Index: trunk/src/tools/gn/pattern.cc |
=================================================================== |
--- trunk/src/tools/gn/pattern.cc (revision 214322) |
+++ trunk/src/tools/gn/pattern.cc (working copy) |
@@ -1,185 +0,0 @@ |
-// Copyright (c) 2013 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 "tools/gn/pattern.h" |
- |
-#include "tools/gn/value.h" |
- |
-namespace { |
- |
-void ParsePattern(const std::string& s, std::vector<Pattern::Subrange>* out) { |
- // Set when the last subrange is a literal so we can just append when we |
- // find another literal. |
- Pattern::Subrange* last_literal = NULL; |
- |
- for (size_t i = 0; i < s.size(); i++) { |
- if (s[i] == '*') { |
- // Don't allow two **. |
- if (out->size() == 0 || |
- (*out)[out->size() - 1].type != Pattern::Subrange::ANYTHING) |
- out->push_back(Pattern::Subrange(Pattern::Subrange::ANYTHING)); |
- last_literal = NULL; |
- } else if (s[i] == '\\') { |
- if (i < s.size() - 1 && s[i + 1] == 'b') { |
- // "\b" means path boundary. |
- i++; |
- out->push_back(Pattern::Subrange(Pattern::Subrange::PATH_BOUNDARY)); |
- last_literal = NULL; |
- } else { |
- // Backslash + anything else means that literal char. |
- if (!last_literal) { |
- out->push_back(Pattern::Subrange(Pattern::Subrange::LITERAL)); |
- last_literal = &(*out)[out->size() - 1]; |
- } |
- if (i < s.size() - 1) { |
- i++; |
- last_literal->literal.push_back(s[i]); |
- } else { |
- // Single backslash at end, use literal backslash. |
- last_literal->literal.push_back('\\'); |
- } |
- } |
- } else { |
- if (!last_literal) { |
- out->push_back(Pattern::Subrange(Pattern::Subrange::LITERAL)); |
- last_literal = &(*out)[out->size() - 1]; |
- } |
- last_literal->literal.push_back(s[i]); |
- } |
- } |
-} |
- |
-} // namespace |
- |
-Pattern::Pattern(const std::string& s) { |
- ParsePattern(s, &subranges_); |
- is_suffix_ = |
- (subranges_.size() == 2 && |
- subranges_[0].type == Subrange::ANYTHING && |
- subranges_[1].type == Subrange::LITERAL); |
-} |
- |
-Pattern::~Pattern() { |
-} |
- |
-bool Pattern::MatchesString(const std::string& s) const { |
- // Empty pattern matches only empty string. |
- if (subranges_.empty()) |
- return s.empty(); |
- |
- if (is_suffix_) { |
- const std::string& suffix = subranges_[1].literal; |
- if (suffix.size() > s.size()) |
- return false; // Too short. |
- return s.compare(s.size() - suffix.size(), suffix.size(), suffix) == 0; |
- } |
- |
- return RecursiveMatch(s, 0, 0, true); |
-} |
- |
-// We assume the number of ranges is small so recursive is always reasonable. |
-// Could be optimized to only be recursive for *. |
-bool Pattern::RecursiveMatch(const std::string& s, |
- size_t begin_char, |
- size_t subrange_index, |
- bool allow_implicit_path_boundary) const { |
- if (subrange_index >= subranges_.size()) { |
- // Hit the end of our subranges, the text should also be at the end for a |
- // match. |
- return begin_char == s.size(); |
- } |
- |
- const Subrange& sr = subranges_[subrange_index]; |
- switch (sr.type) { |
- case Subrange::LITERAL: { |
- if (s.size() - begin_char < sr.literal.size()) |
- return false; // Not enough room. |
- if (s.compare(begin_char, sr.literal.size(), sr.literal) != 0) |
- return false; // Literal doesn't match. |
- |
- // Recursively check the next one. |
- return RecursiveMatch(s, begin_char + sr.literal.size(), |
- subrange_index + 1, true); |
- } |
- |
- case Subrange::PATH_BOUNDARY: { |
- // When we can accept an implicit path boundary, we have to check both |
- // a match of the literal and the implicit one. |
- if (allow_implicit_path_boundary && |
- (begin_char == 0 || begin_char == s.size())) { |
- // At implicit path boundary, see if the rest of the pattern matches. |
- if (RecursiveMatch(s, begin_char, subrange_index + 1, false)) |
- return true; |
- } |
- |
- // Check for a literal "/". |
- if (begin_char < s.size() && s[begin_char] == '/') { |
- // At explicit boundary, see if the rest of the pattern matches. |
- if (RecursiveMatch(s, begin_char + 1, subrange_index + 1, true)) |
- return true; |
- } |
- return false; |
- } |
- |
- case Subrange::ANYTHING: { |
- if (subrange_index == subranges_.size() - 1) |
- return true; // * at the end, consider it matching. |
- |
- size_t min_next_size = sr.MinSize(); |
- |
- // We don't care about exactly what matched as long as there was a match, |
- // so we can do this front-to-back. If we needed the match, we would |
- // normally want "*" to be greedy so would work backwards. |
- for (size_t i = begin_char; i < s.size() - min_next_size; i++) { |
- // Note: this could probably be faster by detecting the type of the |
- // next match in advance and checking for a match in this loop rather |
- // than doing a full recursive call for each character. |
- if (RecursiveMatch(s, i, subrange_index + 1, true)) |
- return true; |
- } |
- return false; |
- } |
- |
- default: |
- NOTREACHED(); |
- } |
- |
- return false; |
-} |
- |
-PatternList::PatternList() { |
-} |
- |
-PatternList::~PatternList() { |
-} |
- |
-void PatternList::SetFromValue(const Value& v, Err* err) { |
- patterns_.clear(); |
- |
- if (v.type() != Value::LIST) { |
- *err = Err(v.origin(), "This value must be a list."); |
- return; |
- } |
- |
- const std::vector<Value>& list = v.list_value(); |
- for (size_t i = 0; i < list.size(); i++) { |
- if (!list[i].VerifyTypeIs(Value::STRING, err)) |
- return; |
- patterns_.push_back(Pattern(list[i].string_value())); |
- } |
-} |
- |
-bool PatternList::MatchesString(const std::string& s) const { |
- for (size_t i = 0; i < patterns_.size(); i++) { |
- if (patterns_[i].MatchesString(s)) |
- return true; |
- } |
- return false; |
-} |
- |
-bool PatternList::MatchesValue(const Value& v) const { |
- if (v.type() == Value::STRING) |
- return MatchesString(v.string_value()); |
- return false; |
-} |