1 | /* This is JavaScriptCore's variant of the PCRE library. While this library |
2 | started out as a copy of PCRE, many of the features of PCRE have been |
3 | removed. This library now supports only the regular expression features |
4 | required by the JavaScript language specification, and has only the functions |
5 | needed by JavaScriptCore and the rest of WebKit. |
6 | |
7 | Originally written by Philip Hazel |
8 | Copyright (c) 1997-2006 University of Cambridge |
9 | Copyright (C) 2002, 2004, 2006, 2007, 2008, 2009 Apple Inc. All rights reserved. |
10 | Copyright (C) 2007 Eric Seidel <eric@webkit.org> |
11 | |
12 | ----------------------------------------------------------------------------- |
13 | Redistribution and use in source and binary forms, with or without |
14 | modification, are permitted provided that the following conditions are met: |
15 | |
16 | * Redistributions of source code must retain the above copyright notice, |
17 | this list of conditions and the following disclaimer. |
18 | |
19 | * Redistributions in binary form must reproduce the above copyright |
20 | notice, this list of conditions and the following disclaimer in the |
21 | documentation and/or other materials provided with the distribution. |
22 | |
23 | * Neither the name of the University of Cambridge nor the names of its |
24 | contributors may be used to endorse or promote products derived from |
25 | this software without specific prior written permission. |
26 | |
27 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
28 | AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
29 | IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
30 | ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
31 | LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
32 | CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
33 | SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
34 | INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
35 | CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
36 | ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
37 | POSSIBILITY OF SUCH DAMAGE. |
38 | ----------------------------------------------------------------------------- |
39 | */ |
40 | |
41 | /* This module contains the external function jsRegExpExecute(), along with |
42 | supporting internal functions that are not used by other modules. */ |
43 | |
44 | #include "config.h" |
45 | |
46 | #include "pcre_internal.h" |
47 | |
48 | #include <string.h> |
49 | #include <wtf/ASCIICType.h> |
50 | #include <wtf/FastMalloc.h> |
51 | |
52 | using namespace WTF; |
53 | |
54 | /* Negative values for the firstchar and reqchar variables */ |
55 | |
56 | #define REQ_UNSET (-2) |
57 | #define REQ_NONE (-1) |
58 | |
59 | /************************************************* |
60 | * Code parameters and static tables * |
61 | *************************************************/ |
62 | |
63 | /* Maximum number of items on the nested bracket stacks at compile time. This |
64 | applies to the nesting of all kinds of parentheses. It does not limit |
65 | un-nested, non-capturing parentheses. This number can be made bigger if |
66 | necessary - it is used to dimension one int and one unsigned char vector at |
67 | compile time. */ |
68 | |
69 | #define BRASTACK_SIZE 200 |
70 | |
71 | /* Table for handling escaped characters in the range '0'-'z'. Positive returns |
72 | are simple data values; negative values are for special things like \d and so |
73 | on. Zero means further processing is needed (for things like \x), or the escape |
74 | is invalid. */ |
75 | |
76 | static const short escapes[] = { |
77 | 0, 0, 0, 0, 0, 0, 0, 0, /* 0 - 7 */ |
78 | 0, 0, ':', ';', '<', '=', '>', '?', /* 8 - ? */ |
79 | '@', 0, -ESC_B, 0, -ESC_D, 0, 0, 0, /* @ - G */ |
80 | 0, 0, 0, 0, 0, 0, 0, 0, /* H - O */ |
81 | 0, 0, 0, -ESC_S, 0, 0, 0, -ESC_W, /* P - W */ |
82 | 0, 0, 0, '[', '\\', ']', '^', '_', /* X - _ */ |
83 | '`', 7, -ESC_b, 0, -ESC_d, 0, '\f', 0, /* ` - g */ |
84 | 0, 0, 0, 0, 0, 0, '\n', 0, /* h - o */ |
85 | 0, 0, '\r', -ESC_s, '\t', 0, '\v', -ESC_w, /* p - w */ |
86 | 0, 0, 0 /* x - z */ |
87 | }; |
88 | |
89 | /* Error code numbers. They are given names so that they can more easily be |
90 | tracked. */ |
91 | |
92 | enum ErrorCode { |
93 | ERR0, ERR1, ERR2, ERR3, ERR4, ERR5, ERR6, ERR7, ERR8, ERR9, |
94 | ERR10, ERR11, ERR12, ERR13, ERR14, ERR15, ERR16, ERR17 |
95 | }; |
96 | |
97 | /* The texts of compile-time error messages. These are "char *" because they |
98 | are passed to the outside world. */ |
99 | |
100 | static const char* errorText(ErrorCode code) |
101 | { |
102 | static const char errorTexts[] = |
103 | /* 1 */ |
104 | "\\ at end of pattern\0" |
105 | "\\c at end of pattern\0" |
106 | "character value in \\x{...} sequence is too large\0" |
107 | "numbers out of order in {} quantifier\0" |
108 | /* 5 */ |
109 | "number too big in {} quantifier\0" |
110 | "missing terminating ] for character class\0" |
111 | "internal error: code overflow\0" |
112 | "range out of order in character class\0" |
113 | "nothing to repeat\0" |
114 | /* 10 */ |
115 | "unmatched parentheses\0" |
116 | "internal error: unexpected repeat\0" |
117 | "unrecognized character after (?\0" |
118 | "failed to get memory\0" |
119 | "missing )\0" |
120 | /* 15 */ |
121 | "reference to non-existent subpattern\0" |
122 | "regular expression too large\0" |
123 | "parentheses nested too deeply" |
124 | ; |
125 | |
126 | int i = code; |
127 | const char* text = errorTexts; |
128 | while (i > 1) |
129 | i -= !*text++; |
130 | return text; |
131 | } |
132 | |
133 | /* Structure for passing "static" information around between the functions |
134 | doing the compiling. */ |
135 | |
136 | struct CompileData { |
137 | CompileData() { |
138 | topBackref = 0; |
139 | backrefMap = 0; |
140 | reqVaryOpt = 0; |
141 | needOuterBracket = false; |
142 | numCapturingBrackets = 0; |
143 | } |
144 | int topBackref; /* Maximum back reference */ |
145 | unsigned backrefMap; /* Bitmap of low back refs */ |
146 | int reqVaryOpt; /* "After variable item" flag for reqByte */ |
147 | bool needOuterBracket; |
148 | int numCapturingBrackets; |
149 | }; |
150 | |
151 | /* Definitions to allow mutual recursion */ |
152 | |
153 | static bool compileBracket(int, int*, unsigned char**, const UChar**, const UChar*, ErrorCode*, int, int*, int*, CompileData&); |
154 | static bool bracketIsAnchored(const unsigned char* code); |
155 | static bool bracketNeedsLineStart(const unsigned char* code, unsigned captureMap, unsigned backrefMap); |
156 | static int bracketFindFirstAssertedCharacter(const unsigned char* code, bool inassert); |
157 | |
158 | /************************************************* |
159 | * Handle escapes * |
160 | *************************************************/ |
161 | |
162 | /* This function is called when a \ has been encountered. It either returns a |
163 | positive value for a simple escape such as \n, or a negative value which |
164 | encodes one of the more complicated things such as \d. When UTF-8 is enabled, |
165 | a positive value greater than 255 may be returned. On entry, ptr is pointing at |
166 | the \. On exit, it is on the final character of the escape sequence. |
167 | |
168 | Arguments: |
169 | ptrPtr points to the pattern position pointer |
170 | errorCodePtr points to the errorcode variable |
171 | bracount number of previous extracting brackets |
172 | options the options bits |
173 | isClass true if inside a character class |
174 | |
175 | Returns: zero or positive => a data character |
176 | negative => a special escape sequence |
177 | on error, errorPtr is set |
178 | */ |
179 | |
180 | static int checkEscape(const UChar** ptrPtr, const UChar* patternEnd, ErrorCode* errorCodePtr, int bracount, bool isClass) |
181 | { |
182 | const UChar* ptr = *ptrPtr + 1; |
183 | |
184 | /* If backslash is at the end of the pattern, it's an error. */ |
185 | if (ptr == patternEnd) { |
186 | *errorCodePtr = ERR1; |
187 | *ptrPtr = ptr; |
188 | return 0; |
189 | } |
190 | |
191 | int c = *ptr; |
192 | |
193 | /* Non-alphamerics are literals. For digits or letters, do an initial lookup in |
194 | a table. A non-zero result is something that can be returned immediately. |
195 | Otherwise further processing may be required. */ |
196 | |
197 | if (c < '0' || c > 'z') { /* Not alphameric */ |
198 | } else if (int escapeValue = escapes[c - '0']) { |
199 | c = escapeValue; |
200 | if (isClass) { |
201 | if (-c == ESC_b) |
202 | c = '\b'; /* \b is backslash in a class */ |
203 | else if (-c == ESC_B) |
204 | c = 'B'; /* and \B is a capital B in a class (in browsers event though ECMAScript 15.10.2.19 says it raises an error) */ |
205 | } |
206 | /* Escapes that need further processing, or are illegal. */ |
207 | |
208 | } else { |
209 | switch (c) { |
210 | case '1': |
211 | case '2': |
212 | case '3': |
213 | case '4': |
214 | case '5': |
215 | case '6': |
216 | case '7': |
217 | case '8': |
218 | case '9': |
219 | /* Escape sequences starting with a non-zero digit are backreferences, |
220 | unless there are insufficient brackets, in which case they are octal |
221 | escape sequences. Those sequences end on the first non-octal character |
222 | or when we overflow 0-255, whichever comes first. */ |
223 | |
224 | if (!isClass) { |
225 | const UChar* oldptr = ptr; |
226 | c -= '0'; |
227 | while ((ptr + 1 < patternEnd) && isASCIIDigit(c: ptr[1]) && c <= bracount) |
228 | c = c * 10 + *(++ptr) - '0'; |
229 | if (c <= bracount) { |
230 | c = -(ESC_REF + c); |
231 | break; |
232 | } |
233 | ptr = oldptr; /* Put the pointer back and fall through */ |
234 | } |
235 | |
236 | /* Handle an octal number following \. If the first digit is 8 or 9, |
237 | this is not octal. */ |
238 | |
239 | if ((c = *ptr) >= '8') { |
240 | c = '\\'; |
241 | ptr -= 1; |
242 | break; |
243 | } |
244 | |
245 | /* \0 always starts an octal number, but we may drop through to here with a |
246 | larger first octal digit. */ |
247 | |
248 | case '0': { |
249 | c -= '0'; |
250 | int i; |
251 | for (i = 1; i <= 2; ++i) { |
252 | if (ptr + i >= patternEnd || ptr[i] < '0' || ptr[i] > '7') |
253 | break; |
254 | int cc = c * 8 + ptr[i] - '0'; |
255 | if (cc > 255) |
256 | break; |
257 | c = cc; |
258 | } |
259 | ptr += i - 1; |
260 | break; |
261 | } |
262 | |
263 | case 'x': { |
264 | c = 0; |
265 | int i; |
266 | for (i = 1; i <= 2; ++i) { |
267 | if (ptr + i >= patternEnd || !isASCIIHexDigit(c: ptr[i])) { |
268 | c = 'x'; |
269 | i = 1; |
270 | break; |
271 | } |
272 | int cc = ptr[i]; |
273 | if (cc >= 'a') |
274 | cc -= 32; /* Convert to upper case */ |
275 | c = c * 16 + cc - ((cc < 'A') ? '0' : ('A' - 10)); |
276 | } |
277 | ptr += i - 1; |
278 | break; |
279 | } |
280 | |
281 | case 'u': { |
282 | c = 0; |
283 | int i; |
284 | for (i = 1; i <= 4; ++i) { |
285 | if (ptr + i >= patternEnd || !isASCIIHexDigit(c: ptr[i])) { |
286 | c = 'u'; |
287 | i = 1; |
288 | break; |
289 | } |
290 | int cc = ptr[i]; |
291 | if (cc >= 'a') |
292 | cc -= 32; /* Convert to upper case */ |
293 | c = c * 16 + cc - ((cc < 'A') ? '0' : ('A' - 10)); |
294 | } |
295 | ptr += i - 1; |
296 | break; |
297 | } |
298 | |
299 | case 'c': |
300 | if (++ptr == patternEnd) { |
301 | *errorCodePtr = ERR2; |
302 | return 0; |
303 | } |
304 | |
305 | c = *ptr; |
306 | |
307 | /* To match Firefox, inside a character class, we also accept |
308 | numbers and '_' as control characters */ |
309 | if ((!isClass && !isASCIIAlpha(c)) || (!isASCIIAlphanumeric(c) && c != '_')) { |
310 | c = '\\'; |
311 | ptr -= 2; |
312 | break; |
313 | } |
314 | |
315 | /* A letter is upper-cased; then the 0x40 bit is flipped. This coding |
316 | is ASCII-specific, but then the whole concept of \cx is ASCII-specific. */ |
317 | c = toASCIIUpper(c) ^ 0x40; |
318 | break; |
319 | } |
320 | } |
321 | |
322 | *ptrPtr = ptr; |
323 | return c; |
324 | } |
325 | |
326 | /************************************************* |
327 | * Check for counted repeat * |
328 | *************************************************/ |
329 | |
330 | /* This function is called when a '{' is encountered in a place where it might |
331 | start a quantifier. It looks ahead to see if it really is a quantifier or not. |
332 | It is only a quantifier if it is one of the forms {ddd} {ddd,} or {ddd,ddd} |
333 | where the ddds are digits. |
334 | |
335 | Arguments: |
336 | p pointer to the first char after '{' |
337 | |
338 | Returns: true or false |
339 | */ |
340 | |
341 | static bool isCountedRepeat(const UChar* p, const UChar* patternEnd) |
342 | { |
343 | if (p >= patternEnd || !isASCIIDigit(c: *p)) |
344 | return false; |
345 | p++; |
346 | while (p < patternEnd && isASCIIDigit(c: *p)) |
347 | p++; |
348 | if (p < patternEnd && *p == '}') |
349 | return true; |
350 | |
351 | if (p >= patternEnd || *p++ != ',') |
352 | return false; |
353 | if (p < patternEnd && *p == '}') |
354 | return true; |
355 | |
356 | if (p >= patternEnd || !isASCIIDigit(c: *p)) |
357 | return false; |
358 | p++; |
359 | while (p < patternEnd && isASCIIDigit(c: *p)) |
360 | p++; |
361 | |
362 | return (p < patternEnd && *p == '}'); |
363 | } |
364 | |
365 | /************************************************* |
366 | * Read repeat counts * |
367 | *************************************************/ |
368 | |
369 | /* Read an item of the form {n,m} and return the values. This is called only |
370 | after isCountedRepeat() has confirmed that a repeat-count quantifier exists, |
371 | so the syntax is guaranteed to be correct, but we need to check the values. |
372 | |
373 | Arguments: |
374 | p pointer to first char after '{' |
375 | minp pointer to int for min |
376 | maxp pointer to int for max |
377 | returned as -1 if no max |
378 | errorCodePtr points to error code variable |
379 | |
380 | Returns: pointer to '}' on success; |
381 | current ptr on error, with errorCodePtr set non-zero |
382 | */ |
383 | |
384 | static const UChar* readRepeatCounts(const UChar* p, int* minp, int* maxp, ErrorCode* errorCodePtr) |
385 | { |
386 | int min = 0; |
387 | int max = -1; |
388 | |
389 | /* Read the minimum value and do a paranoid check: a negative value indicates |
390 | an integer overflow. */ |
391 | |
392 | while (isASCIIDigit(c: *p)) |
393 | min = min * 10 + *p++ - '0'; |
394 | if (min < 0 || min > 65535) { |
395 | *errorCodePtr = ERR5; |
396 | return p; |
397 | } |
398 | |
399 | /* Read the maximum value if there is one, and again do a paranoid on its size. |
400 | Also, max must not be less than min. */ |
401 | |
402 | if (*p == '}') |
403 | max = min; |
404 | else { |
405 | if (*(++p) != '}') { |
406 | max = 0; |
407 | while (isASCIIDigit(c: *p)) |
408 | max = max * 10 + *p++ - '0'; |
409 | if (max < 0 || max > 65535) { |
410 | *errorCodePtr = ERR5; |
411 | return p; |
412 | } |
413 | if (max < min) { |
414 | *errorCodePtr = ERR4; |
415 | return p; |
416 | } |
417 | } |
418 | } |
419 | |
420 | /* Fill in the required variables, and pass back the pointer to the terminating |
421 | '}'. */ |
422 | |
423 | *minp = min; |
424 | *maxp = max; |
425 | return p; |
426 | } |
427 | |
428 | /************************************************* |
429 | * Find first significant op code * |
430 | *************************************************/ |
431 | |
432 | /* This is called by several functions that scan a compiled expression looking |
433 | for a fixed first character, or an anchoring op code etc. It skips over things |
434 | that do not influence this. |
435 | |
436 | Arguments: |
437 | code pointer to the start of the group |
438 | Returns: pointer to the first significant opcode |
439 | */ |
440 | |
441 | static const unsigned char* firstSignificantOpcode(const unsigned char* code) |
442 | { |
443 | while (*code == OP_BRANUMBER) |
444 | code += 3; |
445 | return code; |
446 | } |
447 | |
448 | static const unsigned char* firstSignificantOpcodeSkippingAssertions(const unsigned char* code) |
449 | { |
450 | while (true) { |
451 | switch (*code) { |
452 | case OP_ASSERT_NOT: |
453 | advanceToEndOfBracket(opcodePtr&: code); |
454 | code += 1 + LINK_SIZE; |
455 | break; |
456 | case OP_WORD_BOUNDARY: |
457 | case OP_NOT_WORD_BOUNDARY: |
458 | ++code; |
459 | break; |
460 | case OP_BRANUMBER: |
461 | code += 3; |
462 | break; |
463 | default: |
464 | return code; |
465 | } |
466 | } |
467 | } |
468 | |
469 | /************************************************* |
470 | * Get othercase range * |
471 | *************************************************/ |
472 | |
473 | /* This function is passed the start and end of a class range, in UTF-8 mode |
474 | with UCP support. It searches up the characters, looking for internal ranges of |
475 | characters in the "other" case. Each call returns the next one, updating the |
476 | start address. |
477 | |
478 | Arguments: |
479 | cptr points to starting character value; updated |
480 | d end value |
481 | ocptr where to put start of othercase range |
482 | odptr where to put end of othercase range |
483 | |
484 | Yield: true when range returned; false when no more |
485 | */ |
486 | |
487 | static bool getOthercaseRange(int* cptr, int d, int* ocptr, int* odptr) |
488 | { |
489 | int c, othercase = 0; |
490 | |
491 | for (c = *cptr; c <= d; c++) { |
492 | if ((othercase = jsc_pcre_ucp_othercase(c)) >= 0) |
493 | break; |
494 | } |
495 | |
496 | if (c > d) |
497 | return false; |
498 | |
499 | *ocptr = othercase; |
500 | int next = othercase + 1; |
501 | |
502 | for (++c; c <= d; c++) { |
503 | if (jsc_pcre_ucp_othercase(c) != next) |
504 | break; |
505 | next++; |
506 | } |
507 | |
508 | *odptr = next - 1; |
509 | *cptr = c; |
510 | |
511 | return true; |
512 | } |
513 | |
514 | /************************************************* |
515 | * Convert character value to UTF-8 * |
516 | *************************************************/ |
517 | |
518 | /* This function takes an integer value in the range 0 - 0x7fffffff |
519 | and encodes it as a UTF-8 character in 0 to 6 bytes. |
520 | |
521 | Arguments: |
522 | cvalue the character value |
523 | buffer pointer to buffer for result - at least 6 bytes long |
524 | |
525 | Returns: number of characters placed in the buffer |
526 | */ |
527 | |
528 | static int encodeUTF8(int cvalue, unsigned char *buffer) |
529 | { |
530 | int i; |
531 | for (i = 0; i < jsc_pcre_utf8_table1_size; i++) |
532 | if (cvalue <= jsc_pcre_utf8_table1[i]) |
533 | break; |
534 | buffer += i; |
535 | for (int j = i; j > 0; j--) { |
536 | *buffer-- = 0x80 | (cvalue & 0x3f); |
537 | cvalue >>= 6; |
538 | } |
539 | *buffer = jsc_pcre_utf8_table2[i] | cvalue; |
540 | return i + 1; |
541 | } |
542 | |
543 | /************************************************* |
544 | * Compile one branch * |
545 | *************************************************/ |
546 | |
547 | /* Scan the pattern, compiling it into the code vector. |
548 | |
549 | Arguments: |
550 | options the option bits |
551 | brackets points to number of extracting brackets used |
552 | codePtr points to the pointer to the current code point |
553 | ptrPtr points to the current pattern pointer |
554 | errorCodePtr points to error code variable |
555 | firstbyteptr set to initial literal character, or < 0 (REQ_UNSET, REQ_NONE) |
556 | reqbyteptr set to the last literal character required, else < 0 |
557 | cd contains pointers to tables etc. |
558 | |
559 | Returns: true on success |
560 | false, with *errorCodePtr set non-zero on error |
561 | */ |
562 | |
563 | static inline bool safelyCheckNextChar(const UChar* ptr, const UChar* patternEnd, UChar expected) |
564 | { |
565 | return ((ptr + 1 < patternEnd) && ptr[1] == expected); |
566 | } |
567 | |
568 | static bool |
569 | compileBranch(int options, int* brackets, unsigned char** codePtr, |
570 | const UChar** ptrPtr, const UChar* patternEnd, ErrorCode* errorCodePtr, int *firstbyteptr, |
571 | int* reqbyteptr, CompileData& cd) |
572 | { |
573 | int repeatType, opType; |
574 | int repeatMin = 0, repeat_max = 0; /* To please picky compilers */ |
575 | int bravalue = 0; |
576 | int reqvary, tempreqvary; |
577 | int c; |
578 | unsigned char* code = *codePtr; |
579 | unsigned char* tempcode; |
580 | bool didGroupSetFirstByte = false; |
581 | const UChar* ptr = *ptrPtr; |
582 | unsigned char* previous = NULL; |
583 | unsigned char classbits[32]; |
584 | |
585 | bool class_utf8; |
586 | unsigned char* class_utf8data; |
587 | unsigned char utf8_char[6]; |
588 | |
589 | /* Initialize no first byte, no required byte. REQ_UNSET means "no char |
590 | matching encountered yet". It gets changed to REQ_NONE if we hit something that |
591 | matches a non-fixed char first char; reqByte just remains unset if we never |
592 | find one. |
593 | |
594 | When we hit a repeat whose minimum is zero, we may have to adjust these values |
595 | to take the zero repeat into account. This is implemented by setting them to |
596 | zeroFirstByte and zeroReqByte when such a repeat is encountered. The individual |
597 | item types that can be repeated set these backoff variables appropriately. */ |
598 | |
599 | int firstByte = REQ_UNSET; |
600 | int reqByte = REQ_UNSET; |
601 | int zeroReqByte = REQ_UNSET; |
602 | int zeroFirstByte = REQ_UNSET; |
603 | |
604 | /* The variable reqCaseOpt contains either the REQ_IGNORE_CASE value or zero, |
605 | according to the current setting of the ignores-case flag. REQ_IGNORE_CASE is a bit |
606 | value > 255. It is added into the firstByte or reqByte variables to record the |
607 | case status of the value. This is used only for ASCII characters. */ |
608 | |
609 | int reqCaseOpt = (options & IgnoreCaseOption) ? REQ_IGNORE_CASE : 0; |
610 | |
611 | /* Switch on next character until the end of the branch */ |
612 | |
613 | for (;; ptr++) { |
614 | bool negateClass; |
615 | bool shouldFlipNegation; /* If a negative special such as \S is used, we should negate the whole class to properly support Unicode. */ |
616 | int classCharCount; |
617 | int classLastChar; |
618 | int skipBytes; |
619 | int subReqByte; |
620 | int subFirstByte; |
621 | int mcLength; |
622 | unsigned char mcbuffer[8]; |
623 | |
624 | /* Next byte in the pattern */ |
625 | |
626 | c = ptr < patternEnd ? *ptr : 0; |
627 | |
628 | /* Fill in length of a previous callout, except when the next thing is |
629 | a quantifier. */ |
630 | |
631 | bool isQuantifier = c == '*' || c == '+' || c == '?' || (c == '{' && isCountedRepeat(p: ptr + 1, patternEnd)); |
632 | |
633 | switch (c) { |
634 | /* The branch terminates at end of string, |, or ). */ |
635 | |
636 | case 0: |
637 | if (ptr < patternEnd) |
638 | goto NORMAL_CHAR; |
639 | // End of string; fall through |
640 | case '|': |
641 | case ')': |
642 | *firstbyteptr = firstByte; |
643 | *reqbyteptr = reqByte; |
644 | *codePtr = code; |
645 | *ptrPtr = ptr; |
646 | return true; |
647 | |
648 | /* Handle single-character metacharacters. In multiline mode, ^ disables |
649 | the setting of any following char as a first character. */ |
650 | |
651 | case '^': |
652 | if (options & MatchAcrossMultipleLinesOption) { |
653 | if (firstByte == REQ_UNSET) |
654 | firstByte = REQ_NONE; |
655 | *code++ = OP_BOL; |
656 | } else |
657 | *code++ = OP_CIRC; |
658 | previous = NULL; |
659 | break; |
660 | |
661 | case '$': |
662 | previous = NULL; |
663 | if (options & MatchAcrossMultipleLinesOption) |
664 | *code++ = OP_EOL; |
665 | else |
666 | *code++ = OP_DOLL; |
667 | break; |
668 | |
669 | /* There can never be a first char if '.' is first, whatever happens about |
670 | repeats. The value of reqByte doesn't change either. */ |
671 | |
672 | case '.': |
673 | if (firstByte == REQ_UNSET) |
674 | firstByte = REQ_NONE; |
675 | zeroFirstByte = firstByte; |
676 | zeroReqByte = reqByte; |
677 | previous = code; |
678 | *code++ = OP_NOT_NEWLINE; |
679 | break; |
680 | |
681 | /* Character classes. If the included characters are all < 256, we build a |
682 | 32-byte bitmap of the permitted characters, except in the special case |
683 | where there is only one such character. For negated classes, we build the |
684 | map as usual, then invert it at the end. However, we use a different opcode |
685 | so that data characters > 255 can be handled correctly. |
686 | |
687 | If the class contains characters outside the 0-255 range, a different |
688 | opcode is compiled. It may optionally have a bit map for characters < 256, |
689 | but those above are are explicitly listed afterwards. A flag byte tells |
690 | whether the bitmap is present, and whether this is a negated class or not. |
691 | */ |
692 | |
693 | case '[': { |
694 | previous = code; |
695 | shouldFlipNegation = false; |
696 | |
697 | /* PCRE supports POSIX class stuff inside a class. Perl gives an error if |
698 | they are encountered at the top level, so we'll do that too. */ |
699 | |
700 | /* If the first character is '^', set the negation flag and skip it. */ |
701 | |
702 | if (ptr + 1 >= patternEnd) { |
703 | *errorCodePtr = ERR6; |
704 | return false; |
705 | } |
706 | |
707 | if (ptr[1] == '^') { |
708 | negateClass = true; |
709 | ++ptr; |
710 | } else |
711 | negateClass = false; |
712 | |
713 | /* Keep a count of chars with values < 256 so that we can optimize the case |
714 | of just a single character (as long as it's < 256). For higher valued UTF-8 |
715 | characters, we don't yet do any optimization. */ |
716 | |
717 | classCharCount = 0; |
718 | classLastChar = -1; |
719 | |
720 | class_utf8 = false; /* No chars >= 256 */ |
721 | class_utf8data = code + LINK_SIZE + 34; /* For UTF-8 items */ |
722 | |
723 | /* Initialize the 32-char bit map to all zeros. We have to build the |
724 | map in a temporary bit of store, in case the class contains only 1 |
725 | character (< 256), because in that case the compiled code doesn't use the |
726 | bit map. */ |
727 | |
728 | memset(s: classbits, c: 0, n: 32 * sizeof(unsigned char)); |
729 | |
730 | /* Process characters until ] is reached. The first pass |
731 | through the regex checked the overall syntax, so we don't need to be very |
732 | strict here. At the start of the loop, c contains the first byte of the |
733 | character. */ |
734 | |
735 | while ((++ptr < patternEnd) && (c = *ptr) != ']') { |
736 | /* Backslash may introduce a single character, or it may introduce one |
737 | of the specials, which just set a flag. Escaped items are checked for |
738 | validity in the pre-compiling pass. The sequence \b is a special case. |
739 | Inside a class (and only there) it is treated as backspace. Elsewhere |
740 | it marks a word boundary. Other escapes have preset maps ready to |
741 | or into the one we are building. We assume they have more than one |
742 | character in them, so set classCharCount bigger than one. */ |
743 | |
744 | if (c == '\\') { |
745 | c = checkEscape(ptrPtr: &ptr, patternEnd, errorCodePtr, bracount: cd.numCapturingBrackets, isClass: true); |
746 | if (c < 0) { |
747 | classCharCount += 2; /* Greater than 1 is what matters */ |
748 | switch (-c) { |
749 | case ESC_d: |
750 | for (c = 0; c < 32; c++) |
751 | classbits[c] |= classBitmapForChar(c: c + cbit_digit); |
752 | continue; |
753 | |
754 | case ESC_D: |
755 | shouldFlipNegation = true; |
756 | for (c = 0; c < 32; c++) |
757 | classbits[c] |= ~classBitmapForChar(c: c + cbit_digit); |
758 | continue; |
759 | |
760 | case ESC_w: |
761 | for (c = 0; c < 32; c++) |
762 | classbits[c] |= classBitmapForChar(c: c + cbit_word); |
763 | continue; |
764 | |
765 | case ESC_W: |
766 | shouldFlipNegation = true; |
767 | for (c = 0; c < 32; c++) |
768 | classbits[c] |= ~classBitmapForChar(c: c + cbit_word); |
769 | continue; |
770 | |
771 | case ESC_s: |
772 | for (c = 0; c < 32; c++) |
773 | classbits[c] |= classBitmapForChar(c: c + cbit_space); |
774 | continue; |
775 | |
776 | case ESC_S: |
777 | shouldFlipNegation = true; |
778 | for (c = 0; c < 32; c++) |
779 | classbits[c] |= ~classBitmapForChar(c: c + cbit_space); |
780 | continue; |
781 | |
782 | /* Unrecognized escapes are faulted if PCRE is running in its |
783 | strict mode. By default, for compatibility with Perl, they are |
784 | treated as literals. */ |
785 | |
786 | default: |
787 | c = *ptr; /* The final character */ |
788 | classCharCount -= 2; /* Undo the default count from above */ |
789 | } |
790 | } |
791 | |
792 | /* Fall through if we have a single character (c >= 0). This may be |
793 | > 256 in UTF-8 mode. */ |
794 | |
795 | } /* End of backslash handling */ |
796 | |
797 | /* A single character may be followed by '-' to form a range. However, |
798 | Perl does not permit ']' to be the end of the range. A '-' character |
799 | here is treated as a literal. */ |
800 | |
801 | if ((ptr + 2 < patternEnd) && ptr[1] == '-' && ptr[2] != ']') { |
802 | ptr += 2; |
803 | |
804 | int d = *ptr; |
805 | |
806 | /* The second part of a range can be a single-character escape, but |
807 | not any of the other escapes. Perl 5.6 treats a hyphen as a literal |
808 | in such circumstances. */ |
809 | |
810 | if (d == '\\') { |
811 | const UChar* oldptr = ptr; |
812 | d = checkEscape(ptrPtr: &ptr, patternEnd, errorCodePtr, bracount: cd.numCapturingBrackets, isClass: true); |
813 | |
814 | /* \X is literal X; any other special means the '-' was literal */ |
815 | if (d < 0) { |
816 | ptr = oldptr - 2; |
817 | goto LONE_SINGLE_CHARACTER; /* A few lines below */ |
818 | } |
819 | } |
820 | |
821 | /* The check that the two values are in the correct order happens in |
822 | the pre-pass. Optimize one-character ranges */ |
823 | |
824 | if (d == c) |
825 | goto LONE_SINGLE_CHARACTER; /* A few lines below */ |
826 | |
827 | /* In UTF-8 mode, if the upper limit is > 255, or > 127 for caseless |
828 | matching, we have to use an XCLASS with extra data items. Caseless |
829 | matching for characters > 127 is available only if UCP support is |
830 | available. */ |
831 | |
832 | if ((d > 255 || ((options & IgnoreCaseOption) && d > 127))) { |
833 | class_utf8 = true; |
834 | |
835 | /* With UCP support, we can find the other case equivalents of |
836 | the relevant characters. There may be several ranges. Optimize how |
837 | they fit with the basic range. */ |
838 | |
839 | if (options & IgnoreCaseOption) { |
840 | int occ, ocd; |
841 | int cc = c; |
842 | int origd = d; |
843 | while (getOthercaseRange(cptr: &cc, d: origd, ocptr: &occ, odptr: &ocd)) { |
844 | if (occ >= c && ocd <= d) |
845 | continue; /* Skip embedded ranges */ |
846 | |
847 | if (occ < c && ocd >= c - 1) /* Extend the basic range */ |
848 | { /* if there is overlap, */ |
849 | c = occ; /* noting that if occ < c */ |
850 | continue; /* we can't have ocd > d */ |
851 | } /* because a subrange is */ |
852 | if (ocd > d && occ <= d + 1) /* always shorter than */ |
853 | { /* the basic range. */ |
854 | d = ocd; |
855 | continue; |
856 | } |
857 | |
858 | if (occ == ocd) |
859 | *class_utf8data++ = XCL_SINGLE; |
860 | else { |
861 | *class_utf8data++ = XCL_RANGE; |
862 | class_utf8data += encodeUTF8(cvalue: occ, buffer: class_utf8data); |
863 | } |
864 | class_utf8data += encodeUTF8(cvalue: ocd, buffer: class_utf8data); |
865 | } |
866 | } |
867 | |
868 | /* Now record the original range, possibly modified for UCP caseless |
869 | overlapping ranges. */ |
870 | |
871 | *class_utf8data++ = XCL_RANGE; |
872 | class_utf8data += encodeUTF8(cvalue: c, buffer: class_utf8data); |
873 | class_utf8data += encodeUTF8(cvalue: d, buffer: class_utf8data); |
874 | |
875 | /* With UCP support, we are done. Without UCP support, there is no |
876 | caseless matching for UTF-8 characters > 127; we can use the bit map |
877 | for the smaller ones. */ |
878 | |
879 | continue; /* With next character in the class */ |
880 | } |
881 | |
882 | /* We use the bit map for all cases when not in UTF-8 mode; else |
883 | ranges that lie entirely within 0-127 when there is UCP support; else |
884 | for partial ranges without UCP support. */ |
885 | |
886 | for (; c <= d; c++) { |
887 | classbits[c/8] |= (1 << (c&7)); |
888 | if (options & IgnoreCaseOption) { |
889 | int uc = flipCase(c); |
890 | classbits[uc/8] |= (1 << (uc&7)); |
891 | } |
892 | classCharCount++; /* in case a one-char range */ |
893 | classLastChar = c; |
894 | } |
895 | |
896 | continue; /* Go get the next char in the class */ |
897 | } |
898 | |
899 | /* Handle a lone single character - we can get here for a normal |
900 | non-escape char, or after \ that introduces a single character or for an |
901 | apparent range that isn't. */ |
902 | |
903 | LONE_SINGLE_CHARACTER: |
904 | |
905 | /* Handle a character that cannot go in the bit map */ |
906 | |
907 | if ((c > 255 || ((options & IgnoreCaseOption) && c > 127))) { |
908 | class_utf8 = true; |
909 | *class_utf8data++ = XCL_SINGLE; |
910 | class_utf8data += encodeUTF8(cvalue: c, buffer: class_utf8data); |
911 | |
912 | if (options & IgnoreCaseOption) { |
913 | int othercase; |
914 | if ((othercase = jsc_pcre_ucp_othercase(c)) >= 0) { |
915 | *class_utf8data++ = XCL_SINGLE; |
916 | class_utf8data += encodeUTF8(cvalue: othercase, buffer: class_utf8data); |
917 | } |
918 | } |
919 | } else { |
920 | /* Handle a single-byte character */ |
921 | classbits[c/8] |= (1 << (c&7)); |
922 | if (options & IgnoreCaseOption) { |
923 | c = flipCase(c); |
924 | classbits[c/8] |= (1 << (c&7)); |
925 | } |
926 | classCharCount++; |
927 | classLastChar = c; |
928 | } |
929 | } |
930 | |
931 | /* If classCharCount is 1, we saw precisely one character whose value is |
932 | less than 256. In non-UTF-8 mode we can always optimize. In UTF-8 mode, we |
933 | can optimize the negative case only if there were no characters >= 128 |
934 | because OP_NOT and the related opcodes like OP_NOTSTAR operate on |
935 | single-bytes only. This is an historical hangover. Maybe one day we can |
936 | tidy these opcodes to handle multi-byte characters. |
937 | |
938 | The optimization throws away the bit map. We turn the item into a |
939 | 1-character OP_CHAR[NC] if it's positive, or OP_NOT if it's negative. Note |
940 | that OP_NOT does not support multibyte characters. In the positive case, it |
941 | can cause firstByte to be set. Otherwise, there can be no first char if |
942 | this item is first, whatever repeat count may follow. In the case of |
943 | reqByte, save the previous value for reinstating. */ |
944 | |
945 | if (classCharCount == 1 && (!class_utf8 && (!negateClass || classLastChar < 128))) { |
946 | zeroReqByte = reqByte; |
947 | |
948 | /* The OP_NOT opcode works on one-byte characters only. */ |
949 | |
950 | if (negateClass) { |
951 | if (firstByte == REQ_UNSET) |
952 | firstByte = REQ_NONE; |
953 | zeroFirstByte = firstByte; |
954 | *code++ = OP_NOT; |
955 | *code++ = classLastChar; |
956 | break; |
957 | } |
958 | |
959 | /* For a single, positive character, get the value into c, and |
960 | then we can handle this with the normal one-character code. */ |
961 | |
962 | c = classLastChar; |
963 | goto NORMAL_CHAR; |
964 | } /* End of 1-char optimization */ |
965 | |
966 | /* The general case - not the one-char optimization. If this is the first |
967 | thing in the branch, there can be no first char setting, whatever the |
968 | repeat count. Any reqByte setting must remain unchanged after any kind of |
969 | repeat. */ |
970 | |
971 | if (firstByte == REQ_UNSET) firstByte = REQ_NONE; |
972 | zeroFirstByte = firstByte; |
973 | zeroReqByte = reqByte; |
974 | |
975 | /* If there are characters with values > 255, we have to compile an |
976 | extended class, with its own opcode. If there are no characters < 256, |
977 | we can omit the bitmap. */ |
978 | |
979 | if (class_utf8 && !shouldFlipNegation) { |
980 | *class_utf8data++ = XCL_END; /* Marks the end of extra data */ |
981 | *code++ = OP_XCLASS; |
982 | code += LINK_SIZE; |
983 | *code = negateClass? XCL_NOT : 0; |
984 | |
985 | /* If the map is required, install it, and move on to the end of |
986 | the extra data */ |
987 | |
988 | if (classCharCount > 0) { |
989 | *code++ |= XCL_MAP; |
990 | memcpy(dest: code, src: classbits, n: 32); |
991 | code = class_utf8data; |
992 | } |
993 | |
994 | /* If the map is not required, slide down the extra data. */ |
995 | |
996 | else { |
997 | int len = class_utf8data - (code + 33); |
998 | memmove(dest: code + 1, src: code + 33, n: len); |
999 | code += len + 1; |
1000 | } |
1001 | |
1002 | /* Now fill in the complete length of the item */ |
1003 | |
1004 | putLinkValue(opcodePtr: previous + 1, value: code - previous); |
1005 | break; /* End of class handling */ |
1006 | } |
1007 | |
1008 | /* If there are no characters > 255, negate the 32-byte map if necessary, |
1009 | and copy it into the code vector. If this is the first thing in the branch, |
1010 | there can be no first char setting, whatever the repeat count. Any reqByte |
1011 | setting must remain unchanged after any kind of repeat. */ |
1012 | |
1013 | *code++ = (negateClass == shouldFlipNegation) ? OP_CLASS : OP_NCLASS; |
1014 | if (negateClass) |
1015 | for (c = 0; c < 32; c++) |
1016 | code[c] = ~classbits[c]; |
1017 | else |
1018 | memcpy(dest: code, src: classbits, n: 32); |
1019 | code += 32; |
1020 | break; |
1021 | } |
1022 | |
1023 | /* Various kinds of repeat; '{' is not necessarily a quantifier, but this |
1024 | has been tested above. */ |
1025 | |
1026 | case '{': |
1027 | if (!isQuantifier) |
1028 | goto NORMAL_CHAR; |
1029 | ptr = readRepeatCounts(p: ptr + 1, minp: &repeatMin, maxp: &repeat_max, errorCodePtr); |
1030 | if (*errorCodePtr) |
1031 | goto FAILED; |
1032 | goto REPEAT; |
1033 | |
1034 | case '*': |
1035 | repeatMin = 0; |
1036 | repeat_max = -1; |
1037 | goto REPEAT; |
1038 | |
1039 | case '+': |
1040 | repeatMin = 1; |
1041 | repeat_max = -1; |
1042 | goto REPEAT; |
1043 | |
1044 | case '?': |
1045 | repeatMin = 0; |
1046 | repeat_max = 1; |
1047 | |
1048 | REPEAT: |
1049 | if (!previous) { |
1050 | *errorCodePtr = ERR9; |
1051 | goto FAILED; |
1052 | } |
1053 | |
1054 | if (repeatMin == 0) { |
1055 | firstByte = zeroFirstByte; /* Adjust for zero repeat */ |
1056 | reqByte = zeroReqByte; /* Ditto */ |
1057 | } |
1058 | |
1059 | /* Remember whether this is a variable length repeat */ |
1060 | |
1061 | reqvary = (repeatMin == repeat_max) ? 0 : REQ_VARY; |
1062 | |
1063 | opType = 0; /* Default single-char op codes */ |
1064 | |
1065 | /* Save start of previous item, in case we have to move it up to make space |
1066 | for an inserted OP_ONCE for the additional '+' extension. */ |
1067 | /* FIXME: Probably don't need this because we don't use OP_ONCE. */ |
1068 | |
1069 | tempcode = previous; |
1070 | |
1071 | /* If the next character is '+', we have a possessive quantifier. This |
1072 | implies greediness, whatever the setting of the PCRE_UNGREEDY option. |
1073 | If the next character is '?' this is a minimizing repeat, by default, |
1074 | but if PCRE_UNGREEDY is set, it works the other way round. We change the |
1075 | repeat type to the non-default. */ |
1076 | |
1077 | if (safelyCheckNextChar(ptr, patternEnd, expected: '?')) { |
1078 | repeatType = 1; |
1079 | ptr++; |
1080 | } else |
1081 | repeatType = 0; |
1082 | |
1083 | /* If previous was a character match, abolish the item and generate a |
1084 | repeat item instead. If a char item has a minumum of more than one, ensure |
1085 | that it is set in reqByte - it might not be if a sequence such as x{3} is |
1086 | the first thing in a branch because the x will have gone into firstByte |
1087 | instead. */ |
1088 | |
1089 | if (*previous == OP_CHAR || *previous == OP_CHAR_IGNORING_CASE) { |
1090 | /* Deal with UTF-8 characters that take up more than one byte. It's |
1091 | easier to write this out separately than try to macrify it. Use c to |
1092 | hold the length of the character in bytes, plus 0x80 to flag that it's a |
1093 | length rather than a small character. */ |
1094 | |
1095 | if (code[-1] & 0x80) { |
1096 | unsigned char *lastchar = code - 1; |
1097 | while((*lastchar & 0xc0) == 0x80) |
1098 | lastchar--; |
1099 | c = code - lastchar; /* Length of UTF-8 character */ |
1100 | memcpy(dest: utf8_char, src: lastchar, n: c); /* Save the char */ |
1101 | c |= 0x80; /* Flag c as a length */ |
1102 | } |
1103 | else { |
1104 | c = code[-1]; |
1105 | if (repeatMin > 1) |
1106 | reqByte = c | reqCaseOpt | cd.reqVaryOpt; |
1107 | } |
1108 | |
1109 | goto OUTPUT_SINGLE_REPEAT; /* Code shared with single character types */ |
1110 | } |
1111 | |
1112 | else if (*previous == OP_ASCII_CHAR || *previous == OP_ASCII_LETTER_IGNORING_CASE) { |
1113 | c = previous[1]; |
1114 | if (repeatMin > 1) |
1115 | reqByte = c | reqCaseOpt | cd.reqVaryOpt; |
1116 | goto OUTPUT_SINGLE_REPEAT; |
1117 | } |
1118 | |
1119 | /* If previous was a single negated character ([^a] or similar), we use |
1120 | one of the special opcodes, replacing it. The code is shared with single- |
1121 | character repeats by setting opt_type to add a suitable offset into |
1122 | repeatType. OP_NOT is currently used only for single-byte chars. */ |
1123 | |
1124 | else if (*previous == OP_NOT) { |
1125 | opType = OP_NOTSTAR - OP_STAR; /* Use "not" opcodes */ |
1126 | c = previous[1]; |
1127 | goto OUTPUT_SINGLE_REPEAT; |
1128 | } |
1129 | |
1130 | /* If previous was a character type match (\d or similar), abolish it and |
1131 | create a suitable repeat item. The code is shared with single-character |
1132 | repeats by setting opType to add a suitable offset into repeatType. */ |
1133 | |
1134 | else if (*previous <= OP_NOT_NEWLINE) { |
1135 | opType = OP_TYPESTAR - OP_STAR; /* Use type opcodes */ |
1136 | c = *previous; |
1137 | |
1138 | OUTPUT_SINGLE_REPEAT: |
1139 | int prop_type = -1; |
1140 | int prop_value = -1; |
1141 | |
1142 | unsigned char* oldcode = code; |
1143 | code = previous; /* Usually overwrite previous item */ |
1144 | |
1145 | /* If the maximum is zero then the minimum must also be zero; Perl allows |
1146 | this case, so we do too - by simply omitting the item altogether. */ |
1147 | |
1148 | if (repeat_max == 0) |
1149 | goto END_REPEAT; |
1150 | |
1151 | /* Combine the opType with the repeatType */ |
1152 | |
1153 | repeatType += opType; |
1154 | |
1155 | /* A minimum of zero is handled either as the special case * or ?, or as |
1156 | an UPTO, with the maximum given. */ |
1157 | |
1158 | if (repeatMin == 0) { |
1159 | if (repeat_max == -1) |
1160 | *code++ = OP_STAR + repeatType; |
1161 | else if (repeat_max == 1) |
1162 | *code++ = OP_QUERY + repeatType; |
1163 | else { |
1164 | *code++ = OP_UPTO + repeatType; |
1165 | put2ByteValueAndAdvance(opcodePtr&: code, value: repeat_max); |
1166 | } |
1167 | } |
1168 | |
1169 | /* A repeat minimum of 1 is optimized into some special cases. If the |
1170 | maximum is unlimited, we use OP_PLUS. Otherwise, the original item it |
1171 | left in place and, if the maximum is greater than 1, we use OP_UPTO with |
1172 | one less than the maximum. */ |
1173 | |
1174 | else if (repeatMin == 1) { |
1175 | if (repeat_max == -1) |
1176 | *code++ = OP_PLUS + repeatType; |
1177 | else { |
1178 | code = oldcode; /* leave previous item in place */ |
1179 | if (repeat_max == 1) |
1180 | goto END_REPEAT; |
1181 | *code++ = OP_UPTO + repeatType; |
1182 | put2ByteValueAndAdvance(opcodePtr&: code, value: repeat_max - 1); |
1183 | } |
1184 | } |
1185 | |
1186 | /* The case {n,n} is just an EXACT, while the general case {n,m} is |
1187 | handled as an EXACT followed by an UPTO. */ |
1188 | |
1189 | else { |
1190 | *code++ = OP_EXACT + opType; /* NB EXACT doesn't have repeatType */ |
1191 | put2ByteValueAndAdvance(opcodePtr&: code, value: repeatMin); |
1192 | |
1193 | /* If the maximum is unlimited, insert an OP_STAR. Before doing so, |
1194 | we have to insert the character for the previous code. For a repeated |
1195 | Unicode property match, there are two extra bytes that define the |
1196 | required property. In UTF-8 mode, long characters have their length in |
1197 | c, with the 0x80 bit as a flag. */ |
1198 | |
1199 | if (repeat_max < 0) { |
1200 | if (c >= 128) { |
1201 | memcpy(dest: code, src: utf8_char, n: c & 7); |
1202 | code += c & 7; |
1203 | } else { |
1204 | *code++ = c; |
1205 | if (prop_type >= 0) { |
1206 | *code++ = prop_type; |
1207 | *code++ = prop_value; |
1208 | } |
1209 | } |
1210 | *code++ = OP_STAR + repeatType; |
1211 | } |
1212 | |
1213 | /* Else insert an UPTO if the max is greater than the min, again |
1214 | preceded by the character, for the previously inserted code. */ |
1215 | |
1216 | else if (repeat_max != repeatMin) { |
1217 | if (c >= 128) { |
1218 | memcpy(dest: code, src: utf8_char, n: c & 7); |
1219 | code += c & 7; |
1220 | } else |
1221 | *code++ = c; |
1222 | if (prop_type >= 0) { |
1223 | *code++ = prop_type; |
1224 | *code++ = prop_value; |
1225 | } |
1226 | repeat_max -= repeatMin; |
1227 | *code++ = OP_UPTO + repeatType; |
1228 | put2ByteValueAndAdvance(opcodePtr&: code, value: repeat_max); |
1229 | } |
1230 | } |
1231 | |
1232 | /* The character or character type itself comes last in all cases. */ |
1233 | |
1234 | if (c >= 128) { |
1235 | memcpy(dest: code, src: utf8_char, n: c & 7); |
1236 | code += c & 7; |
1237 | } else |
1238 | *code++ = c; |
1239 | |
1240 | /* For a repeated Unicode property match, there are two extra bytes that |
1241 | define the required property. */ |
1242 | |
1243 | if (prop_type >= 0) { |
1244 | *code++ = prop_type; |
1245 | *code++ = prop_value; |
1246 | } |
1247 | } |
1248 | |
1249 | /* If previous was a character class or a back reference, we put the repeat |
1250 | stuff after it, but just skip the item if the repeat was {0,0}. */ |
1251 | |
1252 | else if (*previous == OP_CLASS || |
1253 | *previous == OP_NCLASS || |
1254 | *previous == OP_XCLASS || |
1255 | *previous == OP_REF) |
1256 | { |
1257 | if (repeat_max == 0) { |
1258 | code = previous; |
1259 | goto END_REPEAT; |
1260 | } |
1261 | |
1262 | if (repeatMin == 0 && repeat_max == -1) |
1263 | *code++ = OP_CRSTAR + repeatType; |
1264 | else if (repeatMin == 1 && repeat_max == -1) |
1265 | *code++ = OP_CRPLUS + repeatType; |
1266 | else if (repeatMin == 0 && repeat_max == 1) |
1267 | *code++ = OP_CRQUERY + repeatType; |
1268 | else { |
1269 | *code++ = OP_CRRANGE + repeatType; |
1270 | put2ByteValueAndAdvance(opcodePtr&: code, value: repeatMin); |
1271 | if (repeat_max == -1) |
1272 | repeat_max = 0; /* 2-byte encoding for max */ |
1273 | put2ByteValueAndAdvance(opcodePtr&: code, value: repeat_max); |
1274 | } |
1275 | } |
1276 | |
1277 | /* If previous was a bracket group, we may have to replicate it in certain |
1278 | cases. */ |
1279 | |
1280 | else if (*previous >= OP_BRA) { |
1281 | int ketoffset = 0; |
1282 | int len = code - previous; |
1283 | unsigned char* bralink = NULL; |
1284 | |
1285 | /* If the maximum repeat count is unlimited, find the end of the bracket |
1286 | by scanning through from the start, and compute the offset back to it |
1287 | from the current code pointer. There may be an OP_OPT setting following |
1288 | the final KET, so we can't find the end just by going back from the code |
1289 | pointer. */ |
1290 | |
1291 | if (repeat_max == -1) { |
1292 | const unsigned char* ket = previous; |
1293 | advanceToEndOfBracket(opcodePtr&: ket); |
1294 | ketoffset = code - ket; |
1295 | } |
1296 | |
1297 | /* The case of a zero minimum is special because of the need to stick |
1298 | OP_BRAZERO in front of it, and because the group appears once in the |
1299 | data, whereas in other cases it appears the minimum number of times. For |
1300 | this reason, it is simplest to treat this case separately, as otherwise |
1301 | the code gets far too messy. There are several special subcases when the |
1302 | minimum is zero. */ |
1303 | |
1304 | if (repeatMin == 0) { |
1305 | /* If the maximum is also zero, we just omit the group from the output |
1306 | altogether. */ |
1307 | |
1308 | if (repeat_max == 0) { |
1309 | code = previous; |
1310 | goto END_REPEAT; |
1311 | } |
1312 | |
1313 | /* If the maximum is 1 or unlimited, we just have to stick in the |
1314 | BRAZERO and do no more at this point. However, we do need to adjust |
1315 | any OP_RECURSE calls inside the group that refer to the group itself or |
1316 | any internal group, because the offset is from the start of the whole |
1317 | regex. Temporarily terminate the pattern while doing this. */ |
1318 | |
1319 | if (repeat_max <= 1) { |
1320 | *code = OP_END; |
1321 | memmove(dest: previous+1, src: previous, n: len); |
1322 | code++; |
1323 | *previous++ = OP_BRAZERO + repeatType; |
1324 | } |
1325 | |
1326 | /* If the maximum is greater than 1 and limited, we have to replicate |
1327 | in a nested fashion, sticking OP_BRAZERO before each set of brackets. |
1328 | The first one has to be handled carefully because it's the original |
1329 | copy, which has to be moved up. The remainder can be handled by code |
1330 | that is common with the non-zero minimum case below. We have to |
1331 | adjust the value of repeat_max, since one less copy is required. */ |
1332 | |
1333 | else { |
1334 | *code = OP_END; |
1335 | memmove(dest: previous + 2 + LINK_SIZE, src: previous, n: len); |
1336 | code += 2 + LINK_SIZE; |
1337 | *previous++ = OP_BRAZERO + repeatType; |
1338 | *previous++ = OP_BRA; |
1339 | |
1340 | /* We chain together the bracket offset fields that have to be |
1341 | filled in later when the ends of the brackets are reached. */ |
1342 | |
1343 | int offset = (!bralink) ? 0 : previous - bralink; |
1344 | bralink = previous; |
1345 | putLinkValueAllowZeroAndAdvance(opcodePtr&: previous, value: offset); |
1346 | } |
1347 | |
1348 | repeat_max--; |
1349 | } |
1350 | |
1351 | /* If the minimum is greater than zero, replicate the group as many |
1352 | times as necessary, and adjust the maximum to the number of subsequent |
1353 | copies that we need. If we set a first char from the group, and didn't |
1354 | set a required char, copy the latter from the former. */ |
1355 | |
1356 | else { |
1357 | if (repeatMin > 1) { |
1358 | if (didGroupSetFirstByte && reqByte < 0) |
1359 | reqByte = firstByte; |
1360 | for (int i = 1; i < repeatMin; i++) { |
1361 | memcpy(dest: code, src: previous, n: len); |
1362 | code += len; |
1363 | } |
1364 | } |
1365 | if (repeat_max > 0) |
1366 | repeat_max -= repeatMin; |
1367 | } |
1368 | |
1369 | /* This code is common to both the zero and non-zero minimum cases. If |
1370 | the maximum is limited, it replicates the group in a nested fashion, |
1371 | remembering the bracket starts on a stack. In the case of a zero minimum, |
1372 | the first one was set up above. In all cases the repeat_max now specifies |
1373 | the number of additional copies needed. */ |
1374 | |
1375 | if (repeat_max >= 0) { |
1376 | for (int i = repeat_max - 1; i >= 0; i--) { |
1377 | *code++ = OP_BRAZERO + repeatType; |
1378 | |
1379 | /* All but the final copy start a new nesting, maintaining the |
1380 | chain of brackets outstanding. */ |
1381 | |
1382 | if (i != 0) { |
1383 | *code++ = OP_BRA; |
1384 | int offset = (!bralink) ? 0 : code - bralink; |
1385 | bralink = code; |
1386 | putLinkValueAllowZeroAndAdvance(opcodePtr&: code, value: offset); |
1387 | } |
1388 | |
1389 | memcpy(dest: code, src: previous, n: len); |
1390 | code += len; |
1391 | } |
1392 | |
1393 | /* Now chain through the pending brackets, and fill in their length |
1394 | fields (which are holding the chain links pro tem). */ |
1395 | |
1396 | while (bralink) { |
1397 | int offset = code - bralink + 1; |
1398 | unsigned char* bra = code - offset; |
1399 | int oldlinkoffset = getLinkValueAllowZero(opcodePtr: bra + 1); |
1400 | bralink = (!oldlinkoffset) ? 0 : bralink - oldlinkoffset; |
1401 | *code++ = OP_KET; |
1402 | putLinkValueAndAdvance(opcodePtr&: code, value: offset); |
1403 | putLinkValue(opcodePtr: bra + 1, value: offset); |
1404 | } |
1405 | } |
1406 | |
1407 | /* If the maximum is unlimited, set a repeater in the final copy. We |
1408 | can't just offset backwards from the current code point, because we |
1409 | don't know if there's been an options resetting after the ket. The |
1410 | correct offset was computed above. */ |
1411 | |
1412 | else |
1413 | code[-ketoffset] = OP_KETRMAX + repeatType; |
1414 | } |
1415 | |
1416 | // A quantifier after an assertion is mostly meaningless, but it |
1417 | // can nullify the assertion if it has a 0 minimum. |
1418 | else if (*previous == OP_ASSERT || *previous == OP_ASSERT_NOT) { |
1419 | if (repeatMin == 0) { |
1420 | code = previous; |
1421 | goto END_REPEAT; |
1422 | } |
1423 | } |
1424 | |
1425 | /* Else there's some kind of shambles */ |
1426 | |
1427 | else { |
1428 | *errorCodePtr = ERR11; |
1429 | goto FAILED; |
1430 | } |
1431 | |
1432 | /* In all case we no longer have a previous item. We also set the |
1433 | "follows varying string" flag for subsequently encountered reqbytes if |
1434 | it isn't already set and we have just passed a varying length item. */ |
1435 | |
1436 | END_REPEAT: |
1437 | previous = NULL; |
1438 | cd.reqVaryOpt |= reqvary; |
1439 | break; |
1440 | |
1441 | /* Start of nested bracket sub-expression, or comment or lookahead or |
1442 | lookbehind or option setting or condition. First deal with special things |
1443 | that can come after a bracket; all are introduced by ?, and the appearance |
1444 | of any of them means that this is not a referencing group. They were |
1445 | checked for validity in the first pass over the string, so we don't have to |
1446 | check for syntax errors here. */ |
1447 | |
1448 | case '(': |
1449 | skipBytes = 0; |
1450 | |
1451 | if (*(++ptr) == '?') { |
1452 | switch (*(++ptr)) { |
1453 | case ':': /* Non-extracting bracket */ |
1454 | bravalue = OP_BRA; |
1455 | ptr++; |
1456 | break; |
1457 | |
1458 | case '=': /* Positive lookahead */ |
1459 | bravalue = OP_ASSERT; |
1460 | ptr++; |
1461 | break; |
1462 | |
1463 | case '!': /* Negative lookahead */ |
1464 | bravalue = OP_ASSERT_NOT; |
1465 | ptr++; |
1466 | break; |
1467 | |
1468 | /* Character after (? not specially recognized */ |
1469 | |
1470 | default: |
1471 | *errorCodePtr = ERR12; |
1472 | goto FAILED; |
1473 | } |
1474 | } |
1475 | |
1476 | /* Else we have a referencing group; adjust the opcode. If the bracket |
1477 | number is greater than EXTRACT_BASIC_MAX, we set the opcode one higher, and |
1478 | arrange for the true number to follow later, in an OP_BRANUMBER item. */ |
1479 | |
1480 | else { |
1481 | if (++(*brackets) > EXTRACT_BASIC_MAX) { |
1482 | bravalue = OP_BRA + EXTRACT_BASIC_MAX + 1; |
1483 | code[1 + LINK_SIZE] = OP_BRANUMBER; |
1484 | put2ByteValue(opcodePtr: code + 2 + LINK_SIZE, value: *brackets); |
1485 | skipBytes = 3; |
1486 | } |
1487 | else |
1488 | bravalue = OP_BRA + *brackets; |
1489 | } |
1490 | |
1491 | /* Process nested bracketed re. We copy code into a non-variable |
1492 | in order to be able to pass its address because some compilers |
1493 | complain otherwise. Pass in a new setting for the ims options |
1494 | if they have changed. */ |
1495 | |
1496 | previous = code; |
1497 | *code = bravalue; |
1498 | tempcode = code; |
1499 | tempreqvary = cd.reqVaryOpt; /* Save value before bracket */ |
1500 | |
1501 | if (!compileBracket( |
1502 | options, |
1503 | brackets, /* Extracting bracket count */ |
1504 | &tempcode, /* Where to put code (updated) */ |
1505 | &ptr, /* Input pointer (updated) */ |
1506 | patternEnd, |
1507 | errorCodePtr, /* Where to put an error message */ |
1508 | skipBytes, /* Skip over OP_BRANUMBER */ |
1509 | &subFirstByte, /* For possible first char */ |
1510 | &subReqByte, /* For possible last char */ |
1511 | cd)) /* Tables block */ |
1512 | goto FAILED; |
1513 | |
1514 | /* At the end of compiling, code is still pointing to the start of the |
1515 | group, while tempcode has been updated to point past the end of the group |
1516 | and any option resetting that may follow it. The pattern pointer (ptr) |
1517 | is on the bracket. */ |
1518 | |
1519 | /* Handle updating of the required and first characters. Update for normal |
1520 | brackets of all kinds, and conditions with two branches (see code above). |
1521 | If the bracket is followed by a quantifier with zero repeat, we have to |
1522 | back off. Hence the definition of zeroReqByte and zeroFirstByte outside the |
1523 | main loop so that they can be accessed for the back off. */ |
1524 | |
1525 | zeroReqByte = reqByte; |
1526 | zeroFirstByte = firstByte; |
1527 | didGroupSetFirstByte = false; |
1528 | |
1529 | if (bravalue >= OP_BRA) { |
1530 | /* If we have not yet set a firstByte in this branch, take it from the |
1531 | subpattern, remembering that it was set here so that a repeat of more |
1532 | than one can replicate it as reqByte if necessary. If the subpattern has |
1533 | no firstByte, set "none" for the whole branch. In both cases, a zero |
1534 | repeat forces firstByte to "none". */ |
1535 | |
1536 | if (firstByte == REQ_UNSET) { |
1537 | if (subFirstByte >= 0) { |
1538 | firstByte = subFirstByte; |
1539 | didGroupSetFirstByte = true; |
1540 | } |
1541 | else |
1542 | firstByte = REQ_NONE; |
1543 | zeroFirstByte = REQ_NONE; |
1544 | } |
1545 | |
1546 | /* If firstByte was previously set, convert the subpattern's firstByte |
1547 | into reqByte if there wasn't one, using the vary flag that was in |
1548 | existence beforehand. */ |
1549 | |
1550 | else if (subFirstByte >= 0 && subReqByte < 0) |
1551 | subReqByte = subFirstByte | tempreqvary; |
1552 | |
1553 | /* If the subpattern set a required byte (or set a first byte that isn't |
1554 | really the first byte - see above), set it. */ |
1555 | |
1556 | if (subReqByte >= 0) |
1557 | reqByte = subReqByte; |
1558 | } |
1559 | |
1560 | /* For a forward assertion, we take the reqByte, if set. This can be |
1561 | helpful if the pattern that follows the assertion doesn't set a different |
1562 | char. For example, it's useful for /(?=abcde).+/. We can't set firstByte |
1563 | for an assertion, however because it leads to incorrect effect for patterns |
1564 | such as /(?=a)a.+/ when the "real" "a" would then become a reqByte instead |
1565 | of a firstByte. This is overcome by a scan at the end if there's no |
1566 | firstByte, looking for an asserted first char. */ |
1567 | |
1568 | else if (bravalue == OP_ASSERT && subReqByte >= 0) |
1569 | reqByte = subReqByte; |
1570 | |
1571 | /* Now update the main code pointer to the end of the group. */ |
1572 | |
1573 | code = tempcode; |
1574 | |
1575 | /* Error if hit end of pattern */ |
1576 | |
1577 | if (ptr >= patternEnd || *ptr != ')') { |
1578 | *errorCodePtr = ERR14; |
1579 | goto FAILED; |
1580 | } |
1581 | break; |
1582 | |
1583 | /* Check \ for being a real metacharacter; if not, fall through and handle |
1584 | it as a data character at the start of a string. Escape items are checked |
1585 | for validity in the pre-compiling pass. */ |
1586 | |
1587 | case '\\': |
1588 | c = checkEscape(ptrPtr: &ptr, patternEnd, errorCodePtr, bracount: cd.numCapturingBrackets, isClass: false); |
1589 | |
1590 | /* Handle metacharacters introduced by \. For ones like \d, the ESC_ values |
1591 | are arranged to be the negation of the corresponding OP_values. For the |
1592 | back references, the values are ESC_REF plus the reference number. Only |
1593 | back references and those types that consume a character may be repeated. |
1594 | We can test for values between ESC_b and ESC_w for the latter; this may |
1595 | have to change if any new ones are ever created. */ |
1596 | |
1597 | if (c < 0) { |
1598 | /* For metasequences that actually match a character, we disable the |
1599 | setting of a first character if it hasn't already been set. */ |
1600 | |
1601 | if (firstByte == REQ_UNSET && -c > ESC_b && -c <= ESC_w) |
1602 | firstByte = REQ_NONE; |
1603 | |
1604 | /* Set values to reset to if this is followed by a zero repeat. */ |
1605 | |
1606 | zeroFirstByte = firstByte; |
1607 | zeroReqByte = reqByte; |
1608 | |
1609 | /* Back references are handled specially */ |
1610 | |
1611 | if (-c >= ESC_REF) { |
1612 | int number = -c - ESC_REF; |
1613 | previous = code; |
1614 | *code++ = OP_REF; |
1615 | put2ByteValueAndAdvance(opcodePtr&: code, value: number); |
1616 | } |
1617 | |
1618 | /* For the rest, we can obtain the OP value by negating the escape |
1619 | value */ |
1620 | |
1621 | else { |
1622 | previous = (-c > ESC_b && -c <= ESC_w) ? code : NULL; |
1623 | *code++ = -c; |
1624 | } |
1625 | continue; |
1626 | } |
1627 | |
1628 | /* Fall through. */ |
1629 | |
1630 | /* Handle a literal character. It is guaranteed not to be whitespace or # |
1631 | when the extended flag is set. If we are in UTF-8 mode, it may be a |
1632 | multi-byte literal character. */ |
1633 | |
1634 | default: |
1635 | NORMAL_CHAR: |
1636 | |
1637 | previous = code; |
1638 | |
1639 | if (c < 128) { |
1640 | mcLength = 1; |
1641 | mcbuffer[0] = c; |
1642 | |
1643 | if ((options & IgnoreCaseOption) && (c | 0x20) >= 'a' && (c | 0x20) <= 'z') { |
1644 | *code++ = OP_ASCII_LETTER_IGNORING_CASE; |
1645 | *code++ = c | 0x20; |
1646 | } else { |
1647 | *code++ = OP_ASCII_CHAR; |
1648 | *code++ = c; |
1649 | } |
1650 | } else { |
1651 | mcLength = encodeUTF8(cvalue: c, buffer: mcbuffer); |
1652 | |
1653 | *code++ = (options & IgnoreCaseOption) ? OP_CHAR_IGNORING_CASE : OP_CHAR; |
1654 | for (c = 0; c < mcLength; c++) |
1655 | *code++ = mcbuffer[c]; |
1656 | } |
1657 | |
1658 | /* Set the first and required bytes appropriately. If no previous first |
1659 | byte, set it from this character, but revert to none on a zero repeat. |
1660 | Otherwise, leave the firstByte value alone, and don't change it on a zero |
1661 | repeat. */ |
1662 | |
1663 | if (firstByte == REQ_UNSET) { |
1664 | zeroFirstByte = REQ_NONE; |
1665 | zeroReqByte = reqByte; |
1666 | |
1667 | /* If the character is more than one byte long, we can set firstByte |
1668 | only if it is not to be matched caselessly. */ |
1669 | |
1670 | if (mcLength == 1 || reqCaseOpt == 0) { |
1671 | firstByte = mcbuffer[0] | reqCaseOpt; |
1672 | if (mcLength != 1) |
1673 | reqByte = code[-1] | cd.reqVaryOpt; |
1674 | } |
1675 | else |
1676 | firstByte = reqByte = REQ_NONE; |
1677 | } |
1678 | |
1679 | /* firstByte was previously set; we can set reqByte only the length is |
1680 | 1 or the matching is caseful. */ |
1681 | |
1682 | else { |
1683 | zeroFirstByte = firstByte; |
1684 | zeroReqByte = reqByte; |
1685 | if (mcLength == 1 || reqCaseOpt == 0) |
1686 | reqByte = code[-1] | reqCaseOpt | cd.reqVaryOpt; |
1687 | } |
1688 | |
1689 | break; /* End of literal character handling */ |
1690 | } |
1691 | } /* end of big loop */ |
1692 | |
1693 | /* Control never reaches here by falling through, only by a goto for all the |
1694 | error states. Pass back the position in the pattern so that it can be displayed |
1695 | to the user for diagnosing the error. */ |
1696 | |
1697 | FAILED: |
1698 | *ptrPtr = ptr; |
1699 | return false; |
1700 | } |
1701 | |
1702 | /************************************************* |
1703 | * Compile sequence of alternatives * |
1704 | *************************************************/ |
1705 | |
1706 | /* On entry, ptr is pointing past the bracket character, but on return |
1707 | it points to the closing bracket, or vertical bar, or end of string. |
1708 | The code variable is pointing at the byte into which the BRA operator has been |
1709 | stored. If the ims options are changed at the start (for a (?ims: group) or |
1710 | during any branch, we need to insert an OP_OPT item at the start of every |
1711 | following branch to ensure they get set correctly at run time, and also pass |
1712 | the new options into every subsequent branch compile. |
1713 | |
1714 | Argument: |
1715 | options option bits, including any changes for this subpattern |
1716 | brackets -> int containing the number of extracting brackets used |
1717 | codePtr -> the address of the current code pointer |
1718 | ptrPtr -> the address of the current pattern pointer |
1719 | errorCodePtr -> pointer to error code variable |
1720 | skipBytes skip this many bytes at start (for OP_BRANUMBER) |
1721 | firstbyteptr place to put the first required character, or a negative number |
1722 | reqbyteptr place to put the last required character, or a negative number |
1723 | cd points to the data block with tables pointers etc. |
1724 | |
1725 | Returns: true on success |
1726 | */ |
1727 | |
1728 | static bool |
1729 | compileBracket(int options, int* brackets, unsigned char** codePtr, |
1730 | const UChar** ptrPtr, const UChar* patternEnd, ErrorCode* errorCodePtr, int skipBytes, |
1731 | int* firstbyteptr, int* reqbyteptr, CompileData& cd) |
1732 | { |
1733 | const UChar* ptr = *ptrPtr; |
1734 | unsigned char* code = *codePtr; |
1735 | unsigned char* lastBranch = code; |
1736 | unsigned char* start_bracket = code; |
1737 | int firstByte = REQ_UNSET; |
1738 | int reqByte = REQ_UNSET; |
1739 | |
1740 | /* Offset is set zero to mark that this bracket is still open */ |
1741 | |
1742 | putLinkValueAllowZero(opcodePtr: code + 1, value: 0); |
1743 | code += 1 + LINK_SIZE + skipBytes; |
1744 | |
1745 | /* Loop for each alternative branch */ |
1746 | |
1747 | while (true) { |
1748 | /* Now compile the branch */ |
1749 | |
1750 | int branchFirstByte; |
1751 | int branchReqByte; |
1752 | if (!compileBranch(options, brackets, codePtr: &code, ptrPtr: &ptr, patternEnd, errorCodePtr, |
1753 | firstbyteptr: &branchFirstByte, reqbyteptr: &branchReqByte, cd)) { |
1754 | *ptrPtr = ptr; |
1755 | return false; |
1756 | } |
1757 | |
1758 | /* If this is the first branch, the firstByte and reqByte values for the |
1759 | branch become the values for the regex. */ |
1760 | |
1761 | if (*lastBranch != OP_ALT) { |
1762 | firstByte = branchFirstByte; |
1763 | reqByte = branchReqByte; |
1764 | } |
1765 | |
1766 | /* If this is not the first branch, the first char and reqByte have to |
1767 | match the values from all the previous branches, except that if the previous |
1768 | value for reqByte didn't have REQ_VARY set, it can still match, and we set |
1769 | REQ_VARY for the regex. */ |
1770 | |
1771 | else { |
1772 | /* If we previously had a firstByte, but it doesn't match the new branch, |
1773 | we have to abandon the firstByte for the regex, but if there was previously |
1774 | no reqByte, it takes on the value of the old firstByte. */ |
1775 | |
1776 | if (firstByte >= 0 && firstByte != branchFirstByte) { |
1777 | if (reqByte < 0) |
1778 | reqByte = firstByte; |
1779 | firstByte = REQ_NONE; |
1780 | } |
1781 | |
1782 | /* If we (now or from before) have no firstByte, a firstByte from the |
1783 | branch becomes a reqByte if there isn't a branch reqByte. */ |
1784 | |
1785 | if (firstByte < 0 && branchFirstByte >= 0 && branchReqByte < 0) |
1786 | branchReqByte = branchFirstByte; |
1787 | |
1788 | /* Now ensure that the reqbytes match */ |
1789 | |
1790 | if ((reqByte & ~REQ_VARY) != (branchReqByte & ~REQ_VARY)) |
1791 | reqByte = REQ_NONE; |
1792 | else |
1793 | reqByte |= branchReqByte; /* To "or" REQ_VARY */ |
1794 | } |
1795 | |
1796 | /* Reached end of expression, either ')' or end of pattern. Go back through |
1797 | the alternative branches and reverse the chain of offsets, with the field in |
1798 | the BRA item now becoming an offset to the first alternative. If there are |
1799 | no alternatives, it points to the end of the group. The length in the |
1800 | terminating ket is always the length of the whole bracketed item. If any of |
1801 | the ims options were changed inside the group, compile a resetting op-code |
1802 | following, except at the very end of the pattern. Return leaving the pointer |
1803 | at the terminating char. */ |
1804 | |
1805 | if (ptr >= patternEnd || *ptr != '|') { |
1806 | int length = code - lastBranch; |
1807 | do { |
1808 | int prevLength = getLinkValueAllowZero(opcodePtr: lastBranch + 1); |
1809 | putLinkValue(opcodePtr: lastBranch + 1, value: length); |
1810 | length = prevLength; |
1811 | lastBranch -= length; |
1812 | } while (length > 0); |
1813 | |
1814 | /* Fill in the ket */ |
1815 | |
1816 | *code = OP_KET; |
1817 | putLinkValue(opcodePtr: code + 1, value: code - start_bracket); |
1818 | code += 1 + LINK_SIZE; |
1819 | |
1820 | /* Set values to pass back */ |
1821 | |
1822 | *codePtr = code; |
1823 | *ptrPtr = ptr; |
1824 | *firstbyteptr = firstByte; |
1825 | *reqbyteptr = reqByte; |
1826 | return true; |
1827 | } |
1828 | |
1829 | /* Another branch follows; insert an "or" node. Its length field points back |
1830 | to the previous branch while the bracket remains open. At the end the chain |
1831 | is reversed. It's done like this so that the start of the bracket has a |
1832 | zero offset until it is closed, making it possible to detect recursion. */ |
1833 | |
1834 | *code = OP_ALT; |
1835 | putLinkValue(opcodePtr: code + 1, value: code - lastBranch); |
1836 | lastBranch = code; |
1837 | code += 1 + LINK_SIZE; |
1838 | ptr++; |
1839 | } |
1840 | ASSERT_NOT_REACHED(); |
1841 | } |
1842 | |
1843 | /************************************************* |
1844 | * Check for anchored expression * |
1845 | *************************************************/ |
1846 | |
1847 | /* Try to find out if this is an anchored regular expression. Consider each |
1848 | alternative branch. If they all start OP_CIRC, or with a bracket |
1849 | all of whose alternatives start OP_CIRC (recurse ad lib), then |
1850 | it's anchored. |
1851 | |
1852 | Arguments: |
1853 | code points to start of expression (the bracket) |
1854 | captureMap a bitmap of which brackets we are inside while testing; this |
1855 | handles up to substring 31; all brackets after that share |
1856 | the zero bit |
1857 | backrefMap the back reference bitmap |
1858 | */ |
1859 | |
1860 | static bool branchIsAnchored(const unsigned char* code) |
1861 | { |
1862 | const unsigned char* scode = firstSignificantOpcode(code); |
1863 | int op = *scode; |
1864 | |
1865 | /* Brackets */ |
1866 | if (op >= OP_BRA || op == OP_ASSERT) |
1867 | return bracketIsAnchored(code: scode); |
1868 | |
1869 | /* Check for explicit anchoring */ |
1870 | return op == OP_CIRC; |
1871 | } |
1872 | |
1873 | static bool bracketIsAnchored(const unsigned char* code) |
1874 | { |
1875 | do { |
1876 | if (!branchIsAnchored(code: code + 1 + LINK_SIZE)) |
1877 | return false; |
1878 | code += getLinkValue(opcodePtr: code + 1); |
1879 | } while (*code == OP_ALT); /* Loop for each alternative */ |
1880 | return true; |
1881 | } |
1882 | |
1883 | /************************************************* |
1884 | * Check for starting with ^ or .* * |
1885 | *************************************************/ |
1886 | |
1887 | /* This is called to find out if every branch starts with ^ or .* so that |
1888 | "first char" processing can be done to speed things up in multiline |
1889 | matching and for non-DOTALL patterns that start with .* (which must start at |
1890 | the beginning or after \n) |
1891 | |
1892 | Except when the .* appears inside capturing parentheses, and there is a |
1893 | subsequent back reference to those parentheses. By keeping a bitmap of the |
1894 | first 31 back references, we can catch some of the more common cases more |
1895 | precisely; all the greater back references share a single bit. |
1896 | |
1897 | Arguments: |
1898 | code points to start of expression (the bracket) |
1899 | captureMap a bitmap of which brackets we are inside while testing; this |
1900 | handles up to substring 31; all brackets after that share |
1901 | the zero bit |
1902 | backrefMap the back reference bitmap |
1903 | */ |
1904 | |
1905 | static bool branchNeedsLineStart(const unsigned char* code, unsigned captureMap, unsigned backrefMap) |
1906 | { |
1907 | const unsigned char* scode = firstSignificantOpcode(code); |
1908 | int op = *scode; |
1909 | |
1910 | /* Capturing brackets */ |
1911 | if (op > OP_BRA) { |
1912 | int captureNum = op - OP_BRA; |
1913 | if (captureNum > EXTRACT_BASIC_MAX) |
1914 | captureNum = get2ByteValue(opcodePtr: scode + 2 + LINK_SIZE); |
1915 | int bracketMask = (captureNum < 32) ? (1 << captureNum) : 1; |
1916 | return bracketNeedsLineStart(code: scode, captureMap: captureMap | bracketMask, backrefMap); |
1917 | } |
1918 | |
1919 | /* Other brackets */ |
1920 | if (op == OP_BRA || op == OP_ASSERT) |
1921 | return bracketNeedsLineStart(code: scode, captureMap, backrefMap); |
1922 | |
1923 | /* .* means "start at start or after \n" if it isn't in brackets that |
1924 | may be referenced. */ |
1925 | |
1926 | if (op == OP_TYPESTAR || op == OP_TYPEMINSTAR) |
1927 | return scode[1] == OP_NOT_NEWLINE && !(captureMap & backrefMap); |
1928 | |
1929 | /* Explicit ^ */ |
1930 | return op == OP_CIRC || op == OP_BOL; |
1931 | } |
1932 | |
1933 | static bool bracketNeedsLineStart(const unsigned char* code, unsigned captureMap, unsigned backrefMap) |
1934 | { |
1935 | do { |
1936 | if (!branchNeedsLineStart(code: code + 1 + LINK_SIZE, captureMap, backrefMap)) |
1937 | return false; |
1938 | code += getLinkValue(opcodePtr: code + 1); |
1939 | } while (*code == OP_ALT); /* Loop for each alternative */ |
1940 | return true; |
1941 | } |
1942 | |
1943 | /************************************************* |
1944 | * Check for asserted fixed first char * |
1945 | *************************************************/ |
1946 | |
1947 | /* During compilation, the "first char" settings from forward assertions are |
1948 | discarded, because they can cause conflicts with actual literals that follow. |
1949 | However, if we end up without a first char setting for an unanchored pattern, |
1950 | it is worth scanning the regex to see if there is an initial asserted first |
1951 | char. If all branches start with the same asserted char, or with a bracket all |
1952 | of whose alternatives start with the same asserted char (recurse ad lib), then |
1953 | we return that char, otherwise -1. |
1954 | |
1955 | Arguments: |
1956 | code points to start of expression (the bracket) |
1957 | options pointer to the options (used to check casing changes) |
1958 | inassert true if in an assertion |
1959 | |
1960 | Returns: -1 or the fixed first char |
1961 | */ |
1962 | |
1963 | static int branchFindFirstAssertedCharacter(const unsigned char* code, bool inassert) |
1964 | { |
1965 | const unsigned char* scode = firstSignificantOpcodeSkippingAssertions(code); |
1966 | int op = *scode; |
1967 | |
1968 | if (op >= OP_BRA) |
1969 | op = OP_BRA; |
1970 | |
1971 | switch (op) { |
1972 | default: |
1973 | return -1; |
1974 | |
1975 | case OP_BRA: |
1976 | case OP_ASSERT: |
1977 | return bracketFindFirstAssertedCharacter(code: scode, inassert: op == OP_ASSERT); |
1978 | |
1979 | case OP_EXACT: |
1980 | scode += 2; |
1981 | /* Fall through */ |
1982 | |
1983 | case OP_CHAR: |
1984 | case OP_CHAR_IGNORING_CASE: |
1985 | case OP_ASCII_CHAR: |
1986 | case OP_ASCII_LETTER_IGNORING_CASE: |
1987 | case OP_PLUS: |
1988 | case OP_MINPLUS: |
1989 | if (!inassert) |
1990 | return -1; |
1991 | return scode[1]; |
1992 | } |
1993 | } |
1994 | |
1995 | static int bracketFindFirstAssertedCharacter(const unsigned char* code, bool inassert) |
1996 | { |
1997 | int c = -1; |
1998 | do { |
1999 | int d = branchFindFirstAssertedCharacter(code: code + 1 + LINK_SIZE, inassert); |
2000 | if (d < 0) |
2001 | return -1; |
2002 | if (c < 0) |
2003 | c = d; |
2004 | else if (c != d) |
2005 | return -1; |
2006 | code += getLinkValue(opcodePtr: code + 1); |
2007 | } while (*code == OP_ALT); |
2008 | return c; |
2009 | } |
2010 | |
2011 | static inline int multiplyWithOverflowCheck(int a, int b) |
2012 | { |
2013 | if (!a || !b) |
2014 | return 0; |
2015 | if (a > MAX_PATTERN_SIZE / b) |
2016 | return -1; |
2017 | return a * b; |
2018 | } |
2019 | |
2020 | static int calculateCompiledPatternLength(const UChar* pattern, int patternLength, JSRegExpIgnoreCaseOption ignoreCase, |
2021 | CompileData& cd, ErrorCode& errorcode) |
2022 | { |
2023 | /* Make a pass over the pattern to compute the |
2024 | amount of store required to hold the compiled code. This does not have to be |
2025 | perfect as long as errors are overestimates. */ |
2026 | |
2027 | if (patternLength > MAX_PATTERN_SIZE) { |
2028 | errorcode = ERR16; |
2029 | return -1; |
2030 | } |
2031 | |
2032 | int length = 1 + LINK_SIZE; /* For initial BRA plus length */ |
2033 | int = 0; |
2034 | int lastitemlength = 0; |
2035 | unsigned brastackptr = 0; |
2036 | int brastack[BRASTACK_SIZE]; |
2037 | unsigned char bralenstack[BRASTACK_SIZE]; |
2038 | int bracount = 0; |
2039 | |
2040 | const UChar* ptr = (const UChar*)(pattern - 1); |
2041 | const UChar* patternEnd = (const UChar*)(pattern + patternLength); |
2042 | |
2043 | while (++ptr < patternEnd) { |
2044 | int minRepeats = 0, maxRepeats = 0; |
2045 | int c = *ptr; |
2046 | |
2047 | switch (c) { |
2048 | /* A backslashed item may be an escaped data character or it may be a |
2049 | character type. */ |
2050 | |
2051 | case '\\': |
2052 | c = checkEscape(ptrPtr: &ptr, patternEnd, errorCodePtr: &errorcode, bracount: cd.numCapturingBrackets, isClass: false); |
2053 | if (errorcode != 0) |
2054 | return -1; |
2055 | |
2056 | lastitemlength = 1; /* Default length of last item for repeats */ |
2057 | |
2058 | if (c >= 0) { /* Data character */ |
2059 | length += 2; /* For a one-byte character */ |
2060 | |
2061 | if (c > 127) { |
2062 | int i; |
2063 | for (i = 0; i < jsc_pcre_utf8_table1_size; i++) |
2064 | if (c <= jsc_pcre_utf8_table1[i]) break; |
2065 | length += i; |
2066 | lastitemlength += i; |
2067 | } |
2068 | |
2069 | continue; |
2070 | } |
2071 | |
2072 | /* Other escapes need one byte */ |
2073 | |
2074 | length++; |
2075 | |
2076 | /* A back reference needs an additional 2 bytes, plus either one or 5 |
2077 | bytes for a repeat. We also need to keep the value of the highest |
2078 | back reference. */ |
2079 | |
2080 | if (c <= -ESC_REF) { |
2081 | int refnum = -c - ESC_REF; |
2082 | cd.backrefMap |= (refnum < 32) ? (1 << refnum) : 1; |
2083 | if (refnum > cd.topBackref) |
2084 | cd.topBackref = refnum; |
2085 | length += 2; /* For single back reference */ |
2086 | if (safelyCheckNextChar(ptr, patternEnd, expected: '{') && isCountedRepeat(p: ptr + 2, patternEnd)) { |
2087 | ptr = readRepeatCounts(p: ptr + 2, minp: &minRepeats, maxp: &maxRepeats, errorCodePtr: &errorcode); |
2088 | if (errorcode) |
2089 | return -1; |
2090 | if ((minRepeats == 0 && (maxRepeats == 1 || maxRepeats == -1)) || |
2091 | (minRepeats == 1 && maxRepeats == -1)) |
2092 | length++; |
2093 | else |
2094 | length += 5; |
2095 | if (safelyCheckNextChar(ptr, patternEnd, expected: '?')) |
2096 | ptr++; |
2097 | } |
2098 | } |
2099 | continue; |
2100 | |
2101 | case '^': /* Single-byte metacharacters */ |
2102 | case '.': |
2103 | case '$': |
2104 | length++; |
2105 | lastitemlength = 1; |
2106 | continue; |
2107 | |
2108 | case '*': /* These repeats won't be after brackets; */ |
2109 | case '+': /* those are handled separately */ |
2110 | case '?': |
2111 | length++; |
2112 | goto POSSESSIVE; |
2113 | |
2114 | /* This covers the cases of braced repeats after a single char, metachar, |
2115 | class, or back reference. */ |
2116 | |
2117 | case '{': |
2118 | if (!isCountedRepeat(p: ptr + 1, patternEnd)) |
2119 | goto NORMAL_CHAR; |
2120 | ptr = readRepeatCounts(p: ptr + 1, minp: &minRepeats, maxp: &maxRepeats, errorCodePtr: &errorcode); |
2121 | if (errorcode != 0) |
2122 | return -1; |
2123 | |
2124 | /* These special cases just insert one extra opcode */ |
2125 | |
2126 | if ((minRepeats == 0 && (maxRepeats == 1 || maxRepeats == -1)) || |
2127 | (minRepeats == 1 && maxRepeats == -1)) |
2128 | length++; |
2129 | |
2130 | /* These cases might insert additional copies of a preceding character. */ |
2131 | |
2132 | else { |
2133 | if (minRepeats != 1) { |
2134 | length -= lastitemlength; /* Uncount the original char or metachar */ |
2135 | if (minRepeats > 0) |
2136 | length += 3 + lastitemlength; |
2137 | } |
2138 | length += lastitemlength + ((maxRepeats > 0) ? 3 : 1); |
2139 | } |
2140 | |
2141 | if (safelyCheckNextChar(ptr, patternEnd, expected: '?')) |
2142 | ptr++; /* Needs no extra length */ |
2143 | |
2144 | POSSESSIVE: /* Test for possessive quantifier */ |
2145 | if (safelyCheckNextChar(ptr, patternEnd, expected: '+')) { |
2146 | ptr++; |
2147 | length += 2 + 2 * LINK_SIZE; /* Allow for atomic brackets */ |
2148 | } |
2149 | continue; |
2150 | |
2151 | /* An alternation contains an offset to the next branch or ket. If any ims |
2152 | options changed in the previous branch(es), and/or if we are in a |
2153 | lookbehind assertion, extra space will be needed at the start of the |
2154 | branch. This is handled by branch_extra. */ |
2155 | |
2156 | case '|': |
2157 | if (brastackptr == 0) |
2158 | cd.needOuterBracket = true; |
2159 | length += 1 + LINK_SIZE + branch_extra; |
2160 | continue; |
2161 | |
2162 | /* A character class uses 33 characters provided that all the character |
2163 | values are less than 256. Otherwise, it uses a bit map for low valued |
2164 | characters, and individual items for others. Don't worry about character |
2165 | types that aren't allowed in classes - they'll get picked up during the |
2166 | compile. A character class that contains only one single-byte character |
2167 | uses 2 or 3 bytes, depending on whether it is negated or not. Notice this |
2168 | where we can. (In UTF-8 mode we can do this only for chars < 128.) */ |
2169 | |
2170 | case '[': { |
2171 | int class_optcount; |
2172 | if (*(++ptr) == '^') { |
2173 | class_optcount = 10; /* Greater than one */ |
2174 | ptr++; |
2175 | } |
2176 | else |
2177 | class_optcount = 0; |
2178 | |
2179 | bool class_utf8 = false; |
2180 | |
2181 | for (; ptr < patternEnd && *ptr != ']'; ++ptr) { |
2182 | /* Check for escapes */ |
2183 | |
2184 | if (*ptr == '\\') { |
2185 | c = checkEscape(ptrPtr: &ptr, patternEnd, errorCodePtr: &errorcode, bracount: cd.numCapturingBrackets, isClass: true); |
2186 | if (errorcode != 0) |
2187 | return -1; |
2188 | |
2189 | /* Handle escapes that turn into characters */ |
2190 | |
2191 | if (c >= 0) |
2192 | goto NON_SPECIAL_CHARACTER; |
2193 | |
2194 | /* Escapes that are meta-things. The normal ones just affect the |
2195 | bit map, but Unicode properties require an XCLASS extended item. */ |
2196 | |
2197 | else |
2198 | class_optcount = 10; /* \d, \s etc; make sure > 1 */ |
2199 | } |
2200 | |
2201 | /* Anything else increments the possible optimization count. We have to |
2202 | detect ranges here so that we can compute the number of extra ranges for |
2203 | caseless wide characters when UCP support is available. If there are wide |
2204 | characters, we are going to have to use an XCLASS, even for single |
2205 | characters. */ |
2206 | |
2207 | else { |
2208 | c = *ptr; |
2209 | |
2210 | /* Come here from handling \ above when it escapes to a char value */ |
2211 | |
2212 | NON_SPECIAL_CHARACTER: |
2213 | class_optcount++; |
2214 | |
2215 | int d = -1; |
2216 | if (safelyCheckNextChar(ptr, patternEnd, expected: '-')) { |
2217 | const UChar* hyptr = ptr++; |
2218 | if (safelyCheckNextChar(ptr, patternEnd, expected: '\\')) { |
2219 | ptr++; |
2220 | d = checkEscape(ptrPtr: &ptr, patternEnd, errorCodePtr: &errorcode, bracount: cd.numCapturingBrackets, isClass: true); |
2221 | if (errorcode != 0) |
2222 | return -1; |
2223 | } |
2224 | else if ((ptr + 1 < patternEnd) && ptr[1] != ']') |
2225 | d = *++ptr; |
2226 | if (d < 0) |
2227 | ptr = hyptr; /* go back to hyphen as data */ |
2228 | } |
2229 | |
2230 | /* If d >= 0 we have a range. In UTF-8 mode, if the end is > 255, or > |
2231 | 127 for caseless matching, we will need to use an XCLASS. */ |
2232 | |
2233 | if (d >= 0) { |
2234 | class_optcount = 10; /* Ensure > 1 */ |
2235 | if (d < c) { |
2236 | errorcode = ERR8; |
2237 | return -1; |
2238 | } |
2239 | |
2240 | if ((d > 255 || (ignoreCase && d > 127))) { |
2241 | unsigned char buffer[6]; |
2242 | if (!class_utf8) /* Allow for XCLASS overhead */ |
2243 | { |
2244 | class_utf8 = true; |
2245 | length += LINK_SIZE + 2; |
2246 | } |
2247 | |
2248 | /* If we have UCP support, find out how many extra ranges are |
2249 | needed to map the other case of characters within this range. We |
2250 | have to mimic the range optimization here, because extending the |
2251 | range upwards might push d over a boundary that makes it use |
2252 | another byte in the UTF-8 representation. */ |
2253 | |
2254 | if (ignoreCase) { |
2255 | int occ, ocd; |
2256 | int cc = c; |
2257 | int origd = d; |
2258 | while (getOthercaseRange(cptr: &cc, d: origd, ocptr: &occ, odptr: &ocd)) { |
2259 | if (occ >= c && ocd <= d) |
2260 | continue; /* Skip embedded */ |
2261 | |
2262 | if (occ < c && ocd >= c - 1) /* Extend the basic range */ |
2263 | { /* if there is overlap, */ |
2264 | c = occ; /* noting that if occ < c */ |
2265 | continue; /* we can't have ocd > d */ |
2266 | } /* because a subrange is */ |
2267 | if (ocd > d && occ <= d + 1) /* always shorter than */ |
2268 | { /* the basic range. */ |
2269 | d = ocd; |
2270 | continue; |
2271 | } |
2272 | |
2273 | /* An extra item is needed */ |
2274 | |
2275 | length += 1 + encodeUTF8(cvalue: occ, buffer) + |
2276 | ((occ == ocd) ? 0 : encodeUTF8(cvalue: ocd, buffer)); |
2277 | } |
2278 | } |
2279 | |
2280 | /* The length of the (possibly extended) range */ |
2281 | |
2282 | length += 1 + encodeUTF8(cvalue: c, buffer) + encodeUTF8(cvalue: d, buffer); |
2283 | } |
2284 | |
2285 | } |
2286 | |
2287 | /* We have a single character. There is nothing to be done unless we |
2288 | are in UTF-8 mode. If the char is > 255, or 127 when caseless, we must |
2289 | allow for an XCL_SINGLE item, doubled for caselessness if there is UCP |
2290 | support. */ |
2291 | |
2292 | else { |
2293 | if ((c > 255 || (ignoreCase && c > 127))) { |
2294 | unsigned char buffer[6]; |
2295 | class_optcount = 10; /* Ensure > 1 */ |
2296 | if (!class_utf8) /* Allow for XCLASS overhead */ |
2297 | { |
2298 | class_utf8 = true; |
2299 | length += LINK_SIZE + 2; |
2300 | } |
2301 | length += (ignoreCase ? 2 : 1) * (1 + encodeUTF8(cvalue: c, buffer)); |
2302 | } |
2303 | } |
2304 | } |
2305 | } |
2306 | |
2307 | if (ptr >= patternEnd) { /* Missing terminating ']' */ |
2308 | errorcode = ERR6; |
2309 | return -1; |
2310 | } |
2311 | |
2312 | /* We can optimize when there was only one optimizable character. |
2313 | Note that this does not detect the case of a negated single character. |
2314 | In that case we do an incorrect length computation, but it's not a serious |
2315 | problem because the computed length is too large rather than too small. */ |
2316 | |
2317 | if (class_optcount == 1) |
2318 | goto NORMAL_CHAR; |
2319 | |
2320 | /* Here, we handle repeats for the class opcodes. */ |
2321 | { |
2322 | length += 33; |
2323 | |
2324 | /* A repeat needs either 1 or 5 bytes. If it is a possessive quantifier, |
2325 | we also need extra for wrapping the whole thing in a sub-pattern. */ |
2326 | |
2327 | if (safelyCheckNextChar(ptr, patternEnd, expected: '{') && isCountedRepeat(p: ptr + 2, patternEnd)) { |
2328 | ptr = readRepeatCounts(p: ptr + 2, minp: &minRepeats, maxp: &maxRepeats, errorCodePtr: &errorcode); |
2329 | if (errorcode != 0) |
2330 | return -1; |
2331 | if ((minRepeats == 0 && (maxRepeats == 1 || maxRepeats == -1)) || |
2332 | (minRepeats == 1 && maxRepeats == -1)) |
2333 | length++; |
2334 | else |
2335 | length += 5; |
2336 | if (safelyCheckNextChar(ptr, patternEnd, expected: '+')) { |
2337 | ptr++; |
2338 | length += 2 + 2 * LINK_SIZE; |
2339 | } else if (safelyCheckNextChar(ptr, patternEnd, expected: '?')) |
2340 | ptr++; |
2341 | } |
2342 | } |
2343 | continue; |
2344 | } |
2345 | |
2346 | /* Brackets may be genuine groups or special things */ |
2347 | |
2348 | case '(': { |
2349 | int = 0; |
2350 | int bracket_length = 1 + LINK_SIZE; |
2351 | bool capturing = false; |
2352 | |
2353 | /* Handle special forms of bracket, which all start (? */ |
2354 | |
2355 | if (safelyCheckNextChar(ptr, patternEnd, expected: '?')) { |
2356 | switch (c = (ptr + 2 < patternEnd ? ptr[2] : 0)) { |
2357 | /* Non-referencing groups and lookaheads just move the pointer on, and |
2358 | then behave like a non-special bracket, except that they don't increment |
2359 | the count of extracting brackets. Ditto for the "once only" bracket, |
2360 | which is in Perl from version 5.005. */ |
2361 | |
2362 | case ':': |
2363 | case '=': |
2364 | case '!': |
2365 | ptr += 2; |
2366 | break; |
2367 | |
2368 | /* Else loop checking valid options until ) is met. Anything else is an |
2369 | error. If we are without any brackets, i.e. at top level, the settings |
2370 | act as if specified in the options, so massage the options immediately. |
2371 | This is for backward compatibility with Perl 5.004. */ |
2372 | |
2373 | default: |
2374 | errorcode = ERR12; |
2375 | return -1; |
2376 | } |
2377 | } else |
2378 | capturing = 1; |
2379 | |
2380 | /* Capturing brackets must be counted so we can process escapes in a |
2381 | Perlish way. If the number exceeds EXTRACT_BASIC_MAX we are going to need |
2382 | an additional 3 bytes of memory per capturing bracket. */ |
2383 | |
2384 | if (capturing) { |
2385 | bracount++; |
2386 | if (bracount > EXTRACT_BASIC_MAX) |
2387 | bracket_length += 3; |
2388 | } |
2389 | |
2390 | /* Save length for computing whole length at end if there's a repeat that |
2391 | requires duplication of the group. Also save the current value of |
2392 | branch_extra, and start the new group with the new value. If non-zero, this |
2393 | will either be 2 for a (?imsx: group, or 3 for a lookbehind assertion. */ |
2394 | |
2395 | if (brastackptr >= sizeof(brastack)/sizeof(int)) { |
2396 | errorcode = ERR17; |
2397 | return -1; |
2398 | } |
2399 | |
2400 | bralenstack[brastackptr] = branch_extra; |
2401 | branch_extra = branch_newextra; |
2402 | |
2403 | brastack[brastackptr++] = length; |
2404 | length += bracket_length; |
2405 | continue; |
2406 | } |
2407 | |
2408 | /* Handle ket. Look for subsequent maxRepeats/minRepeats; for certain sets of values we |
2409 | have to replicate this bracket up to that many times. If brastackptr is |
2410 | 0 this is an unmatched bracket which will generate an error, but take care |
2411 | not to try to access brastack[-1] when computing the length and restoring |
2412 | the branch_extra value. */ |
2413 | |
2414 | case ')': { |
2415 | int duplength; |
2416 | length += 1 + LINK_SIZE; |
2417 | if (brastackptr > 0) { |
2418 | duplength = length - brastack[--brastackptr]; |
2419 | branch_extra = bralenstack[brastackptr]; |
2420 | } |
2421 | else |
2422 | duplength = 0; |
2423 | |
2424 | /* Leave ptr at the final char; for readRepeatCounts this happens |
2425 | automatically; for the others we need an increment. */ |
2426 | |
2427 | if ((ptr + 1 < patternEnd) && (c = ptr[1]) == '{' && isCountedRepeat(p: ptr + 2, patternEnd)) { |
2428 | ptr = readRepeatCounts(p: ptr + 2, minp: &minRepeats, maxp: &maxRepeats, errorCodePtr: &errorcode); |
2429 | if (errorcode) |
2430 | return -1; |
2431 | } else if (c == '*') { |
2432 | minRepeats = 0; |
2433 | maxRepeats = -1; |
2434 | ptr++; |
2435 | } else if (c == '+') { |
2436 | minRepeats = 1; |
2437 | maxRepeats = -1; |
2438 | ptr++; |
2439 | } else if (c == '?') { |
2440 | minRepeats = 0; |
2441 | maxRepeats = 1; |
2442 | ptr++; |
2443 | } else { |
2444 | minRepeats = 1; |
2445 | maxRepeats = 1; |
2446 | } |
2447 | |
2448 | /* If the minimum is zero, we have to allow for an OP_BRAZERO before the |
2449 | group, and if the maximum is greater than zero, we have to replicate |
2450 | maxval-1 times; each replication acquires an OP_BRAZERO plus a nesting |
2451 | bracket set. */ |
2452 | |
2453 | int repeatsLength; |
2454 | if (minRepeats == 0) { |
2455 | length++; |
2456 | if (maxRepeats > 0) { |
2457 | repeatsLength = multiplyWithOverflowCheck(a: maxRepeats - 1, b: duplength + 3 + 2 * LINK_SIZE); |
2458 | if (repeatsLength < 0) { |
2459 | errorcode = ERR16; |
2460 | return -1; |
2461 | } |
2462 | length += repeatsLength; |
2463 | if (length > MAX_PATTERN_SIZE) { |
2464 | errorcode = ERR16; |
2465 | return -1; |
2466 | } |
2467 | } |
2468 | } |
2469 | |
2470 | /* When the minimum is greater than zero, we have to replicate up to |
2471 | minval-1 times, with no additions required in the copies. Then, if there |
2472 | is a limited maximum we have to replicate up to maxval-1 times allowing |
2473 | for a BRAZERO item before each optional copy and nesting brackets for all |
2474 | but one of the optional copies. */ |
2475 | |
2476 | else { |
2477 | repeatsLength = multiplyWithOverflowCheck(a: minRepeats - 1, b: duplength); |
2478 | if (repeatsLength < 0) { |
2479 | errorcode = ERR16; |
2480 | return -1; |
2481 | } |
2482 | length += repeatsLength; |
2483 | if (maxRepeats > minRepeats) { /* Need this test as maxRepeats=-1 means no limit */ |
2484 | repeatsLength = multiplyWithOverflowCheck(a: maxRepeats - minRepeats, b: duplength + 3 + 2 * LINK_SIZE); |
2485 | if (repeatsLength < 0) { |
2486 | errorcode = ERR16; |
2487 | return -1; |
2488 | } |
2489 | length += repeatsLength - (2 + 2 * LINK_SIZE); |
2490 | } |
2491 | if (length > MAX_PATTERN_SIZE) { |
2492 | errorcode = ERR16; |
2493 | return -1; |
2494 | } |
2495 | } |
2496 | |
2497 | /* Allow space for once brackets for "possessive quantifier" */ |
2498 | |
2499 | if (safelyCheckNextChar(ptr, patternEnd, expected: '+')) { |
2500 | ptr++; |
2501 | length += 2 + 2 * LINK_SIZE; |
2502 | } |
2503 | continue; |
2504 | } |
2505 | |
2506 | /* Non-special character. It won't be space or # in extended mode, so it is |
2507 | always a genuine character. If we are in a \Q...\E sequence, check for the |
2508 | end; if not, we have a literal. */ |
2509 | |
2510 | default: |
2511 | NORMAL_CHAR: |
2512 | length += 2; /* For a one-byte character */ |
2513 | lastitemlength = 1; /* Default length of last item for repeats */ |
2514 | |
2515 | if (c > 127) { |
2516 | int i; |
2517 | for (i = 0; i < jsc_pcre_utf8_table1_size; i++) |
2518 | if (c <= jsc_pcre_utf8_table1[i]) |
2519 | break; |
2520 | length += i; |
2521 | lastitemlength += i; |
2522 | } |
2523 | |
2524 | continue; |
2525 | } |
2526 | } |
2527 | |
2528 | length += 2 + LINK_SIZE; /* For final KET and END */ |
2529 | |
2530 | cd.numCapturingBrackets = bracount; |
2531 | return length; |
2532 | } |
2533 | |
2534 | /************************************************* |
2535 | * Compile a Regular Expression * |
2536 | *************************************************/ |
2537 | |
2538 | /* This function takes a string and returns a pointer to a block of store |
2539 | holding a compiled version of the expression. The original API for this |
2540 | function had no error code return variable; it is retained for backwards |
2541 | compatibility. The new function is given a new name. |
2542 | |
2543 | Arguments: |
2544 | pattern the regular expression |
2545 | options various option bits |
2546 | errorCodePtr pointer to error code variable (pcre_compile2() only) |
2547 | can be NULL if you don't want a code value |
2548 | errorPtr pointer to pointer to error text |
2549 | erroroffset ptr offset in pattern where error was detected |
2550 | tables pointer to character tables or NULL |
2551 | |
2552 | Returns: pointer to compiled data block, or NULL on error, |
2553 | with errorPtr and erroroffset set |
2554 | */ |
2555 | |
2556 | static inline JSRegExp* returnError(ErrorCode errorcode, const char** errorPtr) |
2557 | { |
2558 | *errorPtr = errorText(code: errorcode); |
2559 | return 0; |
2560 | } |
2561 | |
2562 | JSRegExp* jsRegExpCompile(const UChar* pattern, int patternLength, |
2563 | JSRegExpIgnoreCaseOption ignoreCase, JSRegExpMultilineOption multiline, |
2564 | unsigned* numSubpatterns, const char** errorPtr) |
2565 | { |
2566 | /* We can't pass back an error message if errorPtr is NULL; I guess the best we |
2567 | can do is just return NULL, but we can set a code value if there is a code pointer. */ |
2568 | if (!errorPtr) |
2569 | return 0; |
2570 | *errorPtr = NULL; |
2571 | |
2572 | CompileData cd; |
2573 | |
2574 | ErrorCode errorcode = ERR0; |
2575 | /* Call this once just to count the brackets. */ |
2576 | calculateCompiledPatternLength(pattern, patternLength, ignoreCase, cd, errorcode); |
2577 | /* Call it again to compute the length. */ |
2578 | int length = calculateCompiledPatternLength(pattern, patternLength, ignoreCase, cd, errorcode); |
2579 | if (errorcode) |
2580 | return returnError(errorcode, errorPtr); |
2581 | |
2582 | if (length > MAX_PATTERN_SIZE) |
2583 | return returnError(errorcode: ERR16, errorPtr); |
2584 | |
2585 | size_t size = length + sizeof(JSRegExp); |
2586 | #if REGEXP_HISTOGRAM |
2587 | size_t stringOffset = (size + sizeof(UChar) - 1) / sizeof(UChar) * sizeof(UChar); |
2588 | size = stringOffset + patternLength * sizeof(UChar); |
2589 | #endif |
2590 | JSRegExp* re = reinterpret_cast<JSRegExp*>(new char[size]); |
2591 | |
2592 | if (!re) |
2593 | return returnError(errorcode: ERR13, errorPtr); |
2594 | |
2595 | re->options = (ignoreCase ? IgnoreCaseOption : 0) | (multiline ? MatchAcrossMultipleLinesOption : 0); |
2596 | |
2597 | /* The starting points of the name/number translation table and of the code are |
2598 | passed around in the compile data block. */ |
2599 | |
2600 | const unsigned char* codeStart = (const unsigned char*)(re + 1); |
2601 | |
2602 | /* Set up a starting, non-extracting bracket, then compile the expression. On |
2603 | error, errorcode will be set non-zero, so we don't need to look at the result |
2604 | of the function here. */ |
2605 | |
2606 | const UChar* ptr = (const UChar*)pattern; |
2607 | const UChar* patternEnd = pattern + patternLength; |
2608 | unsigned char* code = const_cast<unsigned char*>(codeStart); |
2609 | int firstByte, reqByte; |
2610 | int bracketCount = 0; |
2611 | if (!cd.needOuterBracket) |
2612 | compileBranch(options: re->options, brackets: &bracketCount, codePtr: &code, ptrPtr: &ptr, patternEnd, errorCodePtr: &errorcode, firstbyteptr: &firstByte, reqbyteptr: &reqByte, cd); |
2613 | else { |
2614 | *code = OP_BRA; |
2615 | compileBracket(options: re->options, brackets: &bracketCount, codePtr: &code, ptrPtr: &ptr, patternEnd, errorCodePtr: &errorcode, skipBytes: 0, firstbyteptr: &firstByte, reqbyteptr: &reqByte, cd); |
2616 | } |
2617 | re->topBracket = bracketCount; |
2618 | re->topBackref = cd.topBackref; |
2619 | |
2620 | /* If not reached end of pattern on success, there's an excess bracket. */ |
2621 | |
2622 | if (errorcode == 0 && ptr < patternEnd) |
2623 | errorcode = ERR10; |
2624 | |
2625 | /* Fill in the terminating state and check for disastrous overflow, but |
2626 | if debugging, leave the test till after things are printed out. */ |
2627 | |
2628 | *code++ = OP_END; |
2629 | |
2630 | ASSERT(code - codeStart <= length); |
2631 | if (code - codeStart > length) |
2632 | errorcode = ERR7; |
2633 | |
2634 | /* Give an error if there's back reference to a non-existent capturing |
2635 | subpattern. */ |
2636 | |
2637 | if (re->topBackref > re->topBracket) |
2638 | errorcode = ERR15; |
2639 | |
2640 | /* Failed to compile, or error while post-processing */ |
2641 | |
2642 | if (errorcode != ERR0) { |
2643 | delete [] reinterpret_cast<char*>(re); |
2644 | return returnError(errorcode, errorPtr); |
2645 | } |
2646 | |
2647 | /* If the anchored option was not passed, set the flag if we can determine that |
2648 | the pattern is anchored by virtue of ^ characters or \A or anything else (such |
2649 | as starting with .* when DOTALL is set). |
2650 | |
2651 | Otherwise, if we know what the first character has to be, save it, because that |
2652 | speeds up unanchored matches no end. If not, see if we can set the |
2653 | UseMultiLineFirstByteOptimizationOption flag. This is helpful for multiline matches when all branches |
2654 | start with ^. and also when all branches start with .* for non-DOTALL matches. |
2655 | */ |
2656 | |
2657 | if (cd.needOuterBracket ? bracketIsAnchored(code: codeStart) : branchIsAnchored(code: codeStart)) |
2658 | re->options |= IsAnchoredOption; |
2659 | else { |
2660 | if (firstByte < 0) { |
2661 | firstByte = (cd.needOuterBracket |
2662 | ? bracketFindFirstAssertedCharacter(code: codeStart, inassert: false) |
2663 | : branchFindFirstAssertedCharacter(code: codeStart, inassert: false)) |
2664 | | ((re->options & IgnoreCaseOption) ? REQ_IGNORE_CASE : 0); |
2665 | } |
2666 | if (firstByte >= 0) { |
2667 | int ch = firstByte & 255; |
2668 | if (ch < 127) { |
2669 | re->firstByte = ((firstByte & REQ_IGNORE_CASE) && flipCase(c: ch) == ch) ? ch : firstByte; |
2670 | re->options |= UseFirstByteOptimizationOption; |
2671 | } |
2672 | } else { |
2673 | if (cd.needOuterBracket ? bracketNeedsLineStart(code: codeStart, captureMap: 0, backrefMap: cd.backrefMap) : branchNeedsLineStart(code: codeStart, captureMap: 0, backrefMap: cd.backrefMap)) |
2674 | re->options |= UseMultiLineFirstByteOptimizationOption; |
2675 | } |
2676 | } |
2677 | |
2678 | /* For an anchored pattern, we use the "required byte" only if it follows a |
2679 | variable length item in the regex. Remove the caseless flag for non-caseable |
2680 | bytes. */ |
2681 | |
2682 | if (reqByte >= 0 && (!(re->options & IsAnchoredOption) || (reqByte & REQ_VARY))) { |
2683 | int ch = reqByte & 255; |
2684 | if (ch < 127) { |
2685 | re->reqByte = ((reqByte & REQ_IGNORE_CASE) && flipCase(c: ch) == ch) ? (reqByte & ~REQ_IGNORE_CASE) : reqByte; |
2686 | re->options |= UseRequiredByteOptimizationOption; |
2687 | } |
2688 | } |
2689 | |
2690 | #if REGEXP_HISTOGRAM |
2691 | re->stringOffset = stringOffset; |
2692 | re->stringLength = patternLength; |
2693 | memcpy(reinterpret_cast<char*>(re) + stringOffset, pattern, patternLength * 2); |
2694 | #endif |
2695 | |
2696 | if (numSubpatterns) |
2697 | *numSubpatterns = re->topBracket; |
2698 | return re; |
2699 | } |
2700 | |
2701 | void jsRegExpFree(JSRegExp* re) |
2702 | { |
2703 | delete [] reinterpret_cast<char*>(re); |
2704 | } |
2705 | |