Index: ui/base/text/utf16_indexing.cc |
diff --git a/ui/base/text/utf16_indexing.cc b/ui/base/text/utf16_indexing.cc |
new file mode 100644 |
index 0000000000000000000000000000000000000000..d217f5d75c49fd86923619b9ba4081b07a2fb459 |
--- /dev/null |
+++ b/ui/base/text/utf16_indexing.cc |
@@ -0,0 +1,55 @@ |
+// Copyright (c) 2012 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 "ui/base/text/utf16_indexing.h" |
+ |
+#include "base/logging.h" |
+#include "base/third_party/icu/icu_utf.h" |
+ |
+namespace ui { |
+ |
+bool IsValidCodePointIndex(const string16& s, size_t index) { |
+ return index == 0 || index == s.length() || |
+ !(CBU16_IS_TRAIL(s[index]) && CBU16_IS_LEAD(s[index - 1])); |
+} |
+ |
+ptrdiff_t UTF16IndexToOffset(const string16& s, size_t base, size_t pos) { |
+ // The indices point between UTF-16 words (range 0 to s.length() inclusive). |
+ // In order to consistently handle indices that point to the middle of a |
+ // surrogate pair, we count the first word in that surrogate pair and not |
+ // the second. The test "s[i] is not the second half of a surrogate pair" is |
+ // "IsValidCodePointIndex(s, i)". |
+ DCHECK_LE(base, s.length()); |
+ DCHECK_LE(pos, s.length()); |
+ ptrdiff_t delta = 0; |
+ while (base < pos) |
+ delta += IsValidCodePointIndex(s, base++) ? 1 : 0; |
+ while (pos < base) |
+ delta -= IsValidCodePointIndex(s, pos++) ? 1 : 0; |
+ return delta; |
+} |
+ |
+size_t UTF16OffsetToIndex(const string16& s, size_t base, ptrdiff_t offset) { |
+ DCHECK_LE(base, s.length()); |
+ // As in UTF16IndexToOffset, we count the first half of a surrogate pair, not |
+ // the second. When stepping from pos to pos+1 we check s[pos:pos+1] == s[pos] |
+ // (Python syntax), hence pos++. When stepping from pos to pos-1 we check |
+ // s[pos-1], hence --pos. |
+ size_t pos = base; |
+ while (offset > 0 && pos < s.length()) |
+ offset -= IsValidCodePointIndex(s, pos++) ? 1 : 0; |
+ while (offset < 0 && pos > 0) |
+ offset += IsValidCodePointIndex(s, --pos) ? 1 : 0; |
+ // If offset != 0 then we ran off the edge of the string, which is a contract |
+ // violation but is handled anyway (by clamping) in release for safety. |
+ DCHECK_EQ(offset, 0); |
+ // Since the second half of a surrogate pair has "length" zero, there is an |
+ // ambiguity in the returned position. Resolve it by always returning a valid |
+ // index. |
+ if (!IsValidCodePointIndex(s, pos)) |
+ ++pos; |
+ return pos; |
+} |
+ |
+} // namespace ui |