diff options
Diffstat (limited to '')
| -rw-r--r-- | vcl/source/gdi/impglyphitem.cxx | 560 |
1 files changed, 560 insertions, 0 deletions
diff --git a/vcl/source/gdi/impglyphitem.cxx b/vcl/source/gdi/impglyphitem.cxx new file mode 100644 index 000000000..8ec137434 --- /dev/null +++ b/vcl/source/gdi/impglyphitem.cxx @@ -0,0 +1,560 @@ +/* -*- Mode: C++; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */ +/* + * This file is part of the LibreOffice project. + * + * 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/. + * + * This file incorporates work covered by the following license notice: + * + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed + * with this work for additional information regarding copyright + * ownership. The ASF licenses this file to you under the Apache + * License, Version 2.0 (the "License"); you may not use this file + * except in compliance with the License. You may obtain a copy of + * the License at http://www.apache.org/licenses/LICENSE-2.0 . + */ + +#include <impglyphitem.hxx> +#include <vcl/glyphitemcache.hxx> +#include <vcl/vcllayout.hxx> +#include <vcl/lazydelete.hxx> +#include <tools/stream.hxx> +#include <unotools/configmgr.hxx> +#include <TextLayoutCache.hxx> +#include <officecfg/Office/Common.hxx> + +#include <unicode/ubidi.h> +#include <unicode/uchar.h> + +// These need being explicit because of SalLayoutGlyphsImpl being private in vcl. +SalLayoutGlyphs::SalLayoutGlyphs() {} + +SalLayoutGlyphs::~SalLayoutGlyphs() {} + +SalLayoutGlyphs::SalLayoutGlyphs(SalLayoutGlyphs&& rOther) noexcept +{ + std::swap(m_pImpl, rOther.m_pImpl); + std::swap(m_pExtraImpls, rOther.m_pExtraImpls); +} + +SalLayoutGlyphs& SalLayoutGlyphs::operator=(SalLayoutGlyphs&& rOther) noexcept +{ + if (this != &rOther) + { + std::swap(m_pImpl, rOther.m_pImpl); + std::swap(m_pExtraImpls, rOther.m_pExtraImpls); + } + return *this; +} + +bool SalLayoutGlyphs::IsValid() const +{ + if (m_pImpl == nullptr) + return false; + if (!m_pImpl->IsValid()) + return false; + if (m_pExtraImpls) + for (std::unique_ptr<SalLayoutGlyphsImpl> const& impl : *m_pExtraImpls) + if (!impl->IsValid()) + return false; + return true; +} + +void SalLayoutGlyphs::Invalidate() +{ + // Invalidating is in fact simply clearing. + m_pImpl.reset(); + m_pExtraImpls.reset(); +} + +SalLayoutGlyphsImpl* SalLayoutGlyphs::Impl(unsigned int nLevel) const +{ + if (nLevel == 0) + return m_pImpl.get(); + if (m_pExtraImpls != nullptr && nLevel - 1 < m_pExtraImpls->size()) + return (*m_pExtraImpls)[nLevel - 1].get(); + return nullptr; +} + +void SalLayoutGlyphs::AppendImpl(SalLayoutGlyphsImpl* pImpl) +{ + if (!m_pImpl) + m_pImpl.reset(pImpl); + else + { + if (!m_pExtraImpls) + m_pExtraImpls.reset(new std::vector<std::unique_ptr<SalLayoutGlyphsImpl>>); + m_pExtraImpls->emplace_back(pImpl); + } +} + +SalLayoutGlyphsImpl* SalLayoutGlyphsImpl::clone() const { return new SalLayoutGlyphsImpl(*this); } + +// Clone, but only glyphs in the given range in the original text string. +// It is possible the given range may not be cloned, in which case this returns nullptr. +SalLayoutGlyphsImpl* SalLayoutGlyphsImpl::cloneCharRange(sal_Int32 index, sal_Int32 length) const +{ + std::unique_ptr<SalLayoutGlyphsImpl> copy(new SalLayoutGlyphsImpl(*GetFont())); + copy->SetFlags(GetFlags()); + if (empty()) + return copy.release(); + bool rtl = front().IsRTLGlyph(); + // Avoid mixing LTR/RTL or layouts that do not have it set explicitly (BiDiStrong). Otherwise + // the subset may not quite match what would a real layout call give (e.g. some characters with neutral + // direction such as space might have different LTR/RTL flag). It seems bailing out here mostly + // avoid relatively rare corner cases and doesn't matter for performance. + // This is also checked in SalLayoutGlyphsCache::GetLayoutGlyphs() below. + if (!(GetFlags() & SalLayoutFlags::BiDiStrong) + || rtl != bool(GetFlags() & SalLayoutFlags::BiDiRtl)) + return nullptr; + copy->reserve(std::min<size_t>(size(), length)); + sal_Int32 beginPos = index; + sal_Int32 endPos = index + length; + const_iterator pos; + if (rtl) + { + // Glyphs are in reverse order for RTL. + beginPos = index + length - 1; + endPos = index - 1; + // Skip glyphs that are in the string after the given index, i.e. are before the glyphs + // we want. + pos = std::partition_point( + begin(), end(), [beginPos](const GlyphItem& it) { return it.charPos() > beginPos; }); + } + else + { + // Skip glyphs that are in the string before the given index (glyphs are sorted by charPos()). + pos = std::partition_point( + begin(), end(), [beginPos](const GlyphItem& it) { return it.charPos() < beginPos; }); + } + if (pos == end()) + return nullptr; + // Require a start at the exact position given, otherwise bail out. + if (pos->charPos() != beginPos) + return nullptr; + // For RTL make sure we're not cutting in the middle of a multi-character glyph + // (for non-RTL charPos is always the start of a multi-character glyph). + if (rtl && pos->charPos() + pos->charCount() > beginPos + 1) + return nullptr; + if (!isSafeToBreak(pos, rtl)) + return nullptr; + // LinearPos needs adjusting to start at xOffset/yOffset for the first item, + // that's how it's computed in GenericSalLayout::LayoutText(). + DevicePoint zeroPoint = pos->linearPos() - DevicePoint(pos->xOffset(), pos->yOffset()); + // Add and adjust all glyphs until the given length. + // The check is written as 'charPos + charCount <= endPos' rather than 'charPos < endPos' + // (or similarly for RTL) to make sure we include complete glyphs. If a glyph is composed + // from several characters, we should not cut in the middle of those characters, so this + // checks the glyph is entirely in the given character range. If it is not, this will end + // the loop and the later 'pos->charPos() != endPos' check will fail and bail out. + // CppunitTest_sw_layoutwriter's testCombiningCharacterCursorPosition would fail without this. + while (pos != end() + && (rtl ? pos->charPos() - pos->charCount() >= endPos + : pos->charPos() + pos->charCount() <= endPos)) + { + if (pos->IsRTLGlyph() != rtl) + return nullptr; // Don't mix RTL and non-RTL runs. + // HACK: When running CppunitTest_sw_uiwriter3's testTdf104649 on Mac there's glyph + // with id 1232 that has 0 charCount, 0 origWidth and inconsistent xOffset (sometimes 0, + // but sometimes not). Possibly font or Harfbuzz bug? It's extremely rare, so simply bail out. + if (pos->charCount() == 0 && pos->origWidth() == 0) + return nullptr; + copy->push_back(*pos); + copy->back().setLinearPos(copy->back().linearPos() - zeroPoint); + ++pos; + } + if (pos != end()) + { + // Fail if the next character is not at the expected past-end position. For RTL check + // that we're not cutting in the middle of a multi-character glyph. + if (rtl ? pos->charPos() + pos->charCount() != endPos + 1 : pos->charPos() != endPos) + return nullptr; + if (!isSafeToBreak(pos, rtl)) + return nullptr; + } + return copy.release(); +} + +bool SalLayoutGlyphsImpl::isSafeToBreak(const_iterator pos, bool rtl) const +{ + if (rtl) + { + // RTL is more complicated, because HB_GLYPH_FLAG_UNSAFE_TO_BREAK talks about beginning + // of a cluster, which refers to the text, not glyphs. This function is called + // for the first glyph of the subset and the first glyph after the subset, but since + // the glyphs are backwards, and we need the beginning of cluster at the start of the text + // and beginning of the cluster after the text, we need to check glyphs before this position. + if (pos == begin()) + return true; + --pos; + } + // Don't create a subset if it's not safe to break at the beginning or end of the sequence + // (https://harfbuzz.github.io/harfbuzz-hb-buffer.html#hb-glyph-flags-t). + if (pos->IsUnsafeToBreak() || (pos->IsInCluster() && !pos->IsClusterStart())) + return false; + return true; +} + +#ifdef DBG_UTIL +bool SalLayoutGlyphsImpl::isEqual(const SalLayoutGlyphsImpl* other) const +{ + if (!GetFont()->mxFontMetric->CompareDeviceIndependentFontAttributes( + *other->GetFont()->mxFontMetric)) + return false; + if (GetFlags() != other->GetFlags()) + return false; + if (empty() || other->empty()) + return empty() == other->empty(); + if (size() != other->size()) + return false; + for (size_t pos = 0; pos < size(); ++pos) + { + if ((*this)[pos] != (*other)[pos]) + return false; + } + return true; +} +#endif + +bool SalLayoutGlyphsImpl::IsValid() const +{ + if (!m_rFontInstance.is()) + return false; + return true; +} + +void SalLayoutGlyphsCache::clear() { mCachedGlyphs.clear(); } + +SalLayoutGlyphsCache* SalLayoutGlyphsCache::self() +{ + static vcl::DeleteOnDeinit<SalLayoutGlyphsCache> cache( + !utl::ConfigManager::IsFuzzing() + ? officecfg::Office::Common::Cache::Font::GlyphsCacheSize::get() + : 20000); + return cache.get(); +} + +static UBiDiDirection getBiDiDirection(const OUString& text, sal_Int32 index, sal_Int32 len) +{ + // Return whether all character are LTR, RTL, neutral or whether it's mixed. + // This is sort of ubidi_getBaseDirection() and ubidi_getDirection(), + // but it's meant to be fast but also check all characters. + sal_Int32 end = index + len; + UBiDiDirection direction = UBIDI_NEUTRAL; + while (index < end) + { + switch (u_charDirection(text.iterateCodePoints(&index))) + { + // Only characters with strong direction. + case U_LEFT_TO_RIGHT: + if (direction == UBIDI_RTL) + return UBIDI_MIXED; + direction = UBIDI_LTR; + break; + case U_RIGHT_TO_LEFT: + case U_RIGHT_TO_LEFT_ARABIC: + if (direction == UBIDI_LTR) + return UBIDI_MIXED; + direction = UBIDI_RTL; + break; + default: + break; + } + } + return direction; +} + +static SalLayoutGlyphs makeGlyphsSubset(const SalLayoutGlyphs& source, + const OutputDevice* outputDevice, const OUString& text, + sal_Int32 index, sal_Int32 len) +{ + // tdf#149264: We need to check if the text is LTR, RTL or mixed. Apparently + // harfbuzz doesn't give reproducible results (or possibly HB_GLYPH_FLAG_UNSAFE_TO_BREAK + // is not reliable?) when asked to lay out RTL text as LTR. So require that the whole + // subset ir either LTR or RTL. + UBiDiDirection direction = getBiDiDirection(text, index, len); + if (direction == UBIDI_MIXED) + return SalLayoutGlyphs(); + SalLayoutGlyphs ret; + for (int level = 0;; ++level) + { + const SalLayoutGlyphsImpl* sourceLevel = source.Impl(level); + if (sourceLevel == nullptr) + break; + bool sourceRtl = bool(sourceLevel->GetFlags() & SalLayoutFlags::BiDiRtl); + if ((direction == UBIDI_LTR && sourceRtl) || (direction == UBIDI_RTL && !sourceRtl)) + return SalLayoutGlyphs(); + SalLayoutGlyphsImpl* cloned = sourceLevel->cloneCharRange(index, len); + // If the glyphs range cannot be cloned, bail out. + if (cloned == nullptr) + return SalLayoutGlyphs(); + // If the entire string is mixed LTR/RTL but the subset is only LTR, + // then make sure the flags match that, otherwise checkGlyphsEqual() + // would assert on flags being different. + cloned->SetFlags(cloned->GetFlags() + | outputDevice->GetBiDiLayoutFlags(text, index, index + len)); + // SalLayoutFlags::KashidaJustification is set only if any glyph + // in the range has GlyphItemFlags::ALLOW_KASHIDA (otherwise unset it). + if (cloned->GetFlags() & SalLayoutFlags::KashidaJustification) + { + bool hasKashida = false; + for (const GlyphItem& item : *cloned) + { + if (item.AllowKashida()) + { + hasKashida = true; + break; + } + } + if (!hasKashida) + cloned->SetFlags(cloned->GetFlags() & ~SalLayoutFlags::KashidaJustification); + } +#ifdef DBG_UTIL + else + for (const GlyphItem& item : *cloned) + assert(!item.AllowKashida()); +#endif + ret.AppendImpl(cloned); + } + return ret; +} + +#ifdef DBG_UTIL +static void checkGlyphsEqual(const SalLayoutGlyphs& g1, const SalLayoutGlyphs& g2) +{ + for (int level = 0;; ++level) + { + const SalLayoutGlyphsImpl* l1 = g1.Impl(level); + const SalLayoutGlyphsImpl* l2 = g2.Impl(level); + if (l1 == nullptr || l2 == nullptr) + { + assert(l1 == l2); + break; + } + assert(l1->isEqual(l2)); + } +} +#endif + +const SalLayoutGlyphs* +SalLayoutGlyphsCache::GetLayoutGlyphs(VclPtr<const OutputDevice> outputDevice, const OUString& text, + sal_Int32 nIndex, sal_Int32 nLen, tools::Long nLogicWidth, + const vcl::text::TextLayoutCache* layoutCache) +{ + if (nLen == 0) + return nullptr; + const CachedGlyphsKey key(outputDevice, text, nIndex, nLen, nLogicWidth); + GlyphsCache::const_iterator it = mCachedGlyphs.find(key); + if (it != mCachedGlyphs.end()) + { + if (it->second.IsValid()) + return &it->second; + // Do not try to create the layout here. If a cache item exists, it's already + // been attempted and the layout was invalid (this happens with MultiSalLayout). + // So in that case this is a cached failure. + return nullptr; + } + bool resetLastSubstringKey = true; + const sal_Unicode nbSpace = 0xa0; // non-breaking space + // SalLayoutGlyphsImpl::cloneCharRange() requires BiDiStrong, so if not set, do not even try. + bool skipGlyphSubsets + = !(outputDevice->GetLayoutMode() & vcl::text::ComplexTextLayoutFlags::BiDiStrong); + if ((nIndex != 0 || nLen != text.getLength()) && !skipGlyphSubsets) + { + // The glyphs functions are often called first for an entire string + // and then with an increasing starting index until the end of the string. + // Which means it's possible to get the glyphs faster by just copying + // a subset of the full glyphs and adjusting as necessary. + if (mLastTemporaryKey.has_value() && mLastTemporaryKey == key) + return &mLastTemporaryGlyphs; + const CachedGlyphsKey keyWhole(outputDevice, text, 0, text.getLength(), nLogicWidth); + GlyphsCache::const_iterator itWhole = mCachedGlyphs.find(keyWhole); + if (itWhole == mCachedGlyphs.end()) + { + // This function may often be called repeatedly for segments of the same string, + // in which case it is more efficient to cache glyphs for the entire string + // and then return subsets of them. So if a second call either starts at the same + // position or starts at the end of the previous call, cache the entire string. + // This used to do this only for the first two segments of the string, + // but that missed the case when the font slightly changed e.g. because of the first + // part being underlined. Doing this for any two segments allows this optimization + // even when the prefix of the string would use a different font. + // TODO: Can those font differences be ignored? + // Writer layouts tests enable SAL_ABORT_ON_NON_APPLICATION_FONT_USE in order + // to make PrintFontManager::Substitute() abort if font fallback happens. When + // laying out the entire string the chance this happens increases (e.g. testAbi11870 + // normally calls this function only for a part of a string, but this optimization + // lays out the entire string and causes a fallback). Since this optimization + // does not change result of this function, simply disable it for those tests. + static bool bAbortOnFontSubstitute + = getenv("SAL_ABORT_ON_NON_APPLICATION_FONT_USE") != nullptr; + if (mLastSubstringKey.has_value() && !bAbortOnFontSubstitute) + { + sal_Int32 pos = nIndex; + if (mLastSubstringKey->len < pos && text[pos - 1] == nbSpace) + --pos; // Writer skips a non-breaking space, so skip that character too. + if ((mLastSubstringKey->len == pos || mLastSubstringKey->index == nIndex) + && mLastSubstringKey + == CachedGlyphsKey(outputDevice, text, mLastSubstringKey->index, + mLastSubstringKey->len, nLogicWidth)) + { + GetLayoutGlyphs(outputDevice, text, 0, text.getLength(), nLogicWidth, + layoutCache); + itWhole = mCachedGlyphs.find(keyWhole); + } + else + mLastSubstringKey.reset(); + } + if (!mLastSubstringKey.has_value()) + { + mLastSubstringKey = key; + resetLastSubstringKey = false; + } + } + if (itWhole != mCachedGlyphs.end() && itWhole->second.IsValid()) + { + mLastSubstringKey.reset(); + mLastTemporaryGlyphs + = makeGlyphsSubset(itWhole->second, outputDevice, text, nIndex, nLen); + if (mLastTemporaryGlyphs.IsValid()) + { + mLastTemporaryKey = std::move(key); +#ifdef DBG_UTIL + std::shared_ptr<const vcl::text::TextLayoutCache> tmpLayoutCache; + if (layoutCache == nullptr) + { + tmpLayoutCache = vcl::text::TextLayoutCache::Create(text); + layoutCache = tmpLayoutCache.get(); + } + // Check if the subset result really matches what we would get normally, + // to make sure corner cases are handled well (see SalLayoutGlyphsImpl::cloneCharRange()). + std::unique_ptr<SalLayout> layout + = outputDevice->ImplLayout(text, nIndex, nLen, Point(0, 0), nLogicWidth, {}, + SalLayoutFlags::GlyphItemsOnly, layoutCache); + assert(layout); + checkGlyphsEqual(mLastTemporaryGlyphs, layout->GetGlyphs()); +#endif + return &mLastTemporaryGlyphs; + } + } + } + if (resetLastSubstringKey) + { + // Writer does non-breaking space differently (not as part of the string), so in that + // case ignore that call and still allow finding two adjacent substrings that have + // the non-breaking space between them. + if (nLen != 1 || text[nIndex] != nbSpace) + mLastSubstringKey.reset(); + } + + std::shared_ptr<const vcl::text::TextLayoutCache> tmpLayoutCache; + if (layoutCache == nullptr) + { + tmpLayoutCache = vcl::text::TextLayoutCache::Create(text); + layoutCache = tmpLayoutCache.get(); + } + std::unique_ptr<SalLayout> layout + = outputDevice->ImplLayout(text, nIndex, nLen, Point(0, 0), nLogicWidth, {}, + SalLayoutFlags::GlyphItemsOnly, layoutCache); + if (layout) + { + SalLayoutGlyphs glyphs = layout->GetGlyphs(); + if (glyphs.IsValid()) + { + // TODO: Fallbacks do not work reliably (fallback font not included in the key), + // so do not cache (but still return once, using the temporary without a key set). + if (glyphs.Impl(1) != nullptr) + { + mLastTemporaryGlyphs = std::move(glyphs); + mLastTemporaryKey.reset(); + return &mLastTemporaryGlyphs; + } + mCachedGlyphs.insert(std::make_pair(key, layout->GetGlyphs())); + assert(mCachedGlyphs.find(key) + == mCachedGlyphs.begin()); // newly inserted item is first + return &mCachedGlyphs.begin()->second; + } + } + // Failure, cache it too as invalid glyphs. + mCachedGlyphs.insert(std::make_pair(key, SalLayoutGlyphs())); + return nullptr; +} + +SalLayoutGlyphsCache::CachedGlyphsKey::CachedGlyphsKey( + const VclPtr<const OutputDevice>& outputDevice, const OUString& t, sal_Int32 i, sal_Int32 l, + tools::Long w) + : text(t) + , index(i) + , len(l) + , logicWidth(w) + // we also need to save things used in OutputDevice::ImplPrepareLayoutArgs(), in case they + // change in the output device, plus mapMode affects the sizes. + , fontMetric(outputDevice->GetFontMetric()) + // TODO It would be possible to get a better hit ratio if mapMode wasn't part of the key + // and results that differ only in mapmode would have coordinates adjusted based on that. + // That would occasionally lead to rounding errors (at least differences that would + // make checkGlyphsEqual() fail). + , mapMode(outputDevice->GetMapMode()) + , rtl(outputDevice->IsRTLEnabled()) + , layoutMode(outputDevice->GetLayoutMode()) + , digitLanguage(outputDevice->GetDigitLanguage()) +{ + const LogicalFontInstance* fi = outputDevice->GetFontInstance(); + fi->GetScale(&fontScaleX, &fontScaleY); + + const vcl::font::FontSelectPattern& rFSD = fi->GetFontSelectPattern(); + artificialItalic = rFSD.GetItalic() != ITALIC_NONE && fontMetric.GetItalic() == ITALIC_NONE; + artificialBold = rFSD.GetWeight() > WEIGHT_MEDIUM && fontMetric.GetWeight() <= WEIGHT_MEDIUM; + + hashValue = 0; + o3tl::hash_combine(hashValue, vcl::text::FirstCharsStringHash()(text)); + o3tl::hash_combine(hashValue, index); + o3tl::hash_combine(hashValue, len); + o3tl::hash_combine(hashValue, logicWidth); + o3tl::hash_combine(hashValue, outputDevice.get()); + // Need to use IgnoreColor, because sometimes the color changes, but it's irrelevant + // for text layout (and also obsolete in vcl::Font). + o3tl::hash_combine(hashValue, fontMetric.GetHashValueIgnoreColor()); + // For some reason font scale may differ even if vcl::Font is the same, + // so explicitly check it too. + o3tl::hash_combine(hashValue, fontScaleX); + o3tl::hash_combine(hashValue, fontScaleY); + o3tl::hash_combine(hashValue, mapMode.GetHashValue()); + o3tl::hash_combine(hashValue, rtl); + o3tl::hash_combine(hashValue, artificialItalic); + o3tl::hash_combine(hashValue, artificialBold); + o3tl::hash_combine(hashValue, layoutMode); + o3tl::hash_combine(hashValue, digitLanguage.get()); +} + +inline bool SalLayoutGlyphsCache::CachedGlyphsKey::operator==(const CachedGlyphsKey& other) const +{ + return hashValue == other.hashValue && index == other.index && len == other.len + && logicWidth == other.logicWidth && mapMode == other.mapMode && rtl == other.rtl + && artificialItalic == other.artificialItalic && artificialBold == other.artificialBold + && layoutMode == other.layoutMode && digitLanguage == other.digitLanguage + && fontScaleX == other.fontScaleX && fontScaleY == other.fontScaleY + && fontMetric.EqualIgnoreColor(other.fontMetric) + && vcl::text::FastStringCompareEqual()(text, other.text); + // Slower things last in the comparison. +} + +size_t SalLayoutGlyphsCache::GlyphsCost::operator()(const SalLayoutGlyphs& glyphs) const +{ + size_t cost = 0; + for (int level = 0;; ++level) + { + const SalLayoutGlyphsImpl* impl = glyphs.Impl(level); + if (impl == nullptr) + break; + // Count size in bytes, both the SalLayoutGlyphsImpl instance and contained GlyphItem's. + cost += sizeof(*impl); + cost += impl->size() * sizeof(impl->front()); + } + return cost; +} + +/* vim:set shiftwidth=4 softtabstop=4 expandtab: */ |