• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Copyright 2019 the V8 project authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
4 
5 #ifndef V8_REGEXP_SPECIAL_CASE_H_
6 #define V8_REGEXP_SPECIAL_CASE_H_
7 
8 #ifdef V8_INTL_SUPPORT
9 #include "src/base/logging.h"
10 #include "src/common/globals.h"
11 
12 #include "unicode/uchar.h"
13 #include "unicode/uniset.h"
14 #include "unicode/unistr.h"
15 
16 namespace v8 {
17 namespace internal {
18 
19 // Sets of Unicode characters that need special handling under "i" mode
20 
21 // For non-unicode ignoreCase matches (aka "i", not "iu"), ECMA 262
22 // defines slightly different case-folding rules than Unicode. An
23 // input character should match a pattern character if the result of
24 // the Canonicalize algorithm is the same for both characters.
25 //
26 // Roughly speaking, for "i" regexps, Canonicalize(c) is the same as
27 // c.toUpperCase(), unless a) c.toUpperCase() is a multi-character
28 // string, or b) c is non-ASCII, and c.toUpperCase() is ASCII. See
29 // https://tc39.es/ecma262/#sec-runtime-semantics-canonicalize-ch for
30 // the precise definition.
31 //
32 // While compiling such regular expressions, we need to compute the
33 // set of characters that should match a given input character. (See
34 // GetCaseIndependentLetters and CharacterRange::AddCaseEquivalents.)
35 // For almost all characters, this can be efficiently computed using
36 // UnicodeSet::closeOver(USET_CASE_INSENSITIVE). These sets represent
37 // the remaining special cases.
38 //
39 // For a character c, the rules are as follows:
40 //
41 // 1. If c is in neither IgnoreSet nor SpecialAddSet, then calling
42 //    UnicodeSet::closeOver(USET_CASE_INSENSITIVE) on a UnicodeSet
43 //    containing c will produce the set of characters that should
44 //    match /c/i (or /[c]/i), and only those characters.
45 //
46 // 2. If c is in IgnoreSet, then the only character it should match is
47 //    itself. However, closeOver will add additional incorrect
48 //    matches. For example, consider SHARP S: 'ß' (U+00DF) and 'ẞ'
49 //    (U+1E9E). Although closeOver('ß') = "ßẞ", uppercase('ß') is
50 //    "SS".  Step 3.e therefore requires that 'ß' canonicalizes to
51 //    itself, and should not match 'ẞ'. In these cases, we can skip
52 //    the closeOver entirely, because it will never add an equivalent
53 //    character.
54 //
55 // 3. If c is in SpecialAddSet, then it should match at least one
56 //    character other than itself. However, closeOver will add at
57 //    least one additional incorrect match. For example, consider the
58 //    letter 'k'. Closing over 'k' gives "kKK" (lowercase k, uppercase
59 //    K, U+212A KELVIN SIGN). However, because of step 3.g, KELVIN
60 //    SIGN should not match either of the other two characters. As a
61 //    result, "k" and "K" are in SpecialAddSet (and KELVIN SIGN is in
62 //    IgnoreSet). To find the correct matches for characters in
63 //    SpecialAddSet, we closeOver the original character, but filter
64 //    out the results that do not have the same canonical value.
65 //
66 // The contents of these sets are calculated at build time by
67 // src/regexp/gen-regexp-special-case.cc, which generates
68 // gen/src/regexp/special-case.cc. This is done by iterating over the
69 // result of closeOver for each BMP character, and finding sets for
70 // which at least one character has a different canonical value than
71 // another character. Characters that match no other characters in
72 // their equivalence class are added to IgnoreSet. Characters that
73 // match at least one other character are added to SpecialAddSet.
74 
75 class RegExpCaseFolding final : public AllStatic {
76  public:
77   static const icu::UnicodeSet& IgnoreSet();
78   static const icu::UnicodeSet& SpecialAddSet();
79 
80   // This implements ECMAScript 2020 21.2.2.8.2 (Runtime Semantics:
81   // Canonicalize) step 3, which is used to determine whether
82   // characters match when ignoreCase is true and unicode is false.
Canonicalize(UChar32 ch)83   static UChar32 Canonicalize(UChar32 ch) {
84     // a. Assert: ch is a UTF-16 code unit.
85     CHECK_LE(ch, 0xffff);
86 
87     // b. Let s be the String value consisting of the single code unit ch.
88     icu::UnicodeString s(ch);
89 
90     // c. Let u be the same result produced as if by performing the algorithm
91     // for String.prototype.toUpperCase using s as the this value.
92     // d. Assert: Type(u) is String.
93     icu::UnicodeString& u = s.toUpper();
94 
95     // e. If u does not consist of a single code unit, return ch.
96     if (u.length() != 1) {
97       return ch;
98     }
99 
100     // f. Let cu be u's single code unit element.
101     UChar32 cu = u.char32At(0);
102 
103     // g. If the value of ch >= 128 and the value of cu < 128, return ch.
104     if (ch >= 128 && cu < 128) {
105       return ch;
106     }
107 
108     // h. Return cu.
109     return cu;
110   }
111 };
112 
113 }  // namespace internal
114 }  // namespace v8
115 
116 #endif  // V8_INTL_SUPPORT
117 
118 #endif  // V8_REGEXP_SPECIAL_CASE_H_
119