1 | // Copyright 2003-2009 The RE2 Authors. All Rights Reserved. |
2 | // Use of this source code is governed by a BSD-style |
3 | // license that can be found in the LICENSE file. |
4 | |
5 | #ifndef RE2_RE2_H_ |
6 | #define RE2_RE2_H_ |
7 | |
8 | // C++ interface to the re2 regular-expression library. |
9 | // RE2 supports Perl-style regular expressions (with extensions like |
10 | // \d, \w, \s, ...). |
11 | // |
12 | // ----------------------------------------------------------------------- |
13 | // REGEXP SYNTAX: |
14 | // |
15 | // This module uses the re2 library and hence supports |
16 | // its syntax for regular expressions, which is similar to Perl's with |
17 | // some of the more complicated things thrown away. In particular, |
18 | // backreferences and generalized assertions are not available, nor is \Z. |
19 | // |
20 | // See https://github.com/google/re2/wiki/Syntax for the syntax |
21 | // supported by RE2, and a comparison with PCRE and PERL regexps. |
22 | // |
23 | // For those not familiar with Perl's regular expressions, |
24 | // here are some examples of the most commonly used extensions: |
25 | // |
26 | // "hello (\\w+) world" -- \w matches a "word" character |
27 | // "version (\\d+)" -- \d matches a digit |
28 | // "hello\\s+world" -- \s matches any whitespace character |
29 | // "\\b(\\w+)\\b" -- \b matches non-empty string at word boundary |
30 | // "(?i)hello" -- (?i) turns on case-insensitive matching |
31 | // "/\\*(.*?)\\*/" -- .*? matches . minimum no. of times possible |
32 | // |
33 | // The double backslashes are needed when writing C++ string literals. |
34 | // However, they should NOT be used when writing C++11 raw string literals: |
35 | // |
36 | // R"(hello (\w+) world)" -- \w matches a "word" character |
37 | // R"(version (\d+))" -- \d matches a digit |
38 | // R"(hello\s+world)" -- \s matches any whitespace character |
39 | // R"(\b(\w+)\b)" -- \b matches non-empty string at word boundary |
40 | // R"((?i)hello)" -- (?i) turns on case-insensitive matching |
41 | // R"(/\*(.*?)\*/)" -- .*? matches . minimum no. of times possible |
42 | // |
43 | // When using UTF-8 encoding, case-insensitive matching will perform |
44 | // simple case folding, not full case folding. |
45 | // |
46 | // ----------------------------------------------------------------------- |
47 | // MATCHING INTERFACE: |
48 | // |
49 | // The "FullMatch" operation checks that supplied text matches a |
50 | // supplied pattern exactly. |
51 | // |
52 | // Example: successful match |
53 | // ABSL_CHECK(RE2::FullMatch("hello", "h.*o")); |
54 | // |
55 | // Example: unsuccessful match (requires full match): |
56 | // ABSL_CHECK(!RE2::FullMatch("hello", "e")); |
57 | // |
58 | // ----------------------------------------------------------------------- |
59 | // UTF-8 AND THE MATCHING INTERFACE: |
60 | // |
61 | // By default, the pattern and input text are interpreted as UTF-8. |
62 | // The RE2::Latin1 option causes them to be interpreted as Latin-1. |
63 | // |
64 | // Example: |
65 | // ABSL_CHECK(RE2::FullMatch(utf8_string, RE2(utf8_pattern))); |
66 | // ABSL_CHECK(RE2::FullMatch(latin1_string, RE2(latin1_pattern, |
67 | // RE2::Latin1))); |
68 | // |
69 | // ----------------------------------------------------------------------- |
70 | // SUBMATCH EXTRACTION: |
71 | // |
72 | // You can supply extra pointer arguments to extract submatches. |
73 | // On match failure, none of the pointees will have been modified. |
74 | // On match success, the submatches will be converted (as necessary) and |
75 | // their values will be assigned to their pointees until all conversions |
76 | // have succeeded or one conversion has failed. |
77 | // On conversion failure, the pointees will be in an indeterminate state |
78 | // because the caller has no way of knowing which conversion failed. |
79 | // However, conversion cannot fail for types like string and string_view |
80 | // that do not inspect the submatch contents. Hence, in the common case |
81 | // where all of the pointees are of such types, failure is always due to |
82 | // match failure and thus none of the pointees will have been modified. |
83 | // |
84 | // Example: extracts "ruby" into "s" and 1234 into "i" |
85 | // int i; |
86 | // std::string s; |
87 | // ABSL_CHECK(RE2::FullMatch("ruby:1234", "(\\w+):(\\d+)", &s, &i)); |
88 | // |
89 | // Example: extracts "ruby" into "s" and no value into "i" |
90 | // absl::optional<int> i; |
91 | // std::string s; |
92 | // ABSL_CHECK(RE2::FullMatch("ruby", "(\\w+)(?::(\\d+))?", &s, &i)); |
93 | // |
94 | // Example: fails because string cannot be stored in integer |
95 | // ABSL_CHECK(!RE2::FullMatch("ruby", "(.*)", &i)); |
96 | // |
97 | // Example: fails because there aren't enough sub-patterns |
98 | // ABSL_CHECK(!RE2::FullMatch("ruby:1234", "\\w+:\\d+", &s)); |
99 | // |
100 | // Example: does not try to extract any extra sub-patterns |
101 | // ABSL_CHECK(RE2::FullMatch("ruby:1234", "(\\w+):(\\d+)", &s)); |
102 | // |
103 | // Example: does not try to extract into NULL |
104 | // ABSL_CHECK(RE2::FullMatch("ruby:1234", "(\\w+):(\\d+)", NULL, &i)); |
105 | // |
106 | // Example: integer overflow causes failure |
107 | // ABSL_CHECK(!RE2::FullMatch("ruby:1234567891234", "\\w+:(\\d+)", &i)); |
108 | // |
109 | // NOTE(rsc): Asking for submatches slows successful matches quite a bit. |
110 | // This may get a little faster in the future, but right now is slower |
111 | // than PCRE. On the other hand, failed matches run *very* fast (faster |
112 | // than PCRE), as do matches without submatch extraction. |
113 | // |
114 | // ----------------------------------------------------------------------- |
115 | // PARTIAL MATCHES |
116 | // |
117 | // You can use the "PartialMatch" operation when you want the pattern |
118 | // to match any substring of the text. |
119 | // |
120 | // Example: simple search for a string: |
121 | // ABSL_CHECK(RE2::PartialMatch("hello", "ell")); |
122 | // |
123 | // Example: find first number in a string |
124 | // int number; |
125 | // ABSL_CHECK(RE2::PartialMatch("x*100 + 20", "(\\d+)", &number)); |
126 | // ABSL_CHECK_EQ(number, 100); |
127 | // |
128 | // ----------------------------------------------------------------------- |
129 | // PRE-COMPILED REGULAR EXPRESSIONS |
130 | // |
131 | // RE2 makes it easy to use any string as a regular expression, without |
132 | // requiring a separate compilation step. |
133 | // |
134 | // If speed is of the essence, you can create a pre-compiled "RE2" |
135 | // object from the pattern and use it multiple times. If you do so, |
136 | // you can typically parse text faster than with sscanf. |
137 | // |
138 | // Example: precompile pattern for faster matching: |
139 | // RE2 pattern("h.*o"); |
140 | // while (ReadLine(&str)) { |
141 | // if (RE2::FullMatch(str, pattern)) ...; |
142 | // } |
143 | // |
144 | // ----------------------------------------------------------------------- |
145 | // SCANNING TEXT INCREMENTALLY |
146 | // |
147 | // The "Consume" operation may be useful if you want to repeatedly |
148 | // match regular expressions at the front of a string and skip over |
149 | // them as they match. This requires use of the string_view type, |
150 | // which represents a sub-range of a real string. |
151 | // |
152 | // Example: read lines of the form "var = value" from a string. |
153 | // std::string contents = ...; // Fill string somehow |
154 | // absl::string_view input(contents); // Wrap a string_view around it |
155 | // |
156 | // std::string var; |
157 | // int value; |
158 | // while (RE2::Consume(&input, "(\\w+) = (\\d+)\n", &var, &value)) { |
159 | // ...; |
160 | // } |
161 | // |
162 | // Each successful call to "Consume" will set "var/value", and also |
163 | // advance "input" so it points past the matched text. Note that if the |
164 | // regular expression matches an empty string, input will advance |
165 | // by 0 bytes. If the regular expression being used might match |
166 | // an empty string, the loop body must check for this case and either |
167 | // advance the string or break out of the loop. |
168 | // |
169 | // The "FindAndConsume" operation is similar to "Consume" but does not |
170 | // anchor your match at the beginning of the string. For example, you |
171 | // could extract all words from a string by repeatedly calling |
172 | // RE2::FindAndConsume(&input, "(\\w+)", &word) |
173 | // |
174 | // ----------------------------------------------------------------------- |
175 | // USING VARIABLE NUMBER OF ARGUMENTS |
176 | // |
177 | // The above operations require you to know the number of arguments |
178 | // when you write the code. This is not always possible or easy (for |
179 | // example, the regular expression may be calculated at run time). |
180 | // You can use the "N" version of the operations when the number of |
181 | // match arguments are determined at run time. |
182 | // |
183 | // Example: |
184 | // const RE2::Arg* args[10]; |
185 | // int n; |
186 | // // ... populate args with pointers to RE2::Arg values ... |
187 | // // ... set n to the number of RE2::Arg objects ... |
188 | // bool match = RE2::FullMatchN(input, pattern, args, n); |
189 | // |
190 | // The last statement is equivalent to |
191 | // |
192 | // bool match = RE2::FullMatch(input, pattern, |
193 | // *args[0], *args[1], ..., *args[n - 1]); |
194 | // |
195 | // ----------------------------------------------------------------------- |
196 | // PARSING HEX/OCTAL/C-RADIX NUMBERS |
197 | // |
198 | // By default, if you pass a pointer to a numeric value, the |
199 | // corresponding text is interpreted as a base-10 number. You can |
200 | // instead wrap the pointer with a call to one of the operators Hex(), |
201 | // Octal(), or CRadix() to interpret the text in another base. The |
202 | // CRadix operator interprets C-style "0" (base-8) and "0x" (base-16) |
203 | // prefixes, but defaults to base-10. |
204 | // |
205 | // Example: |
206 | // int a, b, c, d; |
207 | // ABSL_CHECK(RE2::FullMatch("100 40 0100 0x40", "(.*) (.*) (.*) (.*)", |
208 | // RE2::Octal(&a), RE2::Hex(&b), RE2::CRadix(&c), RE2::CRadix(&d)); |
209 | // will leave 64 in a, b, c, and d. |
210 | |
211 | #include <stddef.h> |
212 | #include <stdint.h> |
213 | |
214 | #include <algorithm> |
215 | #include <map> |
216 | #include <string> |
217 | #include <type_traits> |
218 | #include <vector> |
219 | |
220 | #include "absl/base/call_once.h" |
221 | #include "absl/strings/string_view.h" |
222 | #include "absl/types/optional.h" |
223 | #include "re2/stringpiece.h" |
224 | |
225 | #if defined(__APPLE__) |
226 | #include <TargetConditionals.h> |
227 | #endif |
228 | |
229 | namespace re2 { |
230 | class Prog; |
231 | class Regexp; |
232 | } // namespace re2 |
233 | |
234 | namespace re2 { |
235 | |
236 | // Interface for regular expression matching. Also corresponds to a |
237 | // pre-compiled regular expression. An "RE2" object is safe for |
238 | // concurrent use by multiple threads. |
239 | class RE2 { |
240 | public: |
241 | // We convert user-passed pointers into special Arg objects |
242 | class Arg; |
243 | class Options; |
244 | |
245 | // Defined in set.h. |
246 | class Set; |
247 | |
248 | enum ErrorCode { |
249 | NoError = 0, |
250 | |
251 | // Unexpected error |
252 | ErrorInternal, |
253 | |
254 | // Parse errors |
255 | ErrorBadEscape, // bad escape sequence |
256 | ErrorBadCharClass, // bad character class |
257 | ErrorBadCharRange, // bad character class range |
258 | ErrorMissingBracket, // missing closing ] |
259 | ErrorMissingParen, // missing closing ) |
260 | ErrorUnexpectedParen, // unexpected closing ) |
261 | ErrorTrailingBackslash, // trailing \ at end of regexp |
262 | ErrorRepeatArgument, // repeat argument missing, e.g. "*" |
263 | ErrorRepeatSize, // bad repetition argument |
264 | ErrorRepeatOp, // bad repetition operator |
265 | ErrorBadPerlOp, // bad perl operator |
266 | ErrorBadUTF8, // invalid UTF-8 in regexp |
267 | ErrorBadNamedCapture, // bad named capture group |
268 | ErrorPatternTooLarge // pattern too large (compile failed) |
269 | }; |
270 | |
271 | // Predefined common options. |
272 | // If you need more complicated things, instantiate |
273 | // an Option class, possibly passing one of these to |
274 | // the Option constructor, change the settings, and pass that |
275 | // Option class to the RE2 constructor. |
276 | enum CannedOptions { |
277 | DefaultOptions = 0, |
278 | Latin1, // treat input as Latin-1 (default UTF-8) |
279 | POSIX, // POSIX syntax, leftmost-longest match |
280 | Quiet // do not log about regexp parse errors |
281 | }; |
282 | |
283 | // Need to have the const char* and const std::string& forms for implicit |
284 | // conversions when passing string literals to FullMatch and PartialMatch. |
285 | // Otherwise the absl::string_view form would be sufficient. |
286 | RE2(const char* pattern); |
287 | RE2(const std::string& pattern); |
288 | RE2(absl::string_view pattern); |
289 | RE2(absl::string_view pattern, const Options& options); |
290 | ~RE2(); |
291 | |
292 | // Not copyable. |
293 | // RE2 objects are expensive. You should probably use std::shared_ptr<RE2> |
294 | // instead. If you really must copy, RE2(first.pattern(), first.options()) |
295 | // effectively does so: it produces a second object that mimics the first. |
296 | RE2(const RE2&) = delete; |
297 | RE2& operator=(const RE2&) = delete; |
298 | // Not movable. |
299 | // RE2 objects are thread-safe and logically immutable. You should probably |
300 | // use std::unique_ptr<RE2> instead. Otherwise, consider std::deque<RE2> if |
301 | // direct emplacement into a container is desired. If you really must move, |
302 | // be prepared to submit a design document along with your feature request. |
303 | RE2(RE2&&) = delete; |
304 | RE2& operator=(RE2&&) = delete; |
305 | |
306 | // Returns whether RE2 was created properly. |
307 | bool ok() const { return error_code() == NoError; } |
308 | |
309 | // The string specification for this RE2. E.g. |
310 | // RE2 re("ab*c?d+"); |
311 | // re.pattern(); // "ab*c?d+" |
312 | const std::string& pattern() const { return *pattern_; } |
313 | |
314 | // If RE2 could not be created properly, returns an error string. |
315 | // Else returns the empty string. |
316 | const std::string& error() const { return *error_; } |
317 | |
318 | // If RE2 could not be created properly, returns an error code. |
319 | // Else returns RE2::NoError (== 0). |
320 | ErrorCode error_code() const { return error_code_; } |
321 | |
322 | // If RE2 could not be created properly, returns the offending |
323 | // portion of the regexp. |
324 | const std::string& error_arg() const { return *error_arg_; } |
325 | |
326 | // Returns the program size, a very approximate measure of a regexp's "cost". |
327 | // Larger numbers are more expensive than smaller numbers. |
328 | int ProgramSize() const; |
329 | int ReverseProgramSize() const; |
330 | |
331 | // If histogram is not null, outputs the program fanout |
332 | // as a histogram bucketed by powers of 2. |
333 | // Returns the number of the largest non-empty bucket. |
334 | int ProgramFanout(std::vector<int>* histogram) const; |
335 | int ReverseProgramFanout(std::vector<int>* histogram) const; |
336 | |
337 | // Returns the underlying Regexp; not for general use. |
338 | // Returns entire_regexp_ so that callers don't need |
339 | // to know about prefix_ and prefix_foldcase_. |
340 | re2::Regexp* Regexp() const { return entire_regexp_; } |
341 | |
342 | /***** The array-based matching interface ******/ |
343 | |
344 | // The functions here have names ending in 'N' and are used to implement |
345 | // the functions whose names are the prefix before the 'N'. It is sometimes |
346 | // useful to invoke them directly, but the syntax is awkward, so the 'N'-less |
347 | // versions should be preferred. |
348 | static bool FullMatchN(absl::string_view text, const RE2& re, |
349 | const Arg* const args[], int n); |
350 | static bool PartialMatchN(absl::string_view text, const RE2& re, |
351 | const Arg* const args[], int n); |
352 | static bool ConsumeN(absl::string_view* input, const RE2& re, |
353 | const Arg* const args[], int n); |
354 | static bool FindAndConsumeN(absl::string_view* input, const RE2& re, |
355 | const Arg* const args[], int n); |
356 | |
357 | private: |
358 | template <typename F, typename SP> |
359 | static inline bool Apply(F f, SP sp, const RE2& re) { |
360 | return f(sp, re, NULL, 0); |
361 | } |
362 | |
363 | template <typename F, typename SP, typename... A> |
364 | static inline bool Apply(F f, SP sp, const RE2& re, const A&... a) { |
365 | const Arg* const args[] = {&a...}; |
366 | const int n = sizeof...(a); |
367 | return f(sp, re, args, n); |
368 | } |
369 | |
370 | public: |
371 | // In order to allow FullMatch() et al. to be called with a varying number |
372 | // of arguments of varying types, we use two layers of variadic templates. |
373 | // The first layer constructs the temporary Arg objects. The second layer |
374 | // (above) constructs the array of pointers to the temporary Arg objects. |
375 | |
376 | /***** The useful part: the matching interface *****/ |
377 | |
378 | // Matches "text" against "re". If pointer arguments are |
379 | // supplied, copies matched sub-patterns into them. |
380 | // |
381 | // You can pass in a "const char*" or a "std::string" for "text". |
382 | // You can pass in a "const char*" or a "std::string" or a "RE2" for "re". |
383 | // |
384 | // The provided pointer arguments can be pointers to any scalar numeric |
385 | // type, or one of: |
386 | // std::string (matched piece is copied to string) |
387 | // absl::string_view (string_view is mutated to point to matched piece) |
388 | // absl::optional<T> (T is a supported numeric or string type as above) |
389 | // T ("bool T::ParseFrom(const char*, size_t)" must exist) |
390 | // (void*)NULL (the corresponding matched sub-pattern is not copied) |
391 | // |
392 | // Returns true iff all of the following conditions are satisfied: |
393 | // a. "text" matches "re" fully - from the beginning to the end of "text". |
394 | // b. The number of matched sub-patterns is >= number of supplied pointers. |
395 | // c. The "i"th argument has a suitable type for holding the |
396 | // string captured as the "i"th sub-pattern. If you pass in |
397 | // NULL for the "i"th argument, or pass fewer arguments than |
398 | // number of sub-patterns, the "i"th captured sub-pattern is |
399 | // ignored. |
400 | // |
401 | // CAVEAT: An optional sub-pattern that does not exist in the |
402 | // matched string is assigned the null string. Therefore, the |
403 | // following returns false because the null string - absence of |
404 | // a string (not even the empty string) - is not a valid number: |
405 | // |
406 | // int number; |
407 | // RE2::FullMatch("abc", "[a-z]+(\\d+)?", &number); |
408 | // |
409 | // Use absl::optional<int> instead to handle this case correctly. |
410 | template <typename... A> |
411 | static bool FullMatch(absl::string_view text, const RE2& re, A&&... a) { |
412 | return Apply(FullMatchN, text, re, Arg(std::forward<A>(a))...); |
413 | } |
414 | |
415 | // Like FullMatch(), except that "re" is allowed to match a substring |
416 | // of "text". |
417 | // |
418 | // Returns true iff all of the following conditions are satisfied: |
419 | // a. "text" matches "re" partially - for some substring of "text". |
420 | // b. The number of matched sub-patterns is >= number of supplied pointers. |
421 | // c. The "i"th argument has a suitable type for holding the |
422 | // string captured as the "i"th sub-pattern. If you pass in |
423 | // NULL for the "i"th argument, or pass fewer arguments than |
424 | // number of sub-patterns, the "i"th captured sub-pattern is |
425 | // ignored. |
426 | template <typename... A> |
427 | static bool PartialMatch(absl::string_view text, const RE2& re, A&&... a) { |
428 | return Apply(PartialMatchN, text, re, Arg(std::forward<A>(a))...); |
429 | } |
430 | |
431 | // Like FullMatch() and PartialMatch(), except that "re" has to match |
432 | // a prefix of the text, and "input" is advanced past the matched |
433 | // text. Note: "input" is modified iff this routine returns true |
434 | // and "re" matched a non-empty substring of "input". |
435 | // |
436 | // Returns true iff all of the following conditions are satisfied: |
437 | // a. "input" matches "re" partially - for some prefix of "input". |
438 | // b. The number of matched sub-patterns is >= number of supplied pointers. |
439 | // c. The "i"th argument has a suitable type for holding the |
440 | // string captured as the "i"th sub-pattern. If you pass in |
441 | // NULL for the "i"th argument, or pass fewer arguments than |
442 | // number of sub-patterns, the "i"th captured sub-pattern is |
443 | // ignored. |
444 | template <typename... A> |
445 | static bool Consume(absl::string_view* input, const RE2& re, A&&... a) { |
446 | return Apply(ConsumeN, input, re, Arg(std::forward<A>(a))...); |
447 | } |
448 | |
449 | // Like Consume(), but does not anchor the match at the beginning of |
450 | // the text. That is, "re" need not start its match at the beginning |
451 | // of "input". For example, "FindAndConsume(s, "(\\w+)", &word)" finds |
452 | // the next word in "s" and stores it in "word". |
453 | // |
454 | // Returns true iff all of the following conditions are satisfied: |
455 | // a. "input" matches "re" partially - for some substring of "input". |
456 | // b. The number of matched sub-patterns is >= number of supplied pointers. |
457 | // c. The "i"th argument has a suitable type for holding the |
458 | // string captured as the "i"th sub-pattern. If you pass in |
459 | // NULL for the "i"th argument, or pass fewer arguments than |
460 | // number of sub-patterns, the "i"th captured sub-pattern is |
461 | // ignored. |
462 | template <typename... A> |
463 | static bool FindAndConsume(absl::string_view* input, const RE2& re, A&&... a) { |
464 | return Apply(FindAndConsumeN, input, re, Arg(std::forward<A>(a))...); |
465 | } |
466 | |
467 | // Replace the first match of "re" in "str" with "rewrite". |
468 | // Within "rewrite", backslash-escaped digits (\1 to \9) can be |
469 | // used to insert text matching corresponding parenthesized group |
470 | // from the pattern. \0 in "rewrite" refers to the entire matching |
471 | // text. E.g., |
472 | // |
473 | // std::string s = "yabba dabba doo"; |
474 | // ABSL_CHECK(RE2::Replace(&s, "b+", "d")); |
475 | // |
476 | // will leave "s" containing "yada dabba doo" |
477 | // |
478 | // Returns true if the pattern matches and a replacement occurs, |
479 | // false otherwise. |
480 | static bool Replace(std::string* str, |
481 | const RE2& re, |
482 | absl::string_view rewrite); |
483 | |
484 | // Like Replace(), except replaces successive non-overlapping occurrences |
485 | // of the pattern in the string with the rewrite. E.g. |
486 | // |
487 | // std::string s = "yabba dabba doo"; |
488 | // ABSL_CHECK(RE2::GlobalReplace(&s, "b+", "d")); |
489 | // |
490 | // will leave "s" containing "yada dada doo" |
491 | // Replacements are not subject to re-matching. |
492 | // |
493 | // Because GlobalReplace only replaces non-overlapping matches, |
494 | // replacing "ana" within "banana" makes only one replacement, not two. |
495 | // |
496 | // Returns the number of replacements made. |
497 | static int GlobalReplace(std::string* str, |
498 | const RE2& re, |
499 | absl::string_view rewrite); |
500 | |
501 | // Like Replace, except that if the pattern matches, "rewrite" |
502 | // is copied into "out" with substitutions. The non-matching |
503 | // portions of "text" are ignored. |
504 | // |
505 | // Returns true iff a match occurred and the extraction happened |
506 | // successfully; if no match occurs, the string is left unaffected. |
507 | // |
508 | // REQUIRES: "text" must not alias any part of "*out". |
509 | static bool (absl::string_view text, |
510 | const RE2& re, |
511 | absl::string_view rewrite, |
512 | std::string* out); |
513 | |
514 | // Escapes all potentially meaningful regexp characters in |
515 | // 'unquoted'. The returned string, used as a regular expression, |
516 | // will match exactly the original string. For example, |
517 | // 1.5-2.0? |
518 | // may become: |
519 | // 1\.5\-2\.0\? |
520 | static std::string QuoteMeta(absl::string_view unquoted); |
521 | |
522 | // Computes range for any strings matching regexp. The min and max can in |
523 | // some cases be arbitrarily precise, so the caller gets to specify the |
524 | // maximum desired length of string returned. |
525 | // |
526 | // Assuming PossibleMatchRange(&min, &max, N) returns successfully, any |
527 | // string s that is an anchored match for this regexp satisfies |
528 | // min <= s && s <= max. |
529 | // |
530 | // Note that PossibleMatchRange() will only consider the first copy of an |
531 | // infinitely repeated element (i.e., any regexp element followed by a '*' or |
532 | // '+' operator). Regexps with "{N}" constructions are not affected, as those |
533 | // do not compile down to infinite repetitions. |
534 | // |
535 | // Returns true on success, false on error. |
536 | bool PossibleMatchRange(std::string* min, std::string* max, |
537 | int maxlen) const; |
538 | |
539 | // Generic matching interface |
540 | |
541 | // Type of match. |
542 | enum Anchor { |
543 | UNANCHORED, // No anchoring |
544 | ANCHOR_START, // Anchor at start only |
545 | ANCHOR_BOTH // Anchor at start and end |
546 | }; |
547 | |
548 | // Return the number of capturing sub-patterns, or -1 if the |
549 | // regexp wasn't valid on construction. The overall match ($0) |
550 | // does not count: if the regexp is "(a)(b)", returns 2. |
551 | int NumberOfCapturingGroups() const { return num_captures_; } |
552 | |
553 | // Return a map from names to capturing indices. |
554 | // The map records the index of the leftmost group |
555 | // with the given name. |
556 | // Only valid until the re is deleted. |
557 | const std::map<std::string, int>& NamedCapturingGroups() const; |
558 | |
559 | // Return a map from capturing indices to names. |
560 | // The map has no entries for unnamed groups. |
561 | // Only valid until the re is deleted. |
562 | const std::map<int, std::string>& CapturingGroupNames() const; |
563 | |
564 | // General matching routine. |
565 | // Match against text starting at offset startpos |
566 | // and stopping the search at offset endpos. |
567 | // Returns true if match found, false if not. |
568 | // On a successful match, fills in submatch[] (up to nsubmatch entries) |
569 | // with information about submatches. |
570 | // I.e. matching RE2("(foo)|(bar)baz") on "barbazbla" will return true, with |
571 | // submatch[0] = "barbaz", submatch[1].data() = NULL, submatch[2] = "bar", |
572 | // submatch[3].data() = NULL, ..., up to submatch[nsubmatch-1].data() = NULL. |
573 | // Caveat: submatch[] may be clobbered even on match failure. |
574 | // |
575 | // Don't ask for more match information than you will use: |
576 | // runs much faster with nsubmatch == 1 than nsubmatch > 1, and |
577 | // runs even faster if nsubmatch == 0. |
578 | // Doesn't make sense to use nsubmatch > 1 + NumberOfCapturingGroups(), |
579 | // but will be handled correctly. |
580 | // |
581 | // Passing text == absl::string_view() will be handled like any other |
582 | // empty string, but note that on return, it will not be possible to tell |
583 | // whether submatch i matched the empty string or did not match: |
584 | // either way, submatch[i].data() == NULL. |
585 | bool Match(absl::string_view text, |
586 | size_t startpos, |
587 | size_t endpos, |
588 | Anchor re_anchor, |
589 | absl::string_view* submatch, |
590 | int nsubmatch) const; |
591 | |
592 | // Check that the given rewrite string is suitable for use with this |
593 | // regular expression. It checks that: |
594 | // * The regular expression has enough parenthesized subexpressions |
595 | // to satisfy all of the \N tokens in rewrite |
596 | // * The rewrite string doesn't have any syntax errors. E.g., |
597 | // '\' followed by anything other than a digit or '\'. |
598 | // A true return value guarantees that Replace() and Extract() won't |
599 | // fail because of a bad rewrite string. |
600 | bool CheckRewriteString(absl::string_view rewrite, |
601 | std::string* error) const; |
602 | |
603 | // Returns the maximum submatch needed for the rewrite to be done by |
604 | // Replace(). E.g. if rewrite == "foo \\2,\\1", returns 2. |
605 | static int MaxSubmatch(absl::string_view rewrite); |
606 | |
607 | // Append the "rewrite" string, with backslash substitutions from "vec", |
608 | // to string "out". |
609 | // Returns true on success. This method can fail because of a malformed |
610 | // rewrite string. CheckRewriteString guarantees that the rewrite will |
611 | // be sucessful. |
612 | bool Rewrite(std::string* out, |
613 | absl::string_view rewrite, |
614 | const absl::string_view* vec, |
615 | int veclen) const; |
616 | |
617 | // Constructor options |
618 | class Options { |
619 | public: |
620 | // The options are (defaults in parentheses): |
621 | // |
622 | // utf8 (true) text and pattern are UTF-8; otherwise Latin-1 |
623 | // posix_syntax (false) restrict regexps to POSIX egrep syntax |
624 | // longest_match (false) search for longest match, not first match |
625 | // log_errors (true) log syntax and execution errors to ERROR |
626 | // max_mem (see below) approx. max memory footprint of RE2 |
627 | // literal (false) interpret string as literal, not regexp |
628 | // never_nl (false) never match \n, even if it is in regexp |
629 | // dot_nl (false) dot matches everything including new line |
630 | // never_capture (false) parse all parens as non-capturing |
631 | // case_sensitive (true) match is case-sensitive (regexp can override |
632 | // with (?i) unless in posix_syntax mode) |
633 | // |
634 | // The following options are only consulted when posix_syntax == true. |
635 | // When posix_syntax == false, these features are always enabled and |
636 | // cannot be turned off; to perform multi-line matching in that case, |
637 | // begin the regexp with (?m). |
638 | // perl_classes (false) allow Perl's \d \s \w \D \S \W |
639 | // word_boundary (false) allow Perl's \b \B (word boundary and not) |
640 | // one_line (false) ^ and $ only match beginning and end of text |
641 | // |
642 | // The max_mem option controls how much memory can be used |
643 | // to hold the compiled form of the regexp (the Prog) and |
644 | // its cached DFA graphs. Code Search placed limits on the number |
645 | // of Prog instructions and DFA states: 10,000 for both. |
646 | // In RE2, those limits would translate to about 240 KB per Prog |
647 | // and perhaps 2.5 MB per DFA (DFA state sizes vary by regexp; RE2 does a |
648 | // better job of keeping them small than Code Search did). |
649 | // Each RE2 has two Progs (one forward, one reverse), and each Prog |
650 | // can have two DFAs (one first match, one longest match). |
651 | // That makes 4 DFAs: |
652 | // |
653 | // forward, first-match - used for UNANCHORED or ANCHOR_START searches |
654 | // if opt.longest_match() == false |
655 | // forward, longest-match - used for all ANCHOR_BOTH searches, |
656 | // and the other two kinds if |
657 | // opt.longest_match() == true |
658 | // reverse, first-match - never used |
659 | // reverse, longest-match - used as second phase for unanchored searches |
660 | // |
661 | // The RE2 memory budget is statically divided between the two |
662 | // Progs and then the DFAs: two thirds to the forward Prog |
663 | // and one third to the reverse Prog. The forward Prog gives half |
664 | // of what it has left over to each of its DFAs. The reverse Prog |
665 | // gives it all to its longest-match DFA. |
666 | // |
667 | // Once a DFA fills its budget, it flushes its cache and starts over. |
668 | // If this happens too often, RE2 falls back on the NFA implementation. |
669 | |
670 | // For now, make the default budget something close to Code Search. |
671 | static const int kDefaultMaxMem = 8<<20; |
672 | |
673 | enum Encoding { |
674 | EncodingUTF8 = 1, |
675 | EncodingLatin1 |
676 | }; |
677 | |
678 | Options() : |
679 | max_mem_(kDefaultMaxMem), |
680 | encoding_(EncodingUTF8), |
681 | posix_syntax_(false), |
682 | longest_match_(false), |
683 | log_errors_(true), |
684 | literal_(false), |
685 | never_nl_(false), |
686 | dot_nl_(false), |
687 | never_capture_(false), |
688 | case_sensitive_(true), |
689 | perl_classes_(false), |
690 | word_boundary_(false), |
691 | one_line_(false) { |
692 | } |
693 | |
694 | /*implicit*/ Options(CannedOptions); |
695 | |
696 | int64_t max_mem() const { return max_mem_; } |
697 | void set_max_mem(int64_t m) { max_mem_ = m; } |
698 | |
699 | Encoding encoding() const { return encoding_; } |
700 | void set_encoding(Encoding encoding) { encoding_ = encoding; } |
701 | |
702 | bool posix_syntax() const { return posix_syntax_; } |
703 | void set_posix_syntax(bool b) { posix_syntax_ = b; } |
704 | |
705 | bool longest_match() const { return longest_match_; } |
706 | void set_longest_match(bool b) { longest_match_ = b; } |
707 | |
708 | bool log_errors() const { return log_errors_; } |
709 | void set_log_errors(bool b) { log_errors_ = b; } |
710 | |
711 | bool literal() const { return literal_; } |
712 | void set_literal(bool b) { literal_ = b; } |
713 | |
714 | bool never_nl() const { return never_nl_; } |
715 | void set_never_nl(bool b) { never_nl_ = b; } |
716 | |
717 | bool dot_nl() const { return dot_nl_; } |
718 | void set_dot_nl(bool b) { dot_nl_ = b; } |
719 | |
720 | bool never_capture() const { return never_capture_; } |
721 | void set_never_capture(bool b) { never_capture_ = b; } |
722 | |
723 | bool case_sensitive() const { return case_sensitive_; } |
724 | void set_case_sensitive(bool b) { case_sensitive_ = b; } |
725 | |
726 | bool perl_classes() const { return perl_classes_; } |
727 | void set_perl_classes(bool b) { perl_classes_ = b; } |
728 | |
729 | bool word_boundary() const { return word_boundary_; } |
730 | void set_word_boundary(bool b) { word_boundary_ = b; } |
731 | |
732 | bool one_line() const { return one_line_; } |
733 | void set_one_line(bool b) { one_line_ = b; } |
734 | |
735 | void Copy(const Options& src) { |
736 | *this = src; |
737 | } |
738 | |
739 | int ParseFlags() const; |
740 | |
741 | private: |
742 | int64_t max_mem_; |
743 | Encoding encoding_; |
744 | bool posix_syntax_; |
745 | bool longest_match_; |
746 | bool log_errors_; |
747 | bool literal_; |
748 | bool never_nl_; |
749 | bool dot_nl_; |
750 | bool never_capture_; |
751 | bool case_sensitive_; |
752 | bool perl_classes_; |
753 | bool word_boundary_; |
754 | bool one_line_; |
755 | }; |
756 | |
757 | // Returns the options set in the constructor. |
758 | const Options& options() const { return options_; } |
759 | |
760 | // Argument converters; see below. |
761 | template <typename T> |
762 | static Arg CRadix(T* ptr); |
763 | template <typename T> |
764 | static Arg Hex(T* ptr); |
765 | template <typename T> |
766 | static Arg Octal(T* ptr); |
767 | |
768 | // Controls the maximum count permitted by GlobalReplace(); -1 is unlimited. |
769 | // FOR FUZZING ONLY. |
770 | static void FUZZING_ONLY_set_maximum_global_replace_count(int i); |
771 | |
772 | private: |
773 | void Init(absl::string_view pattern, const Options& options); |
774 | |
775 | bool DoMatch(absl::string_view text, |
776 | Anchor re_anchor, |
777 | size_t* consumed, |
778 | const Arg* const args[], |
779 | int n) const; |
780 | |
781 | re2::Prog* ReverseProg() const; |
782 | |
783 | // First cache line is relatively cold fields. |
784 | const std::string* pattern_; // string regular expression |
785 | Options options_; // option flags |
786 | re2::Regexp* entire_regexp_; // parsed regular expression |
787 | re2::Regexp* suffix_regexp_; // parsed regular expression, prefix_ removed |
788 | const std::string* error_; // error indicator (or points to empty string) |
789 | const std::string* error_arg_; // fragment of regexp showing error (or ditto) |
790 | |
791 | // Second cache line is relatively hot fields. |
792 | // These are ordered oddly to pack everything. |
793 | int num_captures_; // number of capturing groups |
794 | ErrorCode error_code_ : 29; // error code (29 bits is more than enough) |
795 | bool longest_match_ : 1; // cached copy of options_.longest_match() |
796 | bool is_one_pass_ : 1; // can use prog_->SearchOnePass? |
797 | bool prefix_foldcase_ : 1; // prefix_ is ASCII case-insensitive |
798 | std::string prefix_; // required prefix (before suffix_regexp_) |
799 | re2::Prog* prog_; // compiled program for regexp |
800 | |
801 | // Reverse Prog for DFA execution only |
802 | mutable re2::Prog* rprog_; |
803 | // Map from capture names to indices |
804 | mutable const std::map<std::string, int>* named_groups_; |
805 | // Map from capture indices to names |
806 | mutable const std::map<int, std::string>* group_names_; |
807 | |
808 | mutable absl::once_flag rprog_once_; |
809 | mutable absl::once_flag named_groups_once_; |
810 | mutable absl::once_flag group_names_once_; |
811 | }; |
812 | |
813 | /***** Implementation details *****/ |
814 | |
815 | namespace re2_internal { |
816 | |
817 | // Types for which the 3-ary Parse() function template has specializations. |
818 | template <typename T> struct Parse3ary : public std::false_type {}; |
819 | template <> struct Parse3ary<void> : public std::true_type {}; |
820 | template <> struct Parse3ary<std::string> : public std::true_type {}; |
821 | template <> struct Parse3ary<absl::string_view> : public std::true_type {}; |
822 | template <> struct Parse3ary<char> : public std::true_type {}; |
823 | template <> struct Parse3ary<signed char> : public std::true_type {}; |
824 | template <> struct Parse3ary<unsigned char> : public std::true_type {}; |
825 | template <> struct Parse3ary<float> : public std::true_type {}; |
826 | template <> struct Parse3ary<double> : public std::true_type {}; |
827 | |
828 | template <typename T> |
829 | bool Parse(const char* str, size_t n, T* dest); |
830 | |
831 | // Types for which the 4-ary Parse() function template has specializations. |
832 | template <typename T> struct Parse4ary : public std::false_type {}; |
833 | template <> struct Parse4ary<long> : public std::true_type {}; |
834 | template <> struct Parse4ary<unsigned long> : public std::true_type {}; |
835 | template <> struct Parse4ary<short> : public std::true_type {}; |
836 | template <> struct Parse4ary<unsigned short> : public std::true_type {}; |
837 | template <> struct Parse4ary<int> : public std::true_type {}; |
838 | template <> struct Parse4ary<unsigned int> : public std::true_type {}; |
839 | template <> struct Parse4ary<long long> : public std::true_type {}; |
840 | template <> struct Parse4ary<unsigned long long> : public std::true_type {}; |
841 | |
842 | template <typename T> |
843 | bool Parse(const char* str, size_t n, T* dest, int radix); |
844 | |
845 | // Support absl::optional<T> for all T with a stock parser. |
846 | template <typename T> struct Parse3ary<absl::optional<T>> : public Parse3ary<T> {}; |
847 | template <typename T> struct Parse4ary<absl::optional<T>> : public Parse4ary<T> {}; |
848 | |
849 | template <typename T> |
850 | bool Parse(const char* str, size_t n, absl::optional<T>* dest) { |
851 | if (str == NULL) { |
852 | if (dest != NULL) |
853 | dest->reset(); |
854 | return true; |
855 | } |
856 | T tmp; |
857 | if (Parse(str, n, &tmp)) { |
858 | if (dest != NULL) |
859 | dest->emplace(std::move(tmp)); |
860 | return true; |
861 | } |
862 | return false; |
863 | } |
864 | |
865 | template <typename T> |
866 | bool Parse(const char* str, size_t n, absl::optional<T>* dest, int radix) { |
867 | if (str == NULL) { |
868 | if (dest != NULL) |
869 | dest->reset(); |
870 | return true; |
871 | } |
872 | T tmp; |
873 | if (Parse(str, n, &tmp, radix)) { |
874 | if (dest != NULL) |
875 | dest->emplace(std::move(tmp)); |
876 | return true; |
877 | } |
878 | return false; |
879 | } |
880 | |
881 | } // namespace re2_internal |
882 | |
883 | class RE2::Arg { |
884 | private: |
885 | template <typename T> |
886 | using CanParse3ary = typename std::enable_if< |
887 | re2_internal::Parse3ary<T>::value, |
888 | int>::type; |
889 | |
890 | template <typename T> |
891 | using CanParse4ary = typename std::enable_if< |
892 | re2_internal::Parse4ary<T>::value, |
893 | int>::type; |
894 | |
895 | template <typename T> |
896 | using CanParseFrom = typename std::enable_if< |
897 | std::is_member_function_pointer< |
898 | decltype(static_cast<bool (T::*)(const char*, size_t)>( |
899 | &T::ParseFrom))>::value, |
900 | int>::type; |
901 | |
902 | public: |
903 | Arg() : Arg(nullptr) {} |
904 | Arg(std::nullptr_t ptr) : arg_(ptr), parser_(DoNothing) {} |
905 | |
906 | template <typename T, CanParse3ary<T> = 0> |
907 | Arg(T* ptr) : arg_(ptr), parser_(DoParse3ary<T>) {} |
908 | |
909 | template <typename T, CanParse4ary<T> = 0> |
910 | Arg(T* ptr) : arg_(ptr), parser_(DoParse4ary<T>) {} |
911 | |
912 | template <typename T, CanParseFrom<T> = 0> |
913 | Arg(T* ptr) : arg_(ptr), parser_(DoParseFrom<T>) {} |
914 | |
915 | typedef bool (*Parser)(const char* str, size_t n, void* dest); |
916 | |
917 | template <typename T> |
918 | Arg(T* ptr, Parser parser) : arg_(ptr), parser_(parser) {} |
919 | |
920 | bool Parse(const char* str, size_t n) const { |
921 | return (*parser_)(str, n, arg_); |
922 | } |
923 | |
924 | private: |
925 | static bool DoNothing(const char* /*str*/, size_t /*n*/, void* /*dest*/) { |
926 | return true; |
927 | } |
928 | |
929 | template <typename T> |
930 | static bool DoParse3ary(const char* str, size_t n, void* dest) { |
931 | return re2_internal::Parse(str, n, reinterpret_cast<T*>(dest)); |
932 | } |
933 | |
934 | template <typename T> |
935 | static bool DoParse4ary(const char* str, size_t n, void* dest) { |
936 | return re2_internal::Parse(str, n, reinterpret_cast<T*>(dest), 10); |
937 | } |
938 | |
939 | template <typename T> |
940 | static bool DoParseFrom(const char* str, size_t n, void* dest) { |
941 | if (dest == NULL) return true; |
942 | return reinterpret_cast<T*>(dest)->ParseFrom(str, n); |
943 | } |
944 | |
945 | void* arg_; |
946 | Parser parser_; |
947 | }; |
948 | |
949 | template <typename T> |
950 | inline RE2::Arg RE2::CRadix(T* ptr) { |
951 | return RE2::Arg(ptr, [](const char* str, size_t n, void* dest) -> bool { |
952 | return re2_internal::Parse(str, n, reinterpret_cast<T*>(dest), 0); |
953 | }); |
954 | } |
955 | |
956 | template <typename T> |
957 | inline RE2::Arg RE2::Hex(T* ptr) { |
958 | return RE2::Arg(ptr, [](const char* str, size_t n, void* dest) -> bool { |
959 | return re2_internal::Parse(str, n, reinterpret_cast<T*>(dest), 16); |
960 | }); |
961 | } |
962 | |
963 | template <typename T> |
964 | inline RE2::Arg RE2::Octal(T* ptr) { |
965 | return RE2::Arg(ptr, [](const char* str, size_t n, void* dest) -> bool { |
966 | return re2_internal::Parse(str, n, reinterpret_cast<T*>(dest), 8); |
967 | }); |
968 | } |
969 | |
970 | // Silence warnings about missing initializers for members of LazyRE2. |
971 | #if defined(__GNUC__) |
972 | #pragma GCC diagnostic ignored "-Wmissing-field-initializers" |
973 | #endif |
974 | |
975 | // Helper for writing global or static RE2s safely. |
976 | // Write |
977 | // static LazyRE2 re = {".*"}; |
978 | // and then use *re instead of writing |
979 | // static RE2 re(".*"); |
980 | // The former is more careful about multithreaded |
981 | // situations than the latter. |
982 | // |
983 | // N.B. This class never deletes the RE2 object that |
984 | // it constructs: that's a feature, so that it can be used |
985 | // for global and function static variables. |
986 | class LazyRE2 { |
987 | private: |
988 | struct NoArg {}; |
989 | |
990 | public: |
991 | typedef RE2 element_type; // support std::pointer_traits |
992 | |
993 | // Constructor omitted to preserve braced initialization in C++98. |
994 | |
995 | // Pretend to be a pointer to Type (never NULL due to on-demand creation): |
996 | RE2& operator*() const { return *get(); } |
997 | RE2* operator->() const { return get(); } |
998 | |
999 | // Named accessor/initializer: |
1000 | RE2* get() const { |
1001 | absl::call_once(flag&: once_, fn: &LazyRE2::Init, args: this); |
1002 | return ptr_; |
1003 | } |
1004 | |
1005 | // All data fields must be public to support {"foo"} initialization. |
1006 | const char* pattern_; |
1007 | RE2::CannedOptions options_; |
1008 | NoArg barrier_against_excess_initializers_; |
1009 | |
1010 | mutable RE2* ptr_; |
1011 | mutable absl::once_flag once_; |
1012 | |
1013 | private: |
1014 | static void Init(const LazyRE2* lazy_re2) { |
1015 | lazy_re2->ptr_ = new RE2(lazy_re2->pattern_, lazy_re2->options_); |
1016 | } |
1017 | |
1018 | void operator=(const LazyRE2&); // disallowed |
1019 | }; |
1020 | |
1021 | namespace hooks { |
1022 | |
1023 | // Most platforms support thread_local. Older versions of iOS don't support |
1024 | // thread_local, but for the sake of brevity, we lump together all versions |
1025 | // of Apple platforms that aren't macOS. If an iOS application really needs |
1026 | // the context pointee someday, we can get more specific then... |
1027 | // |
1028 | // As per https://github.com/google/re2/issues/325, thread_local support in |
1029 | // MinGW seems to be buggy. (FWIW, Abseil folks also avoid it.) |
1030 | #define RE2_HAVE_THREAD_LOCAL |
1031 | #if (defined(__APPLE__) && !(defined(TARGET_OS_OSX) && TARGET_OS_OSX)) || defined(__MINGW32__) |
1032 | #undef RE2_HAVE_THREAD_LOCAL |
1033 | #endif |
1034 | |
1035 | // A hook must not make any assumptions regarding the lifetime of the context |
1036 | // pointee beyond the current invocation of the hook. Pointers and references |
1037 | // obtained via the context pointee should be considered invalidated when the |
1038 | // hook returns. Hence, any data about the context pointee (e.g. its pattern) |
1039 | // would have to be copied in order for it to be kept for an indefinite time. |
1040 | // |
1041 | // A hook must not use RE2 for matching. Control flow reentering RE2::Match() |
1042 | // could result in infinite mutual recursion. To discourage that possibility, |
1043 | // RE2 will not maintain the context pointer correctly when used in that way. |
1044 | #ifdef RE2_HAVE_THREAD_LOCAL |
1045 | extern thread_local const RE2* context; |
1046 | #endif |
1047 | |
1048 | struct DFAStateCacheReset { |
1049 | int64_t state_budget; |
1050 | size_t state_cache_size; |
1051 | }; |
1052 | |
1053 | struct DFASearchFailure { |
1054 | // Nothing yet... |
1055 | }; |
1056 | |
1057 | #define DECLARE_HOOK(type) \ |
1058 | using type##Callback = void(const type&); \ |
1059 | void Set##type##Hook(type##Callback* cb); \ |
1060 | type##Callback* Get##type##Hook(); |
1061 | |
1062 | DECLARE_HOOK(DFAStateCacheReset) |
1063 | DECLARE_HOOK(DFASearchFailure) |
1064 | |
1065 | #undef DECLARE_HOOK |
1066 | |
1067 | } // namespace hooks |
1068 | |
1069 | } // namespace re2 |
1070 | |
1071 | using re2::RE2; |
1072 | using re2::LazyRE2; |
1073 | |
1074 | #endif // RE2_RE2_H_ |
1075 | |