gecko-dev/intl/lwbrk/Segmenter.cpp
Jan-Niklas Jaeschke 062cfc14f2 Bug 1928931: Speed up nsFind by reusing the word segmenter. r=emilio
Turns out that `nsFind::BreakInBetween()` is called a lot,
especially for large documents.
If `nsFind::Find()` is then called in a loop (e.g. for creating a text directive),
the alloc cost for the word segmenter shows up in profiles.
This can be avoided by having the segmenter be a member of `nsFind`.

Differential Revision: https://phabricator.services.mozilla.com/D227804
2024-11-04 12:44:49 +00:00

557 lines
15 KiB
C++

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this file,
* You can obtain one at http://mozilla.org/MPL/2.0/. */
/* Classes to iterate over grapheme, word, sentence, or line. */
#include "mozilla/intl/Segmenter.h"
#include "mozilla/intl/LineBreaker.h"
#include "mozilla/intl/WordBreaker.h"
#include "mozilla/intl/UnicodeProperties.h"
#include "mozilla/StaticPrefs_intl.h"
#include "nsUnicodeProperties.h"
#include "nsCharTraits.h"
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
# include "ICU4XDataProvider.h"
# include "ICU4XGraphemeClusterSegmenter.h"
# include "ICU4XLineSegmenter.h"
# include "ICU4XSentenceSegmenter.h"
# include "ICU4XWordSegmenter.h"
# include "mozilla/ClearOnShutdown.h"
# include "mozilla/intl/ICU4XGeckoDataProvider.h"
# include "nsThreadUtils.h"
# include <mutex>
#endif
using namespace mozilla::unicode;
namespace mozilla::intl {
SegmentIteratorUtf16::SegmentIteratorUtf16(Span<const char16_t> aText)
: mText(aText) {}
Maybe<uint32_t> SegmentIteratorUtf16::Seek(uint32_t aPos) {
if (mPos < aPos) {
mPos = aPos;
}
return Next();
}
LineBreakIteratorUtf16::LineBreakIteratorUtf16(Span<const char16_t> aText,
const LineBreakOptions& aOptions)
: SegmentIteratorUtf16(aText), mOptions(aOptions) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (!StaticPrefs::intl_icu4x_segmenter_enabled()) {
return;
}
auto result =
capi::ICU4XLineSegmenter_create_auto(mozilla::intl::GetDataProvider());
MOZ_RELEASE_ASSERT(result.is_ok);
mSegmenter = result.ok;
mIterator = capi::ICU4XLineSegmenter_segment_utf16(
mSegmenter, mText.Elements(), mText.Length());
#endif
}
LineBreakIteratorUtf16::~LineBreakIteratorUtf16() {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
capi::ICU4XLineBreakIteratorUtf16_destroy(mIterator);
}
if (mSegmenter) {
capi::ICU4XLineSegmenter_destroy(mSegmenter);
}
#endif
}
Maybe<uint32_t> LineBreakIteratorUtf16::Next() {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
const int32_t nextPos = capi::ICU4XLineBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
if (!nextPos) {
return Next();
}
mPos = nextPos;
return Some(mPos);
}
#endif
const int32_t nextPos =
LineBreaker::Next(mText.Elements(), mText.Length(), mPos);
if (nextPos == NS_LINEBREAKER_NEED_MORE_TEXT) {
return Nothing();
}
mPos = nextPos;
return Some(mPos);
}
Maybe<uint32_t> LineBreakIteratorUtf16::Seek(uint32_t aPos) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
if (mPos >= aPos) {
return Next();
}
while (mPos < aPos) {
const int32_t nextPos = capi::ICU4XLineBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
mPos = static_cast<uint32_t>(nextPos);
}
if (aPos < mPos) {
return Some(mPos);
}
return Next();
}
#endif
return SegmentIteratorUtf16::Seek(aPos);
}
WordBreakIteratorUtf16::WordBreakIteratorUtf16(Span<const char16_t> aText)
: SegmentIteratorUtf16(aText) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (!StaticPrefs::intl_icu4x_segmenter_enabled()) {
return;
}
auto result =
capi::ICU4XWordSegmenter_create_auto(mozilla::intl::GetDataProvider());
MOZ_RELEASE_ASSERT(result.is_ok);
mSegmenter = result.ok;
mIterator = capi::ICU4XWordSegmenter_segment_utf16(
mSegmenter, mText.Elements(), mText.Length());
#endif
}
WordBreakIteratorUtf16::~WordBreakIteratorUtf16() {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
capi::ICU4XWordBreakIteratorUtf16_destroy(mIterator);
}
if (mSegmenter) {
capi::ICU4XWordSegmenter_destroy(mSegmenter);
}
#endif
}
void WordBreakIteratorUtf16::Reset(Span<const char16_t> aText) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
mText = aText;
if (mIterator) {
capi::ICU4XWordBreakIteratorUtf16_destroy(mIterator);
}
mIterator = capi::ICU4XWordSegmenter_segment_utf16(
mSegmenter, mText.Elements(), mText.Length());
#endif
}
Maybe<uint32_t> WordBreakIteratorUtf16::Next() {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
const int32_t nextPos = capi::ICU4XWordBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
if (!nextPos) {
return Next();
}
mPos = nextPos;
return Some(mPos);
}
#endif
const int32_t nextPos =
WordBreaker::Next(mText.Elements(), mText.Length(), mPos);
if (nextPos == NS_WORDBREAKER_NEED_MORE_TEXT) {
return Nothing();
}
mPos = nextPos;
return Some(mPos);
}
Maybe<uint32_t> WordBreakIteratorUtf16::Seek(uint32_t aPos) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
if (mPos >= aPos) {
return Next();
}
while (mPos < aPos) {
const int32_t nextPos = capi::ICU4XWordBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
mPos = static_cast<uint32_t>(nextPos);
}
if (aPos < mPos) {
return Some(mPos);
}
return Next();
}
#endif
return SegmentIteratorUtf16::Seek(aPos);
}
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
capi::ICU4XGraphemeClusterSegmenter*
GraphemeClusterBreakIteratorUtf16::sSegmenter = nullptr;
#endif
GraphemeClusterBreakIteratorUtf16::GraphemeClusterBreakIteratorUtf16(
Span<const char16_t> aText)
: SegmentIteratorUtf16(aText) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (!StaticPrefs::intl_icu4x_segmenter_enabled()) {
return;
}
static std::once_flag sOnce;
std::call_once(sOnce, [] {
auto result = capi::ICU4XGraphemeClusterSegmenter_create(
mozilla::intl::GetDataProvider());
MOZ_RELEASE_ASSERT(result.is_ok);
sSegmenter = result.ok;
NS_DispatchToMainThread(
NS_NewRunnableFunction("GraphemeClusterBreakIteratorUtf16", [] {
RunOnShutdown([] {
capi::ICU4XGraphemeClusterSegmenter_destroy(sSegmenter);
sSegmenter = nullptr;
});
}));
});
MOZ_RELEASE_ASSERT(sSegmenter);
mIterator = capi::ICU4XGraphemeClusterSegmenter_segment_utf16(
sSegmenter, mText.Elements(), mText.Length());
#endif
}
GraphemeClusterBreakIteratorUtf16::~GraphemeClusterBreakIteratorUtf16() {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
capi::ICU4XGraphemeClusterBreakIteratorUtf16_destroy(mIterator);
}
#endif
}
enum HSType {
HST_NONE = U_HST_NOT_APPLICABLE,
HST_L = U_HST_LEADING_JAMO,
HST_V = U_HST_VOWEL_JAMO,
HST_T = U_HST_TRAILING_JAMO,
HST_LV = U_HST_LV_SYLLABLE,
HST_LVT = U_HST_LVT_SYLLABLE
};
static HSType GetHangulSyllableType(uint32_t aCh) {
return HSType(UnicodeProperties::GetIntPropertyValue(
aCh, UnicodeProperties::IntProperty::HangulSyllableType));
}
Maybe<uint32_t> GraphemeClusterBreakIteratorUtf16::Next() {
const auto len = mText.Length();
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
const int32_t nextPos =
capi::ICU4XGraphemeClusterBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
if (!nextPos) {
return Next();
}
mPos = nextPos;
return Some(mPos);
}
#endif
if (mPos >= len) {
// The iterator has already reached the end.
return Nothing();
}
uint32_t ch = mText[mPos++];
if (mPos < len && NS_IS_SURROGATE_PAIR(ch, mText[mPos])) {
ch = SURROGATE_TO_UCS4(ch, mText[mPos++]);
} else if ((ch & ~0xff) == 0x1100 || (ch >= 0xa960 && ch <= 0xa97f) ||
(ch >= 0xac00 && ch <= 0xd7ff)) {
// Handle conjoining Jamo that make Hangul syllables
HSType hangulState = GetHangulSyllableType(ch);
while (mPos < len) {
ch = mText[mPos];
HSType hangulType = GetHangulSyllableType(ch);
switch (hangulType) {
case HST_L:
case HST_LV:
case HST_LVT:
if (hangulState == HST_L) {
hangulState = hangulType;
mPos++;
continue;
}
break;
case HST_V:
if ((hangulState != HST_NONE) && (hangulState != HST_T) &&
(hangulState != HST_LVT)) {
hangulState = hangulType;
mPos++;
continue;
}
break;
case HST_T:
if (hangulState != HST_NONE && hangulState != HST_L) {
hangulState = hangulType;
mPos++;
continue;
}
break;
default:
break;
}
break;
}
}
const uint32_t kVS16 = 0xfe0f;
const uint32_t kZWJ = 0x200d;
// UTF-16 surrogate values for Fitzpatrick type modifiers
const uint32_t kFitzpatrickHigh = 0xD83C;
const uint32_t kFitzpatrickLowFirst = 0xDFFB;
const uint32_t kFitzpatrickLowLast = 0xDFFF;
// Checking the emoji-presentation property of the base character is a bit
// expensive, so we do it lazily.
enum class EmojiStatus : uint8_t {
No,
Yes,
Unknown,
} baseIsEmojiStatus = EmojiStatus::Unknown;
// Remember the base character and the position of the next, in case we need
// to evaluate its emoji status.
uint32_t baseCh = ch;
uint32_t afterBase = mPos;
auto isFitzpatrickModifierAt = [&](uint32_t aPos) -> bool {
return aPos + 1 < len && mText[aPos] == kFitzpatrickHigh &&
mText[aPos + 1] >= kFitzpatrickLowFirst &&
mText[aPos + 1] <= kFitzpatrickLowLast;
};
auto baseIsEmoji = [&]() -> bool {
if (baseIsEmojiStatus == EmojiStatus::Unknown) {
auto basePresentation = GetEmojiPresentation(baseCh);
baseIsEmojiStatus =
basePresentation == EmojiDefault ||
(basePresentation == TextDefault &&
((afterBase < len && mText[afterBase] == kVS16) ||
isFitzpatrickModifierAt(afterBase)))
? EmojiStatus::Yes
: EmojiStatus::No;
}
return baseIsEmojiStatus == EmojiStatus::Yes;
};
bool prevWasZwj = false;
while (mPos < len) {
ch = mText[mPos];
size_t chLen = 1;
// Check for surrogate pairs; note that isolated surrogates will just
// be treated as generic (non-cluster-extending) characters here,
// which is fine for cluster-iterating purposes
if (mPos < len - 1 && NS_IS_SURROGATE_PAIR(ch, mText[mPos + 1])) {
ch = SURROGATE_TO_UCS4(ch, mText[mPos + 1]);
chLen = 2;
}
bool extendCluster =
IsClusterExtender(ch) ||
(prevWasZwj && baseIsEmoji() &&
((GetEmojiPresentation(ch) == EmojiDefault) ||
(GetEmojiPresentation(ch) == TextDefault && mPos + chLen < len &&
mText[mPos + chLen] == kVS16)));
if (!extendCluster) {
break;
}
prevWasZwj = (ch == kZWJ);
mPos += chLen;
}
MOZ_ASSERT(mPos <= len, "Next() has overshot the string!");
return Some(mPos);
}
Maybe<uint32_t> GraphemeClusterBreakIteratorUtf16::Seek(uint32_t aPos) {
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (mIterator) {
if (mPos >= aPos) {
return Next();
}
while (mPos < aPos) {
const int32_t nextPos =
capi::ICU4XGraphemeClusterBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
mPos = static_cast<uint32_t>(nextPos);
}
if (aPos < mPos) {
return Some(mPos);
}
return Next();
}
#endif
return SegmentIteratorUtf16::Seek(aPos);
}
GraphemeClusterBreakReverseIteratorUtf16::
GraphemeClusterBreakReverseIteratorUtf16(Span<const char16_t> aText)
: SegmentIteratorUtf16(aText) {
mPos = mText.Length();
}
Maybe<uint32_t> GraphemeClusterBreakReverseIteratorUtf16::Next() {
if (mPos == 0) {
return Nothing();
}
uint32_t ch;
do {
ch = mText[--mPos];
if (mPos > 0 && NS_IS_SURROGATE_PAIR(mText[mPos - 1], ch)) {
ch = SURROGATE_TO_UCS4(mText[--mPos], ch);
}
if (!IsClusterExtender(ch)) {
break;
}
} while (mPos > 0);
// XXX May need to handle conjoining Jamo
return Some(mPos);
}
Maybe<uint32_t> GraphemeClusterBreakReverseIteratorUtf16::Seek(uint32_t aPos) {
if (mPos > aPos) {
mPos = aPos;
}
return Next();
}
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
SentenceBreakIteratorUtf16::SentenceBreakIteratorUtf16(
Span<const char16_t> aText)
: SegmentIteratorUtf16(aText) {
auto result =
capi::ICU4XSentenceSegmenter_create(mozilla::intl::GetDataProvider());
MOZ_RELEASE_ASSERT(result.is_ok);
mSegmenter = result.ok;
mIterator = capi::ICU4XSentenceSegmenter_segment_utf16(
mSegmenter, mText.Elements(), mText.Length());
}
SentenceBreakIteratorUtf16::~SentenceBreakIteratorUtf16() {
if (mIterator) {
capi::ICU4XSentenceBreakIteratorUtf16_destroy(mIterator);
}
if (mSegmenter) {
capi::ICU4XSentenceSegmenter_destroy(mSegmenter);
}
}
Maybe<uint32_t> SentenceBreakIteratorUtf16::Seek(uint32_t aPos) {
if (!mIterator) {
return Nothing();
}
if (mPos >= aPos) {
return Next();
}
while (mPos < aPos) {
const int32_t nextPos =
capi::ICU4XSentenceBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
mPos = static_cast<uint32_t>(nextPos);
}
if (aPos < mPos) {
return Some(mPos);
}
return Next();
}
Maybe<uint32_t> SentenceBreakIteratorUtf16::Next() {
if (!mIterator) {
return Nothing();
}
const int32_t nextPos = capi::ICU4XSentenceBreakIteratorUtf16_next(mIterator);
if (nextPos < 0) {
return Nothing();
}
if (!nextPos) {
return Next();
}
mPos = nextPos;
return Some(mPos);
}
#endif
Result<UniquePtr<Segmenter>, ICUError> Segmenter::TryCreate(
Span<const char> aLocale, const SegmenterOptions& aOptions) {
#if !defined(MOZ_ICU4X) || !defined(JS_HAS_INTL_API)
if (aOptions.mGranularity == SegmenterGranularity::Sentence) {
// Grapheme and Sentence iterator are not yet implemented.
return Err(ICUError::InternalError);
}
#endif
return MakeUnique<Segmenter>(aLocale, aOptions);
}
UniquePtr<SegmentIteratorUtf16> Segmenter::Segment(
Span<const char16_t> aText) const {
switch (mOptions.mGranularity) {
case SegmenterGranularity::Grapheme:
return MakeUnique<GraphemeClusterBreakIteratorUtf16>(aText);
case SegmenterGranularity::Sentence:
#if defined(MOZ_ICU4X) && defined(JS_HAS_INTL_API)
if (StaticPrefs::intl_icu4x_segmenter_enabled()) {
return MakeUnique<SentenceBreakIteratorUtf16>(aText);
}
#endif
MOZ_ASSERT_UNREACHABLE("Unimplemented yet!");
return nullptr;
case SegmenterGranularity::Word:
return MakeUnique<WordBreakIteratorUtf16>(aText);
case SegmenterGranularity::Line:
return MakeUnique<LineBreakIteratorUtf16>(aText);
}
MOZ_ASSERT_UNREACHABLE("All granularities must be handled!");
return nullptr;
}
} // namespace mozilla::intl