCRE_UNGREEDY) != 0);
5594: greedy_non_default = greedy_default ^ 1;
1.2 misha 5595: req_caseopt = ((newoptions & PCRE_CASELESS) != 0)? REQ_CASELESS : 0;
1.1 misha 5596: }
5597:
1.2 misha 5598: /* Change options at this level, and pass them back for use
5599: in subsequent branches. When not at the start of the pattern, this
5600: information is also necessary so that a resetting item can be
5601: compiled at the end of a group (if we are in a group). */
5602:
5603: *optionsptr = options = newoptions;
1.1 misha 5604: previous = NULL; /* This item can't be repeated */
5605: continue; /* It is complete */
5606: }
5607:
5608: /* If the options ended with ':' we are heading into a nested group
5609: with possible change of options. Such groups are non-capturing and are
5610: not assertions of any kind. All we need to do is skip over the ':';
5611: the newoptions value is handled below. */
5612:
5613: bravalue = OP_BRA;
5614: ptr++;
5615: } /* End of switch for character following (? */
5616: } /* End of (? handling */
5617:
1.4 ! misha 5618: /* Opening parenthesis not followed by '*' or '?'. If PCRE_NO_AUTO_CAPTURE
! 5619: is set, all unadorned brackets become non-capturing and behave like (?:...)
1.1 misha 5620: brackets. */
5621:
5622: else if ((options & PCRE_NO_AUTO_CAPTURE) != 0)
5623: {
5624: bravalue = OP_BRA;
5625: }
5626:
5627: /* Else we have a capturing group. */
5628:
5629: else
5630: {
5631: NUMBERED_GROUP:
5632: cd->bracount += 1;
5633: PUT2(code, 1+LINK_SIZE, cd->bracount);
5634: skipbytes = 2;
5635: }
5636:
5637: /* Process nested bracketed regex. Assertions may not be repeated, but
5638: other kinds can be. All their opcodes are >= OP_ONCE. We copy code into a
5639: non-register variable in order to be able to pass its address because some
5640: compilers complain otherwise. Pass in a new setting for the ims options if
5641: they have changed. */
5642:
5643: previous = (bravalue >= OP_ONCE)? code : NULL;
5644: *code = bravalue;
5645: tempcode = code;
5646: tempreqvary = cd->req_varyopt; /* Save value before bracket */
5647: length_prevgroup = 0; /* Initialize for pre-compile phase */
5648:
5649: if (!compile_regex(
5650: newoptions, /* The complete new option state */
5651: options & PCRE_IMS, /* The previous ims option state */
5652: &tempcode, /* Where to put code (updated) */
5653: &ptr, /* Input pointer (updated) */
5654: errorcodeptr, /* Where to put an error message */
5655: (bravalue == OP_ASSERTBACK ||
5656: bravalue == OP_ASSERTBACK_NOT), /* TRUE if back assert */
5657: reset_bracount, /* True if (?| group */
5658: skipbytes, /* Skip over bracket number */
5659: &subfirstbyte, /* For possible first char */
5660: &subreqbyte, /* For possible last char */
5661: bcptr, /* Current branch chain */
5662: cd, /* Tables block */
5663: (lengthptr == NULL)? NULL : /* Actual compile phase */
5664: &length_prevgroup /* Pre-compile phase */
5665: ))
5666: goto FAILED;
5667:
5668: /* At the end of compiling, code is still pointing to the start of the
5669: group, while tempcode has been updated to point past the end of the group
5670: and any option resetting that may follow it. The pattern pointer (ptr)
5671: is on the bracket. */
5672:
5673: /* If this is a conditional bracket, check that there are no more than
5674: two branches in the group, or just one if it's a DEFINE group. We do this
5675: in the real compile phase, not in the pre-pass, where the whole group may
5676: not be available. */
5677:
5678: if (bravalue == OP_COND && lengthptr == NULL)
5679: {
5680: uschar *tc = code;
5681: int condcount = 0;
5682:
5683: do {
5684: condcount++;
5685: tc += GET(tc,1);
5686: }
5687: while (*tc != OP_KET);
5688:
5689: /* A DEFINE group is never obeyed inline (the "condition" is always
5690: false). It must have only one branch. */
5691:
5692: if (code[LINK_SIZE+1] == OP_DEF)
5693: {
5694: if (condcount > 1)
5695: {
5696: *errorcodeptr = ERR54;
5697: goto FAILED;
5698: }
5699: bravalue = OP_DEF; /* Just a flag to suppress char handling below */
5700: }
5701:
5702: /* A "normal" conditional group. If there is just one branch, we must not
5703: make use of its firstbyte or reqbyte, because this is equivalent to an
5704: empty second branch. */
5705:
5706: else
5707: {
5708: if (condcount > 2)
5709: {
5710: *errorcodeptr = ERR27;
5711: goto FAILED;
5712: }
5713: if (condcount == 1) subfirstbyte = subreqbyte = REQ_NONE;
5714: }
5715: }
5716:
5717: /* Error if hit end of pattern */
5718:
1.3 misha 5719: if (*ptr != CHAR_RIGHT_PARENTHESIS)
1.1 misha 5720: {
5721: *errorcodeptr = ERR14;
5722: goto FAILED;
5723: }
5724:
5725: /* In the pre-compile phase, update the length by the length of the group,
5726: less the brackets at either end. Then reduce the compiled code to just a
5727: set of non-capturing brackets so that it doesn't use much memory if it is
5728: duplicated by a quantifier.*/
5729:
5730: if (lengthptr != NULL)
5731: {
5732: if (OFLOW_MAX - *lengthptr < length_prevgroup - 2 - 2*LINK_SIZE)
5733: {
5734: *errorcodeptr = ERR20;
5735: goto FAILED;
5736: }
5737: *lengthptr += length_prevgroup - 2 - 2*LINK_SIZE;
5738: *code++ = OP_BRA;
5739: PUTINC(code, 0, 1 + LINK_SIZE);
5740: *code++ = OP_KET;
5741: PUTINC(code, 0, 1 + LINK_SIZE);
5742: break; /* No need to waste time with special character handling */
5743: }
5744:
5745: /* Otherwise update the main code pointer to the end of the group. */
5746:
5747: code = tempcode;
5748:
5749: /* For a DEFINE group, required and first character settings are not
5750: relevant. */
5751:
5752: if (bravalue == OP_DEF) break;
5753:
5754: /* Handle updating of the required and first characters for other types of
5755: group. Update for normal brackets of all kinds, and conditions with two
5756: branches (see code above). If the bracket is followed by a quantifier with
5757: zero repeat, we have to back off. Hence the definition of zeroreqbyte and
5758: zerofirstbyte outside the main loop so that they can be accessed for the
5759: back off. */
5760:
5761: zeroreqbyte = reqbyte;
5762: zerofirstbyte = firstbyte;
5763: groupsetfirstbyte = FALSE;
5764:
5765: if (bravalue >= OP_ONCE)
5766: {
5767: /* If we have not yet set a firstbyte in this branch, take it from the
5768: subpattern, remembering that it was set here so that a repeat of more
5769: than one can replicate it as reqbyte if necessary. If the subpattern has
5770: no firstbyte, set "none" for the whole branch. In both cases, a zero
5771: repeat forces firstbyte to "none". */
5772:
5773: if (firstbyte == REQ_UNSET)
5774: {
5775: if (subfirstbyte >= 0)
5776: {
5777: firstbyte = subfirstbyte;
5778: groupsetfirstbyte = TRUE;
5779: }
5780: else firstbyte = REQ_NONE;
5781: zerofirstbyte = REQ_NONE;
5782: }
5783:
5784: /* If firstbyte was previously set, convert the subpattern's firstbyte
5785: into reqbyte if there wasn't one, using the vary flag that was in
5786: existence beforehand. */
5787:
5788: else if (subfirstbyte >= 0 && subreqbyte < 0)
5789: subreqbyte = subfirstbyte | tempreqvary;
5790:
5791: /* If the subpattern set a required byte (or set a first byte that isn't
5792: really the first byte - see above), set it. */
5793:
5794: if (subreqbyte >= 0) reqbyte = subreqbyte;
5795: }
5796:
5797: /* For a forward assertion, we take the reqbyte, if set. This can be
5798: helpful if the pattern that follows the assertion doesn't set a different
5799: char. For example, it's useful for /(?=abcde).+/. We can't set firstbyte
5800: for an assertion, however because it leads to incorrect effect for patterns
5801: such as /(?=a)a.+/ when the "real" "a" would then become a reqbyte instead
5802: of a firstbyte. This is overcome by a scan at the end if there's no
5803: firstbyte, looking for an asserted first char. */
5804:
5805: else if (bravalue == OP_ASSERT && subreqbyte >= 0) reqbyte = subreqbyte;
5806: break; /* End of processing '(' */
5807:
5808:
5809: /* ===================================================================*/
5810: /* Handle metasequences introduced by \. For ones like \d, the ESC_ values
1.4 ! misha 5811: are arranged to be the negation of the corresponding OP_values in the
! 5812: default case when PCRE_UCP is not set. For the back references, the values
! 5813: are ESC_REF plus the reference number. Only back references and those types
! 5814: that consume a character may be repeated. We can test for values between
! 5815: ESC_b and ESC_Z for the latter; this may have to change if any new ones are
! 5816: ever created. */
1.1 misha 5817:
1.3 misha 5818: case CHAR_BACKSLASH:
1.1 misha 5819: tempptr = ptr;
5820: c = check_escape(&ptr, errorcodeptr, cd->bracount, options, FALSE);
5821: if (*errorcodeptr != 0) goto FAILED;
5822:
5823: if (c < 0)
5824: {
5825: if (-c == ESC_Q) /* Handle start of quoted string */
5826: {
1.3 misha 5827: if (ptr[1] == CHAR_BACKSLASH && ptr[2] == CHAR_E)
5828: ptr += 2; /* avoid empty string */
5829: else inescq = TRUE;
1.1 misha 5830: continue;
5831: }
5832:
5833: if (-c == ESC_E) continue; /* Perl ignores an orphan \E */
5834:
5835: /* For metasequences that actually match a character, we disable the
5836: setting of a first character if it hasn't already been set. */
5837:
5838: if (firstbyte == REQ_UNSET && -c > ESC_b && -c < ESC_Z)
5839: firstbyte = REQ_NONE;
5840:
5841: /* Set values to reset to if this is followed by a zero repeat. */
5842:
5843: zerofirstbyte = firstbyte;
5844: zeroreqbyte = reqbyte;
5845:
5846: /* \g<name> or \g'name' is a subroutine call by name and \g<n> or \g'n'
5847: is a subroutine call by number (Oniguruma syntax). In fact, the value
5848: -ESC_g is returned only for these cases. So we don't need to check for <
5849: or ' if the value is -ESC_g. For the Perl syntax \g{n} the value is
5850: -ESC_REF+n, and for the Perl syntax \g{name} the result is -ESC_k (as
5851: that is a synonym for a named back reference). */
5852:
5853: if (-c == ESC_g)
5854: {
5855: const uschar *p;
5856: save_hwm = cd->hwm; /* Normally this is set when '(' is read */
1.3 misha 5857: terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5858: CHAR_GREATER_THAN_SIGN : CHAR_APOSTROPHE;
1.1 misha 5859:
5860: /* These two statements stop the compiler for warning about possibly
5861: unset variables caused by the jump to HANDLE_NUMERICAL_RECURSION. In
5862: fact, because we actually check for a number below, the paths that
5863: would actually be in error are never taken. */
5864:
5865: skipbytes = 0;
5866: reset_bracount = FALSE;
5867:
5868: /* Test for a name */
5869:
1.3 misha 5870: if (ptr[1] != CHAR_PLUS && ptr[1] != CHAR_MINUS)
1.1 misha 5871: {
5872: BOOL isnumber = TRUE;
5873: for (p = ptr + 1; *p != 0 && *p != terminator; p++)
5874: {
5875: if ((cd->ctypes[*p] & ctype_digit) == 0) isnumber = FALSE;
5876: if ((cd->ctypes[*p] & ctype_word) == 0) break;
5877: }
5878: if (*p != terminator)
5879: {
5880: *errorcodeptr = ERR57;
5881: break;
5882: }
5883: if (isnumber)
5884: {
5885: ptr++;
5886: goto HANDLE_NUMERICAL_RECURSION;
5887: }
5888: is_recurse = TRUE;
5889: goto NAMED_REF_OR_RECURSE;
5890: }
5891:
5892: /* Test a signed number in angle brackets or quotes. */
5893:
5894: p = ptr + 2;
5895: while ((digitab[*p] & ctype_digit) != 0) p++;
5896: if (*p != terminator)
5897: {
5898: *errorcodeptr = ERR57;
5899: break;
5900: }
5901: ptr++;
5902: goto HANDLE_NUMERICAL_RECURSION;
5903: }
5904:
5905: /* \k<name> or \k'name' is a back reference by name (Perl syntax).
5906: We also support \k{name} (.NET syntax) */
5907:
1.3 misha 5908: if (-c == ESC_k && (ptr[1] == CHAR_LESS_THAN_SIGN ||
5909: ptr[1] == CHAR_APOSTROPHE || ptr[1] == CHAR_LEFT_CURLY_BRACKET))
1.1 misha 5910: {
5911: is_recurse = FALSE;
1.3 misha 5912: terminator = (*(++ptr) == CHAR_LESS_THAN_SIGN)?
5913: CHAR_GREATER_THAN_SIGN : (*ptr == CHAR_APOSTROPHE)?
5914: CHAR_APOSTROPHE : CHAR_RIGHT_CURLY_BRACKET;
1.1 misha 5915: goto NAMED_REF_OR_RECURSE;
5916: }
5917:
5918: /* Back references are handled specially; must disable firstbyte if
5919: not set to cope with cases like (?=(\w+))\1: which would otherwise set
5920: ':' later. */
5921:
5922: if (-c >= ESC_REF)
5923: {
1.4 ! misha 5924: open_capitem *oc;
1.1 misha 5925: recno = -c - ESC_REF;
5926:
5927: HANDLE_REFERENCE: /* Come here from named backref handling */
5928: if (firstbyte == REQ_UNSET) firstbyte = REQ_NONE;
5929: previous = code;
5930: *code++ = OP_REF;
5931: PUT2INC(code, 0, recno);
5932: cd->backref_map |= (recno < 32)? (1 << recno) : 1;
5933: if (recno > cd->top_backref) cd->top_backref = recno;
1.4 ! misha 5934:
! 5935: /* Check to see if this back reference is recursive, that it, it
! 5936: is inside the group that it references. A flag is set so that the
! 5937: group can be made atomic. */
! 5938:
! 5939: for (oc = cd->open_caps; oc != NULL; oc = oc->next)
! 5940: {
! 5941: if (oc->number == recno)
! 5942: {
! 5943: oc->flag = TRUE;
! 5944: break;
! 5945: }
! 5946: }
1.1 misha 5947: }
5948:
5949: /* So are Unicode property matches, if supported. */
5950:
5951: #ifdef SUPPORT_UCP
5952: else if (-c == ESC_P || -c == ESC_p)
5953: {
5954: BOOL negated;
5955: int pdata;
5956: int ptype = get_ucp(&ptr, &negated, &pdata, errorcodeptr);
5957: if (ptype < 0) goto FAILED;
5958: previous = code;
5959: *code++ = ((-c == ESC_p) != negated)? OP_PROP : OP_NOTPROP;
5960: *code++ = ptype;
5961: *code++ = pdata;
5962: }
5963: #else
5964:
5965: /* If Unicode properties are not supported, \X, \P, and \p are not
5966: allowed. */
5967:
5968: else if (-c == ESC_X || -c == ESC_P || -c == ESC_p)
5969: {
5970: *errorcodeptr = ERR45;
5971: goto FAILED;
5972: }
5973: #endif
5974:
5975: /* For the rest (including \X when Unicode properties are supported), we
1.4 ! misha 5976: can obtain the OP value by negating the escape value in the default
! 5977: situation when PCRE_UCP is not set. When it *is* set, we substitute
! 5978: Unicode property tests. */
1.1 misha 5979:
5980: else
5981: {
1.4 ! misha 5982: #ifdef SUPPORT_UCP
! 5983: if (-c >= ESC_DU && -c <= ESC_wu)
! 5984: {
! 5985: nestptr = ptr + 1; /* Where to resume */
! 5986: ptr = substitutes[-c - ESC_DU] - 1; /* Just before substitute */
! 5987: }
! 5988: else
! 5989: #endif
! 5990: {
! 5991: previous = (-c > ESC_b && -c < ESC_Z)? code : NULL;
! 5992: *code++ = -c;
! 5993: }
1.1 misha 5994: }
5995: continue;
5996: }
5997:
5998: /* We have a data character whose value is in c. In UTF-8 mode it may have
5999: a value > 127. We set its representation in the length/buffer, and then
6000: handle it as a data character. */
6001:
6002: #ifdef SUPPORT_UTF8
6003: if (utf8 && c > 127)
6004: mclength = _pcre_ord2utf8(c, mcbuffer);
6005: else
6006: #endif
6007:
6008: {
6009: mcbuffer[0] = c;
6010: mclength = 1;
6011: }
6012: goto ONE_CHAR;
6013:
6014:
6015: /* ===================================================================*/
6016: /* Handle a literal character. It is guaranteed not to be whitespace or #
6017: when the extended flag is set. If we are in UTF-8 mode, it may be a
6018: multi-byte literal character. */
6019:
6020: default:
6021: NORMAL_CHAR:
6022: mclength = 1;
6023: mcbuffer[0] = c;
6024:
6025: #ifdef SUPPORT_UTF8
6026: if (utf8 && c >= 0xc0)
6027: {
6028: while ((ptr[1] & 0xc0) == 0x80)
6029: mcbuffer[mclength++] = *(++ptr);
6030: }
6031: #endif
6032:
6033: /* At this point we have the character's bytes in mcbuffer, and the length
6034: in mclength. When not in UTF-8 mode, the length is always 1. */
6035:
6036: ONE_CHAR:
6037: previous = code;
6038: *code++ = ((options & PCRE_CASELESS) != 0)? OP_CHARNC : OP_CHAR;
6039: for (c = 0; c < mclength; c++) *code++ = mcbuffer[c];
6040:
6041: /* Remember if \r or \n were seen */
6042:
1.3 misha 6043: if (mcbuffer[0] == CHAR_CR || mcbuffer[0] == CHAR_NL)
1.1 misha 6044: cd->external_flags |= PCRE_HASCRORLF;
6045:
6046: /* Set the first and required bytes appropriately. If no previous first
6047: byte, set it from this character, but revert to none on a zero repeat.
6048: Otherwise, leave the firstbyte value alone, and don't change it on a zero
6049: repeat. */
6050:
6051: if (firstbyte == REQ_UNSET)
6052: {
6053: zerofirstbyte = REQ_NONE;
6054: zeroreqbyte = reqbyte;
6055:
6056: /* If the character is more than one byte long, we can set firstbyte
6057: only if it is not to be matched caselessly. */
6058:
6059: if (mclength == 1 || req_caseopt == 0)
6060: {
6061: firstbyte = mcbuffer[0] | req_caseopt;
6062: if (mclength != 1) reqbyte = code[-1] | cd->req_varyopt;
6063: }
6064: else firstbyte = reqbyte = REQ_NONE;
6065: }
6066:
6067: /* firstbyte was previously set; we can set reqbyte only the length is
6068: 1 or the matching is caseful. */
6069:
6070: else
6071: {
6072: zerofirstbyte = firstbyte;
6073: zeroreqbyte = reqbyte;
6074: if (mclength == 1 || req_caseopt == 0)
6075: reqbyte = code[-1] | req_caseopt | cd->req_varyopt;
6076: }
6077:
6078: break; /* End of literal character handling */
6079: }
6080: } /* end of big loop */
6081:
6082:
6083: /* Control never reaches here by falling through, only by a goto for all the
6084: error states. Pass back the position in the pattern so that it can be displayed
6085: to the user for diagnosing the error. */
6086:
6087: FAILED:
6088: *ptrptr = ptr;
6089: return FALSE;
6090: }
6091:
6092:
6093:
6094:
6095: /*************************************************
6096: * Compile sequence of alternatives *
6097: *************************************************/
6098:
6099: /* On entry, ptr is pointing past the bracket character, but on return it
6100: points to the closing bracket, or vertical bar, or end of string. The code
6101: variable is pointing at the byte into which the BRA operator has been stored.
6102: If the ims options are changed at the start (for a (?ims: group) or during any
6103: branch, we need to insert an OP_OPT item at the start of every following branch
6104: to ensure they get set correctly at run time, and also pass the new options
6105: into every subsequent branch compile.
6106:
6107: This function is used during the pre-compile phase when we are trying to find
6108: out the amount of memory needed, as well as during the real compile phase. The
6109: value of lengthptr distinguishes the two phases.
6110:
6111: Arguments:
6112: options option bits, including any changes for this subpattern
6113: oldims previous settings of ims option bits
6114: codeptr -> the address of the current code pointer
6115: ptrptr -> the address of the current pattern pointer
6116: errorcodeptr -> pointer to error code variable
6117: lookbehind TRUE if this is a lookbehind assertion
6118: reset_bracount TRUE to reset the count for each branch
6119: skipbytes skip this many bytes at start (for brackets and OP_COND)
6120: firstbyteptr place to put the first required character, or a negative number
6121: reqbyteptr place to put the last required character, or a negative number
6122: bcptr pointer to the chain of currently open branches
6123: cd points to the data block with tables pointers etc.
6124: lengthptr NULL during the real compile phase
6125: points to length accumulator during pre-compile phase
6126:
6127: Returns: TRUE on success
6128: */
6129:
6130: static BOOL
6131: compile_regex(int options, int oldims, uschar **codeptr, const uschar **ptrptr,
6132: int *errorcodeptr, BOOL lookbehind, BOOL reset_bracount, int skipbytes,
6133: int *firstbyteptr, int *reqbyteptr, branch_chain *bcptr, compile_data *cd,
6134: int *lengthptr)
6135: {
6136: const uschar *ptr = *ptrptr;
6137: uschar *code = *codeptr;
6138: uschar *last_branch = code;
6139: uschar *start_bracket = code;
6140: uschar *reverse_count = NULL;
1.4 ! misha 6141: open_capitem capitem;
! 6142: int capnumber = 0;
1.1 misha 6143: int firstbyte, reqbyte;
6144: int branchfirstbyte, branchreqbyte;
6145: int length;
6146: int orig_bracount;
6147: int max_bracount;
1.4 ! misha 6148: int old_external_options = cd->external_options;
1.1 misha 6149: branch_chain bc;
6150:
6151: bc.outer = bcptr;
1.4 ! misha 6152: bc.current_branch = code;
1.1 misha 6153:
6154: firstbyte = reqbyte = REQ_UNSET;
6155:
6156: /* Accumulate the length for use in the pre-compile phase. Start with the
6157: length of the BRA and KET and any extra bytes that are required at the
6158: beginning. We accumulate in a local variable to save frequent testing of
6159: lenthptr for NULL. We cannot do this by looking at the value of code at the
6160: start and end of each alternative, because compiled items are discarded during
6161: the pre-compile phase so that the work space is not exceeded. */
6162:
6163: length = 2 + 2*LINK_SIZE + skipbytes;
6164:
6165: /* WARNING: If the above line is changed for any reason, you must also change
6166: the code that abstracts option settings at the start of the pattern and makes
6167: them global. It tests the value of length for (2 + 2*LINK_SIZE) in the
6168: pre-compile phase to find out whether anything has yet been compiled or not. */
6169:
1.4 ! misha 6170: /* If this is a capturing subpattern, add to the chain of open capturing items
! 6171: so that we can detect them if (*ACCEPT) is encountered. This is also used to
! 6172: detect groups that contain recursive back references to themselves. */
! 6173:
! 6174: if (*code == OP_CBRA)
! 6175: {
! 6176: capnumber = GET2(code, 1 + LINK_SIZE);
! 6177: capitem.number = capnumber;
! 6178: capitem.next = cd->open_caps;
! 6179: capitem.flag = FALSE;
! 6180: cd->open_caps = &capitem;
! 6181: }
! 6182:
1.1 misha 6183: /* Offset is set zero to mark that this bracket is still open */
6184:
6185: PUT(code, 1, 0);
6186: code += 1 + LINK_SIZE + skipbytes;
6187:
6188: /* Loop for each alternative branch */
6189:
6190: orig_bracount = max_bracount = cd->bracount;
6191: for (;;)
6192: {
6193: /* For a (?| group, reset the capturing bracket count so that each branch
6194: uses the same numbers. */
6195:
6196: if (reset_bracount) cd->bracount = orig_bracount;
6197:
6198: /* Handle a change of ims options at the start of the branch */
6199:
6200: if ((options & PCRE_IMS) != oldims)
6201: {
6202: *code++ = OP_OPT;
6203: *code++ = options & PCRE_IMS;
6204: length += 2;
6205: }
6206:
6207: /* Set up dummy OP_REVERSE if lookbehind assertion */
6208:
6209: if (lookbehind)
6210: {
6211: *code++ = OP_REVERSE;
6212: reverse_count = code;
6213: PUTINC(code, 0, 0);
6214: length += 1 + LINK_SIZE;
6215: }
6216:
6217: /* Now compile the branch; in the pre-compile phase its length gets added
6218: into the length. */
6219:
6220: if (!compile_branch(&options, &code, &ptr, errorcodeptr, &branchfirstbyte,
6221: &branchreqbyte, &bc, cd, (lengthptr == NULL)? NULL : &length))
6222: {
6223: *ptrptr = ptr;
6224: return FALSE;
6225: }
6226:
1.4 ! misha 6227: /* If the external options have changed during this branch, it means that we
! 6228: are at the top level, and a leading option setting has been encountered. We
! 6229: need to re-set the original option values to take account of this so that,
! 6230: during the pre-compile phase, we know to allow for a re-set at the start of
! 6231: subsequent branches. */
! 6232:
! 6233: if (old_external_options != cd->external_options)
! 6234: oldims = cd->external_options & PCRE_IMS;
! 6235:
1.1 misha 6236: /* Keep the highest bracket count in case (?| was used and some branch
6237: has fewer than the rest. */
6238:
6239: if (cd->bracount > max_bracount) max_bracount = cd->bracount;
6240:
6241: /* In the real compile phase, there is some post-processing to be done. */
6242:
6243: if (lengthptr == NULL)
6244: {
6245: /* If this is the first branch, the firstbyte and reqbyte values for the
6246: branch become the values for the regex. */
6247:
6248: if (*last_branch != OP_ALT)
6249: {
6250: firstbyte = branchfirstbyte;
6251: reqbyte = branchreqbyte;
6252: }
6253:
6254: /* If this is not the first branch, the first char and reqbyte have to
6255: match the values from all the previous branches, except that if the
6256: previous value for reqbyte didn't have REQ_VARY set, it can still match,
6257: and we set REQ_VARY for the regex. */
6258:
6259: else
6260: {
6261: /* If we previously had a firstbyte, but it doesn't match the new branch,
6262: we have to abandon the firstbyte for the regex, but if there was
6263: previously no reqbyte, it takes on the value of the old firstbyte. */
6264:
6265: if (firstbyte >= 0 && firstbyte != branchfirstbyte)
6266: {
6267: if (reqbyte < 0) reqbyte = firstbyte;
6268: firstbyte = REQ_NONE;
6269: }
6270:
6271: /* If we (now or from before) have no firstbyte, a firstbyte from the
6272: branch becomes a reqbyte if there isn't a branch reqbyte. */
6273:
6274: if (firstbyte < 0 && branchfirstbyte >= 0 && branchreqbyte < 0)
6275: branchreqbyte = branchfirstbyte;
6276:
6277: /* Now ensure that the reqbytes match */
6278:
6279: if ((reqbyte & ~REQ_VARY) != (branchreqbyte & ~REQ_VARY))
6280: reqbyte = REQ_NONE;
6281: else reqbyte |= branchreqbyte; /* To "or" REQ_VARY */
6282: }
6283:
6284: /* If lookbehind, check that this branch matches a fixed-length string, and
6285: put the length into the OP_REVERSE item. Temporarily mark the end of the
1.4 ! misha 6286: branch with OP_END. If the branch contains OP_RECURSE, the result is -3
! 6287: because there may be forward references that we can't check here. Set a
! 6288: flag to cause another lookbehind check at the end. Why not do it all at the
! 6289: end? Because common, erroneous checks are picked up here and the offset of
! 6290: the problem can be shown. */
1.1 misha 6291:
6292: if (lookbehind)
6293: {
6294: int fixed_length;
6295: *code = OP_END;
1.4 ! misha 6296: fixed_length = find_fixedlength(last_branch, options, FALSE, cd);
1.1 misha 6297: DPRINTF(("fixed length = %d\n", fixed_length));
1.4 ! misha 6298: if (fixed_length == -3)
! 6299: {
! 6300: cd->check_lookbehind = TRUE;
! 6301: }
! 6302: else if (fixed_length < 0)
1.1 misha 6303: {
6304: *errorcodeptr = (fixed_length == -2)? ERR36 : ERR25;
6305: *ptrptr = ptr;
6306: return FALSE;
6307: }
1.4 ! misha 6308: else { PUT(reverse_count, 0, fixed_length); }
1.1 misha 6309: }
6310: }
6311:
6312: /* Reached end of expression, either ')' or end of pattern. In the real
6313: compile phase, go back through the alternative branches and reverse the chain
6314: of offsets, with the field in the BRA item now becoming an offset to the
6315: first alternative. If there are no alternatives, it points to the end of the
6316: group. The length in the terminating ket is always the length of the whole
6317: bracketed item. If any of the ims options were changed inside the group,
6318: compile a resetting op-code following, except at the very end of the pattern.
6319: Return leaving the pointer at the terminating char. */
6320:
1.3 misha 6321: if (*ptr != CHAR_VERTICAL_LINE)
1.1 misha 6322: {
6323: if (lengthptr == NULL)
6324: {
1.4 ! misha 6325: int branch_length = (int)(code - last_branch);
1.1 misha 6326: do
6327: {
6328: int prev_length = GET(last_branch, 1);
6329: PUT(last_branch, 1, branch_length);
6330: branch_length = prev_length;
6331: last_branch -= branch_length;
6332: }
6333: while (branch_length > 0);
6334: }
6335:
6336: /* Fill in the ket */
6337:
6338: *code = OP_KET;
1.4 ! misha 6339: PUT(code, 1, (int)(code - start_bracket));
1.1 misha 6340: code += 1 + LINK_SIZE;
6341:
1.4 ! misha 6342: /* If it was a capturing subpattern, check to see if it contained any
! 6343: recursive back references. If so, we must wrap it in atomic brackets.
! 6344: In any event, remove the block from the chain. */
! 6345:
! 6346: if (capnumber > 0)
! 6347: {
! 6348: if (cd->open_caps->flag)
! 6349: {
! 6350: memmove(start_bracket + 1 + LINK_SIZE, start_bracket,
! 6351: code - start_bracket);
! 6352: *start_bracket = OP_ONCE;
! 6353: code += 1 + LINK_SIZE;
! 6354: PUT(start_bracket, 1, (int)(code - start_bracket));
! 6355: *code = OP_KET;
! 6356: PUT(code, 1, (int)(code - start_bracket));
! 6357: code += 1 + LINK_SIZE;
! 6358: length += 2 + 2*LINK_SIZE;
! 6359: }
! 6360: cd->open_caps = cd->open_caps->next;
! 6361: }
! 6362:
! 6363: /* Reset options if needed. */
1.1 misha 6364:
1.3 misha 6365: if ((options & PCRE_IMS) != oldims && *ptr == CHAR_RIGHT_PARENTHESIS)
1.1 misha 6366: {
6367: *code++ = OP_OPT;
6368: *code++ = oldims;
6369: length += 2;
6370: }
6371:
6372: /* Retain the highest bracket number, in case resetting was used. */
6373:
6374: cd->bracount = max_bracount;
6375:
6376: /* Set values to pass back */
6377:
6378: *codeptr = code;
6379: *ptrptr = ptr;
6380: *firstbyteptr = firstbyte;
6381: *reqbyteptr = reqbyte;
6382: if (lengthptr != NULL)
6383: {
6384: if (OFLOW_MAX - *lengthptr < length)
6385: {
6386: *errorcodeptr = ERR20;
6387: return FALSE;
6388: }
6389: *lengthptr += length;
6390: }
6391: return TRUE;
6392: }
6393:
6394: /* Another branch follows. In the pre-compile phase, we can move the code
6395: pointer back to where it was for the start of the first branch. (That is,
6396: pretend that each branch is the only one.)
6397:
6398: In the real compile phase, insert an ALT node. Its length field points back
6399: to the previous branch while the bracket remains open. At the end the chain
6400: is reversed. It's done like this so that the start of the bracket has a
6401: zero offset until it is closed, making it possible to detect recursion. */
6402:
6403: if (lengthptr != NULL)
6404: {
6405: code = *codeptr + 1 + LINK_SIZE + skipbytes;
6406: length += 1 + LINK_SIZE;
6407: }
6408: else
6409: {
6410: *code = OP_ALT;
1.4 ! misha 6411: PUT(code, 1, (int)(code - last_branch));
! 6412: bc.current_branch = last_branch = code;
1.1 misha 6413: code += 1 + LINK_SIZE;
6414: }
6415:
6416: ptr++;
6417: }
6418: /* Control never reaches here */
6419: }
6420:
6421:
6422:
6423:
6424: /*************************************************
6425: * Check for anchored expression *
6426: *************************************************/
6427:
6428: /* Try to find out if this is an anchored regular expression. Consider each
6429: alternative branch. If they all start with OP_SOD or OP_CIRC, or with a bracket
6430: all of whose alternatives start with OP_SOD or OP_CIRC (recurse ad lib), then
6431: it's anchored. However, if this is a multiline pattern, then only OP_SOD
6432: counts, since OP_CIRC can match in the middle.
6433:
6434: We can also consider a regex to be anchored if OP_SOM starts all its branches.
6435: This is the code for \G, which means "match at start of match position, taking
6436: into account the match offset".
6437:
6438: A branch is also implicitly anchored if it starts with .* and DOTALL is set,
6439: because that will try the rest of the pattern at all possible matching points,
6440: so there is no point trying again.... er ....
6441:
6442: .... except when the .* appears inside capturing parentheses, and there is a
6443: subsequent back reference to those parentheses. We haven't enough information
6444: to catch that case precisely.
6445:
6446: At first, the best we could do was to detect when .* was in capturing brackets
6447: and the highest back reference was greater than or equal to that level.
6448: However, by keeping a bitmap of the first 31 back references, we can catch some
6449: of the more common cases more precisely.
6450:
6451: Arguments:
6452: code points to start of expression (the bracket)
6453: options points to the options setting
6454: bracket_map a bitmap of which brackets we are inside while testing; this
6455: handles up to substring 31; after that we just have to take
6456: the less precise approach
6457: backref_map the back reference bitmap
6458:
6459: Returns: TRUE or FALSE
6460: */
6461:
6462: static BOOL
6463: is_anchored(register const uschar *code, int *options, unsigned int bracket_map,
6464: unsigned int backref_map)
6465: {
6466: do {
6467: const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6468: options, PCRE_MULTILINE, FALSE);
6469: register int op = *scode;
6470:
6471: /* Non-capturing brackets */
6472:
6473: if (op == OP_BRA)
6474: {
6475: if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6476: }
6477:
6478: /* Capturing brackets */
6479:
6480: else if (op == OP_CBRA)
6481: {
6482: int n = GET2(scode, 1+LINK_SIZE);
6483: int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6484: if (!is_anchored(scode, options, new_map, backref_map)) return FALSE;
6485: }
6486:
6487: /* Other brackets */
6488:
6489: else if (op == OP_ASSERT || op == OP_ONCE || op == OP_COND)
6490: {
6491: if (!is_anchored(scode, options, bracket_map, backref_map)) return FALSE;
6492: }
6493:
6494: /* .* is not anchored unless DOTALL is set (which generates OP_ALLANY) and
6495: it isn't in brackets that are or may be referenced. */
6496:
6497: else if ((op == OP_TYPESTAR || op == OP_TYPEMINSTAR ||
6498: op == OP_TYPEPOSSTAR))
6499: {
6500: if (scode[1] != OP_ALLANY || (bracket_map & backref_map) != 0)
6501: return FALSE;
6502: }
6503:
6504: /* Check for explicit anchoring */
6505:
6506: else if (op != OP_SOD && op != OP_SOM &&
6507: ((*options & PCRE_MULTILINE) != 0 || op != OP_CIRC))
6508: return FALSE;
6509: code += GET(code, 1);
6510: }
6511: while (*code == OP_ALT); /* Loop for each alternative */
6512: return TRUE;
6513: }
6514:
6515:
6516:
6517: /*************************************************
6518: * Check for starting with ^ or .* *
6519: *************************************************/
6520:
6521: /* This is called to find out if every branch starts with ^ or .* so that
6522: "first char" processing can be done to speed things up in multiline
6523: matching and for non-DOTALL patterns that start with .* (which must start at
6524: the beginning or after \n). As in the case of is_anchored() (see above), we
6525: have to take account of back references to capturing brackets that contain .*
6526: because in that case we can't make the assumption.
6527:
6528: Arguments:
6529: code points to start of expression (the bracket)
6530: bracket_map a bitmap of which brackets we are inside while testing; this
6531: handles up to substring 31; after that we just have to take
6532: the less precise approach
6533: backref_map the back reference bitmap
6534:
6535: Returns: TRUE or FALSE
6536: */
6537:
6538: static BOOL
6539: is_startline(const uschar *code, unsigned int bracket_map,
6540: unsigned int backref_map)
6541: {
6542: do {
6543: const uschar *scode = first_significant_code(code + _pcre_OP_lengths[*code],
6544: NULL, 0, FALSE);
6545: register int op = *scode;
6546:
1.3 misha 6547: /* If we are at the start of a conditional assertion group, *both* the
6548: conditional assertion *and* what follows the condition must satisfy the test
6549: for start of line. Other kinds of condition fail. Note that there may be an
6550: auto-callout at the start of a condition. */
6551:
6552: if (op == OP_COND)
6553: {
6554: scode += 1 + LINK_SIZE;
6555: if (*scode == OP_CALLOUT) scode += _pcre_OP_lengths[OP_CALLOUT];
6556: switch (*scode)
6557: {
6558: case OP_CREF:
1.4 ! misha 6559: case OP_NCREF:
1.3 misha 6560: case OP_RREF:
1.4 ! misha 6561: case OP_NRREF:
1.3 misha 6562: case OP_DEF:
6563: return FALSE;
6564:
6565: default: /* Assertion */
6566: if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6567: do scode += GET(scode, 1); while (*scode == OP_ALT);
6568: scode += 1 + LINK_SIZE;
6569: break;
6570: }
6571: scode = first_significant_code(scode, NULL, 0, FALSE);
6572: op = *scode;
6573: }
6574:
1.1 misha 6575: /* Non-capturing brackets */
6576:
6577: if (op == OP_BRA)
6578: {
6579: if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6580: }
6581:
6582: /* Capturing brackets */
6583:
6584: else if (op == OP_CBRA)
6585: {
6586: int n = GET2(scode, 1+LINK_SIZE);
6587: int new_map = bracket_map | ((n < 32)? (1 << n) : 1);
6588: if (!is_startline(scode, new_map, backref_map)) return FALSE;
6589: }
6590:
6591: /* Other brackets */
6592:
1.3 misha 6593: else if (op == OP_ASSERT || op == OP_ONCE)
6594: {
6595: if (!is_startline(scode, bracket_map, backref_map)) return FALSE;
6596: }
1.1 misha 6597:
6598: /* .* means "start at start or after \n" if it isn't in brackets that
6599: may be referenced. */
6600:
6601: else if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR || op == OP_TYPEPOSSTAR)
6602: {
6603: if (scode[1] != OP_ANY || (bracket_map & backref_map) != 0) return FALSE;
6604: }
6605:
6606: /* Check for explicit circumflex */
6607:
6608: else if (op != OP_CIRC) return FALSE;
6609:
6610: /* Move on to the next alternative */
6611:
6612: code += GET(code, 1);
6613: }
6614: while (*code == OP_ALT); /* Loop for each alternative */
6615: return TRUE;
6616: }
6617:
6618:
6619:
6620: /*************************************************
6621: * Check for asserted fixed first char *
6622: *************************************************/
6623:
6624: /* During compilation, the "first char" settings from forward assertions are
6625: discarded, because they can cause conflicts with actual literals that follow.
6626: However, if we end up without a first char setting for an unanchored pattern,
6627: it is worth scanning the regex to see if there is an initial asserted first
6628: char. If all branches start with the same asserted char, or with a bracket all
6629: of whose alternatives start with the same asserted char (recurse ad lib), then
6630: we return that char, otherwise -1.
6631:
6632: Arguments:
6633: code points to start of expression (the bracket)
6634: options pointer to the options (used to check casing changes)
6635: inassert TRUE if in an assertion
6636:
6637: Returns: -1 or the fixed first char
6638: */
6639:
6640: static int
6641: find_firstassertedchar(const uschar *code, int *options, BOOL inassert)
6642: {
6643: register int c = -1;
6644: do {
6645: int d;
6646: const uschar *scode =
6647: first_significant_code(code + 1+LINK_SIZE, options, PCRE_CASELESS, TRUE);
6648: register int op = *scode;
6649:
6650: switch(op)
6651: {
6652: default:
6653: return -1;
6654:
6655: case OP_BRA:
6656: case OP_CBRA:
6657: case OP_ASSERT:
6658: case OP_ONCE:
6659: case OP_COND:
6660: if ((d = find_firstassertedchar(scode, options, op == OP_ASSERT)) < 0)
6661: return -1;
6662: if (c < 0) c = d; else if (c != d) return -1;
6663: break;
6664:
6665: case OP_EXACT: /* Fall through */
6666: scode += 2;
6667:
6668: case OP_CHAR:
6669: case OP_CHARNC:
6670: case OP_PLUS:
6671: case OP_MINPLUS:
6672: case OP_POSPLUS:
6673: if (!inassert) return -1;
6674: if (c < 0)
6675: {
6676: c = scode[1];
6677: if ((*options & PCRE_CASELESS) != 0) c |= REQ_CASELESS;
6678: }
6679: else if (c != scode[1]) return -1;
6680: break;
6681: }
6682:
6683: code += GET(code, 1);
6684: }
6685: while (*code == OP_ALT);
6686: return c;
6687: }
6688:
6689:
6690:
6691: /*************************************************
6692: * Compile a Regular Expression *
6693: *************************************************/
6694:
6695: /* This function takes a string and returns a pointer to a block of store
6696: holding a compiled version of the expression. The original API for this
6697: function had no error code return variable; it is retained for backwards
6698: compatibility. The new function is given a new name.
6699:
6700: Arguments:
6701: pattern the regular expression
6702: options various option bits
6703: errorcodeptr pointer to error code variable (pcre_compile2() only)
6704: can be NULL if you don't want a code value
6705: errorptr pointer to pointer to error text
6706: erroroffset ptr offset in pattern where error was detected
6707: tables pointer to character tables or NULL
6708:
6709: Returns: pointer to compiled data block, or NULL on error,
6710: with errorptr and erroroffset set
6711: */
6712:
1.2 misha 6713: PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
1.1 misha 6714: pcre_compile(const char *pattern, int options, const char **errorptr,
6715: int *erroroffset, const unsigned char *tables)
6716: {
6717: return pcre_compile2(pattern, options, NULL, errorptr, erroroffset, tables);
6718: }
6719:
6720:
1.2 misha 6721: PCRE_EXP_DEFN pcre * PCRE_CALL_CONVENTION
1.1 misha 6722: pcre_compile2(const char *pattern, int options, int *errorcodeptr,
6723: const char **errorptr, int *erroroffset, const unsigned char *tables)
6724: {
6725: real_pcre *re;
6726: int length = 1; /* For final END opcode */
6727: int firstbyte, reqbyte, newline;
6728: int errorcode = 0;
6729: int skipatstart = 0;
6730: BOOL utf8;
6731: size_t size;
6732: uschar *code;
6733: const uschar *codestart;
6734: const uschar *ptr;
6735: compile_data compile_block;
6736: compile_data *cd = &compile_block;
6737:
6738: /* This space is used for "compiling" into during the first phase, when we are
6739: computing the amount of memory that is needed. Compiled items are thrown away
6740: as soon as possible, so that a fairly large buffer should be sufficient for
6741: this purpose. The same space is used in the second phase for remembering where
6742: to fill in forward references to subpatterns. */
6743:
6744: uschar cworkspace[COMPILE_WORK_SIZE];
6745:
6746: /* Set this early so that early errors get offset 0. */
6747:
6748: ptr = (const uschar *)pattern;
6749:
6750: /* We can't pass back an error message if errorptr is NULL; I guess the best we
6751: can do is just return NULL, but we can set a code value if there is a code
6752: pointer. */
6753:
6754: if (errorptr == NULL)
6755: {
6756: if (errorcodeptr != NULL) *errorcodeptr = 99;
6757: return NULL;
6758: }
6759:
6760: *errorptr = NULL;
6761: if (errorcodeptr != NULL) *errorcodeptr = ERR0;
6762:
6763: /* However, we can give a message for this error */
6764:
6765: if (erroroffset == NULL)
6766: {
6767: errorcode = ERR16;
6768: goto PCRE_EARLY_ERROR_RETURN2;
6769: }
6770:
6771: *erroroffset = 0;
6772:
6773: /* Set up pointers to the individual character tables */
6774:
6775: if (tables == NULL) tables = _pcre_default_tables;
6776: cd->lcc = tables + lcc_offset;
6777: cd->fcc = tables + fcc_offset;
6778: cd->cbits = tables + cbits_offset;
6779: cd->ctypes = tables + ctypes_offset;
6780:
1.3 misha 6781: /* Check that all undefined public option bits are zero */
6782:
6783: if ((options & ~PUBLIC_COMPILE_OPTIONS) != 0)
6784: {
6785: errorcode = ERR17;
6786: goto PCRE_EARLY_ERROR_RETURN;
6787: }
6788:
1.1 misha 6789: /* Check for global one-time settings at the start of the pattern, and remember
6790: the offset for later. */
6791:
1.3 misha 6792: while (ptr[skipatstart] == CHAR_LEFT_PARENTHESIS &&
6793: ptr[skipatstart+1] == CHAR_ASTERISK)
1.1 misha 6794: {
6795: int newnl = 0;
6796: int newbsr = 0;
6797:
1.3 misha 6798: if (strncmp((char *)(ptr+skipatstart+2), STRING_UTF8_RIGHTPAR, 5) == 0)
6799: { skipatstart += 7; options |= PCRE_UTF8; continue; }
1.4 ! misha 6800: else if (strncmp((char *)(ptr+skipatstart+2), STRING_UCP_RIGHTPAR, 4) == 0)
! 6801: { skipatstart += 6; options |= PCRE_UCP; continue; }
1.3 misha 6802:
6803: if (strncmp((char *)(ptr+skipatstart+2), STRING_CR_RIGHTPAR, 3) == 0)
1.1 misha 6804: { skipatstart += 5; newnl = PCRE_NEWLINE_CR; }
1.3 misha 6805: else if (strncmp((char *)(ptr+skipatstart+2), STRING_LF_RIGHTPAR, 3) == 0)
1.1 misha 6806: { skipatstart += 5; newnl = PCRE_NEWLINE_LF; }
1.3 misha 6807: else if (strncmp((char *)(ptr+skipatstart+2), STRING_CRLF_RIGHTPAR, 5) == 0)
1.1 misha 6808: { skipatstart += 7; newnl = PCRE_NEWLINE_CR + PCRE_NEWLINE_LF; }
1.3 misha 6809: else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANY_RIGHTPAR, 4) == 0)
1.1 misha 6810: { skipatstart += 6; newnl = PCRE_NEWLINE_ANY; }
1.3 misha 6811: else if (strncmp((char *)(ptr+skipatstart+2), STRING_ANYCRLF_RIGHTPAR, 8) == 0)
1.1 misha 6812: { skipatstart += 10; newnl = PCRE_NEWLINE_ANYCRLF; }
6813:
1.3 misha 6814: else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_ANYCRLF_RIGHTPAR, 12) == 0)
1.1 misha 6815: { skipatstart += 14; newbsr = PCRE_BSR_ANYCRLF; }
1.3 misha 6816: else if (strncmp((char *)(ptr+skipatstart+2), STRING_BSR_UNICODE_RIGHTPAR, 12) == 0)
1.1 misha 6817: { skipatstart += 14; newbsr = PCRE_BSR_UNICODE; }
6818:
6819: if (newnl != 0)
6820: options = (options & ~PCRE_NEWLINE_BITS) | newnl;
6821: else if (newbsr != 0)
6822: options = (options & ~(PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE)) | newbsr;
6823: else break;
6824: }
6825:
1.4 ! misha 6826: utf8 = (options & PCRE_UTF8) != 0;
! 6827:
1.3 misha 6828: /* Can't support UTF8 unless PCRE has been compiled to include the code. */
6829:
6830: #ifdef SUPPORT_UTF8
6831: if (utf8 && (options & PCRE_NO_UTF8_CHECK) == 0 &&
1.4 ! misha 6832: (*erroroffset = _pcre_valid_utf8((USPTR)pattern, -1)) >= 0)
1.3 misha 6833: {
6834: errorcode = ERR44;
6835: goto PCRE_EARLY_ERROR_RETURN2;
6836: }
6837: #else
1.4 ! misha 6838: if (utf8)
1.3 misha 6839: {
6840: errorcode = ERR32;
6841: goto PCRE_EARLY_ERROR_RETURN;
6842: }
6843: #endif
6844:
1.4 ! misha 6845: /* Can't support UCP unless PCRE has been compiled to include the code. */
! 6846:
! 6847: #ifndef SUPPORT_UCP
! 6848: if ((options & PCRE_UCP) != 0)
! 6849: {
! 6850: errorcode = ERR67;
! 6851: goto PCRE_EARLY_ERROR_RETURN;
! 6852: }
! 6853: #endif
! 6854:
1.1 misha 6855: /* Check validity of \R options. */
6856:
6857: switch (options & (PCRE_BSR_ANYCRLF|PCRE_BSR_UNICODE))
6858: {
6859: case 0:
6860: case PCRE_BSR_ANYCRLF:
6861: case PCRE_BSR_UNICODE:
6862: break;
6863: default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6864: }
6865:
6866: /* Handle different types of newline. The three bits give seven cases. The
6867: current code allows for fixed one- or two-byte sequences, plus "any" and
6868: "anycrlf". */
6869:
6870: switch (options & PCRE_NEWLINE_BITS)
6871: {
6872: case 0: newline = NEWLINE; break; /* Build-time default */
1.3 misha 6873: case PCRE_NEWLINE_CR: newline = CHAR_CR; break;
6874: case PCRE_NEWLINE_LF: newline = CHAR_NL; break;
1.1 misha 6875: case PCRE_NEWLINE_CR+
1.3 misha 6876: PCRE_NEWLINE_LF: newline = (CHAR_CR << 8) | CHAR_NL; break;
1.1 misha 6877: case PCRE_NEWLINE_ANY: newline = -1; break;
6878: case PCRE_NEWLINE_ANYCRLF: newline = -2; break;
6879: default: errorcode = ERR56; goto PCRE_EARLY_ERROR_RETURN;
6880: }
6881:
6882: if (newline == -2)
6883: {
6884: cd->nltype = NLTYPE_ANYCRLF;
6885: }
6886: else if (newline < 0)
6887: {
6888: cd->nltype = NLTYPE_ANY;
6889: }
6890: else
6891: {
6892: cd->nltype = NLTYPE_FIXED;
6893: if (newline > 255)
6894: {
6895: cd->nllen = 2;
6896: cd->nl[0] = (newline >> 8) & 255;
6897: cd->nl[1] = newline & 255;
6898: }
6899: else
6900: {
6901: cd->nllen = 1;
6902: cd->nl[0] = newline;
6903: }
6904: }
6905:
6906: /* Maximum back reference and backref bitmap. The bitmap records up to 31 back
6907: references to help in deciding whether (.*) can be treated as anchored or not.
6908: */
6909:
6910: cd->top_backref = 0;
6911: cd->backref_map = 0;
6912:
6913: /* Reflect pattern for debugging output */
6914:
6915: DPRINTF(("------------------------------------------------------------------\n"));
6916: DPRINTF(("%s\n", pattern));
6917:
6918: /* Pretend to compile the pattern while actually just accumulating the length
6919: of memory required. This behaviour is triggered by passing a non-NULL final
6920: argument to compile_regex(). We pass a block of workspace (cworkspace) for it
6921: to compile parts of the pattern into; the compiled code is discarded when it is
6922: no longer needed, so hopefully this workspace will never overflow, though there
6923: is a test for its doing so. */
6924:
6925: cd->bracount = cd->final_bracount = 0;
6926: cd->names_found = 0;
6927: cd->name_entry_size = 0;
6928: cd->name_table = NULL;
6929: cd->start_workspace = cworkspace;
6930: cd->start_code = cworkspace;
6931: cd->hwm = cworkspace;
6932: cd->start_pattern = (const uschar *)pattern;
6933: cd->end_pattern = (const uschar *)(pattern + strlen(pattern));
6934: cd->req_varyopt = 0;
6935: cd->external_options = options;
6936: cd->external_flags = 0;
1.4 ! misha 6937: cd->open_caps = NULL;
1.1 misha 6938:
6939: /* Now do the pre-compile. On error, errorcode will be set non-zero, so we
6940: don't need to look at the result of the function here. The initial options have
6941: been put into the cd block so that they can be changed if an option setting is
6942: found within the regex right at the beginning. Bringing initial option settings
6943: outside can help speed up starting point checks. */
6944:
6945: ptr += skipatstart;
6946: code = cworkspace;
6947: *code = OP_BRA;
6948: (void)compile_regex(cd->external_options, cd->external_options & PCRE_IMS,
6949: &code, &ptr, &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd,
6950: &length);
6951: if (errorcode != 0) goto PCRE_EARLY_ERROR_RETURN;
6952:
6953: DPRINTF(("end pre-compile: length=%d workspace=%d\n", length,
6954: cd->hwm - cworkspace));
6955:
6956: if (length > MAX_PATTERN_SIZE)
6957: {
6958: errorcode = ERR20;
6959: goto PCRE_EARLY_ERROR_RETURN;
6960: }
6961:
6962: /* Compute the size of data block needed and get it, either from malloc or
6963: externally provided function. Integer overflow should no longer be possible
6964: because nowadays we limit the maximum value of cd->names_found and
6965: cd->name_entry_size. */
6966:
6967: size = length + sizeof(real_pcre) + cd->names_found * (cd->name_entry_size + 3);
6968: re = (real_pcre *)(pcre_malloc)(size);
6969:
6970: if (re == NULL)
6971: {
6972: errorcode = ERR21;
6973: goto PCRE_EARLY_ERROR_RETURN;
6974: }
6975:
6976: /* Put in the magic number, and save the sizes, initial options, internal
6977: flags, and character table pointer. NULL is used for the default character
6978: tables. The nullpad field is at the end; it's there to help in the case when a
6979: regex compiled on a system with 4-byte pointers is run on another with 8-byte
6980: pointers. */
6981:
6982: re->magic_number = MAGIC_NUMBER;
1.4 ! misha 6983: re->size = (int)size;
1.1 misha 6984: re->options = cd->external_options;
6985: re->flags = cd->external_flags;
6986: re->dummy1 = 0;
6987: re->first_byte = 0;
6988: re->req_byte = 0;
6989: re->name_table_offset = sizeof(real_pcre);
6990: re->name_entry_size = cd->name_entry_size;
6991: re->name_count = cd->names_found;
6992: re->ref_count = 0;
6993: re->tables = (tables == _pcre_default_tables)? NULL : tables;
6994: re->nullpad = NULL;
6995:
6996: /* The starting points of the name/number translation table and of the code are
6997: passed around in the compile data block. The start/end pattern and initial
6998: options are already set from the pre-compile phase, as is the name_entry_size
6999: field. Reset the bracket count and the names_found field. Also reset the hwm
7000: field; this time it's used for remembering forward references to subpatterns.
7001: */
7002:
7003: cd->final_bracount = cd->bracount; /* Save for checking forward references */
7004: cd->bracount = 0;
7005: cd->names_found = 0;
7006: cd->name_table = (uschar *)re + re->name_table_offset;
7007: codestart = cd->name_table + re->name_entry_size * re->name_count;
7008: cd->start_code = codestart;
7009: cd->hwm = cworkspace;
7010: cd->req_varyopt = 0;
7011: cd->had_accept = FALSE;
1.4 ! misha 7012: cd->check_lookbehind = FALSE;
! 7013: cd->open_caps = NULL;
1.1 misha 7014:
7015: /* Set up a starting, non-extracting bracket, then compile the expression. On
7016: error, errorcode will be set non-zero, so we don't need to look at the result
7017: of the function here. */
7018:
7019: ptr = (const uschar *)pattern + skipatstart;
7020: code = (uschar *)codestart;
7021: *code = OP_BRA;
7022: (void)compile_regex(re->options, re->options & PCRE_IMS, &code, &ptr,
7023: &errorcode, FALSE, FALSE, 0, &firstbyte, &reqbyte, NULL, cd, NULL);
7024: re->top_bracket = cd->bracount;
7025: re->top_backref = cd->top_backref;
7026: re->flags = cd->external_flags;
7027:
7028: if (cd->had_accept) reqbyte = -1; /* Must disable after (*ACCEPT) */
7029:
7030: /* If not reached end of pattern on success, there's an excess bracket. */
7031:
7032: if (errorcode == 0 && *ptr != 0) errorcode = ERR22;
7033:
7034: /* Fill in the terminating state and check for disastrous overflow, but
7035: if debugging, leave the test till after things are printed out. */
7036:
7037: *code++ = OP_END;
7038:
1.4 ! misha 7039: #ifndef PCRE_DEBUG
1.1 misha 7040: if (code - codestart > length) errorcode = ERR23;
7041: #endif
7042:
7043: /* Fill in any forward references that are required. */
7044:
7045: while (errorcode == 0 && cd->hwm > cworkspace)
7046: {
7047: int offset, recno;
7048: const uschar *groupptr;
7049: cd->hwm -= LINK_SIZE;
7050: offset = GET(cd->hwm, 0);
7051: recno = GET(codestart, offset);
1.4 ! misha 7052: groupptr = _pcre_find_bracket(codestart, utf8, recno);
1.1 misha 7053: if (groupptr == NULL) errorcode = ERR53;
1.4 ! misha 7054: else PUT(((uschar *)codestart), offset, (int)(groupptr - codestart));
1.1 misha 7055: }
7056:
7057: /* Give an error if there's back reference to a non-existent capturing
7058: subpattern. */
7059:
7060: if (errorcode == 0 && re->top_backref > re->top_bracket) errorcode = ERR15;
7061:
1.4 ! misha 7062: /* If there were any lookbehind assertions that contained OP_RECURSE
! 7063: (recursions or subroutine calls), a flag is set for them to be checked here,
! 7064: because they may contain forward references. Actual recursions can't be fixed
! 7065: length, but subroutine calls can. It is done like this so that those without
! 7066: OP_RECURSE that are not fixed length get a diagnosic with a useful offset. The
! 7067: exceptional ones forgo this. We scan the pattern to check that they are fixed
! 7068: length, and set their lengths. */
! 7069:
! 7070: if (cd->check_lookbehind)
! 7071: {
! 7072: uschar *cc = (uschar *)codestart;
! 7073:
! 7074: /* Loop, searching for OP_REVERSE items, and process those that do not have
! 7075: their length set. (Actually, it will also re-process any that have a length
! 7076: of zero, but that is a pathological case, and it does no harm.) When we find
! 7077: one, we temporarily terminate the branch it is in while we scan it. */
! 7078:
! 7079: for (cc = (uschar *)_pcre_find_bracket(codestart, utf8, -1);
! 7080: cc != NULL;
! 7081: cc = (uschar *)_pcre_find_bracket(cc, utf8, -1))
! 7082: {
! 7083: if (GET(cc, 1) == 0)
! 7084: {
! 7085: int fixed_length;
! 7086: uschar *be = cc - 1 - LINK_SIZE + GET(cc, -LINK_SIZE);
! 7087: int end_op = *be;
! 7088: *be = OP_END;
! 7089: fixed_length = find_fixedlength(cc, re->options, TRUE, cd);
! 7090: *be = end_op;
! 7091: DPRINTF(("fixed length = %d\n", fixed_length));
! 7092: if (fixed_length < 0)
! 7093: {
! 7094: errorcode = (fixed_length == -2)? ERR36 : ERR25;
! 7095: break;
! 7096: }
! 7097: PUT(cc, 1, fixed_length);
! 7098: }
! 7099: cc += 1 + LINK_SIZE;
! 7100: }
! 7101: }
! 7102:
1.1 misha 7103: /* Failed to compile, or error while post-processing */
7104:
7105: if (errorcode != 0)
7106: {
7107: (pcre_free)(re);
7108: PCRE_EARLY_ERROR_RETURN:
1.4 ! misha 7109: *erroroffset = (int)(ptr - (const uschar *)pattern);
1.1 misha 7110: PCRE_EARLY_ERROR_RETURN2:
7111: *errorptr = find_error_text(errorcode);
7112: if (errorcodeptr != NULL) *errorcodeptr = errorcode;
7113: return NULL;
7114: }
7115:
7116: /* If the anchored option was not passed, set the flag if we can determine that
7117: the pattern is anchored by virtue of ^ characters or \A or anything else (such
7118: as starting with .* when DOTALL is set).
7119:
7120: Otherwise, if we know what the first byte has to be, save it, because that
7121: speeds up unanchored matches no end. If not, see if we can set the
7122: PCRE_STARTLINE flag. This is helpful for multiline matches when all branches
7123: start with ^. and also when all branches start with .* for non-DOTALL matches.
7124: */
7125:
7126: if ((re->options & PCRE_ANCHORED) == 0)
7127: {
7128: int temp_options = re->options; /* May get changed during these scans */
7129: if (is_anchored(codestart, &temp_options, 0, cd->backref_map))
7130: re->options |= PCRE_ANCHORED;
7131: else
7132: {
7133: if (firstbyte < 0)
7134: firstbyte = find_firstassertedchar(codestart, &temp_options, FALSE);
7135: if (firstbyte >= 0) /* Remove caseless flag for non-caseable chars */
7136: {
7137: int ch = firstbyte & 255;
7138: re->first_byte = ((firstbyte & REQ_CASELESS) != 0 &&
7139: cd->fcc[ch] == ch)? ch : firstbyte;
7140: re->flags |= PCRE_FIRSTSET;
7141: }
7142: else if (is_startline(codestart, 0, cd->backref_map))
7143: re->flags |= PCRE_STARTLINE;
7144: }
7145: }
7146:
7147: /* For an anchored pattern, we use the "required byte" only if it follows a
7148: variable length item in the regex. Remove the caseless flag for non-caseable
7149: bytes. */
7150:
7151: if (reqbyte >= 0 &&
7152: ((re->options & PCRE_ANCHORED) == 0 || (reqbyte & REQ_VARY) != 0))
7153: {
7154: int ch = reqbyte & 255;
7155: re->req_byte = ((reqbyte & REQ_CASELESS) != 0 &&
7156: cd->fcc[ch] == ch)? (reqbyte & ~REQ_CASELESS) : reqbyte;
7157: re->flags |= PCRE_REQCHSET;
7158: }
7159:
7160: /* Print out the compiled data if debugging is enabled. This is never the
7161: case when building a production library. */
7162:
1.4 ! misha 7163: #ifdef PCRE_DEBUG
1.1 misha 7164: printf("Length = %d top_bracket = %d top_backref = %d\n",
7165: length, re->top_bracket, re->top_backref);
7166:
7167: printf("Options=%08x\n", re->options);
7168:
7169: if ((re->flags & PCRE_FIRSTSET) != 0)
7170: {
7171: int ch = re->first_byte & 255;
7172: const char *caseless = ((re->first_byte & REQ_CASELESS) == 0)?
7173: "" : " (caseless)";
7174: if (isprint(ch)) printf("First char = %c%s\n", ch, caseless);
7175: else printf("First char = \\x%02x%s\n", ch, caseless);
7176: }
7177:
7178: if ((re->flags & PCRE_REQCHSET) != 0)
7179: {
7180: int ch = re->req_byte & 255;
7181: const char *caseless = ((re->req_byte & REQ_CASELESS) == 0)?
7182: "" : " (caseless)";
7183: if (isprint(ch)) printf("Req char = %c%s\n", ch, caseless);
7184: else printf("Req char = \\x%02x%s\n", ch, caseless);
7185: }
7186:
7187: pcre_printint(re, stdout, TRUE);
7188:
7189: /* This check is done here in the debugging case so that the code that
7190: was compiled can be seen. */
7191:
7192: if (code - codestart > length)
7193: {
7194: (pcre_free)(re);
7195: *errorptr = find_error_text(ERR23);
7196: *erroroffset = ptr - (uschar *)pattern;
7197: if (errorcodeptr != NULL) *errorcodeptr = ERR23;
7198: return NULL;
7199: }
1.4 ! misha 7200: #endif /* PCRE_DEBUG */
1.1 misha 7201:
7202: return (pcre *)re;
7203: }
7204:
7205: /* End of pcre_compile.c */
E-mail:
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to pcre_compile.c CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [parser3project] / win32 / pcre