mirror of
https://github.com/klzgrad/naiveproxy.git
synced 2024-12-01 01:36:09 +03:00
183 lines
7.1 KiB
C
183 lines
7.1 KiB
C
|
// Copyright (c) 2011 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.
|
||
|
|
||
|
#ifndef BASE_I18N_BREAK_ITERATOR_H_
|
||
|
#define BASE_I18N_BREAK_ITERATOR_H_
|
||
|
|
||
|
#include <stddef.h>
|
||
|
|
||
|
#include "base/i18n/base_i18n_export.h"
|
||
|
#include "base/macros.h"
|
||
|
#include "base/strings/string16.h"
|
||
|
#include "base/strings/string_piece.h"
|
||
|
|
||
|
// The BreakIterator class iterates through the words, word breaks, and
|
||
|
// line breaks in a UTF-16 string.
|
||
|
//
|
||
|
// It provides several modes, BREAK_WORD, BREAK_LINE, and BREAK_NEWLINE,
|
||
|
// which modify how characters are aggregated into the returned string.
|
||
|
//
|
||
|
// Under BREAK_WORD mode, once a word is encountered any non-word
|
||
|
// characters are not included in the returned string (e.g. in the
|
||
|
// UTF-16 equivalent of the string " foo bar! ", the word breaks are at
|
||
|
// the periods in ". .foo. .bar.!. .").
|
||
|
// Note that Chinese/Japanese/Thai do not use spaces between words so that
|
||
|
// boundaries can fall in the middle of a continuous run of non-space /
|
||
|
// non-punctuation characters.
|
||
|
//
|
||
|
// Under BREAK_LINE mode, once a line breaking opportunity is encountered,
|
||
|
// any non-word characters are included in the returned string, breaking
|
||
|
// only when a space-equivalent character or a line breaking opportunity
|
||
|
// is encountered (e.g. in the UTF16-equivalent of the string " foo bar! ",
|
||
|
// the breaks are at the periods in ". .foo .bar! .").
|
||
|
//
|
||
|
// Note that lines can be broken at any character/syllable/grapheme cluster
|
||
|
// boundary in Chinese/Japanese/Korean and at word boundaries in Thai
|
||
|
// (Thai does not use spaces between words). Therefore, this is NOT the same
|
||
|
// as breaking only at space-equivalent characters where its former
|
||
|
// name (BREAK_SPACE) implied.
|
||
|
//
|
||
|
// Under BREAK_NEWLINE mode, all characters are included in the returned
|
||
|
// string, breaking only when a newline-equivalent character is encountered
|
||
|
// (eg. in the UTF-16 equivalent of the string "foo\nbar!\n\n", the line
|
||
|
// breaks are at the periods in ".foo\n.bar\n.\n.").
|
||
|
//
|
||
|
// To extract the words from a string, move a BREAK_WORD BreakIterator
|
||
|
// through the string and test whether IsWord() is true. E.g.,
|
||
|
// BreakIterator iter(str, BreakIterator::BREAK_WORD);
|
||
|
// if (!iter.Init())
|
||
|
// return false;
|
||
|
// while (iter.Advance()) {
|
||
|
// if (iter.IsWord()) {
|
||
|
// // Region [iter.prev(), iter.pos()) contains a word.
|
||
|
// VLOG(1) << "word: " << iter.GetString();
|
||
|
// }
|
||
|
// }
|
||
|
|
||
|
namespace base {
|
||
|
namespace i18n {
|
||
|
|
||
|
class BASE_I18N_EXPORT BreakIterator {
|
||
|
public:
|
||
|
enum BreakType {
|
||
|
BREAK_WORD,
|
||
|
BREAK_LINE,
|
||
|
// TODO(jshin): Remove this after reviewing call sites.
|
||
|
// If call sites really need break only on space-like characters
|
||
|
// implement it separately.
|
||
|
BREAK_SPACE = BREAK_LINE,
|
||
|
BREAK_NEWLINE,
|
||
|
BREAK_CHARACTER,
|
||
|
// But don't remove this one!
|
||
|
RULE_BASED,
|
||
|
};
|
||
|
|
||
|
enum WordBreakStatus {
|
||
|
// The end of text that the iterator recognizes as word characters.
|
||
|
// Non-word characters are things like punctuation and spaces.
|
||
|
IS_WORD_BREAK,
|
||
|
// Characters that the iterator can skip past, such as punctuation,
|
||
|
// whitespace, and, if using RULE_BASED mode, characters from another
|
||
|
// character set.
|
||
|
IS_SKIPPABLE_WORD,
|
||
|
// Only used if not in BREAK_WORD or RULE_BASED mode. This is returned for
|
||
|
// newlines, line breaks, and character breaks.
|
||
|
IS_LINE_OR_CHAR_BREAK
|
||
|
};
|
||
|
|
||
|
// Requires |str| to live as long as the BreakIterator does.
|
||
|
BreakIterator(const StringPiece16& str, BreakType break_type);
|
||
|
// Make a rule-based iterator. BreakType == RULE_BASED is implied.
|
||
|
// TODO(andrewhayden): This signature could easily be misinterpreted as
|
||
|
// "(const string16& str, const string16& locale)". We should do something
|
||
|
// better.
|
||
|
BreakIterator(const StringPiece16& str, const string16& rules);
|
||
|
~BreakIterator();
|
||
|
|
||
|
// Init() must be called before any of the iterators are valid.
|
||
|
// Returns false if ICU failed to initialize.
|
||
|
bool Init();
|
||
|
|
||
|
// Advance to the next break. Returns false if we've run past the end of
|
||
|
// the string. (Note that the very last "break" is after the final
|
||
|
// character in the string, and when we advance to that position it's the
|
||
|
// last time Advance() returns true.)
|
||
|
bool Advance();
|
||
|
|
||
|
// Updates the text used by the iterator, resetting the iterator as if
|
||
|
// if Init() had been called again. Any old state is lost. Returns true
|
||
|
// unless there is an error setting the text.
|
||
|
bool SetText(const base::char16* text, const size_t length);
|
||
|
|
||
|
// Under BREAK_WORD mode, returns true if the break we just hit is the
|
||
|
// end of a word. (Otherwise, the break iterator just skipped over e.g.
|
||
|
// whitespace or punctuation.) Under BREAK_LINE and BREAK_NEWLINE modes,
|
||
|
// this distinction doesn't apply and it always returns false.
|
||
|
bool IsWord() const;
|
||
|
|
||
|
// Under BREAK_WORD mode:
|
||
|
// - Returns IS_SKIPPABLE_WORD if non-word characters, such as punctuation or
|
||
|
// spaces, are found.
|
||
|
// - Returns IS_WORD_BREAK if the break we just hit is the end of a sequence
|
||
|
// of word characters.
|
||
|
// Under RULE_BASED mode:
|
||
|
// - Returns IS_SKIPPABLE_WORD if characters outside the rules' character set
|
||
|
// or non-word characters, such as punctuation or spaces, are found.
|
||
|
// - Returns IS_WORD_BREAK if the break we just hit is the end of a sequence
|
||
|
// of word characters that are in the rules' character set.
|
||
|
// Not under BREAK_WORD or RULE_BASED mode:
|
||
|
// - Returns IS_LINE_OR_CHAR_BREAK.
|
||
|
BreakIterator::WordBreakStatus GetWordBreakStatus() const;
|
||
|
|
||
|
// Under BREAK_WORD mode, returns true if |position| is at the end of word or
|
||
|
// at the start of word. It always returns false under BREAK_LINE and
|
||
|
// BREAK_NEWLINE modes.
|
||
|
bool IsEndOfWord(size_t position) const;
|
||
|
bool IsStartOfWord(size_t position) const;
|
||
|
|
||
|
// Under BREAK_CHARACTER mode, returns whether |position| is a Unicode
|
||
|
// grapheme boundary.
|
||
|
bool IsGraphemeBoundary(size_t position) const;
|
||
|
|
||
|
// Returns the string between prev() and pos().
|
||
|
// Advance() must have been called successfully at least once for pos() to
|
||
|
// have advanced to somewhere useful.
|
||
|
string16 GetString() const;
|
||
|
|
||
|
StringPiece16 GetStringPiece() const;
|
||
|
|
||
|
// Returns the value of pos() returned before Advance() was last called.
|
||
|
size_t prev() const { return prev_; }
|
||
|
|
||
|
// Returns the current break position within the string,
|
||
|
// or BreakIterator::npos when done.
|
||
|
size_t pos() const { return pos_; }
|
||
|
|
||
|
private:
|
||
|
// ICU iterator, avoiding ICU ubrk.h dependence.
|
||
|
// This is actually an ICU UBreakiterator* type, which turns out to be
|
||
|
// a typedef for a void* in the ICU headers. Using void* directly prevents
|
||
|
// callers from needing access to the ICU public headers directory.
|
||
|
void* iter_;
|
||
|
|
||
|
// The string we're iterating over. Can be changed with SetText(...)
|
||
|
StringPiece16 string_;
|
||
|
|
||
|
// Rules for our iterator. Mutually exclusive with break_type_.
|
||
|
const string16 rules_;
|
||
|
|
||
|
// The breaking style (word/space/newline). Mutually exclusive with rules_
|
||
|
BreakType break_type_;
|
||
|
|
||
|
// Previous and current iterator positions.
|
||
|
size_t prev_, pos_;
|
||
|
|
||
|
DISALLOW_COPY_AND_ASSIGN(BreakIterator);
|
||
|
};
|
||
|
|
||
|
} // namespace i18n
|
||
|
} // namespace base
|
||
|
|
||
|
#endif // BASE_I18N_BREAK_ITERATOR_H_
|