mirror of
https://github.com/klzgrad/naiveproxy.git
synced 2024-11-24 06:16:30 +03:00
156 lines
5.0 KiB
C++
156 lines
5.0 KiB
C++
// Copyright 2015 The Chromium Authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file.
|
|
|
|
#include "base/strings/pattern.h"
|
|
|
|
#include "base/third_party/icu/icu_utf.h"
|
|
|
|
namespace base {
|
|
|
|
namespace {
|
|
|
|
constexpr bool IsWildcard(base_icu::UChar32 character) {
|
|
return character == '*' || character == '?';
|
|
}
|
|
|
|
// Searches for the next subpattern of |pattern| in |string|, up to the given
|
|
// |maximum_distance|. The subpattern extends from the start of |pattern| up to
|
|
// the first wildcard character (or the end of the string). If the value of
|
|
// |maximum_distance| is negative, the maximum distance is considered infinite.
|
|
template <typename CHAR, typename NEXT>
|
|
constexpr bool SearchForChars(const CHAR** pattern,
|
|
const CHAR* pattern_end,
|
|
const CHAR** string,
|
|
const CHAR* string_end,
|
|
int maximum_distance,
|
|
NEXT next) {
|
|
const CHAR* pattern_start = *pattern;
|
|
const CHAR* string_start = *string;
|
|
bool escape = false;
|
|
while (true) {
|
|
if (*pattern == pattern_end) {
|
|
// If this is the end of the pattern, only accept the end of the string;
|
|
// anything else falls through to the mismatch case.
|
|
if (*string == string_end)
|
|
return true;
|
|
} else {
|
|
// If we have found a wildcard, we're done.
|
|
if (!escape && IsWildcard(**pattern))
|
|
return true;
|
|
|
|
// Check if the escape character is found. If so, skip it and move to the
|
|
// next character.
|
|
if (!escape && **pattern == '\\') {
|
|
escape = true;
|
|
next(pattern, pattern_end);
|
|
continue;
|
|
}
|
|
|
|
escape = false;
|
|
|
|
if (*string == string_end)
|
|
return false;
|
|
|
|
// Check if the chars match, if so, increment the ptrs.
|
|
const CHAR* pattern_next = *pattern;
|
|
const CHAR* string_next = *string;
|
|
base_icu::UChar32 pattern_char = next(&pattern_next, pattern_end);
|
|
if (pattern_char == next(&string_next, string_end) &&
|
|
pattern_char != CBU_SENTINEL) {
|
|
*pattern = pattern_next;
|
|
*string = string_next;
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// Mismatch. If we have reached the maximum distance, return false,
|
|
// otherwise restart at the beginning of the pattern with the next character
|
|
// in the string.
|
|
// TODO(bauerb): This is a naive implementation of substring search, which
|
|
// could be implemented with a more efficient algorithm, e.g.
|
|
// Knuth-Morris-Pratt (at the expense of requiring preprocessing).
|
|
if (maximum_distance == 0)
|
|
return false;
|
|
|
|
// Because unlimited distance is represented as -1, this will never reach 0
|
|
// and therefore fail the match above.
|
|
maximum_distance--;
|
|
*pattern = pattern_start;
|
|
next(&string_start, string_end);
|
|
*string = string_start;
|
|
}
|
|
}
|
|
|
|
// Consumes consecutive wildcard characters (? or *). Returns the maximum number
|
|
// of characters matched by the sequence of wildcards, or -1 if the wildcards
|
|
// match an arbitrary number of characters (which is the case if it contains at
|
|
// least one *).
|
|
template <typename CHAR, typename NEXT>
|
|
constexpr int EatWildcards(const CHAR** pattern, const CHAR* end, NEXT next) {
|
|
int num_question_marks = 0;
|
|
bool has_asterisk = false;
|
|
while (*pattern != end) {
|
|
if (**pattern == '?') {
|
|
num_question_marks++;
|
|
} else if (**pattern == '*') {
|
|
has_asterisk = true;
|
|
} else {
|
|
break;
|
|
}
|
|
|
|
next(pattern, end);
|
|
}
|
|
return has_asterisk ? -1 : num_question_marks;
|
|
}
|
|
|
|
template <typename CHAR, typename NEXT>
|
|
constexpr bool MatchPatternT(const CHAR* eval,
|
|
const CHAR* eval_end,
|
|
const CHAR* pattern,
|
|
const CHAR* pattern_end,
|
|
NEXT next) {
|
|
do {
|
|
int maximum_wildcard_length = EatWildcards(&pattern, pattern_end, next);
|
|
if (!SearchForChars(&pattern, pattern_end, &eval, eval_end,
|
|
maximum_wildcard_length, next)) {
|
|
return false;
|
|
}
|
|
} while (pattern != pattern_end);
|
|
return true;
|
|
}
|
|
|
|
struct NextCharUTF8 {
|
|
base_icu::UChar32 operator()(const char** p, const char* end) {
|
|
base_icu::UChar32 c;
|
|
int offset = 0;
|
|
CBU8_NEXT(*p, offset, end - *p, c);
|
|
*p += offset;
|
|
return c;
|
|
}
|
|
};
|
|
|
|
struct NextCharUTF16 {
|
|
base_icu::UChar32 operator()(const char16** p, const char16* end) {
|
|
base_icu::UChar32 c;
|
|
int offset = 0;
|
|
CBU16_NEXT(*p, offset, end - *p, c);
|
|
*p += offset;
|
|
return c;
|
|
}
|
|
};
|
|
|
|
} // namespace
|
|
|
|
bool MatchPattern(StringPiece eval, StringPiece pattern) {
|
|
return MatchPatternT(eval.data(), eval.data() + eval.size(), pattern.data(),
|
|
pattern.data() + pattern.size(), NextCharUTF8());
|
|
}
|
|
|
|
bool MatchPattern(StringPiece16 eval, StringPiece16 pattern) {
|
|
return MatchPatternT(eval.data(), eval.data() + eval.size(), pattern.data(),
|
|
pattern.data() + pattern.size(), NextCharUTF16());
|
|
}
|
|
|
|
} // namespace base
|