mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-10-30 05:35:31 +00:00
5805 lines
204 KiB
C++
5805 lines
204 KiB
C++
/* -*- Mode: C++; tab-width: 20; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
|
|
/* 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/. */
|
|
|
|
#ifdef MOZ_LOGGING
|
|
#define FORCE_PR_LOG /* Allow logging in the release build */
|
|
#endif
|
|
#include "prlog.h"
|
|
|
|
#include "nsServiceManagerUtils.h"
|
|
#include "nsReadableUtils.h"
|
|
#include "nsExpirationTracker.h"
|
|
#include "nsILanguageAtomService.h"
|
|
#include "nsITimer.h"
|
|
|
|
#include "gfxFont.h"
|
|
#include "gfxPlatform.h"
|
|
#include "nsGkAtoms.h"
|
|
|
|
#include "gfxTypes.h"
|
|
#include "nsAlgorithm.h"
|
|
#include "gfxContext.h"
|
|
#include "gfxFontMissingGlyphs.h"
|
|
#include "gfxUserFontSet.h"
|
|
#include "gfxPlatformFontList.h"
|
|
#include "gfxScriptItemizer.h"
|
|
#include "nsUnicodeProperties.h"
|
|
#include "nsMathUtils.h"
|
|
#include "nsBidiUtils.h"
|
|
#include "nsUnicodeRange.h"
|
|
#include "nsStyleConsts.h"
|
|
#include "mozilla/Preferences.h"
|
|
#include "mozilla/Services.h"
|
|
#include "mozilla/Telemetry.h"
|
|
#include "gfxSVGGlyphs.h"
|
|
|
|
#include "cairo.h"
|
|
#include "gfxFontTest.h"
|
|
|
|
#include "harfbuzz/hb.h"
|
|
|
|
#include "nsCRT.h"
|
|
#include "sampler.h"
|
|
|
|
#include <algorithm>
|
|
|
|
using namespace mozilla;
|
|
using namespace mozilla::gfx;
|
|
using namespace mozilla::unicode;
|
|
using mozilla::services::GetObserverService;
|
|
|
|
gfxFontCache *gfxFontCache::gGlobalCache = nullptr;
|
|
|
|
#ifdef DEBUG_roc
|
|
#define DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
#endif
|
|
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
static uint32_t gTextRunStorageHighWaterMark = 0;
|
|
static uint32_t gTextRunStorage = 0;
|
|
static uint32_t gFontCount = 0;
|
|
static uint32_t gGlyphExtentsCount = 0;
|
|
static uint32_t gGlyphExtentsWidthsTotalSize = 0;
|
|
static uint32_t gGlyphExtentsSetupEagerSimple = 0;
|
|
static uint32_t gGlyphExtentsSetupEagerTight = 0;
|
|
static uint32_t gGlyphExtentsSetupLazyTight = 0;
|
|
static uint32_t gGlyphExtentsSetupFallBackToTight = 0;
|
|
#endif
|
|
|
|
void
|
|
gfxCharacterMap::NotifyReleased()
|
|
{
|
|
gfxPlatformFontList *fontlist = gfxPlatformFontList::PlatformFontList();
|
|
if (mShared) {
|
|
fontlist->RemoveCmap(this);
|
|
}
|
|
delete this;
|
|
}
|
|
|
|
gfxFontEntry::~gfxFontEntry()
|
|
{
|
|
if (mSVGGlyphs) {
|
|
delete mSVGGlyphs;
|
|
}
|
|
delete mUserFontData;
|
|
}
|
|
|
|
bool gfxFontEntry::IsSymbolFont()
|
|
{
|
|
return mSymbolFont;
|
|
}
|
|
|
|
bool gfxFontEntry::TestCharacterMap(uint32_t aCh)
|
|
{
|
|
if (!mCharacterMap) {
|
|
ReadCMAP();
|
|
NS_ASSERTION(mCharacterMap, "failed to initialize character map");
|
|
}
|
|
return mCharacterMap->test(aCh);
|
|
}
|
|
|
|
nsresult gfxFontEntry::InitializeUVSMap()
|
|
{
|
|
// mUVSOffset will not be initialized
|
|
// until cmap is initialized.
|
|
if (!mCharacterMap) {
|
|
ReadCMAP();
|
|
NS_ASSERTION(mCharacterMap, "failed to initialize character map");
|
|
}
|
|
|
|
if (!mUVSOffset) {
|
|
return NS_ERROR_FAILURE;
|
|
}
|
|
|
|
if (!mUVSData) {
|
|
const uint32_t kCmapTag = TRUETYPE_TAG('c','m','a','p');
|
|
AutoFallibleTArray<uint8_t,16384> buffer;
|
|
if (GetFontTable(kCmapTag, buffer) != NS_OK) {
|
|
mUVSOffset = 0; // don't bother to read the table again
|
|
return NS_ERROR_FAILURE;
|
|
}
|
|
|
|
uint8_t* uvsData;
|
|
nsresult rv = gfxFontUtils::ReadCMAPTableFormat14(
|
|
buffer.Elements() + mUVSOffset,
|
|
buffer.Length() - mUVSOffset,
|
|
uvsData);
|
|
if (NS_FAILED(rv)) {
|
|
mUVSOffset = 0; // don't bother to read the table again
|
|
return rv;
|
|
}
|
|
|
|
mUVSData = uvsData;
|
|
}
|
|
|
|
return NS_OK;
|
|
}
|
|
|
|
uint16_t gfxFontEntry::GetUVSGlyph(uint32_t aCh, uint32_t aVS)
|
|
{
|
|
InitializeUVSMap();
|
|
|
|
if (mUVSData) {
|
|
return gfxFontUtils::MapUVSToGlyphFormat14(mUVSData, aCh, aVS);
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
nsresult gfxFontEntry::ReadCMAP()
|
|
{
|
|
NS_ASSERTION(false, "using default no-op implementation of ReadCMAP");
|
|
mCharacterMap = new gfxCharacterMap();
|
|
return NS_OK;
|
|
}
|
|
|
|
nsString gfxFontEntry::FamilyName() const
|
|
{
|
|
NS_ASSERTION(mFamily, "orphaned font entry");
|
|
if (mFamily) {
|
|
return mFamily->Name();
|
|
} else {
|
|
return nsString();
|
|
}
|
|
}
|
|
|
|
nsString
|
|
gfxFontEntry::RealFaceName()
|
|
{
|
|
FallibleTArray<uint8_t> nameTable;
|
|
nsresult rv = GetFontTable(TRUETYPE_TAG('n','a','m','e'), nameTable);
|
|
if (NS_SUCCEEDED(rv)) {
|
|
nsAutoString name;
|
|
rv = gfxFontUtils::GetFullNameFromTable(nameTable, name);
|
|
if (NS_SUCCEEDED(rv)) {
|
|
return name;
|
|
}
|
|
}
|
|
return Name();
|
|
}
|
|
|
|
already_AddRefed<gfxFont>
|
|
gfxFontEntry::FindOrMakeFont(const gfxFontStyle *aStyle, bool aNeedsBold)
|
|
{
|
|
// the font entry name is the psname, not the family name
|
|
nsRefPtr<gfxFont> font = gfxFontCache::GetCache()->Lookup(this, aStyle);
|
|
|
|
if (!font) {
|
|
gfxFont *newFont = CreateFontInstance(aStyle, aNeedsBold);
|
|
if (!newFont)
|
|
return nullptr;
|
|
if (!newFont->Valid()) {
|
|
delete newFont;
|
|
return nullptr;
|
|
}
|
|
font = newFont;
|
|
gfxFontCache::GetCache()->AddNew(font);
|
|
}
|
|
gfxFont *f = nullptr;
|
|
font.swap(f);
|
|
return f;
|
|
}
|
|
|
|
bool
|
|
gfxFontEntry::HasSVGGlyph(uint32_t aGlyphId)
|
|
{
|
|
NS_ASSERTION(mSVGInitialized, "SVG data has not yet been loaded. TryGetSVGData() first.");
|
|
return mSVGGlyphs->HasSVGGlyph(aGlyphId);
|
|
}
|
|
|
|
bool
|
|
gfxFontEntry::GetSVGGlyphExtents(gfxContext *aContext, uint32_t aGlyphId,
|
|
gfxRect *aResult)
|
|
{
|
|
NS_ABORT_IF_FALSE(mSVGInitialized, "SVG data has not yet been loaded. TryGetSVGData() first.");
|
|
|
|
gfxContextAutoSaveRestore matrixRestore(aContext);
|
|
cairo_matrix_t fontMatrix;
|
|
cairo_get_font_matrix(aContext->GetCairo(), &fontMatrix);
|
|
|
|
gfxMatrix svgToAppSpace = *reinterpret_cast<gfxMatrix*>(&fontMatrix);
|
|
svgToAppSpace.Scale(1.0f / gfxSVGGlyphs::SVG_UNITS_PER_EM,
|
|
1.0f / gfxSVGGlyphs::SVG_UNITS_PER_EM);
|
|
|
|
return mSVGGlyphs->GetGlyphExtents(aGlyphId, svgToAppSpace, aResult);
|
|
}
|
|
|
|
bool
|
|
gfxFontEntry::RenderSVGGlyph(gfxContext *aContext, uint32_t aGlyphId,
|
|
int aDrawMode, gfxTextObjectPaint *aObjectPaint)
|
|
{
|
|
NS_ASSERTION(mSVGInitialized, "SVG data has not yet been loaded. TryGetSVGData() first.");
|
|
return mSVGGlyphs->RenderGlyph(aContext, aGlyphId, gfxFont::DrawMode(aDrawMode),
|
|
aObjectPaint);
|
|
}
|
|
|
|
bool
|
|
gfxFontEntry::TryGetSVGData()
|
|
{
|
|
if (!mSVGInitialized) {
|
|
mSVGInitialized = true;
|
|
|
|
bool svgEnabled;
|
|
nsresult rv =
|
|
Preferences::GetBool("gfx.font_rendering.opentype_svg.enabled",
|
|
&svgEnabled);
|
|
|
|
if (NS_FAILED(rv) || !svgEnabled) {
|
|
return false;
|
|
}
|
|
|
|
FallibleTArray<uint8_t> svgTable;
|
|
rv = GetFontTable(TRUETYPE_TAG('S', 'V', 'G', ' '), svgTable);
|
|
if (NS_FAILED(rv)) {
|
|
return false;
|
|
}
|
|
|
|
FallibleTArray<uint8_t> cmapTable;
|
|
rv = GetFontTable(TRUETYPE_TAG('c', 'm', 'a', 'p'), cmapTable);
|
|
NS_ENSURE_SUCCESS(rv, false);
|
|
|
|
mSVGGlyphs = new gfxSVGGlyphs(svgTable, cmapTable);
|
|
}
|
|
|
|
return !!mSVGGlyphs;
|
|
}
|
|
|
|
/**
|
|
* FontTableBlobData
|
|
*
|
|
* See FontTableHashEntry for the general strategy.
|
|
*/
|
|
|
|
class gfxFontEntry::FontTableBlobData {
|
|
public:
|
|
// Adopts the content of aBuffer.
|
|
// Pass a non-null aHashEntry only if it should be cleared if/when this
|
|
// FontTableBlobData is deleted.
|
|
FontTableBlobData(FallibleTArray<uint8_t>& aBuffer,
|
|
FontTableHashEntry *aHashEntry)
|
|
: mHashEntry(aHashEntry), mHashtable()
|
|
{
|
|
MOZ_COUNT_CTOR(FontTableBlobData);
|
|
mTableData.SwapElements(aBuffer);
|
|
}
|
|
|
|
~FontTableBlobData() {
|
|
MOZ_COUNT_DTOR(FontTableBlobData);
|
|
if (mHashEntry) {
|
|
if (mHashtable) {
|
|
mHashtable->RemoveEntry(mHashEntry->GetKey());
|
|
} else {
|
|
mHashEntry->Clear();
|
|
}
|
|
}
|
|
}
|
|
|
|
// Useful for creating blobs
|
|
const char *GetTable() const
|
|
{
|
|
return reinterpret_cast<const char*>(mTableData.Elements());
|
|
}
|
|
uint32_t GetTableLength() const { return mTableData.Length(); }
|
|
|
|
// Tell this FontTableBlobData to remove the HashEntry when this is
|
|
// destroyed.
|
|
void ManageHashEntry(nsTHashtable<FontTableHashEntry> *aHashtable)
|
|
{
|
|
mHashtable = aHashtable;
|
|
}
|
|
|
|
// Disconnect from the HashEntry (because the blob has already been
|
|
// removed from the hashtable).
|
|
void ForgetHashEntry()
|
|
{
|
|
mHashEntry = nullptr;
|
|
}
|
|
|
|
size_t SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf) const {
|
|
return mTableData.SizeOfExcludingThis(aMallocSizeOf);
|
|
}
|
|
size_t SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf) const {
|
|
return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf);
|
|
}
|
|
|
|
private:
|
|
// The font table data block, owned (via adoption)
|
|
FallibleTArray<uint8_t> mTableData;
|
|
// The blob destroy function needs to know the hashtable entry,
|
|
FontTableHashEntry *mHashEntry;
|
|
// and the owning hashtable, so that it can remove the entry.
|
|
nsTHashtable<FontTableHashEntry> *mHashtable;
|
|
|
|
// not implemented
|
|
FontTableBlobData(const FontTableBlobData&);
|
|
};
|
|
|
|
void
|
|
gfxFontEntry::FontTableHashEntry::SaveTable(FallibleTArray<uint8_t>& aTable)
|
|
{
|
|
Clear();
|
|
// adopts elements of aTable
|
|
FontTableBlobData *data = new FontTableBlobData(aTable, nullptr);
|
|
mBlob = hb_blob_create(data->GetTable(), data->GetTableLength(),
|
|
HB_MEMORY_MODE_READONLY,
|
|
data, DeleteFontTableBlobData);
|
|
}
|
|
|
|
hb_blob_t *
|
|
gfxFontEntry::FontTableHashEntry::
|
|
ShareTableAndGetBlob(FallibleTArray<uint8_t>& aTable,
|
|
nsTHashtable<FontTableHashEntry> *aHashtable)
|
|
{
|
|
Clear();
|
|
// adopts elements of aTable
|
|
mSharedBlobData = new FontTableBlobData(aTable, this);
|
|
mBlob = hb_blob_create(mSharedBlobData->GetTable(),
|
|
mSharedBlobData->GetTableLength(),
|
|
HB_MEMORY_MODE_READONLY,
|
|
mSharedBlobData, DeleteFontTableBlobData);
|
|
if (!mSharedBlobData) {
|
|
// The FontTableBlobData was destroyed during hb_blob_create().
|
|
// The (empty) blob is still be held in the hashtable with a strong
|
|
// reference.
|
|
return hb_blob_reference(mBlob);
|
|
}
|
|
|
|
// Tell the FontTableBlobData to remove this hash entry when destroyed.
|
|
// The hashtable does not keep a strong reference.
|
|
mSharedBlobData->ManageHashEntry(aHashtable);
|
|
return mBlob;
|
|
}
|
|
|
|
void
|
|
gfxFontEntry::FontTableHashEntry::Clear()
|
|
{
|
|
// If the FontTableBlobData is managing the hash entry, then the blob is
|
|
// not owned by this HashEntry; otherwise there is strong reference to the
|
|
// blob that must be removed.
|
|
if (mSharedBlobData) {
|
|
mSharedBlobData->ForgetHashEntry();
|
|
mSharedBlobData = nullptr;
|
|
} else if (mBlob) {
|
|
hb_blob_destroy(mBlob);
|
|
}
|
|
mBlob = nullptr;
|
|
}
|
|
|
|
// a hb_destroy_func for hb_blob_create
|
|
|
|
/* static */ void
|
|
gfxFontEntry::FontTableHashEntry::DeleteFontTableBlobData(void *aBlobData)
|
|
{
|
|
delete static_cast<FontTableBlobData*>(aBlobData);
|
|
}
|
|
|
|
hb_blob_t *
|
|
gfxFontEntry::FontTableHashEntry::GetBlob() const
|
|
{
|
|
return hb_blob_reference(mBlob);
|
|
}
|
|
|
|
bool
|
|
gfxFontEntry::GetExistingFontTable(uint32_t aTag, hb_blob_t **aBlob)
|
|
{
|
|
if (!mFontTableCache.IsInitialized()) {
|
|
// we do this here rather than on fontEntry construction
|
|
// because not all shapers will access the table cache at all
|
|
mFontTableCache.Init(10);
|
|
}
|
|
|
|
FontTableHashEntry *entry = mFontTableCache.GetEntry(aTag);
|
|
if (!entry) {
|
|
return false;
|
|
}
|
|
|
|
*aBlob = entry->GetBlob();
|
|
return true;
|
|
}
|
|
|
|
hb_blob_t *
|
|
gfxFontEntry::ShareFontTableAndGetBlob(uint32_t aTag,
|
|
FallibleTArray<uint8_t>* aBuffer)
|
|
{
|
|
if (NS_UNLIKELY(!mFontTableCache.IsInitialized())) {
|
|
// we do this here rather than on fontEntry construction
|
|
// because not all shapers will access the table cache at all
|
|
mFontTableCache.Init(10);
|
|
}
|
|
|
|
FontTableHashEntry *entry = mFontTableCache.PutEntry(aTag);
|
|
if (NS_UNLIKELY(!entry)) { // OOM
|
|
return nullptr;
|
|
}
|
|
|
|
if (!aBuffer) {
|
|
// ensure the entry is null
|
|
entry->Clear();
|
|
return nullptr;
|
|
}
|
|
|
|
return entry->ShareTableAndGetBlob(*aBuffer, &mFontTableCache);
|
|
}
|
|
|
|
#ifdef MOZ_GRAPHITE
|
|
void
|
|
gfxFontEntry::CheckForGraphiteTables()
|
|
{
|
|
AutoFallibleTArray<uint8_t,16384> buffer;
|
|
mHasGraphiteTables =
|
|
NS_SUCCEEDED(GetFontTable(TRUETYPE_TAG('S','i','l','f'), buffer));
|
|
}
|
|
#endif
|
|
|
|
/* static */ size_t
|
|
gfxFontEntry::FontTableHashEntry::SizeOfEntryExcludingThis
|
|
(FontTableHashEntry *aEntry,
|
|
nsMallocSizeOfFun aMallocSizeOf,
|
|
void* aUserArg)
|
|
{
|
|
FontListSizes *sizes = static_cast<FontListSizes*>(aUserArg);
|
|
if (aEntry->mBlob) {
|
|
sizes->mFontTableCacheSize += aMallocSizeOf(aEntry->mBlob);
|
|
}
|
|
if (aEntry->mSharedBlobData) {
|
|
sizes->mFontTableCacheSize +=
|
|
aEntry->mSharedBlobData->SizeOfIncludingThis(aMallocSizeOf);
|
|
}
|
|
|
|
// the size of the table is recorded in the FontListSizes record,
|
|
// so we return 0 here for the function result
|
|
return 0;
|
|
}
|
|
|
|
void
|
|
gfxFontEntry::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontListSizes* aSizes) const
|
|
{
|
|
aSizes->mFontListSize += mName.SizeOfExcludingThisIfUnshared(aMallocSizeOf);
|
|
|
|
// cmaps are shared so only non-shared cmaps are included here
|
|
if (mCharacterMap && mCharacterMap->mBuildOnTheFly) {
|
|
aSizes->mCharMapsSize +=
|
|
mCharacterMap->SizeOfIncludingThis(aMallocSizeOf);
|
|
}
|
|
aSizes->mFontTableCacheSize +=
|
|
mFontTableCache.SizeOfExcludingThis(
|
|
FontTableHashEntry::SizeOfEntryExcludingThis,
|
|
aMallocSizeOf, aSizes);
|
|
}
|
|
|
|
void
|
|
gfxFontEntry::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontListSizes* aSizes) const
|
|
{
|
|
aSizes->mFontListSize += aMallocSizeOf(this);
|
|
SizeOfExcludingThis(aMallocSizeOf, aSizes);
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////////
|
|
//
|
|
// class gfxFontFamily
|
|
//
|
|
//////////////////////////////////////////////////////////////////////////////
|
|
|
|
// we consider faces with mStandardFace == true to be "greater than" those with false,
|
|
// because during style matching, later entries will replace earlier ones
|
|
class FontEntryStandardFaceComparator {
|
|
public:
|
|
bool Equals(const nsRefPtr<gfxFontEntry>& a, const nsRefPtr<gfxFontEntry>& b) const {
|
|
return a->mStandardFace == b->mStandardFace;
|
|
}
|
|
bool LessThan(const nsRefPtr<gfxFontEntry>& a, const nsRefPtr<gfxFontEntry>& b) const {
|
|
return (a->mStandardFace == false && b->mStandardFace == true);
|
|
}
|
|
};
|
|
|
|
void
|
|
gfxFontFamily::SortAvailableFonts()
|
|
{
|
|
mAvailableFonts.Sort(FontEntryStandardFaceComparator());
|
|
}
|
|
|
|
bool
|
|
gfxFontFamily::HasOtherFamilyNames()
|
|
{
|
|
// need to read in other family names to determine this
|
|
if (!mOtherFamilyNamesInitialized) {
|
|
ReadOtherFamilyNames(gfxPlatformFontList::PlatformFontList()); // sets mHasOtherFamilyNames
|
|
}
|
|
return mHasOtherFamilyNames;
|
|
}
|
|
|
|
gfxFontEntry*
|
|
gfxFontFamily::FindFontForStyle(const gfxFontStyle& aFontStyle,
|
|
bool& aNeedsSyntheticBold)
|
|
{
|
|
if (!mHasStyles)
|
|
FindStyleVariations(); // collect faces for the family, if not already done
|
|
|
|
NS_ASSERTION(mAvailableFonts.Length() > 0, "font family with no faces!");
|
|
|
|
aNeedsSyntheticBold = false;
|
|
|
|
int8_t baseWeight = aFontStyle.ComputeWeight();
|
|
bool wantBold = baseWeight >= 6;
|
|
|
|
// If the family has only one face, we simply return it; no further checking needed
|
|
if (mAvailableFonts.Length() == 1) {
|
|
gfxFontEntry *fe = mAvailableFonts[0];
|
|
aNeedsSyntheticBold = wantBold && !fe->IsBold();
|
|
return fe;
|
|
}
|
|
|
|
bool wantItalic = (aFontStyle.style &
|
|
(NS_FONT_STYLE_ITALIC | NS_FONT_STYLE_OBLIQUE)) != 0;
|
|
|
|
// Most families are "simple", having just Regular/Bold/Italic/BoldItalic,
|
|
// or some subset of these. In this case, we have exactly 4 entries in mAvailableFonts,
|
|
// stored in the above order; note that some of the entries may be NULL.
|
|
// We can then pick the required entry based on whether the request is for
|
|
// bold or non-bold, italic or non-italic, without running the more complex
|
|
// matching algorithm used for larger families with many weights and/or widths.
|
|
|
|
if (mIsSimpleFamily) {
|
|
// Family has no more than the "standard" 4 faces, at fixed indexes;
|
|
// calculate which one we want.
|
|
// Note that we cannot simply return it as not all 4 faces are necessarily present.
|
|
uint8_t faceIndex = (wantItalic ? kItalicMask : 0) |
|
|
(wantBold ? kBoldMask : 0);
|
|
|
|
// if the desired style is available, return it directly
|
|
gfxFontEntry *fe = mAvailableFonts[faceIndex];
|
|
if (fe) {
|
|
// no need to set aNeedsSyntheticBold here as we matched the boldness request
|
|
return fe;
|
|
}
|
|
|
|
// order to check fallback faces in a simple family, depending on requested style
|
|
static const uint8_t simpleFallbacks[4][3] = {
|
|
{ kBoldFaceIndex, kItalicFaceIndex, kBoldItalicFaceIndex }, // fallbacks for Regular
|
|
{ kRegularFaceIndex, kBoldItalicFaceIndex, kItalicFaceIndex },// Bold
|
|
{ kBoldItalicFaceIndex, kRegularFaceIndex, kBoldFaceIndex }, // Italic
|
|
{ kItalicFaceIndex, kBoldFaceIndex, kRegularFaceIndex } // BoldItalic
|
|
};
|
|
const uint8_t *order = simpleFallbacks[faceIndex];
|
|
|
|
for (uint8_t trial = 0; trial < 3; ++trial) {
|
|
// check remaining faces in order of preference to find the first that actually exists
|
|
fe = mAvailableFonts[order[trial]];
|
|
if (fe) {
|
|
aNeedsSyntheticBold = wantBold && !fe->IsBold();
|
|
return fe;
|
|
}
|
|
}
|
|
|
|
// this can't happen unless we have totally broken the font-list manager!
|
|
NS_NOTREACHED("no face found in simple font family!");
|
|
return nullptr;
|
|
}
|
|
|
|
// This is a large/rich font family, so we do full style- and weight-matching:
|
|
// first collect a list of weights that are the best match for the requested
|
|
// font-stretch and font-style, then pick the best weight match among those
|
|
// available.
|
|
|
|
gfxFontEntry *weightList[10] = { 0 };
|
|
bool foundWeights = FindWeightsForStyle(weightList, wantItalic, aFontStyle.stretch);
|
|
if (!foundWeights) {
|
|
return nullptr;
|
|
}
|
|
|
|
// First find a match for the best weight
|
|
int8_t matchBaseWeight = 0;
|
|
int8_t i = baseWeight;
|
|
|
|
// Need to special case when normal face doesn't exist but medium does.
|
|
// In that case, use medium otherwise weights < 400
|
|
if (baseWeight == 4 && !weightList[4]) {
|
|
i = 5; // medium
|
|
}
|
|
|
|
// Loop through weights, since one exists loop will terminate
|
|
int8_t direction = (baseWeight > 5) ? 1 : -1;
|
|
for (; ; i += direction) {
|
|
if (weightList[i]) {
|
|
matchBaseWeight = i;
|
|
break;
|
|
}
|
|
|
|
// If we've reached one side without finding a font,
|
|
// start over and go the other direction until we find a match
|
|
if (i == 1 || i == 9) {
|
|
i = baseWeight;
|
|
direction = -direction;
|
|
}
|
|
}
|
|
|
|
NS_ASSERTION(matchBaseWeight != 0,
|
|
"weight mapping should always find at least one font in a family");
|
|
|
|
gfxFontEntry *matchFE = weightList[matchBaseWeight];
|
|
|
|
NS_ASSERTION(matchFE,
|
|
"weight mapping should always find at least one font in a family");
|
|
|
|
if (!matchFE->IsBold() && baseWeight >= 6)
|
|
{
|
|
aNeedsSyntheticBold = true;
|
|
}
|
|
|
|
return matchFE;
|
|
}
|
|
|
|
void
|
|
gfxFontFamily::CheckForSimpleFamily()
|
|
{
|
|
uint32_t count = mAvailableFonts.Length();
|
|
if (count > 4 || count == 0) {
|
|
return; // can't be "simple" if there are >4 faces;
|
|
// if none then the family is unusable anyway
|
|
}
|
|
|
|
if (count == 1) {
|
|
mIsSimpleFamily = true;
|
|
return;
|
|
}
|
|
|
|
int16_t firstStretch = mAvailableFonts[0]->Stretch();
|
|
|
|
gfxFontEntry *faces[4] = { 0 };
|
|
for (uint8_t i = 0; i < count; ++i) {
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
if (fe->Stretch() != firstStretch) {
|
|
return; // font-stretch doesn't match, don't treat as simple family
|
|
}
|
|
uint8_t faceIndex = (fe->IsItalic() ? kItalicMask : 0) |
|
|
(fe->Weight() >= 600 ? kBoldMask : 0);
|
|
if (faces[faceIndex]) {
|
|
return; // two faces resolve to the same slot; family isn't "simple"
|
|
}
|
|
faces[faceIndex] = fe;
|
|
}
|
|
|
|
// we have successfully slotted the available faces into the standard
|
|
// 4-face framework
|
|
mAvailableFonts.SetLength(4);
|
|
for (uint8_t i = 0; i < 4; ++i) {
|
|
if (mAvailableFonts[i].get() != faces[i]) {
|
|
mAvailableFonts[i].swap(faces[i]);
|
|
}
|
|
}
|
|
|
|
mIsSimpleFamily = true;
|
|
}
|
|
|
|
static inline uint32_t
|
|
StyleDistance(gfxFontEntry *aFontEntry,
|
|
bool anItalic, int16_t aStretch)
|
|
{
|
|
// Compute a measure of the "distance" between the requested style
|
|
// and the given fontEntry,
|
|
// considering italicness and font-stretch but not weight.
|
|
|
|
int32_t distance = 0;
|
|
if (aStretch != aFontEntry->mStretch) {
|
|
// stretch values are in the range -4 .. +4
|
|
// if aStretch is positive, we prefer more-positive values;
|
|
// if zero or negative, prefer more-negative
|
|
if (aStretch > 0) {
|
|
distance = (aFontEntry->mStretch - aStretch) * 2;
|
|
} else {
|
|
distance = (aStretch - aFontEntry->mStretch) * 2;
|
|
}
|
|
// if the computed "distance" here is negative, it means that
|
|
// aFontEntry lies in the "non-preferred" direction from aStretch,
|
|
// so we treat that as larger than any preferred-direction distance
|
|
// (max possible is 8) by adding an extra 10 to the absolute value
|
|
if (distance < 0) {
|
|
distance = -distance + 10;
|
|
}
|
|
}
|
|
if (aFontEntry->IsItalic() != anItalic) {
|
|
distance += 1;
|
|
}
|
|
return uint32_t(distance);
|
|
}
|
|
|
|
bool
|
|
gfxFontFamily::FindWeightsForStyle(gfxFontEntry* aFontsForWeights[],
|
|
bool anItalic, int16_t aStretch)
|
|
{
|
|
uint32_t foundWeights = 0;
|
|
uint32_t bestMatchDistance = 0xffffffff;
|
|
|
|
uint32_t count = mAvailableFonts.Length();
|
|
for (uint32_t i = 0; i < count; i++) {
|
|
// this is not called for "simple" families, and therefore it does not
|
|
// need to check the mAvailableFonts entries for NULL
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
uint32_t distance = StyleDistance(fe, anItalic, aStretch);
|
|
if (distance <= bestMatchDistance) {
|
|
int8_t wt = fe->mWeight / 100;
|
|
NS_ASSERTION(wt >= 1 && wt < 10, "invalid weight in fontEntry");
|
|
if (!aFontsForWeights[wt]) {
|
|
// record this as a possible candidate for weight matching
|
|
aFontsForWeights[wt] = fe;
|
|
++foundWeights;
|
|
} else {
|
|
uint32_t prevDistance =
|
|
StyleDistance(aFontsForWeights[wt], anItalic, aStretch);
|
|
if (prevDistance >= distance) {
|
|
// replacing a weight we already found,
|
|
// so don't increment foundWeights
|
|
aFontsForWeights[wt] = fe;
|
|
}
|
|
}
|
|
bestMatchDistance = distance;
|
|
}
|
|
}
|
|
|
|
NS_ASSERTION(foundWeights > 0, "Font family containing no faces?");
|
|
|
|
if (foundWeights == 1) {
|
|
// no need to cull entries if we only found one weight
|
|
return true;
|
|
}
|
|
|
|
// we might have recorded some faces that were a partial style match, but later found
|
|
// others that were closer; in this case, we need to cull the poorer matches from the
|
|
// weight list we'll return
|
|
for (uint32_t i = 0; i < 10; ++i) {
|
|
if (aFontsForWeights[i] &&
|
|
StyleDistance(aFontsForWeights[i], anItalic, aStretch) > bestMatchDistance)
|
|
{
|
|
aFontsForWeights[i] = 0;
|
|
}
|
|
}
|
|
|
|
return (foundWeights > 0);
|
|
}
|
|
|
|
|
|
void gfxFontFamily::LocalizedName(nsAString& aLocalizedName)
|
|
{
|
|
// just return the primary name; subclasses should override
|
|
aLocalizedName = mName;
|
|
}
|
|
|
|
// metric for how close a given font matches a style
|
|
static int32_t
|
|
CalcStyleMatch(gfxFontEntry *aFontEntry, const gfxFontStyle *aStyle)
|
|
{
|
|
int32_t rank = 0;
|
|
if (aStyle) {
|
|
// italics
|
|
bool wantItalic =
|
|
(aStyle->style & (NS_FONT_STYLE_ITALIC | NS_FONT_STYLE_OBLIQUE)) != 0;
|
|
if (aFontEntry->IsItalic() == wantItalic) {
|
|
rank += 10;
|
|
}
|
|
|
|
// measure of closeness of weight to the desired value
|
|
rank += 9 - abs(aFontEntry->Weight() / 100 - aStyle->ComputeWeight());
|
|
} else {
|
|
// if no font to match, prefer non-bold, non-italic fonts
|
|
if (!aFontEntry->IsItalic()) {
|
|
rank += 3;
|
|
}
|
|
if (!aFontEntry->IsBold()) {
|
|
rank += 2;
|
|
}
|
|
}
|
|
|
|
return rank;
|
|
}
|
|
|
|
#define RANK_MATCHED_CMAP 20
|
|
|
|
void
|
|
gfxFontFamily::FindFontForChar(GlobalFontMatch *aMatchData)
|
|
{
|
|
if (mFamilyCharacterMapInitialized && !TestCharacterMap(aMatchData->mCh)) {
|
|
// none of the faces in the family support the required char,
|
|
// so bail out immediately
|
|
return;
|
|
}
|
|
|
|
bool needsBold;
|
|
gfxFontStyle normal;
|
|
gfxFontEntry *fe = FindFontForStyle(
|
|
(aMatchData->mStyle == nullptr) ? *aMatchData->mStyle : normal,
|
|
needsBold);
|
|
|
|
if (fe && !fe->SkipDuringSystemFallback()) {
|
|
int32_t rank = 0;
|
|
|
|
if (fe->TestCharacterMap(aMatchData->mCh)) {
|
|
rank += RANK_MATCHED_CMAP;
|
|
aMatchData->mCount++;
|
|
#ifdef PR_LOGGING
|
|
PRLogModuleInfo *log = gfxPlatform::GetLog(eGfxLog_textrun);
|
|
|
|
if (NS_UNLIKELY(log)) {
|
|
uint32_t charRange = gfxFontUtils::CharRangeBit(aMatchData->mCh);
|
|
uint32_t unicodeRange = FindCharUnicodeRange(aMatchData->mCh);
|
|
uint32_t script = GetScriptCode(aMatchData->mCh);
|
|
PR_LOG(log, PR_LOG_DEBUG,\
|
|
("(textrun-systemfallback-fonts) char: u+%6.6x "
|
|
"char-range: %d unicode-range: %d script: %d match: [%s]\n",
|
|
aMatchData->mCh,
|
|
charRange, unicodeRange, script,
|
|
NS_ConvertUTF16toUTF8(fe->Name()).get()));
|
|
}
|
|
#endif
|
|
}
|
|
|
|
aMatchData->mCmapsTested++;
|
|
if (rank == 0) {
|
|
return;
|
|
}
|
|
|
|
// omitting from original windows code -- family name, lang group, pitch
|
|
// not available in current FontEntry implementation
|
|
rank += CalcStyleMatch(fe, aMatchData->mStyle);
|
|
|
|
// xxx - add whether AAT font with morphing info for specific lang groups
|
|
|
|
if (rank > aMatchData->mMatchRank
|
|
|| (rank == aMatchData->mMatchRank &&
|
|
Compare(fe->Name(), aMatchData->mBestMatch->Name()) > 0))
|
|
{
|
|
aMatchData->mBestMatch = fe;
|
|
aMatchData->mMatchRank = rank;
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxFontFamily::SearchAllFontsForChar(GlobalFontMatch *aMatchData)
|
|
{
|
|
uint32_t i, numFonts = mAvailableFonts.Length();
|
|
for (i = 0; i < numFonts; i++) {
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
if (fe && fe->TestCharacterMap(aMatchData->mCh)) {
|
|
int32_t rank = RANK_MATCHED_CMAP;
|
|
rank += CalcStyleMatch(fe, aMatchData->mStyle);
|
|
if (rank > aMatchData->mMatchRank
|
|
|| (rank == aMatchData->mMatchRank &&
|
|
Compare(fe->Name(), aMatchData->mBestMatch->Name()) > 0))
|
|
{
|
|
aMatchData->mBestMatch = fe;
|
|
aMatchData->mMatchRank = rank;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// returns true if other names were found, false otherwise
|
|
bool
|
|
gfxFontFamily::ReadOtherFamilyNamesForFace(gfxPlatformFontList *aPlatformFontList,
|
|
FallibleTArray<uint8_t>& aNameTable,
|
|
bool useFullName)
|
|
{
|
|
const uint8_t *nameData = aNameTable.Elements();
|
|
uint32_t dataLength = aNameTable.Length();
|
|
const gfxFontUtils::NameHeader *nameHeader =
|
|
reinterpret_cast<const gfxFontUtils::NameHeader*>(nameData);
|
|
|
|
uint32_t nameCount = nameHeader->count;
|
|
if (nameCount * sizeof(gfxFontUtils::NameRecord) > dataLength) {
|
|
NS_WARNING("invalid font (name records)");
|
|
return false;
|
|
}
|
|
|
|
const gfxFontUtils::NameRecord *nameRecord =
|
|
reinterpret_cast<const gfxFontUtils::NameRecord*>(nameData + sizeof(gfxFontUtils::NameHeader));
|
|
uint32_t stringsBase = uint32_t(nameHeader->stringOffset);
|
|
|
|
bool foundNames = false;
|
|
for (uint32_t i = 0; i < nameCount; i++, nameRecord++) {
|
|
uint32_t nameLen = nameRecord->length;
|
|
uint32_t nameOff = nameRecord->offset; // offset from base of string storage
|
|
|
|
if (stringsBase + nameOff + nameLen > dataLength) {
|
|
NS_WARNING("invalid font (name table strings)");
|
|
return false;
|
|
}
|
|
|
|
uint16_t nameID = nameRecord->nameID;
|
|
if ((useFullName && nameID == gfxFontUtils::NAME_ID_FULL) ||
|
|
(!useFullName && (nameID == gfxFontUtils::NAME_ID_FAMILY ||
|
|
nameID == gfxFontUtils::NAME_ID_PREFERRED_FAMILY))) {
|
|
nsAutoString otherFamilyName;
|
|
bool ok = gfxFontUtils::DecodeFontName(nameData + stringsBase + nameOff,
|
|
nameLen,
|
|
uint32_t(nameRecord->platformID),
|
|
uint32_t(nameRecord->encodingID),
|
|
uint32_t(nameRecord->languageID),
|
|
otherFamilyName);
|
|
// add if not same as canonical family name
|
|
if (ok && otherFamilyName != mName) {
|
|
aPlatformFontList->AddOtherFamilyName(this, otherFamilyName);
|
|
foundNames = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
return foundNames;
|
|
}
|
|
|
|
|
|
void
|
|
gfxFontFamily::ReadOtherFamilyNames(gfxPlatformFontList *aPlatformFontList)
|
|
{
|
|
if (mOtherFamilyNamesInitialized)
|
|
return;
|
|
mOtherFamilyNamesInitialized = true;
|
|
|
|
FindStyleVariations();
|
|
|
|
// read in other family names for the first face in the list
|
|
uint32_t i, numFonts = mAvailableFonts.Length();
|
|
const uint32_t kNAME = TRUETYPE_TAG('n','a','m','e');
|
|
AutoFallibleTArray<uint8_t,8192> buffer;
|
|
|
|
for (i = 0; i < numFonts; ++i) {
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
if (!fe)
|
|
continue;
|
|
|
|
if (fe->GetFontTable(kNAME, buffer) != NS_OK)
|
|
continue;
|
|
|
|
mHasOtherFamilyNames = ReadOtherFamilyNamesForFace(aPlatformFontList,
|
|
buffer);
|
|
break;
|
|
}
|
|
|
|
// read in other names for the first face in the list with the assumption
|
|
// that if extra names don't exist in that face then they don't exist in
|
|
// other faces for the same font
|
|
if (!mHasOtherFamilyNames)
|
|
return;
|
|
|
|
// read in names for all faces, needed to catch cases where fonts have
|
|
// family names for individual weights (e.g. Hiragino Kaku Gothic Pro W6)
|
|
for ( ; i < numFonts; i++) {
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
if (!fe)
|
|
continue;
|
|
|
|
if (fe->GetFontTable(kNAME, buffer) != NS_OK)
|
|
continue;
|
|
|
|
ReadOtherFamilyNamesForFace(aPlatformFontList, buffer);
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxFontFamily::ReadFaceNames(gfxPlatformFontList *aPlatformFontList,
|
|
bool aNeedFullnamePostscriptNames)
|
|
{
|
|
// if all needed names have already been read, skip
|
|
if (mOtherFamilyNamesInitialized &&
|
|
(mFaceNamesInitialized || !aNeedFullnamePostscriptNames))
|
|
return;
|
|
|
|
FindStyleVariations();
|
|
|
|
uint32_t i, numFonts = mAvailableFonts.Length();
|
|
const uint32_t kNAME = TRUETYPE_TAG('n','a','m','e');
|
|
AutoFallibleTArray<uint8_t,8192> buffer;
|
|
nsAutoString fullname, psname;
|
|
|
|
bool firstTime = true, readAllFaces = false;
|
|
for (i = 0; i < numFonts; ++i) {
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
if (!fe)
|
|
continue;
|
|
|
|
if (fe->GetFontTable(kNAME, buffer) != NS_OK)
|
|
continue;
|
|
|
|
if (aNeedFullnamePostscriptNames) {
|
|
if (gfxFontUtils::ReadCanonicalName(
|
|
buffer, gfxFontUtils::NAME_ID_FULL, fullname) == NS_OK)
|
|
{
|
|
aPlatformFontList->AddFullname(fe, fullname);
|
|
}
|
|
|
|
if (gfxFontUtils::ReadCanonicalName(
|
|
buffer, gfxFontUtils::NAME_ID_POSTSCRIPT, psname) == NS_OK)
|
|
{
|
|
aPlatformFontList->AddPostscriptName(fe, psname);
|
|
}
|
|
}
|
|
|
|
if (!mOtherFamilyNamesInitialized && (firstTime || readAllFaces)) {
|
|
bool foundOtherName = ReadOtherFamilyNamesForFace(aPlatformFontList,
|
|
buffer);
|
|
|
|
// if the first face has a different name, scan all faces, otherwise
|
|
// assume the family doesn't have other names
|
|
if (firstTime && foundOtherName) {
|
|
mHasOtherFamilyNames = true;
|
|
readAllFaces = true;
|
|
}
|
|
firstTime = false;
|
|
}
|
|
|
|
// if not reading in any more names, skip other faces
|
|
if (!readAllFaces && !aNeedFullnamePostscriptNames)
|
|
break;
|
|
}
|
|
|
|
mFaceNamesInitialized = true;
|
|
mOtherFamilyNamesInitialized = true;
|
|
}
|
|
|
|
|
|
gfxFontEntry*
|
|
gfxFontFamily::FindFont(const nsAString& aPostscriptName)
|
|
{
|
|
// find the font using a simple linear search
|
|
uint32_t numFonts = mAvailableFonts.Length();
|
|
for (uint32_t i = 0; i < numFonts; i++) {
|
|
gfxFontEntry *fe = mAvailableFonts[i].get();
|
|
if (fe && fe->Name() == aPostscriptName)
|
|
return fe;
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
void
|
|
gfxFontFamily::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontListSizes* aSizes) const
|
|
{
|
|
aSizes->mFontListSize +=
|
|
mName.SizeOfExcludingThisIfUnshared(aMallocSizeOf);
|
|
aSizes->mCharMapsSize +=
|
|
mFamilyCharacterMap.SizeOfExcludingThis(aMallocSizeOf);
|
|
|
|
aSizes->mFontListSize +=
|
|
mAvailableFonts.SizeOfExcludingThis(aMallocSizeOf);
|
|
for (uint32_t i = 0; i < mAvailableFonts.Length(); ++i) {
|
|
gfxFontEntry *fe = mAvailableFonts[i];
|
|
if (fe) {
|
|
fe->SizeOfIncludingThis(aMallocSizeOf, aSizes);
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxFontFamily::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontListSizes* aSizes) const
|
|
{
|
|
aSizes->mFontListSize += aMallocSizeOf(this);
|
|
SizeOfExcludingThis(aMallocSizeOf, aSizes);
|
|
}
|
|
|
|
/*
|
|
* gfxFontCache - global cache of gfxFont instances.
|
|
* Expires unused fonts after a short interval;
|
|
* notifies fonts to age their cached shaped-word records;
|
|
* observes memory-pressure notification and tells fonts to clear their
|
|
* shaped-word caches to free up memory.
|
|
*/
|
|
|
|
NS_IMPL_ISUPPORTS1(gfxFontCache::MemoryReporter, nsIMemoryMultiReporter)
|
|
|
|
NS_MEMORY_REPORTER_MALLOC_SIZEOF_FUN(FontCacheMallocSizeOf, "font-cache")
|
|
|
|
NS_IMETHODIMP
|
|
gfxFontCache::MemoryReporter::GetName(nsACString &aName)
|
|
{
|
|
aName.AssignLiteral("font-cache");
|
|
return NS_OK;
|
|
}
|
|
|
|
NS_IMETHODIMP
|
|
gfxFontCache::MemoryReporter::CollectReports
|
|
(nsIMemoryMultiReporterCallback* aCb,
|
|
nsISupports* aClosure)
|
|
{
|
|
FontCacheSizes sizes;
|
|
|
|
gfxFontCache::GetCache()->SizeOfIncludingThis(&FontCacheMallocSizeOf,
|
|
&sizes);
|
|
|
|
aCb->Callback(EmptyCString(),
|
|
NS_LITERAL_CSTRING("explicit/gfx/font-cache"),
|
|
nsIMemoryReporter::KIND_HEAP, nsIMemoryReporter::UNITS_BYTES,
|
|
sizes.mFontInstances,
|
|
NS_LITERAL_CSTRING("Memory used for active font instances."),
|
|
aClosure);
|
|
|
|
aCb->Callback(EmptyCString(),
|
|
NS_LITERAL_CSTRING("explicit/gfx/font-shaped-words"),
|
|
nsIMemoryReporter::KIND_HEAP, nsIMemoryReporter::UNITS_BYTES,
|
|
sizes.mShapedWords,
|
|
NS_LITERAL_CSTRING("Memory used to cache shaped glyph data."),
|
|
aClosure);
|
|
|
|
return NS_OK;
|
|
}
|
|
|
|
NS_IMETHODIMP
|
|
gfxFontCache::MemoryReporter::GetExplicitNonHeap(int64_t* aAmount)
|
|
{
|
|
// This reporter only measures heap memory.
|
|
*aAmount = 0;
|
|
return NS_OK;
|
|
}
|
|
|
|
// Observer for the memory-pressure notification, to trigger
|
|
// flushing of the shaped-word caches
|
|
class MemoryPressureObserver MOZ_FINAL : public nsIObserver,
|
|
public nsSupportsWeakReference
|
|
{
|
|
public:
|
|
NS_DECL_ISUPPORTS
|
|
NS_DECL_NSIOBSERVER
|
|
};
|
|
|
|
NS_IMPL_ISUPPORTS2(MemoryPressureObserver, nsIObserver, nsISupportsWeakReference)
|
|
|
|
NS_IMETHODIMP
|
|
MemoryPressureObserver::Observe(nsISupports *aSubject,
|
|
const char *aTopic,
|
|
const PRUnichar *someData)
|
|
{
|
|
if (!nsCRT::strcmp(aTopic, "memory-pressure")) {
|
|
gfxFontCache *fontCache = gfxFontCache::GetCache();
|
|
if (fontCache) {
|
|
fontCache->FlushShapedWordCaches();
|
|
}
|
|
}
|
|
return NS_OK;
|
|
}
|
|
|
|
nsresult
|
|
gfxFontCache::Init()
|
|
{
|
|
NS_ASSERTION(!gGlobalCache, "Where did this come from?");
|
|
gGlobalCache = new gfxFontCache();
|
|
if (!gGlobalCache) {
|
|
return NS_ERROR_OUT_OF_MEMORY;
|
|
}
|
|
NS_RegisterMemoryMultiReporter(new MemoryReporter);
|
|
return NS_OK;
|
|
}
|
|
|
|
void
|
|
gfxFontCache::Shutdown()
|
|
{
|
|
delete gGlobalCache;
|
|
gGlobalCache = nullptr;
|
|
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
printf("Textrun storage high water mark=%d\n", gTextRunStorageHighWaterMark);
|
|
printf("Total number of fonts=%d\n", gFontCount);
|
|
printf("Total glyph extents allocated=%d (size %d)\n", gGlyphExtentsCount,
|
|
int(gGlyphExtentsCount*sizeof(gfxGlyphExtents)));
|
|
printf("Total glyph extents width-storage size allocated=%d\n", gGlyphExtentsWidthsTotalSize);
|
|
printf("Number of simple glyph extents eagerly requested=%d\n", gGlyphExtentsSetupEagerSimple);
|
|
printf("Number of tight glyph extents eagerly requested=%d\n", gGlyphExtentsSetupEagerTight);
|
|
printf("Number of tight glyph extents lazily requested=%d\n", gGlyphExtentsSetupLazyTight);
|
|
printf("Number of simple glyph extent setups that fell back to tight=%d\n", gGlyphExtentsSetupFallBackToTight);
|
|
#endif
|
|
}
|
|
|
|
gfxFontCache::gfxFontCache()
|
|
: nsExpirationTracker<gfxFont,3>(FONT_TIMEOUT_SECONDS * 1000)
|
|
{
|
|
mFonts.Init();
|
|
|
|
nsCOMPtr<nsIObserverService> obs = GetObserverService();
|
|
if (obs) {
|
|
obs->AddObserver(new MemoryPressureObserver, "memory-pressure", false);
|
|
}
|
|
|
|
#if 0 // disabled due to crashiness, see bug 717175
|
|
mWordCacheExpirationTimer = do_CreateInstance("@mozilla.org/timer;1");
|
|
if (mWordCacheExpirationTimer) {
|
|
mWordCacheExpirationTimer->
|
|
InitWithFuncCallback(WordCacheExpirationTimerCallback, this,
|
|
SHAPED_WORD_TIMEOUT_SECONDS * 1000,
|
|
nsITimer::TYPE_REPEATING_SLACK);
|
|
}
|
|
#endif
|
|
}
|
|
|
|
gfxFontCache::~gfxFontCache()
|
|
{
|
|
if (mWordCacheExpirationTimer) {
|
|
mWordCacheExpirationTimer->Cancel();
|
|
mWordCacheExpirationTimer = nullptr;
|
|
}
|
|
|
|
// Expire everything that has a zero refcount, so we don't leak them.
|
|
AgeAllGenerations();
|
|
// All fonts should be gone.
|
|
NS_WARN_IF_FALSE(mFonts.Count() == 0,
|
|
"Fonts still alive while shutting down gfxFontCache");
|
|
// Note that we have to delete everything through the expiration
|
|
// tracker, since there might be fonts not in the hashtable but in
|
|
// the tracker.
|
|
}
|
|
|
|
bool
|
|
gfxFontCache::HashEntry::KeyEquals(const KeyTypePointer aKey) const
|
|
{
|
|
return aKey->mFontEntry == mFont->GetFontEntry() &&
|
|
aKey->mStyle->Equals(*mFont->GetStyle());
|
|
}
|
|
|
|
already_AddRefed<gfxFont>
|
|
gfxFontCache::Lookup(const gfxFontEntry *aFontEntry,
|
|
const gfxFontStyle *aStyle)
|
|
{
|
|
Key key(aFontEntry, aStyle);
|
|
HashEntry *entry = mFonts.GetEntry(key);
|
|
|
|
Telemetry::Accumulate(Telemetry::FONT_CACHE_HIT, entry != nullptr);
|
|
if (!entry)
|
|
return nullptr;
|
|
|
|
gfxFont *font = entry->mFont;
|
|
NS_ADDREF(font);
|
|
return font;
|
|
}
|
|
|
|
void
|
|
gfxFontCache::AddNew(gfxFont *aFont)
|
|
{
|
|
Key key(aFont->GetFontEntry(), aFont->GetStyle());
|
|
HashEntry *entry = mFonts.PutEntry(key);
|
|
if (!entry)
|
|
return;
|
|
gfxFont *oldFont = entry->mFont;
|
|
entry->mFont = aFont;
|
|
// If someone's asked us to replace an existing font entry, then that's a
|
|
// bit weird, but let it happen, and expire the old font if it's not used.
|
|
if (oldFont && oldFont->GetExpirationState()->IsTracked()) {
|
|
// if oldFont == aFont, recount should be > 0,
|
|
// so we shouldn't be here.
|
|
NS_ASSERTION(aFont != oldFont, "new font is tracked for expiry!");
|
|
NotifyExpired(oldFont);
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxFontCache::NotifyReleased(gfxFont *aFont)
|
|
{
|
|
nsresult rv = AddObject(aFont);
|
|
if (NS_FAILED(rv)) {
|
|
// We couldn't track it for some reason. Kill it now.
|
|
DestroyFont(aFont);
|
|
}
|
|
// Note that we might have fonts that aren't in the hashtable, perhaps because
|
|
// of OOM adding to the hashtable or because someone did an AddNew where
|
|
// we already had a font. These fonts are added to the expiration tracker
|
|
// anyway, even though Lookup can't resurrect them. Eventually they will
|
|
// expire and be deleted.
|
|
}
|
|
|
|
void
|
|
gfxFontCache::NotifyExpired(gfxFont *aFont)
|
|
{
|
|
aFont->ClearCachedWords();
|
|
RemoveObject(aFont);
|
|
DestroyFont(aFont);
|
|
}
|
|
|
|
void
|
|
gfxFontCache::DestroyFont(gfxFont *aFont)
|
|
{
|
|
Key key(aFont->GetFontEntry(), aFont->GetStyle());
|
|
HashEntry *entry = mFonts.GetEntry(key);
|
|
if (entry && entry->mFont == aFont)
|
|
mFonts.RemoveEntry(key);
|
|
NS_ASSERTION(aFont->GetRefCount() == 0,
|
|
"Destroying with non-zero ref count!");
|
|
delete aFont;
|
|
}
|
|
|
|
/*static*/
|
|
PLDHashOperator
|
|
gfxFontCache::AgeCachedWordsForFont(HashEntry* aHashEntry, void* aUserData)
|
|
{
|
|
aHashEntry->mFont->AgeCachedWords();
|
|
return PL_DHASH_NEXT;
|
|
}
|
|
|
|
/*static*/
|
|
void
|
|
gfxFontCache::WordCacheExpirationTimerCallback(nsITimer* aTimer, void* aCache)
|
|
{
|
|
gfxFontCache* cache = static_cast<gfxFontCache*>(aCache);
|
|
cache->mFonts.EnumerateEntries(AgeCachedWordsForFont, nullptr);
|
|
}
|
|
|
|
/*static*/
|
|
PLDHashOperator
|
|
gfxFontCache::ClearCachedWordsForFont(HashEntry* aHashEntry, void* aUserData)
|
|
{
|
|
aHashEntry->mFont->ClearCachedWords();
|
|
return PL_DHASH_NEXT;
|
|
}
|
|
|
|
/*static*/
|
|
size_t
|
|
gfxFontCache::SizeOfFontEntryExcludingThis(HashEntry* aHashEntry,
|
|
nsMallocSizeOfFun aMallocSizeOf,
|
|
void* aUserArg)
|
|
{
|
|
HashEntry *entry = static_cast<HashEntry*>(aHashEntry);
|
|
FontCacheSizes *sizes = static_cast<FontCacheSizes*>(aUserArg);
|
|
entry->mFont->SizeOfExcludingThis(aMallocSizeOf, sizes);
|
|
|
|
// The font records its size in the |sizes| parameter, so we return zero
|
|
// here to the hashtable enumerator.
|
|
return 0;
|
|
}
|
|
|
|
void
|
|
gfxFontCache::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontCacheSizes* aSizes) const
|
|
{
|
|
// TODO: add the overhead of the expiration tracker (generation arrays)
|
|
|
|
mFonts.SizeOfExcludingThis(SizeOfFontEntryExcludingThis,
|
|
aMallocSizeOf, aSizes);
|
|
}
|
|
|
|
void
|
|
gfxFontCache::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontCacheSizes* aSizes) const
|
|
{
|
|
aSizes->mFontInstances += aMallocSizeOf(this);
|
|
SizeOfExcludingThis(aMallocSizeOf, aSizes);
|
|
}
|
|
|
|
/* static */ bool
|
|
gfxFontShaper::MergeFontFeatures(
|
|
const nsTArray<gfxFontFeature>& aStyleRuleFeatures,
|
|
const nsTArray<gfxFontFeature>& aFontFeatures,
|
|
bool aDisableLigatures,
|
|
nsDataHashtable<nsUint32HashKey,uint32_t>& aMergedFeatures)
|
|
{
|
|
// bail immediately if nothing to do
|
|
if (aStyleRuleFeatures.IsEmpty() &&
|
|
aFontFeatures.IsEmpty() &&
|
|
!aDisableLigatures) {
|
|
return false;
|
|
}
|
|
|
|
aMergedFeatures.Init();
|
|
|
|
// Ligature features are enabled by default in the generic shaper,
|
|
// so we explicitly turn them off if necessary (for letter-spacing)
|
|
if (aDisableLigatures) {
|
|
aMergedFeatures.Put(HB_TAG('l','i','g','a'), 0);
|
|
aMergedFeatures.Put(HB_TAG('c','l','i','g'), 0);
|
|
}
|
|
|
|
// add feature values from font
|
|
uint32_t i, count;
|
|
|
|
count = aFontFeatures.Length();
|
|
for (i = 0; i < count; i++) {
|
|
const gfxFontFeature& feature = aFontFeatures.ElementAt(i);
|
|
aMergedFeatures.Put(feature.mTag, feature.mValue);
|
|
}
|
|
|
|
// add feature values from style rules
|
|
count = aStyleRuleFeatures.Length();
|
|
for (i = 0; i < count; i++) {
|
|
const gfxFontFeature& feature = aStyleRuleFeatures.ElementAt(i);
|
|
aMergedFeatures.Put(feature.mTag, feature.mValue);
|
|
}
|
|
|
|
return aMergedFeatures.Count() != 0;
|
|
}
|
|
|
|
void
|
|
gfxFont::RunMetrics::CombineWith(const RunMetrics& aOther, bool aOtherIsOnLeft)
|
|
{
|
|
mAscent = NS_MAX(mAscent, aOther.mAscent);
|
|
mDescent = NS_MAX(mDescent, aOther.mDescent);
|
|
if (aOtherIsOnLeft) {
|
|
mBoundingBox =
|
|
(mBoundingBox + gfxPoint(aOther.mAdvanceWidth, 0)).Union(aOther.mBoundingBox);
|
|
} else {
|
|
mBoundingBox =
|
|
mBoundingBox.Union(aOther.mBoundingBox + gfxPoint(mAdvanceWidth, 0));
|
|
}
|
|
mAdvanceWidth += aOther.mAdvanceWidth;
|
|
}
|
|
|
|
gfxFont::gfxFont(gfxFontEntry *aFontEntry, const gfxFontStyle *aFontStyle,
|
|
AntialiasOption anAAOption, cairo_scaled_font_t *aScaledFont) :
|
|
mScaledFont(aScaledFont),
|
|
mFontEntry(aFontEntry), mIsValid(true),
|
|
mApplySyntheticBold(false),
|
|
mStyle(*aFontStyle),
|
|
mAdjustedSize(0.0),
|
|
mFUnitsConvFactor(0.0f),
|
|
mAntialiasOption(anAAOption)
|
|
{
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
++gFontCount;
|
|
#endif
|
|
}
|
|
|
|
gfxFont::~gfxFont()
|
|
{
|
|
uint32_t i, count = mGlyphExtentsArray.Length();
|
|
// We destroy the contents of mGlyphExtentsArray explicitly instead of
|
|
// using nsAutoPtr because VC++ can't deal with nsTArrays of nsAutoPtrs
|
|
// of classes that lack a proper copy constructor
|
|
for (i = 0; i < count; ++i) {
|
|
delete mGlyphExtentsArray[i];
|
|
}
|
|
}
|
|
|
|
/*static*/
|
|
PLDHashOperator
|
|
gfxFont::AgeCacheEntry(CacheHashEntry *aEntry, void *aUserData)
|
|
{
|
|
if (!aEntry->mShapedWord) {
|
|
NS_ASSERTION(aEntry->mShapedWord, "cache entry has no gfxShapedWord!");
|
|
return PL_DHASH_REMOVE;
|
|
}
|
|
if (aEntry->mShapedWord->IncrementAge() == kShapedWordCacheMaxAge) {
|
|
return PL_DHASH_REMOVE;
|
|
}
|
|
return PL_DHASH_NEXT;
|
|
}
|
|
|
|
hb_blob_t *
|
|
gfxFont::GetFontTable(uint32_t aTag) {
|
|
hb_blob_t *blob;
|
|
if (mFontEntry->GetExistingFontTable(aTag, &blob))
|
|
return blob;
|
|
|
|
FallibleTArray<uint8_t> buffer;
|
|
bool haveTable = NS_SUCCEEDED(mFontEntry->GetFontTable(aTag, buffer));
|
|
|
|
return mFontEntry->ShareFontTableAndGetBlob(aTag,
|
|
haveTable ? &buffer : nullptr);
|
|
}
|
|
|
|
/**
|
|
* A helper function in case we need to do any rounding or other
|
|
* processing here.
|
|
*/
|
|
#define ToDeviceUnits(aAppUnits, aDevUnitsPerAppUnit) \
|
|
(double(aAppUnits)*double(aDevUnitsPerAppUnit))
|
|
|
|
struct GlyphBuffer {
|
|
#define GLYPH_BUFFER_SIZE (2048/sizeof(cairo_glyph_t))
|
|
cairo_glyph_t mGlyphBuffer[GLYPH_BUFFER_SIZE];
|
|
unsigned int mNumGlyphs;
|
|
|
|
GlyphBuffer()
|
|
: mNumGlyphs(0) { }
|
|
|
|
cairo_glyph_t *AppendGlyph() {
|
|
return &mGlyphBuffer[mNumGlyphs++];
|
|
}
|
|
|
|
void Flush(cairo_t *aCR, gfxFont::DrawMode aDrawMode, bool aReverse,
|
|
gfxTextObjectPaint *aObjectPaint,
|
|
const gfxMatrix& aGlobalMatrix, bool aFinish = false) {
|
|
// Ensure there's enough room for a glyph to be added to the buffer
|
|
// and we actually have glyphs to draw
|
|
if ((!aFinish && mNumGlyphs < GLYPH_BUFFER_SIZE) || !mNumGlyphs) {
|
|
return;
|
|
}
|
|
|
|
if (aReverse) {
|
|
for (uint32_t i = 0; i < mNumGlyphs/2; ++i) {
|
|
cairo_glyph_t tmp = mGlyphBuffer[i];
|
|
mGlyphBuffer[i] = mGlyphBuffer[mNumGlyphs - 1 - i];
|
|
mGlyphBuffer[mNumGlyphs - 1 - i] = tmp;
|
|
}
|
|
}
|
|
|
|
if (aDrawMode == gfxFont::GLYPH_PATH) {
|
|
cairo_glyph_path(aCR, mGlyphBuffer, mNumGlyphs);
|
|
} else {
|
|
if (aDrawMode & gfxFont::GLYPH_FILL) {
|
|
SAMPLE_LABEL("GlyphBuffer", "cairo_show_glyphs");
|
|
nsRefPtr<gfxPattern> pattern;
|
|
if (aObjectPaint &&
|
|
!!(pattern = aObjectPaint->GetFillPattern(aGlobalMatrix))) {
|
|
cairo_save(aCR);
|
|
cairo_set_source(aCR, pattern->CairoPattern());
|
|
}
|
|
|
|
cairo_show_glyphs(aCR, mGlyphBuffer, mNumGlyphs);
|
|
|
|
if (pattern) {
|
|
cairo_restore(aCR);
|
|
}
|
|
}
|
|
|
|
if (aDrawMode & gfxFont::GLYPH_STROKE) {
|
|
nsRefPtr<gfxPattern> pattern;
|
|
if (aObjectPaint &&
|
|
!!(pattern = aObjectPaint->GetStrokePattern(aGlobalMatrix))) {
|
|
cairo_save(aCR);
|
|
cairo_set_source(aCR, pattern->CairoPattern());
|
|
}
|
|
|
|
cairo_new_path(aCR);
|
|
cairo_glyph_path(aCR, mGlyphBuffer, mNumGlyphs);
|
|
cairo_stroke(aCR);
|
|
|
|
if (pattern) {
|
|
cairo_restore(aCR);
|
|
}
|
|
}
|
|
}
|
|
|
|
mNumGlyphs = 0;
|
|
}
|
|
#undef GLYPH_BUFFER_SIZE
|
|
};
|
|
|
|
struct GlyphBufferAzure {
|
|
#define GLYPH_BUFFER_SIZE (2048/sizeof(Glyph))
|
|
Glyph mGlyphBuffer[GLYPH_BUFFER_SIZE];
|
|
unsigned int mNumGlyphs;
|
|
|
|
GlyphBufferAzure()
|
|
: mNumGlyphs(0) { }
|
|
|
|
Glyph *AppendGlyph() {
|
|
return &mGlyphBuffer[mNumGlyphs++];
|
|
}
|
|
|
|
void Flush(DrawTarget *aDT, gfxTextObjectPaint *aObjectPaint, ScaledFont *aFont,
|
|
gfxFont::DrawMode aDrawMode, bool aReverse, const GlyphRenderingOptions *aOptions,
|
|
gfxContext *aThebesContext, const Matrix *aInvFontMatrix, bool aFinish = false)
|
|
{
|
|
// Ensure there's enough room for a glyph to be added to the buffer
|
|
if (!aFinish && mNumGlyphs < GLYPH_BUFFER_SIZE || !mNumGlyphs) {
|
|
return;
|
|
}
|
|
|
|
if (aReverse) {
|
|
Glyph *begin = &mGlyphBuffer[0];
|
|
Glyph *end = &mGlyphBuffer[mNumGlyphs];
|
|
std::reverse(begin, end);
|
|
}
|
|
|
|
gfx::GlyphBuffer buf;
|
|
buf.mGlyphs = mGlyphBuffer;
|
|
buf.mNumGlyphs = mNumGlyphs;
|
|
|
|
gfxContext::AzureState state = aThebesContext->CurrentState();
|
|
if (aDrawMode & gfxFont::GLYPH_FILL) {
|
|
if (state.pattern || aObjectPaint) {
|
|
Pattern *pat;
|
|
|
|
nsRefPtr<gfxPattern> fillPattern;
|
|
if (!aObjectPaint ||
|
|
!(fillPattern = aObjectPaint->GetFillPattern(aThebesContext->CurrentMatrix()))) {
|
|
pat = state.pattern->GetPattern(aDT, state.patternTransformChanged ? &state.patternTransform : nullptr);
|
|
} else {
|
|
pat = fillPattern->GetPattern(aDT);
|
|
}
|
|
|
|
Matrix saved;
|
|
Matrix *mat = nullptr;
|
|
if (aInvFontMatrix) {
|
|
// The brush matrix needs to be multiplied with the inverted matrix
|
|
// as well, to move the brush into the space of the glyphs. Before
|
|
// the render target transformation
|
|
|
|
// This relies on the returned Pattern not to be reused by
|
|
// others, but regenerated on GetPattern calls. This is true!
|
|
if (pat->GetType() == PATTERN_LINEAR_GRADIENT) {
|
|
mat = &static_cast<LinearGradientPattern*>(pat)->mMatrix;
|
|
} else if (pat->GetType() == PATTERN_RADIAL_GRADIENT) {
|
|
mat = &static_cast<RadialGradientPattern*>(pat)->mMatrix;
|
|
} else if (pat->GetType() == PATTERN_SURFACE) {
|
|
mat = &static_cast<SurfacePattern*>(pat)->mMatrix;
|
|
}
|
|
|
|
if (mat) {
|
|
saved = *mat;
|
|
*mat = (*mat) * (*aInvFontMatrix);
|
|
}
|
|
}
|
|
|
|
aDT->FillGlyphs(aFont, buf, *pat,
|
|
DrawOptions(), aOptions);
|
|
|
|
if (mat) {
|
|
*mat = saved;
|
|
}
|
|
} else if (state.sourceSurface) {
|
|
aDT->FillGlyphs(aFont, buf, SurfacePattern(state.sourceSurface,
|
|
EXTEND_CLAMP,
|
|
state.surfTransform),
|
|
DrawOptions(), aOptions);
|
|
} else {
|
|
aDT->FillGlyphs(aFont, buf, ColorPattern(state.color),
|
|
DrawOptions(), aOptions);
|
|
}
|
|
}
|
|
if (aDrawMode & gfxFont::GLYPH_PATH) {
|
|
aThebesContext->EnsurePathBuilder();
|
|
aFont->CopyGlyphsToBuilder(buf, aThebesContext->mPathBuilder);
|
|
}
|
|
if (aDrawMode & gfxFont::GLYPH_STROKE) {
|
|
RefPtr<Path> path = aFont->GetPathForGlyphs(buf, aDT);
|
|
if (aObjectPaint) {
|
|
nsRefPtr<gfxPattern> strokePattern =
|
|
aObjectPaint->GetStrokePattern(aThebesContext->CurrentMatrix());
|
|
if (strokePattern) {
|
|
aDT->Stroke(path, *strokePattern->GetPattern(aDT), state.strokeOptions);
|
|
}
|
|
}
|
|
}
|
|
|
|
mNumGlyphs = 0;
|
|
}
|
|
#undef GLYPH_BUFFER_SIZE
|
|
};
|
|
|
|
// Bug 674909. When synthetic bolding text by drawing twice, need to
|
|
// render using a pixel offset in device pixels, otherwise text
|
|
// doesn't appear bolded, it appears as if a bad text shadow exists
|
|
// when a non-identity transform exists. Use an offset factor so that
|
|
// the second draw occurs at a constant offset in device pixels.
|
|
|
|
double
|
|
gfxFont::CalcXScale(gfxContext *aContext)
|
|
{
|
|
// determine magnitude of a 1px x offset in device space
|
|
gfxSize t = aContext->UserToDevice(gfxSize(1.0, 0.0));
|
|
if (t.width == 1.0 && t.height == 0.0) {
|
|
// short-circuit the most common case to avoid sqrt() and division
|
|
return 1.0;
|
|
}
|
|
|
|
double m = sqrt(t.width * t.width + t.height * t.height);
|
|
|
|
NS_ASSERTION(m != 0.0, "degenerate transform while synthetic bolding");
|
|
if (m == 0.0) {
|
|
return 0.0; // effectively disables offset
|
|
}
|
|
|
|
// scale factor so that offsets are 1px in device pixels
|
|
return 1.0 / m;
|
|
}
|
|
|
|
void
|
|
gfxFont::Draw(gfxTextRun *aTextRun, uint32_t aStart, uint32_t aEnd,
|
|
gfxContext *aContext, DrawMode aDrawMode, gfxPoint *aPt,
|
|
Spacing *aSpacing, gfxTextObjectPaint *aObjectPaint)
|
|
{
|
|
NS_ASSERTION(aDrawMode <= gfxFont::GLYPH_PATH, "GLYPH_PATH cannot be used with GLYPH_FILL or GLYPH_STROKE");
|
|
|
|
if (aStart >= aEnd)
|
|
return;
|
|
|
|
const gfxTextRun::CompressedGlyph *charGlyphs = aTextRun->GetCharacterGlyphs();
|
|
const uint32_t appUnitsPerDevUnit = aTextRun->GetAppUnitsPerDevUnit();
|
|
const double devUnitsPerAppUnit = 1.0/double(appUnitsPerDevUnit);
|
|
bool isRTL = aTextRun->IsRightToLeft();
|
|
double direction = aTextRun->GetDirection();
|
|
gfxMatrix globalMatrix = aContext->CurrentMatrix();
|
|
|
|
bool haveSVGGlyphs = GetFontEntry()->TryGetSVGData();
|
|
nsAutoPtr<gfxTextObjectPaint> objectPaint;
|
|
if (haveSVGGlyphs && !aObjectPaint) {
|
|
// If no pattern is specified for fill, use the current pattern
|
|
NS_ASSERTION((aDrawMode & GLYPH_STROKE) == 0, "no pattern supplied for stroking text");
|
|
nsRefPtr<gfxPattern> fillPattern = aContext->GetPattern();
|
|
objectPaint = new SimpleTextObjectPaint(fillPattern, nullptr,
|
|
aContext->CurrentMatrix());
|
|
aObjectPaint = objectPaint;
|
|
}
|
|
|
|
// synthetic-bold strikes are each offset one device pixel in run direction
|
|
// (these values are only needed if IsSyntheticBold() is true)
|
|
double synBoldOnePixelOffset = 0;
|
|
int32_t strikes = 0;
|
|
if (IsSyntheticBold()) {
|
|
double xscale = CalcXScale(aContext);
|
|
synBoldOnePixelOffset = direction * xscale;
|
|
// use as many strikes as needed for the the increased advance
|
|
strikes = NS_lroundf(GetSyntheticBoldOffset() / xscale);
|
|
}
|
|
|
|
uint32_t i;
|
|
// Current position in appunits
|
|
double x = aPt->x;
|
|
double y = aPt->y;
|
|
|
|
cairo_t *cr = aContext->GetCairo();
|
|
RefPtr<DrawTarget> dt = aContext->GetDrawTarget();
|
|
|
|
if (aContext->IsCairo()) {
|
|
bool success = SetupCairoFont(aContext);
|
|
if (NS_UNLIKELY(!success))
|
|
return;
|
|
|
|
::GlyphBuffer glyphs;
|
|
cairo_glyph_t *glyph;
|
|
|
|
if (aSpacing) {
|
|
x += direction*aSpacing[0].mBefore;
|
|
}
|
|
for (i = aStart; i < aEnd; ++i) {
|
|
const gfxTextRun::CompressedGlyph *glyphData = &charGlyphs[i];
|
|
if (glyphData->IsSimpleGlyph()) {
|
|
double advance = glyphData->GetSimpleAdvance();
|
|
double glyphX;
|
|
if (isRTL) {
|
|
x -= advance;
|
|
glyphX = x;
|
|
} else {
|
|
glyphX = x;
|
|
x += advance;
|
|
}
|
|
|
|
if (haveSVGGlyphs) {
|
|
gfxPoint point(ToDeviceUnits(glyphX, devUnitsPerAppUnit),
|
|
ToDeviceUnits(y, devUnitsPerAppUnit));
|
|
if (RenderSVGGlyph(aContext, point, aDrawMode,
|
|
glyphData->GetSimpleGlyph(),
|
|
aObjectPaint)) {
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// Perhaps we should put a scale in the cairo context instead of
|
|
// doing this scaling here...
|
|
// Multiplying by the reciprocal may introduce tiny error here,
|
|
// but we assume cairo is going to round coordinates at some stage
|
|
// and this is faster
|
|
glyph = glyphs.AppendGlyph();
|
|
glyph->index = glyphData->GetSimpleGlyph();
|
|
glyph->x = ToDeviceUnits(glyphX, devUnitsPerAppUnit);
|
|
glyph->y = ToDeviceUnits(y, devUnitsPerAppUnit);
|
|
glyphs.Flush(cr, aDrawMode, isRTL, aObjectPaint, globalMatrix);
|
|
|
|
// synthetic bolding by multi-striking with 1-pixel offsets
|
|
// at least once, more if there's room (large font sizes)
|
|
if (IsSyntheticBold()) {
|
|
double strikeOffset = synBoldOnePixelOffset;
|
|
int32_t strikeCount = strikes;
|
|
do {
|
|
cairo_glyph_t *doubleglyph;
|
|
doubleglyph = glyphs.AppendGlyph();
|
|
doubleglyph->index = glyph->index;
|
|
doubleglyph->x =
|
|
ToDeviceUnits(glyphX + strikeOffset * appUnitsPerDevUnit,
|
|
devUnitsPerAppUnit);
|
|
doubleglyph->y = glyph->y;
|
|
strikeOffset += synBoldOnePixelOffset;
|
|
glyphs.Flush(cr, aDrawMode, isRTL, aObjectPaint, globalMatrix);
|
|
} while (--strikeCount > 0);
|
|
}
|
|
} else {
|
|
uint32_t glyphCount = glyphData->GetGlyphCount();
|
|
if (glyphCount > 0) {
|
|
const gfxTextRun::DetailedGlyph *details =
|
|
aTextRun->GetDetailedGlyphs(i);
|
|
NS_ASSERTION(details, "detailedGlyph should not be missing!");
|
|
for (uint32_t j = 0; j < glyphCount; ++j, ++details) {
|
|
double advance = details->mAdvance;
|
|
if (glyphData->IsMissing()) {
|
|
// default ignorable characters will have zero advance width.
|
|
// we don't have to draw the hexbox for them
|
|
if (aDrawMode != gfxFont::GLYPH_PATH && advance > 0) {
|
|
double glyphX = x;
|
|
if (isRTL) {
|
|
glyphX -= advance;
|
|
}
|
|
gfxPoint pt(ToDeviceUnits(glyphX, devUnitsPerAppUnit),
|
|
ToDeviceUnits(y, devUnitsPerAppUnit));
|
|
gfxFloat advanceDevUnits = ToDeviceUnits(advance, devUnitsPerAppUnit);
|
|
gfxFloat height = GetMetrics().maxAscent;
|
|
gfxRect glyphRect(pt.x, pt.y - height, advanceDevUnits, height);
|
|
gfxFontMissingGlyphs::DrawMissingGlyph(aContext,
|
|
glyphRect,
|
|
details->mGlyphID);
|
|
}
|
|
} else {
|
|
double glyphX = x + details->mXOffset;
|
|
if (isRTL) {
|
|
glyphX -= advance;
|
|
}
|
|
|
|
gfxPoint point(ToDeviceUnits(glyphX, devUnitsPerAppUnit),
|
|
ToDeviceUnits(y, devUnitsPerAppUnit));
|
|
|
|
if (!haveSVGGlyphs ||
|
|
!RenderSVGGlyph(aContext, point, aDrawMode,
|
|
details->mGlyphID, aObjectPaint)) {
|
|
glyph = glyphs.AppendGlyph();
|
|
glyph->index = details->mGlyphID;
|
|
glyph->x = ToDeviceUnits(glyphX, devUnitsPerAppUnit);
|
|
glyph->y = ToDeviceUnits(y + details->mYOffset, devUnitsPerAppUnit);
|
|
glyphs.Flush(cr, aDrawMode, isRTL, aObjectPaint, globalMatrix);
|
|
|
|
if (IsSyntheticBold()) {
|
|
double strikeOffset = synBoldOnePixelOffset;
|
|
int32_t strikeCount = strikes;
|
|
do {
|
|
cairo_glyph_t *doubleglyph;
|
|
doubleglyph = glyphs.AppendGlyph();
|
|
doubleglyph->index = glyph->index;
|
|
doubleglyph->x =
|
|
ToDeviceUnits(glyphX + strikeOffset *
|
|
appUnitsPerDevUnit,
|
|
devUnitsPerAppUnit);
|
|
doubleglyph->y = glyph->y;
|
|
strikeOffset += synBoldOnePixelOffset;
|
|
glyphs.Flush(cr, aDrawMode, isRTL, aObjectPaint, globalMatrix);
|
|
} while (--strikeCount > 0);
|
|
}
|
|
}
|
|
}
|
|
x += direction*advance;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (aSpacing) {
|
|
double space = aSpacing[i - aStart].mAfter;
|
|
if (i + 1 < aEnd) {
|
|
space += aSpacing[i + 1 - aStart].mBefore;
|
|
}
|
|
x += direction*space;
|
|
}
|
|
}
|
|
|
|
if (gfxFontTestStore::CurrentStore()) {
|
|
/* This assumes that the tests won't have anything that results
|
|
* in more than GLYPH_BUFFER_SIZE glyphs. Do this before we
|
|
* flush, since that'll blow away the num_glyphs.
|
|
*/
|
|
gfxFontTestStore::CurrentStore()->AddItem(GetName(),
|
|
glyphs.mGlyphBuffer,
|
|
glyphs.mNumGlyphs);
|
|
}
|
|
|
|
// draw any remaining glyphs
|
|
glyphs.Flush(cr, aDrawMode, isRTL, aObjectPaint, globalMatrix, true);
|
|
|
|
} else {
|
|
RefPtr<ScaledFont> scaledFont = GetScaledFont(dt);
|
|
|
|
if (!scaledFont) {
|
|
return;
|
|
}
|
|
|
|
bool oldSubpixelAA = dt->GetPermitSubpixelAA();
|
|
|
|
if (!AllowSubpixelAA()) {
|
|
dt->SetPermitSubpixelAA(false);
|
|
}
|
|
|
|
GlyphBufferAzure glyphs;
|
|
Glyph *glyph;
|
|
|
|
Matrix mat, matInv;
|
|
Matrix oldMat = dt->GetTransform();
|
|
|
|
// This is NULL when we have inverse-transformed glyphs and we need to
|
|
// transform the Brush inside flush.
|
|
Matrix *passedInvMatrix = nullptr;
|
|
|
|
RefPtr<GlyphRenderingOptions> renderingOptions =
|
|
GetGlyphRenderingOptions();
|
|
|
|
if (mScaledFont) {
|
|
cairo_matrix_t matrix;
|
|
cairo_scaled_font_get_font_matrix(mScaledFont, &matrix);
|
|
if (matrix.xy != 0) {
|
|
// If this matrix applies a skew, which can happen when drawing
|
|
// oblique fonts, we will set the DrawTarget matrix to apply the
|
|
// skew. We'll need to move the glyphs by the inverse of the skew to
|
|
// get the glyphs positioned correctly in the new device space
|
|
// though, since the font matrix should only be applied to drawing
|
|
// the glyphs, and not to their position.
|
|
mat = ToMatrix(*reinterpret_cast<gfxMatrix*>(&matrix));
|
|
|
|
mat._11 = mat._22 = 1.0;
|
|
mat._21 /= mAdjustedSize;
|
|
|
|
dt->SetTransform(mat * oldMat);
|
|
|
|
matInv = mat;
|
|
matInv.Invert();
|
|
|
|
passedInvMatrix = &matInv;
|
|
}
|
|
}
|
|
|
|
if (aSpacing) {
|
|
x += direction*aSpacing[0].mBefore;
|
|
}
|
|
for (i = aStart; i < aEnd; ++i) {
|
|
const gfxTextRun::CompressedGlyph *glyphData = &charGlyphs[i];
|
|
if (glyphData->IsSimpleGlyph()) {
|
|
double advance = glyphData->GetSimpleAdvance();
|
|
double glyphX;
|
|
if (isRTL) {
|
|
x -= advance;
|
|
glyphX = x;
|
|
} else {
|
|
glyphX = x;
|
|
x += advance;
|
|
}
|
|
|
|
if (haveSVGGlyphs) {
|
|
gfxPoint point(ToDeviceUnits(glyphX, devUnitsPerAppUnit),
|
|
ToDeviceUnits(y, devUnitsPerAppUnit));
|
|
if (RenderSVGGlyph(aContext, point, aDrawMode,
|
|
glyphData->GetSimpleGlyph(),
|
|
aObjectPaint)) {
|
|
continue;
|
|
}
|
|
}
|
|
|
|
// Perhaps we should put a scale in the cairo context instead of
|
|
// doing this scaling here...
|
|
// Multiplying by the reciprocal may introduce tiny error here,
|
|
// but we assume cairo is going to round coordinates at some stage
|
|
// and this is faster
|
|
glyph = glyphs.AppendGlyph();
|
|
glyph->mIndex = glyphData->GetSimpleGlyph();
|
|
glyph->mPosition.x = ToDeviceUnits(glyphX, devUnitsPerAppUnit);
|
|
glyph->mPosition.y = ToDeviceUnits(y, devUnitsPerAppUnit);
|
|
glyph->mPosition = matInv * glyph->mPosition;
|
|
glyphs.Flush(dt, aObjectPaint, scaledFont,
|
|
aDrawMode, isRTL, renderingOptions,
|
|
aContext, passedInvMatrix);
|
|
|
|
// synthetic bolding by multi-striking with 1-pixel offsets
|
|
// at least once, more if there's room (large font sizes)
|
|
if (IsSyntheticBold()) {
|
|
double strikeOffset = synBoldOnePixelOffset;
|
|
int32_t strikeCount = strikes;
|
|
do {
|
|
Glyph *doubleglyph;
|
|
doubleglyph = glyphs.AppendGlyph();
|
|
doubleglyph->mIndex = glyph->mIndex;
|
|
doubleglyph->mPosition.x =
|
|
ToDeviceUnits(glyphX + strikeOffset * appUnitsPerDevUnit,
|
|
devUnitsPerAppUnit);
|
|
doubleglyph->mPosition.y = glyph->mPosition.y;
|
|
doubleglyph->mPosition = matInv * doubleglyph->mPosition;
|
|
strikeOffset += synBoldOnePixelOffset;
|
|
glyphs.Flush(dt, aObjectPaint, scaledFont,
|
|
aDrawMode, isRTL, renderingOptions,
|
|
aContext, passedInvMatrix);
|
|
} while (--strikeCount > 0);
|
|
}
|
|
} else {
|
|
uint32_t glyphCount = glyphData->GetGlyphCount();
|
|
if (glyphCount > 0) {
|
|
const gfxTextRun::DetailedGlyph *details =
|
|
aTextRun->GetDetailedGlyphs(i);
|
|
NS_ASSERTION(details, "detailedGlyph should not be missing!");
|
|
for (uint32_t j = 0; j < glyphCount; ++j, ++details) {
|
|
double advance = details->mAdvance;
|
|
if (glyphData->IsMissing()) {
|
|
// default ignorable characters will have zero advance width.
|
|
// we don't have to draw the hexbox for them
|
|
if (aDrawMode != gfxFont::GLYPH_PATH && advance > 0) {
|
|
double glyphX = x;
|
|
if (isRTL) {
|
|
glyphX -= advance;
|
|
}
|
|
gfxPoint pt(ToDeviceUnits(glyphX, devUnitsPerAppUnit),
|
|
ToDeviceUnits(y, devUnitsPerAppUnit));
|
|
gfxFloat advanceDevUnits = ToDeviceUnits(advance, devUnitsPerAppUnit);
|
|
gfxFloat height = GetMetrics().maxAscent;
|
|
gfxRect glyphRect(pt.x, pt.y - height, advanceDevUnits, height);
|
|
gfxFontMissingGlyphs::DrawMissingGlyph(aContext,
|
|
glyphRect,
|
|
details->mGlyphID);
|
|
}
|
|
} else {
|
|
double glyphX = x + details->mXOffset;
|
|
if (isRTL) {
|
|
glyphX -= advance;
|
|
}
|
|
|
|
gfxPoint point(ToDeviceUnits(glyphX, devUnitsPerAppUnit),
|
|
ToDeviceUnits(y, devUnitsPerAppUnit));
|
|
|
|
if (!haveSVGGlyphs ||
|
|
!RenderSVGGlyph(aContext, point, aDrawMode,
|
|
details->mGlyphID, aObjectPaint)) {
|
|
glyph = glyphs.AppendGlyph();
|
|
glyph->mIndex = details->mGlyphID;
|
|
glyph->mPosition.x = ToDeviceUnits(glyphX, devUnitsPerAppUnit);
|
|
glyph->mPosition.y = ToDeviceUnits(y + details->mYOffset, devUnitsPerAppUnit);
|
|
glyph->mPosition = matInv * glyph->mPosition;
|
|
glyphs.Flush(dt, aObjectPaint, scaledFont, aDrawMode,
|
|
isRTL, renderingOptions, aContext, passedInvMatrix);
|
|
|
|
if (IsSyntheticBold()) {
|
|
double strikeOffset = synBoldOnePixelOffset;
|
|
int32_t strikeCount = strikes;
|
|
do {
|
|
Glyph *doubleglyph;
|
|
doubleglyph = glyphs.AppendGlyph();
|
|
doubleglyph->mIndex = glyph->mIndex;
|
|
doubleglyph->mPosition.x =
|
|
ToDeviceUnits(glyphX + strikeOffset *
|
|
appUnitsPerDevUnit,
|
|
devUnitsPerAppUnit);
|
|
doubleglyph->mPosition.y = glyph->mPosition.y;
|
|
strikeOffset += synBoldOnePixelOffset;
|
|
doubleglyph->mPosition = matInv * doubleglyph->mPosition;
|
|
glyphs.Flush(dt, aObjectPaint, scaledFont,
|
|
aDrawMode, isRTL, renderingOptions,
|
|
aContext, passedInvMatrix);
|
|
} while (--strikeCount > 0);
|
|
}
|
|
}
|
|
}
|
|
x += direction*advance;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (aSpacing) {
|
|
double space = aSpacing[i - aStart].mAfter;
|
|
if (i + 1 < aEnd) {
|
|
space += aSpacing[i + 1 - aStart].mBefore;
|
|
}
|
|
x += direction*space;
|
|
}
|
|
}
|
|
|
|
glyphs.Flush(dt, aObjectPaint, scaledFont, aDrawMode, isRTL,
|
|
renderingOptions, aContext, passedInvMatrix, true);
|
|
|
|
dt->SetTransform(oldMat);
|
|
|
|
dt->SetPermitSubpixelAA(oldSubpixelAA);
|
|
}
|
|
|
|
*aPt = gfxPoint(x, y);
|
|
}
|
|
|
|
bool
|
|
gfxFont::RenderSVGGlyph(gfxContext *aContext, gfxPoint aPoint, DrawMode aDrawMode,
|
|
uint32_t aGlyphId, gfxTextObjectPaint *aObjectPaint)
|
|
{
|
|
if (!GetFontEntry()->HasSVGGlyph(aGlyphId)) {
|
|
return false;
|
|
}
|
|
|
|
const gfxFloat devUnitsPerSVGUnit = GetStyle()->size / gfxSVGGlyphs::SVG_UNITS_PER_EM;
|
|
gfxContextMatrixAutoSaveRestore matrixRestore(aContext);
|
|
|
|
aContext->Translate(gfxPoint(aPoint.x, aPoint.y));
|
|
aContext->Scale(devUnitsPerSVGUnit, devUnitsPerSVGUnit);
|
|
|
|
aObjectPaint->InitStrokeGeometry(aContext, devUnitsPerSVGUnit);
|
|
|
|
return GetFontEntry()->RenderSVGGlyph(aContext, aGlyphId, aDrawMode,
|
|
aObjectPaint);
|
|
}
|
|
|
|
static void
|
|
UnionRange(gfxFloat aX, gfxFloat* aDestMin, gfxFloat* aDestMax)
|
|
{
|
|
*aDestMin = NS_MIN(*aDestMin, aX);
|
|
*aDestMax = NS_MAX(*aDestMax, aX);
|
|
}
|
|
|
|
// We get precise glyph extents if the textrun creator requested them, or
|
|
// if the font is a user font --- in which case the author may be relying
|
|
// on overflowing glyphs.
|
|
static bool
|
|
NeedsGlyphExtents(gfxFont *aFont, gfxTextRun *aTextRun)
|
|
{
|
|
return (aTextRun->GetFlags() & gfxTextRunFactory::TEXT_NEED_BOUNDING_BOX) ||
|
|
aFont->GetFontEntry()->IsUserFont();
|
|
}
|
|
|
|
static bool
|
|
NeedsGlyphExtents(gfxTextRun *aTextRun)
|
|
{
|
|
if (aTextRun->GetFlags() & gfxTextRunFactory::TEXT_NEED_BOUNDING_BOX)
|
|
return true;
|
|
uint32_t numRuns;
|
|
const gfxTextRun::GlyphRun *glyphRuns = aTextRun->GetGlyphRuns(&numRuns);
|
|
for (uint32_t i = 0; i < numRuns; ++i) {
|
|
if (glyphRuns[i].mFont->GetFontEntry()->IsUserFont())
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
gfxFont::RunMetrics
|
|
gfxFont::Measure(gfxTextRun *aTextRun,
|
|
uint32_t aStart, uint32_t aEnd,
|
|
BoundingBoxType aBoundingBoxType,
|
|
gfxContext *aRefContext,
|
|
Spacing *aSpacing)
|
|
{
|
|
// If aBoundingBoxType is TIGHT_HINTED_OUTLINE_EXTENTS
|
|
// and the underlying cairo font may be antialiased,
|
|
// we need to create a copy in order to avoid getting cached extents.
|
|
// This is only used by MathML layout at present.
|
|
if (aBoundingBoxType == TIGHT_HINTED_OUTLINE_EXTENTS &&
|
|
mAntialiasOption != kAntialiasNone) {
|
|
if (!mNonAAFont) {
|
|
mNonAAFont = CopyWithAntialiasOption(kAntialiasNone);
|
|
}
|
|
// if font subclass doesn't implement CopyWithAntialiasOption(),
|
|
// it will return null and we'll proceed to use the existing font
|
|
if (mNonAAFont) {
|
|
return mNonAAFont->Measure(aTextRun, aStart, aEnd,
|
|
TIGHT_HINTED_OUTLINE_EXTENTS,
|
|
aRefContext, aSpacing);
|
|
}
|
|
}
|
|
|
|
const uint32_t appUnitsPerDevUnit = aTextRun->GetAppUnitsPerDevUnit();
|
|
// Current position in appunits
|
|
const gfxFont::Metrics& fontMetrics = GetMetrics();
|
|
|
|
RunMetrics metrics;
|
|
metrics.mAscent = fontMetrics.maxAscent*appUnitsPerDevUnit;
|
|
metrics.mDescent = fontMetrics.maxDescent*appUnitsPerDevUnit;
|
|
if (aStart == aEnd) {
|
|
// exit now before we look at aSpacing[0], which is undefined
|
|
metrics.mBoundingBox = gfxRect(0, -metrics.mAscent, 0, metrics.mAscent + metrics.mDescent);
|
|
return metrics;
|
|
}
|
|
|
|
gfxFloat advanceMin = 0, advanceMax = 0;
|
|
const gfxTextRun::CompressedGlyph *charGlyphs = aTextRun->GetCharacterGlyphs();
|
|
bool isRTL = aTextRun->IsRightToLeft();
|
|
double direction = aTextRun->GetDirection();
|
|
bool needsGlyphExtents = NeedsGlyphExtents(this, aTextRun);
|
|
gfxGlyphExtents *extents =
|
|
(aBoundingBoxType == LOOSE_INK_EXTENTS &&
|
|
!needsGlyphExtents &&
|
|
!aTextRun->HasDetailedGlyphs()) ? nullptr
|
|
: GetOrCreateGlyphExtents(aTextRun->GetAppUnitsPerDevUnit());
|
|
double x = 0;
|
|
if (aSpacing) {
|
|
x += direction*aSpacing[0].mBefore;
|
|
}
|
|
uint32_t i;
|
|
for (i = aStart; i < aEnd; ++i) {
|
|
const gfxTextRun::CompressedGlyph *glyphData = &charGlyphs[i];
|
|
if (glyphData->IsSimpleGlyph()) {
|
|
double advance = glyphData->GetSimpleAdvance();
|
|
// Only get the real glyph horizontal extent if we were asked
|
|
// for the tight bounding box or we're in quality mode
|
|
if ((aBoundingBoxType != LOOSE_INK_EXTENTS || needsGlyphExtents) &&
|
|
extents) {
|
|
uint32_t glyphIndex = glyphData->GetSimpleGlyph();
|
|
uint16_t extentsWidth = extents->GetContainedGlyphWidthAppUnits(glyphIndex);
|
|
if (extentsWidth != gfxGlyphExtents::INVALID_WIDTH &&
|
|
aBoundingBoxType == LOOSE_INK_EXTENTS) {
|
|
UnionRange(x, &advanceMin, &advanceMax);
|
|
UnionRange(x + direction*extentsWidth, &advanceMin, &advanceMax);
|
|
} else {
|
|
gfxRect glyphRect;
|
|
if (!extents->GetTightGlyphExtentsAppUnits(this,
|
|
aRefContext, glyphIndex, &glyphRect)) {
|
|
glyphRect = gfxRect(0, metrics.mBoundingBox.Y(),
|
|
advance, metrics.mBoundingBox.Height());
|
|
}
|
|
if (isRTL) {
|
|
glyphRect -= gfxPoint(advance, 0);
|
|
}
|
|
glyphRect += gfxPoint(x, 0);
|
|
metrics.mBoundingBox = metrics.mBoundingBox.Union(glyphRect);
|
|
}
|
|
}
|
|
x += direction*advance;
|
|
} else {
|
|
uint32_t glyphCount = glyphData->GetGlyphCount();
|
|
if (glyphCount > 0) {
|
|
const gfxTextRun::DetailedGlyph *details =
|
|
aTextRun->GetDetailedGlyphs(i);
|
|
NS_ASSERTION(details != nullptr,
|
|
"detaiedGlyph record should not be missing!");
|
|
uint32_t j;
|
|
for (j = 0; j < glyphCount; ++j, ++details) {
|
|
uint32_t glyphIndex = details->mGlyphID;
|
|
gfxPoint glyphPt(x + details->mXOffset, details->mYOffset);
|
|
double advance = details->mAdvance;
|
|
gfxRect glyphRect;
|
|
if (glyphData->IsMissing() || !extents ||
|
|
!extents->GetTightGlyphExtentsAppUnits(this,
|
|
aRefContext, glyphIndex, &glyphRect)) {
|
|
// We might have failed to get glyph extents due to
|
|
// OOM or something
|
|
glyphRect = gfxRect(0, -metrics.mAscent,
|
|
advance, metrics.mAscent + metrics.mDescent);
|
|
}
|
|
if (isRTL) {
|
|
glyphRect -= gfxPoint(advance, 0);
|
|
}
|
|
glyphRect += gfxPoint(x, 0);
|
|
metrics.mBoundingBox = metrics.mBoundingBox.Union(glyphRect);
|
|
x += direction*advance;
|
|
}
|
|
}
|
|
}
|
|
// Every other glyph type is ignored
|
|
if (aSpacing) {
|
|
double space = aSpacing[i - aStart].mAfter;
|
|
if (i + 1 < aEnd) {
|
|
space += aSpacing[i + 1 - aStart].mBefore;
|
|
}
|
|
x += direction*space;
|
|
}
|
|
}
|
|
|
|
if (aBoundingBoxType == LOOSE_INK_EXTENTS) {
|
|
UnionRange(x, &advanceMin, &advanceMax);
|
|
gfxRect fontBox(advanceMin, -metrics.mAscent,
|
|
advanceMax - advanceMin, metrics.mAscent + metrics.mDescent);
|
|
metrics.mBoundingBox = metrics.mBoundingBox.Union(fontBox);
|
|
}
|
|
if (isRTL) {
|
|
metrics.mBoundingBox -= gfxPoint(x, 0);
|
|
}
|
|
|
|
metrics.mAdvanceWidth = x*direction;
|
|
return metrics;
|
|
}
|
|
|
|
#define MAX_SHAPING_LENGTH 32760 // slightly less than 32K, trying to avoid
|
|
// over-stressing platform shapers
|
|
|
|
#define BACKTRACK_LIMIT 1024 // If we can't find a space or a cluster start
|
|
// within 1K chars, just chop arbitrarily.
|
|
// Limiting backtrack here avoids pathological
|
|
// behavior on long runs with no whitespace.
|
|
|
|
static bool
|
|
IsBoundarySpace(PRUnichar aChar, PRUnichar aNextChar)
|
|
{
|
|
return (aChar == ' ' || aChar == 0x00A0) && !IsClusterExtender(aNextChar);
|
|
}
|
|
|
|
static inline uint32_t
|
|
HashMix(uint32_t aHash, PRUnichar aCh)
|
|
{
|
|
return (aHash >> 28) ^ (aHash << 4) ^ aCh;
|
|
}
|
|
|
|
template<typename T>
|
|
gfxShapedWord*
|
|
gfxFont::GetShapedWord(gfxContext *aContext,
|
|
const T *aText,
|
|
uint32_t aLength,
|
|
uint32_t aHash,
|
|
int32_t aRunScript,
|
|
int32_t aAppUnitsPerDevUnit,
|
|
uint32_t aFlags)
|
|
{
|
|
// if the cache is getting too big, flush it and start over
|
|
if (mWordCache.Count() > 10000) {
|
|
NS_WARNING("flushing shaped-word cache");
|
|
ClearCachedWords();
|
|
}
|
|
|
|
// if there's a cached entry for this word, just return it
|
|
CacheHashKey key(aText, aLength, aHash,
|
|
aRunScript,
|
|
aAppUnitsPerDevUnit,
|
|
aFlags);
|
|
|
|
CacheHashEntry *entry = mWordCache.PutEntry(key);
|
|
if (!entry) {
|
|
NS_WARNING("failed to create word cache entry - expect missing text");
|
|
return nullptr;
|
|
}
|
|
gfxShapedWord *sw = entry->mShapedWord;
|
|
|
|
if (sw) {
|
|
sw->ResetAge();
|
|
Telemetry::Accumulate(Telemetry::WORD_CACHE_HITS, aLength);
|
|
return sw;
|
|
}
|
|
|
|
Telemetry::Accumulate(Telemetry::WORD_CACHE_MISSES, aLength);
|
|
sw = entry->mShapedWord = gfxShapedWord::Create(aText, aLength,
|
|
aRunScript,
|
|
aAppUnitsPerDevUnit,
|
|
aFlags);
|
|
if (!sw) {
|
|
NS_WARNING("failed to create gfxShapedWord - expect missing text");
|
|
return nullptr;
|
|
}
|
|
|
|
DebugOnly<bool> ok = false;
|
|
if (sizeof(T) == sizeof(PRUnichar)) {
|
|
ok = ShapeWord(aContext, sw, (const PRUnichar*)aText);
|
|
} else {
|
|
nsAutoString utf16;
|
|
AppendASCIItoUTF16(nsDependentCSubstring((const char*)aText, aLength),
|
|
utf16);
|
|
if (utf16.Length() == aLength) {
|
|
ok = ShapeWord(aContext, sw, utf16.BeginReading());
|
|
}
|
|
}
|
|
NS_WARN_IF_FALSE(ok, "failed to shape word - expect garbled text");
|
|
|
|
for (uint32_t i = 0; i < aLength; ++i) {
|
|
if (aText[i] == ' ') {
|
|
sw->SetIsSpace(i);
|
|
} else if (i > 0 &&
|
|
NS_IS_HIGH_SURROGATE(aText[i - 1]) &&
|
|
NS_IS_LOW_SURROGATE(aText[i])) {
|
|
sw->SetIsLowSurrogate(i);
|
|
}
|
|
}
|
|
|
|
return sw;
|
|
}
|
|
|
|
bool
|
|
gfxFont::CacheHashEntry::KeyEquals(const KeyTypePointer aKey) const
|
|
{
|
|
const gfxShapedWord *sw = mShapedWord;
|
|
if (!sw) {
|
|
return false;
|
|
}
|
|
if (sw->Length() != aKey->mLength ||
|
|
sw->Flags() != aKey->mFlags ||
|
|
sw->AppUnitsPerDevUnit() != aKey->mAppUnitsPerDevUnit ||
|
|
sw->Script() != aKey->mScript) {
|
|
return false;
|
|
}
|
|
if (sw->TextIs8Bit()) {
|
|
if (aKey->mTextIs8Bit) {
|
|
return (0 == memcmp(sw->Text8Bit(), aKey->mText.mSingle,
|
|
aKey->mLength * sizeof(uint8_t)));
|
|
}
|
|
// The key has 16-bit text, even though all the characters are < 256,
|
|
// so the TEXT_IS_8BIT flag was set and the cached ShapedWord we're
|
|
// comparing with will have 8-bit text.
|
|
const uint8_t *s1 = sw->Text8Bit();
|
|
const PRUnichar *s2 = aKey->mText.mDouble;
|
|
const PRUnichar *s2end = s2 + aKey->mLength;
|
|
while (s2 < s2end) {
|
|
if (*s1++ != *s2++) {
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
NS_ASSERTION((aKey->mFlags & gfxTextRunFactory::TEXT_IS_8BIT) == 0 &&
|
|
!aKey->mTextIs8Bit, "didn't expect 8-bit text here");
|
|
return (0 == memcmp(sw->TextUnicode(), aKey->mText.mDouble,
|
|
aKey->mLength * sizeof(PRUnichar)));
|
|
}
|
|
|
|
bool
|
|
gfxFont::ShapeWord(gfxContext *aContext,
|
|
gfxShapedWord *aShapedWord,
|
|
const PRUnichar *aText,
|
|
bool aPreferPlatformShaping)
|
|
{
|
|
bool ok = false;
|
|
|
|
#ifdef MOZ_GRAPHITE
|
|
if (mGraphiteShaper && gfxPlatform::GetPlatform()->UseGraphiteShaping()) {
|
|
ok = mGraphiteShaper->ShapeWord(aContext, aShapedWord, aText);
|
|
}
|
|
#endif
|
|
|
|
if (!ok && mHarfBuzzShaper && !aPreferPlatformShaping) {
|
|
if (gfxPlatform::GetPlatform()->UseHarfBuzzForScript(aShapedWord->Script())) {
|
|
ok = mHarfBuzzShaper->ShapeWord(aContext, aShapedWord, aText);
|
|
}
|
|
}
|
|
|
|
if (!ok) {
|
|
if (!mPlatformShaper) {
|
|
CreatePlatformShaper();
|
|
NS_ASSERTION(mPlatformShaper, "no platform shaper available!");
|
|
}
|
|
if (mPlatformShaper) {
|
|
ok = mPlatformShaper->ShapeWord(aContext, aShapedWord, aText);
|
|
}
|
|
}
|
|
|
|
if (ok && IsSyntheticBold()) {
|
|
float synBoldOffset =
|
|
GetSyntheticBoldOffset() * CalcXScale(aContext);
|
|
aShapedWord->AdjustAdvancesForSyntheticBold(synBoldOffset);
|
|
}
|
|
|
|
return ok;
|
|
}
|
|
|
|
inline static bool IsChar8Bit(uint8_t /*aCh*/) { return true; }
|
|
inline static bool IsChar8Bit(PRUnichar aCh) { return aCh < 0x100; }
|
|
|
|
template<typename T>
|
|
bool
|
|
gfxFont::SplitAndInitTextRun(gfxContext *aContext,
|
|
gfxTextRun *aTextRun,
|
|
const T *aString,
|
|
uint32_t aRunStart,
|
|
uint32_t aRunLength,
|
|
int32_t aRunScript)
|
|
{
|
|
if (aRunLength == 0) {
|
|
return true;
|
|
}
|
|
|
|
InitWordCache();
|
|
|
|
// the only flags we care about for ShapedWord construction/caching
|
|
uint32_t flags = aTextRun->GetFlags() &
|
|
(gfxTextRunFactory::TEXT_IS_RTL |
|
|
gfxTextRunFactory::TEXT_DISABLE_OPTIONAL_LIGATURES);
|
|
if (sizeof(T) == sizeof(uint8_t)) {
|
|
flags |= gfxTextRunFactory::TEXT_IS_8BIT;
|
|
}
|
|
|
|
const T *text = aString + aRunStart;
|
|
uint32_t wordStart = 0;
|
|
uint32_t hash = 0;
|
|
bool wordIs8Bit = true;
|
|
int32_t appUnitsPerDevUnit = aTextRun->GetAppUnitsPerDevUnit();
|
|
|
|
T nextCh = text[0];
|
|
for (uint32_t i = 0; i <= aRunLength; ++i) {
|
|
T ch = nextCh;
|
|
nextCh = (i < aRunLength - 1) ? text[i + 1] : '\n';
|
|
bool boundary = IsBoundarySpace(ch, nextCh);
|
|
bool invalid = !boundary && gfxFontGroup::IsInvalidChar(ch);
|
|
uint32_t length = i - wordStart;
|
|
|
|
// break into separate ShapedWords when we hit an invalid char,
|
|
// or a boundary space (always handled individually),
|
|
// or the first non-space after a space
|
|
bool breakHere = boundary || invalid;
|
|
|
|
if (!breakHere) {
|
|
// if we're approaching the max ShapedWord length, break anyway...
|
|
if (sizeof(T) == sizeof(uint8_t)) {
|
|
// in 8-bit text, no clusters or surrogates to worry about
|
|
if (length >= gfxShapedWord::kMaxLength) {
|
|
breakHere = true;
|
|
}
|
|
} else {
|
|
// try to avoid breaking before combining mark or low surrogate
|
|
if (length >= gfxShapedWord::kMaxLength - 15) {
|
|
if (!NS_IS_LOW_SURROGATE(ch)) {
|
|
if (!IsClusterExtender(ch)) {
|
|
breakHere = true;
|
|
}
|
|
}
|
|
if (!breakHere && length >= gfxShapedWord::kMaxLength - 3) {
|
|
if (!NS_IS_LOW_SURROGATE(ch)) {
|
|
breakHere = true;
|
|
}
|
|
}
|
|
if (!breakHere && length >= gfxShapedWord::kMaxLength) {
|
|
breakHere = true;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (!breakHere) {
|
|
if (!IsChar8Bit(ch)) {
|
|
wordIs8Bit = false;
|
|
}
|
|
// include this character in the hash, and move on to next
|
|
hash = HashMix(hash, ch);
|
|
continue;
|
|
}
|
|
|
|
// We've decided to break here (i.e. we're at the end of a "word",
|
|
// or the word is becoming excessively long): shape the word and
|
|
// add it to the textrun
|
|
if (length > 0) {
|
|
uint32_t wordFlags = flags;
|
|
// in the 8-bit version of this method, TEXT_IS_8BIT was
|
|
// already set as part of |flags|, so no need for a per-word
|
|
// adjustment here
|
|
if (sizeof(T) == sizeof(PRUnichar)) {
|
|
if (wordIs8Bit) {
|
|
wordFlags |= gfxTextRunFactory::TEXT_IS_8BIT;
|
|
}
|
|
}
|
|
gfxShapedWord *sw = GetShapedWord(aContext,
|
|
text + wordStart, length,
|
|
hash, aRunScript,
|
|
appUnitsPerDevUnit,
|
|
wordFlags);
|
|
if (sw) {
|
|
aTextRun->CopyGlyphDataFrom(sw, aRunStart + wordStart);
|
|
} else {
|
|
return false; // failed, presumably out of memory?
|
|
}
|
|
}
|
|
|
|
if (boundary) {
|
|
// word was terminated by a space: add that to the textrun
|
|
if (!aTextRun->SetSpaceGlyphIfSimple(this, aContext,
|
|
aRunStart + i, ch))
|
|
{
|
|
static const uint8_t space = ' ';
|
|
gfxShapedWord *sw =
|
|
GetShapedWord(aContext,
|
|
&space, 1,
|
|
HashMix(0, ' '), aRunScript,
|
|
appUnitsPerDevUnit,
|
|
flags | gfxTextRunFactory::TEXT_IS_8BIT);
|
|
if (sw) {
|
|
aTextRun->CopyGlyphDataFrom(sw, aRunStart + i);
|
|
} else {
|
|
return false;
|
|
}
|
|
}
|
|
hash = 0;
|
|
wordStart = i + 1;
|
|
wordIs8Bit = true;
|
|
continue;
|
|
}
|
|
|
|
if (i == aRunLength) {
|
|
break;
|
|
}
|
|
|
|
if (invalid) {
|
|
// word was terminated by an invalid char: skip it,
|
|
// but record where TAB or NEWLINE occur
|
|
if (ch == '\t') {
|
|
aTextRun->SetIsTab(aRunStart + i);
|
|
} else if (ch == '\n') {
|
|
aTextRun->SetIsNewline(aRunStart + i);
|
|
}
|
|
hash = 0;
|
|
wordStart = i + 1;
|
|
wordIs8Bit = true;
|
|
continue;
|
|
}
|
|
|
|
// word was forcibly broken, so current char will begin next word
|
|
hash = HashMix(0, ch);
|
|
wordStart = i;
|
|
wordIs8Bit = IsChar8Bit(ch);
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
gfxGlyphExtents *
|
|
gfxFont::GetOrCreateGlyphExtents(uint32_t aAppUnitsPerDevUnit) {
|
|
uint32_t i, count = mGlyphExtentsArray.Length();
|
|
for (i = 0; i < count; ++i) {
|
|
if (mGlyphExtentsArray[i]->GetAppUnitsPerDevUnit() == aAppUnitsPerDevUnit)
|
|
return mGlyphExtentsArray[i];
|
|
}
|
|
gfxGlyphExtents *glyphExtents = new gfxGlyphExtents(aAppUnitsPerDevUnit);
|
|
if (glyphExtents) {
|
|
mGlyphExtentsArray.AppendElement(glyphExtents);
|
|
// Initialize the extents of a space glyph, assuming that spaces don't
|
|
// render anything!
|
|
glyphExtents->SetContainedGlyphWidthAppUnits(GetSpaceGlyph(), 0);
|
|
}
|
|
return glyphExtents;
|
|
}
|
|
|
|
void
|
|
gfxFont::SetupGlyphExtents(gfxContext *aContext, uint32_t aGlyphID, bool aNeedTight,
|
|
gfxGlyphExtents *aExtents)
|
|
{
|
|
gfxContextMatrixAutoSaveRestore matrixRestore(aContext);
|
|
aContext->IdentityMatrix();
|
|
cairo_glyph_t glyph;
|
|
glyph.index = aGlyphID;
|
|
glyph.x = 0;
|
|
glyph.y = 0;
|
|
cairo_text_extents_t extents;
|
|
cairo_glyph_extents(aContext->GetCairo(), &glyph, 1, &extents);
|
|
|
|
const Metrics& fontMetrics = GetMetrics();
|
|
uint32_t appUnitsPerDevUnit = aExtents->GetAppUnitsPerDevUnit();
|
|
if (!aNeedTight && extents.x_bearing >= 0 &&
|
|
extents.y_bearing >= -fontMetrics.maxAscent &&
|
|
extents.height + extents.y_bearing <= fontMetrics.maxDescent) {
|
|
uint32_t appUnitsWidth =
|
|
uint32_t(ceil((extents.x_bearing + extents.width)*appUnitsPerDevUnit));
|
|
if (appUnitsWidth < gfxGlyphExtents::INVALID_WIDTH) {
|
|
aExtents->SetContainedGlyphWidthAppUnits(aGlyphID, uint16_t(appUnitsWidth));
|
|
return;
|
|
}
|
|
}
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
if (!aNeedTight) {
|
|
++gGlyphExtentsSetupFallBackToTight;
|
|
}
|
|
#endif
|
|
|
|
gfxFloat d2a = appUnitsPerDevUnit;
|
|
gfxRect bounds(extents.x_bearing*d2a, extents.y_bearing*d2a,
|
|
extents.width*d2a, extents.height*d2a);
|
|
|
|
gfxRect svgBounds;
|
|
if (mFontEntry->TryGetSVGData() &&
|
|
mFontEntry->HasSVGGlyph(aGlyphID) &&
|
|
mFontEntry->GetSVGGlyphExtents(aContext, aGlyphID, &svgBounds)) {
|
|
|
|
bounds = bounds.Union(gfxRect(svgBounds.x * d2a, svgBounds.y * d2a,
|
|
svgBounds.width * d2a, svgBounds.height * d2a));
|
|
}
|
|
|
|
aExtents->SetTightGlyphExtents(aGlyphID, bounds);
|
|
}
|
|
|
|
// Try to initialize font metrics by reading sfnt tables directly;
|
|
// set mIsValid=TRUE and return TRUE on success.
|
|
// Return FALSE if the gfxFontEntry subclass does not
|
|
// implement GetFontTable(), or for non-sfnt fonts where tables are
|
|
// not available.
|
|
bool
|
|
gfxFont::InitMetricsFromSfntTables(Metrics& aMetrics)
|
|
{
|
|
mIsValid = false; // font is NOT valid in case of early return
|
|
|
|
const uint32_t kHeadTableTag = TRUETYPE_TAG('h','e','a','d');
|
|
const uint32_t kHheaTableTag = TRUETYPE_TAG('h','h','e','a');
|
|
const uint32_t kPostTableTag = TRUETYPE_TAG('p','o','s','t');
|
|
const uint32_t kOS_2TableTag = TRUETYPE_TAG('O','S','/','2');
|
|
|
|
if (mFUnitsConvFactor == 0.0) {
|
|
// If the conversion factor from FUnits is not yet set,
|
|
// 'head' table is required; otherwise we cannot read any metrics
|
|
// because we don't know unitsPerEm
|
|
AutoFallibleTArray<uint8_t,sizeof(HeadTable)> headData;
|
|
if (NS_FAILED(mFontEntry->GetFontTable(kHeadTableTag, headData)) ||
|
|
headData.Length() < sizeof(HeadTable)) {
|
|
return false; // no 'head' table -> not an sfnt
|
|
}
|
|
HeadTable *head = reinterpret_cast<HeadTable*>(headData.Elements());
|
|
uint32_t unitsPerEm = head->unitsPerEm;
|
|
if (!unitsPerEm) {
|
|
return true; // is an sfnt, but not valid
|
|
}
|
|
mFUnitsConvFactor = mAdjustedSize / unitsPerEm;
|
|
}
|
|
|
|
// 'hhea' table is required to get vertical extents
|
|
AutoFallibleTArray<uint8_t,sizeof(HheaTable)> hheaData;
|
|
if (NS_FAILED(mFontEntry->GetFontTable(kHheaTableTag, hheaData)) ||
|
|
hheaData.Length() < sizeof(HheaTable)) {
|
|
return false; // no 'hhea' table -> not an sfnt
|
|
}
|
|
HheaTable *hhea = reinterpret_cast<HheaTable*>(hheaData.Elements());
|
|
|
|
#define SET_UNSIGNED(field,src) aMetrics.field = uint16_t(src) * mFUnitsConvFactor
|
|
#define SET_SIGNED(field,src) aMetrics.field = int16_t(src) * mFUnitsConvFactor
|
|
|
|
SET_UNSIGNED(maxAdvance, hhea->advanceWidthMax);
|
|
SET_SIGNED(maxAscent, hhea->ascender);
|
|
SET_SIGNED(maxDescent, -int16_t(hhea->descender));
|
|
SET_SIGNED(externalLeading, hhea->lineGap);
|
|
|
|
// 'post' table is required for underline metrics
|
|
AutoFallibleTArray<uint8_t,sizeof(PostTable)> postData;
|
|
if (NS_FAILED(mFontEntry->GetFontTable(kPostTableTag, postData))) {
|
|
return true; // no 'post' table -> sfnt is not valid
|
|
}
|
|
if (postData.Length() <
|
|
offsetof(PostTable, underlineThickness) + sizeof(uint16_t)) {
|
|
return true; // bad post table -> sfnt is not valid
|
|
}
|
|
PostTable *post = reinterpret_cast<PostTable*>(postData.Elements());
|
|
|
|
SET_SIGNED(underlineOffset, post->underlinePosition);
|
|
SET_UNSIGNED(underlineSize, post->underlineThickness);
|
|
|
|
// 'OS/2' table is optional, if not found we'll estimate xHeight
|
|
// and aveCharWidth by measuring glyphs
|
|
AutoFallibleTArray<uint8_t,sizeof(OS2Table)> os2data;
|
|
if (NS_SUCCEEDED(mFontEntry->GetFontTable(kOS_2TableTag, os2data))) {
|
|
OS2Table *os2 = reinterpret_cast<OS2Table*>(os2data.Elements());
|
|
|
|
if (os2data.Length() >= offsetof(OS2Table, sxHeight) +
|
|
sizeof(int16_t) &&
|
|
uint16_t(os2->version) >= 2) {
|
|
// version 2 and later includes the x-height field
|
|
SET_SIGNED(xHeight, os2->sxHeight);
|
|
// NS_ABS because of negative xHeight seen in Kokonor (Tibetan) font
|
|
aMetrics.xHeight = NS_ABS(aMetrics.xHeight);
|
|
}
|
|
// this should always be present
|
|
if (os2data.Length() >= offsetof(OS2Table, yStrikeoutPosition) +
|
|
sizeof(int16_t)) {
|
|
SET_SIGNED(aveCharWidth, os2->xAvgCharWidth);
|
|
SET_SIGNED(subscriptOffset, os2->ySubscriptYOffset);
|
|
SET_SIGNED(superscriptOffset, os2->ySuperscriptYOffset);
|
|
SET_SIGNED(strikeoutSize, os2->yStrikeoutSize);
|
|
SET_SIGNED(strikeoutOffset, os2->yStrikeoutPosition);
|
|
}
|
|
}
|
|
|
|
mIsValid = true;
|
|
|
|
return true;
|
|
}
|
|
|
|
static double
|
|
RoundToNearestMultiple(double aValue, double aFraction)
|
|
{
|
|
return floor(aValue/aFraction + 0.5) * aFraction;
|
|
}
|
|
|
|
void gfxFont::CalculateDerivedMetrics(Metrics& aMetrics)
|
|
{
|
|
aMetrics.maxAscent =
|
|
ceil(RoundToNearestMultiple(aMetrics.maxAscent, 1/1024.0));
|
|
aMetrics.maxDescent =
|
|
ceil(RoundToNearestMultiple(aMetrics.maxDescent, 1/1024.0));
|
|
|
|
if (aMetrics.xHeight <= 0) {
|
|
// only happens if we couldn't find either font metrics
|
|
// or a char to measure;
|
|
// pick an arbitrary value that's better than zero
|
|
aMetrics.xHeight = aMetrics.maxAscent * DEFAULT_XHEIGHT_FACTOR;
|
|
}
|
|
|
|
aMetrics.maxHeight = aMetrics.maxAscent + aMetrics.maxDescent;
|
|
|
|
if (aMetrics.maxHeight - aMetrics.emHeight > 0.0) {
|
|
aMetrics.internalLeading = aMetrics.maxHeight - aMetrics.emHeight;
|
|
} else {
|
|
aMetrics.internalLeading = 0.0;
|
|
}
|
|
|
|
aMetrics.emAscent = aMetrics.maxAscent * aMetrics.emHeight
|
|
/ aMetrics.maxHeight;
|
|
aMetrics.emDescent = aMetrics.emHeight - aMetrics.emAscent;
|
|
|
|
if (GetFontEntry()->IsFixedPitch()) {
|
|
// Some Quartz fonts are fixed pitch, but there's some glyph with a bigger
|
|
// advance than the average character width... this forces
|
|
// those fonts to be recognized like fixed pitch fonts by layout.
|
|
aMetrics.maxAdvance = aMetrics.aveCharWidth;
|
|
}
|
|
|
|
if (!aMetrics.subscriptOffset) {
|
|
aMetrics.subscriptOffset = aMetrics.xHeight;
|
|
}
|
|
if (!aMetrics.superscriptOffset) {
|
|
aMetrics.superscriptOffset = aMetrics.xHeight;
|
|
}
|
|
|
|
if (!aMetrics.strikeoutOffset) {
|
|
aMetrics.strikeoutOffset = aMetrics.xHeight * 0.5;
|
|
}
|
|
if (!aMetrics.strikeoutSize) {
|
|
aMetrics.strikeoutSize = aMetrics.underlineSize;
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxFont::SanitizeMetrics(gfxFont::Metrics *aMetrics, bool aIsBadUnderlineFont)
|
|
{
|
|
// Even if this font size is zero, this font is created with non-zero size.
|
|
// However, for layout and others, we should return the metrics of zero size font.
|
|
if (mStyle.size == 0.0) {
|
|
memset(aMetrics, 0, sizeof(gfxFont::Metrics));
|
|
return;
|
|
}
|
|
|
|
// MS (P)Gothic and MS (P)Mincho are not having suitable values in their super script offset.
|
|
// If the values are not suitable, we should use x-height instead of them.
|
|
// See https://bugzilla.mozilla.org/show_bug.cgi?id=353632
|
|
if (aMetrics->superscriptOffset <= 0 ||
|
|
aMetrics->superscriptOffset >= aMetrics->maxAscent) {
|
|
aMetrics->superscriptOffset = aMetrics->xHeight;
|
|
}
|
|
// And also checking the case of sub script offset. The old gfx for win has checked this too.
|
|
if (aMetrics->subscriptOffset <= 0 ||
|
|
aMetrics->subscriptOffset >= aMetrics->maxAscent) {
|
|
aMetrics->subscriptOffset = aMetrics->xHeight;
|
|
}
|
|
|
|
aMetrics->underlineSize = NS_MAX(1.0, aMetrics->underlineSize);
|
|
aMetrics->strikeoutSize = NS_MAX(1.0, aMetrics->strikeoutSize);
|
|
|
|
aMetrics->underlineOffset = NS_MIN(aMetrics->underlineOffset, -1.0);
|
|
|
|
if (aMetrics->maxAscent < 1.0) {
|
|
// We cannot draw strikeout line and overline in the ascent...
|
|
aMetrics->underlineSize = 0;
|
|
aMetrics->underlineOffset = 0;
|
|
aMetrics->strikeoutSize = 0;
|
|
aMetrics->strikeoutOffset = 0;
|
|
return;
|
|
}
|
|
|
|
/**
|
|
* Some CJK fonts have bad underline offset. Therefore, if this is such font,
|
|
* we need to lower the underline offset to bottom of *em* descent.
|
|
* However, if this is system font, we should not do this for the rendering compatibility with
|
|
* another application's UI on the platform.
|
|
* XXX Should not use this hack if the font size is too small?
|
|
* Such text cannot be read, this might be used for tight CSS rendering? (E.g., Acid2)
|
|
*/
|
|
if (!mStyle.systemFont && aIsBadUnderlineFont) {
|
|
// First, we need 2 pixels between baseline and underline at least. Because many CJK characters
|
|
// put their glyphs on the baseline, so, 1 pixel is too close for CJK characters.
|
|
aMetrics->underlineOffset = NS_MIN(aMetrics->underlineOffset, -2.0);
|
|
|
|
// Next, we put the underline to bottom of below of the descent space.
|
|
if (aMetrics->internalLeading + aMetrics->externalLeading > aMetrics->underlineSize) {
|
|
aMetrics->underlineOffset = NS_MIN(aMetrics->underlineOffset, -aMetrics->emDescent);
|
|
} else {
|
|
aMetrics->underlineOffset = NS_MIN(aMetrics->underlineOffset,
|
|
aMetrics->underlineSize - aMetrics->emDescent);
|
|
}
|
|
}
|
|
// If underline positioned is too far from the text, descent position is preferred so that underline
|
|
// will stay within the boundary.
|
|
else if (aMetrics->underlineSize - aMetrics->underlineOffset > aMetrics->maxDescent) {
|
|
if (aMetrics->underlineSize > aMetrics->maxDescent)
|
|
aMetrics->underlineSize = NS_MAX(aMetrics->maxDescent, 1.0);
|
|
// The max underlineOffset is 1px (the min underlineSize is 1px, and min maxDescent is 0px.)
|
|
aMetrics->underlineOffset = aMetrics->underlineSize - aMetrics->maxDescent;
|
|
}
|
|
|
|
// If strikeout line is overflowed from the ascent, the line should be resized and moved for
|
|
// that being in the ascent space.
|
|
// Note that the strikeoutOffset is *middle* of the strikeout line position.
|
|
gfxFloat halfOfStrikeoutSize = floor(aMetrics->strikeoutSize / 2.0 + 0.5);
|
|
if (halfOfStrikeoutSize + aMetrics->strikeoutOffset > aMetrics->maxAscent) {
|
|
if (aMetrics->strikeoutSize > aMetrics->maxAscent) {
|
|
aMetrics->strikeoutSize = NS_MAX(aMetrics->maxAscent, 1.0);
|
|
halfOfStrikeoutSize = floor(aMetrics->strikeoutSize / 2.0 + 0.5);
|
|
}
|
|
gfxFloat ascent = floor(aMetrics->maxAscent + 0.5);
|
|
aMetrics->strikeoutOffset = NS_MAX(halfOfStrikeoutSize, ascent / 2.0);
|
|
}
|
|
|
|
// If overline is larger than the ascent, the line should be resized.
|
|
if (aMetrics->underlineSize > aMetrics->maxAscent) {
|
|
aMetrics->underlineSize = aMetrics->maxAscent;
|
|
}
|
|
}
|
|
|
|
gfxFloat
|
|
gfxFont::SynthesizeSpaceWidth(uint32_t aCh)
|
|
{
|
|
// return an appropriate width for various Unicode space characters
|
|
// that we "fake" if they're not actually present in the font;
|
|
// returns negative value if the char is not a known space.
|
|
switch (aCh) {
|
|
case 0x2000: // en quad
|
|
case 0x2002: return GetAdjustedSize() / 2; // en space
|
|
case 0x2001: // em quad
|
|
case 0x2003: return GetAdjustedSize(); // em space
|
|
case 0x2004: return GetAdjustedSize() / 3; // three-per-em space
|
|
case 0x2005: return GetAdjustedSize() / 4; // four-per-em space
|
|
case 0x2006: return GetAdjustedSize() / 6; // six-per-em space
|
|
case 0x2007: return GetMetrics().zeroOrAveCharWidth; // figure space
|
|
case 0x2008: return GetMetrics().spaceWidth; // punctuation space
|
|
case 0x2009: return GetAdjustedSize() / 5; // thin space
|
|
case 0x200a: return GetAdjustedSize() / 10; // hair space
|
|
case 0x202f: return GetAdjustedSize() / 5; // narrow no-break space
|
|
default: return -1.0;
|
|
}
|
|
}
|
|
|
|
/*static*/ size_t
|
|
gfxFont::WordCacheEntrySizeOfExcludingThis(CacheHashEntry* aHashEntry,
|
|
nsMallocSizeOfFun aMallocSizeOf,
|
|
void* aUserArg)
|
|
{
|
|
return aMallocSizeOf(aHashEntry->mShapedWord.get());
|
|
}
|
|
|
|
void
|
|
gfxFont::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontCacheSizes* aSizes) const
|
|
{
|
|
for (uint32_t i = 0; i < mGlyphExtentsArray.Length(); ++i) {
|
|
aSizes->mFontInstances +=
|
|
mGlyphExtentsArray[i]->SizeOfIncludingThis(aMallocSizeOf);
|
|
}
|
|
aSizes->mShapedWords +=
|
|
mWordCache.SizeOfExcludingThis(WordCacheEntrySizeOfExcludingThis,
|
|
aMallocSizeOf);
|
|
}
|
|
|
|
void
|
|
gfxFont::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf,
|
|
FontCacheSizes* aSizes) const
|
|
{
|
|
aSizes->mFontInstances += aMallocSizeOf(this);
|
|
SizeOfExcludingThis(aMallocSizeOf, aSizes);
|
|
}
|
|
|
|
gfxGlyphExtents::~gfxGlyphExtents()
|
|
{
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
gGlyphExtentsWidthsTotalSize +=
|
|
mContainedGlyphWidths.SizeOfExcludingThis(&FontCacheMallocSizeOf);
|
|
gGlyphExtentsCount++;
|
|
#endif
|
|
MOZ_COUNT_DTOR(gfxGlyphExtents);
|
|
}
|
|
|
|
bool
|
|
gfxGlyphExtents::GetTightGlyphExtentsAppUnits(gfxFont *aFont,
|
|
gfxContext *aContext, uint32_t aGlyphID, gfxRect *aExtents)
|
|
{
|
|
HashEntry *entry = mTightGlyphExtents.GetEntry(aGlyphID);
|
|
if (!entry) {
|
|
if (!aContext) {
|
|
NS_WARNING("Could not get glyph extents (no aContext)");
|
|
return false;
|
|
}
|
|
|
|
if (aFont->SetupCairoFont(aContext)) {
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
++gGlyphExtentsSetupLazyTight;
|
|
#endif
|
|
aFont->SetupGlyphExtents(aContext, aGlyphID, true, this);
|
|
entry = mTightGlyphExtents.GetEntry(aGlyphID);
|
|
}
|
|
if (!entry) {
|
|
NS_WARNING("Could not get glyph extents");
|
|
return false;
|
|
}
|
|
}
|
|
|
|
*aExtents = gfxRect(entry->x, entry->y, entry->width, entry->height);
|
|
return true;
|
|
}
|
|
|
|
gfxGlyphExtents::GlyphWidths::~GlyphWidths()
|
|
{
|
|
uint32_t i, count = mBlocks.Length();
|
|
for (i = 0; i < count; ++i) {
|
|
uintptr_t bits = mBlocks[i];
|
|
if (bits && !(bits & 0x1)) {
|
|
delete[] reinterpret_cast<uint16_t *>(bits);
|
|
}
|
|
}
|
|
}
|
|
|
|
uint32_t
|
|
gfxGlyphExtents::GlyphWidths::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf) const
|
|
{
|
|
uint32_t i;
|
|
uint32_t size = mBlocks.SizeOfExcludingThis(aMallocSizeOf);
|
|
for (i = 0; i < mBlocks.Length(); ++i) {
|
|
uintptr_t bits = mBlocks[i];
|
|
if (bits && !(bits & 0x1)) {
|
|
size += aMallocSizeOf(reinterpret_cast<void*>(bits));
|
|
}
|
|
}
|
|
return size;
|
|
}
|
|
|
|
void
|
|
gfxGlyphExtents::GlyphWidths::Set(uint32_t aGlyphID, uint16_t aWidth)
|
|
{
|
|
uint32_t block = aGlyphID >> BLOCK_SIZE_BITS;
|
|
uint32_t len = mBlocks.Length();
|
|
if (block >= len) {
|
|
uintptr_t *elems = mBlocks.AppendElements(block + 1 - len);
|
|
if (!elems)
|
|
return;
|
|
memset(elems, 0, sizeof(uintptr_t)*(block + 1 - len));
|
|
}
|
|
|
|
uintptr_t bits = mBlocks[block];
|
|
uint32_t glyphOffset = aGlyphID & (BLOCK_SIZE - 1);
|
|
if (!bits) {
|
|
mBlocks[block] = MakeSingle(glyphOffset, aWidth);
|
|
return;
|
|
}
|
|
|
|
uint16_t *newBlock;
|
|
if (bits & 0x1) {
|
|
// Expand the block to a real block. We could avoid this by checking
|
|
// glyphOffset == GetGlyphOffset(bits), but that never happens so don't bother
|
|
newBlock = new uint16_t[BLOCK_SIZE];
|
|
if (!newBlock)
|
|
return;
|
|
uint32_t i;
|
|
for (i = 0; i < BLOCK_SIZE; ++i) {
|
|
newBlock[i] = INVALID_WIDTH;
|
|
}
|
|
newBlock[GetGlyphOffset(bits)] = GetWidth(bits);
|
|
mBlocks[block] = reinterpret_cast<uintptr_t>(newBlock);
|
|
} else {
|
|
newBlock = reinterpret_cast<uint16_t *>(bits);
|
|
}
|
|
newBlock[glyphOffset] = aWidth;
|
|
}
|
|
|
|
void
|
|
gfxGlyphExtents::SetTightGlyphExtents(uint32_t aGlyphID, const gfxRect& aExtentsAppUnits)
|
|
{
|
|
HashEntry *entry = mTightGlyphExtents.PutEntry(aGlyphID);
|
|
if (!entry)
|
|
return;
|
|
entry->x = aExtentsAppUnits.X();
|
|
entry->y = aExtentsAppUnits.Y();
|
|
entry->width = aExtentsAppUnits.Width();
|
|
entry->height = aExtentsAppUnits.Height();
|
|
}
|
|
|
|
size_t
|
|
gfxGlyphExtents::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf) const
|
|
{
|
|
return mContainedGlyphWidths.SizeOfExcludingThis(aMallocSizeOf) +
|
|
mTightGlyphExtents.SizeOfExcludingThis(nullptr, aMallocSizeOf);
|
|
}
|
|
|
|
size_t
|
|
gfxGlyphExtents::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf) const
|
|
{
|
|
return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf);
|
|
}
|
|
|
|
gfxFontGroup::gfxFontGroup(const nsAString& aFamilies, const gfxFontStyle *aStyle, gfxUserFontSet *aUserFontSet)
|
|
: mFamilies(aFamilies), mStyle(*aStyle), mUnderlineOffset(UNDERLINE_OFFSET_NOT_SET)
|
|
{
|
|
mUserFontSet = nullptr;
|
|
SetUserFontSet(aUserFontSet);
|
|
|
|
mSkipDrawing = false;
|
|
|
|
mPageLang = gfxPlatform::GetFontPrefLangFor(mStyle.language);
|
|
BuildFontList();
|
|
}
|
|
|
|
void
|
|
gfxFontGroup::BuildFontList()
|
|
{
|
|
// "#if" to be removed once all platforms are moved to gfxPlatformFontList interface
|
|
// and subclasses of gfxFontGroup eliminated
|
|
#if defined(XP_MACOSX) || defined(XP_WIN) || defined(ANDROID)
|
|
ForEachFont(FindPlatformFont, this);
|
|
|
|
if (mFonts.Length() == 0) {
|
|
bool needsBold;
|
|
gfxPlatformFontList *pfl = gfxPlatformFontList::PlatformFontList();
|
|
gfxFontEntry *defaultFont = pfl->GetDefaultFont(&mStyle, needsBold);
|
|
NS_ASSERTION(defaultFont, "invalid default font returned by GetDefaultFont");
|
|
|
|
if (defaultFont) {
|
|
nsRefPtr<gfxFont> font = defaultFont->FindOrMakeFont(&mStyle,
|
|
needsBold);
|
|
if (font) {
|
|
mFonts.AppendElement(font);
|
|
}
|
|
}
|
|
|
|
if (mFonts.Length() == 0) {
|
|
// Try for a "font of last resort...."
|
|
// Because an empty font list would be Really Bad for later code
|
|
// that assumes it will be able to get valid metrics for layout,
|
|
// just look for the first usable font and put in the list.
|
|
// (see bug 554544)
|
|
nsAutoTArray<nsRefPtr<gfxFontFamily>,200> families;
|
|
pfl->GetFontFamilyList(families);
|
|
uint32_t count = families.Length();
|
|
for (uint32_t i = 0; i < count; ++i) {
|
|
gfxFontEntry *fe = families[i]->FindFontForStyle(mStyle,
|
|
needsBold);
|
|
if (fe) {
|
|
nsRefPtr<gfxFont> font = fe->FindOrMakeFont(&mStyle,
|
|
needsBold);
|
|
if (font) {
|
|
mFonts.AppendElement(font);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if (mFonts.Length() == 0) {
|
|
// an empty font list at this point is fatal; we're not going to
|
|
// be able to do even the most basic layout operations
|
|
char msg[256]; // CHECK buffer length if revising message below
|
|
sprintf(msg, "unable to find a usable font (%.220s)",
|
|
NS_ConvertUTF16toUTF8(mFamilies).get());
|
|
NS_RUNTIMEABORT(msg);
|
|
}
|
|
}
|
|
|
|
if (!mStyle.systemFont) {
|
|
uint32_t count = mFonts.Length();
|
|
for (uint32_t i = 0; i < count; ++i) {
|
|
gfxFont* font = mFonts[i];
|
|
if (font->GetFontEntry()->mIsBadUnderlineFont) {
|
|
gfxFloat first = mFonts[0]->GetMetrics().underlineOffset;
|
|
gfxFloat bad = font->GetMetrics().underlineOffset;
|
|
mUnderlineOffset = NS_MIN(first, bad);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
#endif
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::FindPlatformFont(const nsAString& aName,
|
|
const nsACString& aGenericName,
|
|
bool aUseFontSet,
|
|
void *aClosure)
|
|
{
|
|
gfxFontGroup *fontGroup = static_cast<gfxFontGroup*>(aClosure);
|
|
const gfxFontStyle *fontStyle = fontGroup->GetStyle();
|
|
|
|
bool needsBold;
|
|
gfxFontEntry *fe = nullptr;
|
|
|
|
bool foundFamily = false;
|
|
if (aUseFontSet) {
|
|
// First, look up in the user font set...
|
|
// If the fontSet matches the family, we must not look for a platform
|
|
// font of the same name, even if we fail to actually get a fontEntry
|
|
// here; we'll fall back to the next name in the CSS font-family list.
|
|
gfxUserFontSet *fs = fontGroup->GetUserFontSet();
|
|
if (fs) {
|
|
// If the fontSet matches the family, but the font has not yet finished
|
|
// loading (nor has its load timeout fired), the fontGroup should wait
|
|
// for the download, and not actually draw its text yet.
|
|
bool waitForUserFont = false;
|
|
fe = fs->FindFontEntry(aName, *fontStyle, foundFamily,
|
|
needsBold, waitForUserFont);
|
|
if (!fe && waitForUserFont) {
|
|
fontGroup->mSkipDrawing = true;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Not known in the user font set ==> check system fonts
|
|
if (!foundFamily) {
|
|
fe = gfxPlatformFontList::PlatformFontList()->
|
|
FindFontForFamily(aName, fontStyle, needsBold);
|
|
}
|
|
|
|
// add to the font group, unless it's already there
|
|
if (fe && !fontGroup->HasFont(fe)) {
|
|
nsRefPtr<gfxFont> font = fe->FindOrMakeFont(fontStyle, needsBold);
|
|
if (font) {
|
|
fontGroup->mFonts.AppendElement(font);
|
|
}
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::HasFont(const gfxFontEntry *aFontEntry)
|
|
{
|
|
uint32_t count = mFonts.Length();
|
|
for (uint32_t i = 0; i < count; ++i) {
|
|
if (mFonts.ElementAt(i)->GetFontEntry() == aFontEntry)
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
gfxFontGroup::~gfxFontGroup() {
|
|
mFonts.Clear();
|
|
SetUserFontSet(nullptr);
|
|
}
|
|
|
|
gfxFontGroup *
|
|
gfxFontGroup::Copy(const gfxFontStyle *aStyle)
|
|
{
|
|
return new gfxFontGroup(mFamilies, aStyle, mUserFontSet);
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::IsInvalidChar(uint8_t ch)
|
|
{
|
|
return ((ch & 0x7f) < 0x20);
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::IsInvalidChar(PRUnichar ch)
|
|
{
|
|
// All printable 7-bit ASCII values are OK
|
|
if (ch >= ' ' && ch < 0x80) {
|
|
return false;
|
|
}
|
|
// No point in sending non-printing control chars through font shaping
|
|
if (ch <= 0x9f) {
|
|
return true;
|
|
}
|
|
return ((ch & 0xFF00) == 0x2000 /* Unicode control character */ &&
|
|
(ch == 0x200B/*ZWSP*/ || ch == 0x2028/*LSEP*/ || ch == 0x2029/*PSEP*/ ||
|
|
IS_BIDI_CONTROL_CHAR(ch)));
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::ForEachFont(FontCreationCallback fc,
|
|
void *closure)
|
|
{
|
|
return ForEachFontInternal(mFamilies, mStyle.language,
|
|
true, true, true, fc, closure);
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::ForEachFont(const nsAString& aFamilies,
|
|
nsIAtom *aLanguage,
|
|
FontCreationCallback fc,
|
|
void *closure)
|
|
{
|
|
return ForEachFontInternal(aFamilies, aLanguage,
|
|
false, true, true, fc, closure);
|
|
}
|
|
|
|
struct ResolveData {
|
|
ResolveData(gfxFontGroup::FontCreationCallback aCallback,
|
|
nsACString& aGenericFamily,
|
|
bool aUseFontSet,
|
|
void *aClosure) :
|
|
mCallback(aCallback),
|
|
mGenericFamily(aGenericFamily),
|
|
mUseFontSet(aUseFontSet),
|
|
mClosure(aClosure) {
|
|
}
|
|
gfxFontGroup::FontCreationCallback mCallback;
|
|
nsCString mGenericFamily;
|
|
bool mUseFontSet;
|
|
void *mClosure;
|
|
};
|
|
|
|
bool
|
|
gfxFontGroup::ForEachFontInternal(const nsAString& aFamilies,
|
|
nsIAtom *aLanguage,
|
|
bool aResolveGeneric,
|
|
bool aResolveFontName,
|
|
bool aUseFontSet,
|
|
FontCreationCallback fc,
|
|
void *closure)
|
|
{
|
|
const PRUnichar kSingleQuote = PRUnichar('\'');
|
|
const PRUnichar kDoubleQuote = PRUnichar('\"');
|
|
const PRUnichar kComma = PRUnichar(',');
|
|
|
|
nsIAtom *groupAtom = nullptr;
|
|
nsAutoCString groupString;
|
|
if (aLanguage) {
|
|
if (!gLangService) {
|
|
CallGetService(NS_LANGUAGEATOMSERVICE_CONTRACTID, &gLangService);
|
|
}
|
|
if (gLangService) {
|
|
nsresult rv;
|
|
groupAtom = gLangService->GetLanguageGroup(aLanguage, &rv);
|
|
}
|
|
}
|
|
if (!groupAtom) {
|
|
groupAtom = nsGkAtoms::Unicode;
|
|
}
|
|
groupAtom->ToUTF8String(groupString);
|
|
|
|
nsPromiseFlatString families(aFamilies);
|
|
const PRUnichar *p, *p_end;
|
|
families.BeginReading(p);
|
|
families.EndReading(p_end);
|
|
nsAutoString family;
|
|
nsAutoCString lcFamily;
|
|
nsAutoString genericFamily;
|
|
|
|
while (p < p_end) {
|
|
while (nsCRT::IsAsciiSpace(*p) || *p == kComma)
|
|
if (++p == p_end)
|
|
return true;
|
|
|
|
bool generic;
|
|
if (*p == kSingleQuote || *p == kDoubleQuote) {
|
|
// quoted font family
|
|
PRUnichar quoteMark = *p;
|
|
if (++p == p_end)
|
|
return true;
|
|
const PRUnichar *nameStart = p;
|
|
|
|
// XXX What about CSS character escapes?
|
|
while (*p != quoteMark)
|
|
if (++p == p_end)
|
|
return true;
|
|
|
|
family = Substring(nameStart, p);
|
|
generic = false;
|
|
genericFamily.SetIsVoid(true);
|
|
|
|
while (++p != p_end && *p != kComma)
|
|
/* nothing */ ;
|
|
|
|
} else {
|
|
// unquoted font family
|
|
const PRUnichar *nameStart = p;
|
|
while (++p != p_end && *p != kComma)
|
|
/* nothing */ ;
|
|
|
|
family = Substring(nameStart, p);
|
|
family.CompressWhitespace(false, true);
|
|
|
|
if (aResolveGeneric &&
|
|
(family.LowerCaseEqualsLiteral("serif") ||
|
|
family.LowerCaseEqualsLiteral("sans-serif") ||
|
|
family.LowerCaseEqualsLiteral("monospace") ||
|
|
family.LowerCaseEqualsLiteral("cursive") ||
|
|
family.LowerCaseEqualsLiteral("fantasy")))
|
|
{
|
|
generic = true;
|
|
|
|
ToLowerCase(NS_LossyConvertUTF16toASCII(family), lcFamily);
|
|
|
|
nsAutoCString prefName("font.name.");
|
|
prefName.Append(lcFamily);
|
|
prefName.AppendLiteral(".");
|
|
prefName.Append(groupString);
|
|
|
|
nsAdoptingString value = Preferences::GetString(prefName.get());
|
|
if (value) {
|
|
CopyASCIItoUTF16(lcFamily, genericFamily);
|
|
family = value;
|
|
}
|
|
} else {
|
|
generic = false;
|
|
genericFamily.SetIsVoid(true);
|
|
}
|
|
}
|
|
|
|
NS_LossyConvertUTF16toASCII gf(genericFamily);
|
|
if (generic) {
|
|
ForEachFontInternal(family, groupAtom, false,
|
|
aResolveFontName, false,
|
|
fc, closure);
|
|
} else if (!family.IsEmpty()) {
|
|
if (aResolveFontName) {
|
|
ResolveData data(fc, gf, aUseFontSet, closure);
|
|
bool aborted = false, needsBold;
|
|
nsresult rv = NS_OK;
|
|
bool foundFamily = false;
|
|
bool waitForUserFont = false;
|
|
if (aUseFontSet && mUserFontSet &&
|
|
mUserFontSet->FindFontEntry(family, mStyle, foundFamily,
|
|
needsBold, waitForUserFont))
|
|
{
|
|
gfxFontGroup::FontResolverProc(family, &data);
|
|
} else {
|
|
if (waitForUserFont) {
|
|
mSkipDrawing = true;
|
|
}
|
|
if (!foundFamily) {
|
|
gfxPlatform *pf = gfxPlatform::GetPlatform();
|
|
rv = pf->ResolveFontName(family,
|
|
gfxFontGroup::FontResolverProc,
|
|
&data, aborted);
|
|
}
|
|
}
|
|
if (NS_FAILED(rv) || aborted)
|
|
return false;
|
|
}
|
|
else {
|
|
if (!fc(family, gf, aUseFontSet, closure))
|
|
return false;
|
|
}
|
|
}
|
|
|
|
if (generic && aResolveGeneric) {
|
|
nsAutoCString prefName("font.name-list.");
|
|
prefName.Append(lcFamily);
|
|
prefName.AppendLiteral(".");
|
|
prefName.Append(groupString);
|
|
nsAdoptingString value = Preferences::GetString(prefName.get());
|
|
if (value) {
|
|
ForEachFontInternal(value, groupAtom, false,
|
|
aResolveFontName, false,
|
|
fc, closure);
|
|
}
|
|
}
|
|
|
|
++p; // may advance past p_end
|
|
}
|
|
|
|
return true;
|
|
}
|
|
|
|
bool
|
|
gfxFontGroup::FontResolverProc(const nsAString& aName, void *aClosure)
|
|
{
|
|
ResolveData *data = reinterpret_cast<ResolveData*>(aClosure);
|
|
return (data->mCallback)(aName, data->mGenericFamily, data->mUseFontSet,
|
|
data->mClosure);
|
|
}
|
|
|
|
gfxTextRun *
|
|
gfxFontGroup::MakeEmptyTextRun(const Parameters *aParams, uint32_t aFlags)
|
|
{
|
|
aFlags |= TEXT_IS_8BIT | TEXT_IS_ASCII | TEXT_IS_PERSISTENT;
|
|
return gfxTextRun::Create(aParams, 0, this, aFlags);
|
|
}
|
|
|
|
gfxTextRun *
|
|
gfxFontGroup::MakeSpaceTextRun(const Parameters *aParams, uint32_t aFlags)
|
|
{
|
|
aFlags |= TEXT_IS_8BIT | TEXT_IS_ASCII | TEXT_IS_PERSISTENT;
|
|
|
|
gfxTextRun *textRun = gfxTextRun::Create(aParams, 1, this, aFlags);
|
|
if (!textRun) {
|
|
return nullptr;
|
|
}
|
|
|
|
gfxFont *font = GetFontAt(0);
|
|
if (NS_UNLIKELY(GetStyle()->size == 0)) {
|
|
// Short-circuit for size-0 fonts, as Windows and ATSUI can't handle
|
|
// them, and always create at least size 1 fonts, i.e. they still
|
|
// render something for size 0 fonts.
|
|
textRun->AddGlyphRun(font, gfxTextRange::kFontGroup, 0, false);
|
|
}
|
|
else {
|
|
textRun->SetSpaceGlyph(font, aParams->mContext, 0);
|
|
}
|
|
|
|
// Note that the gfxGlyphExtents glyph bounds storage for the font will
|
|
// always contain an entry for the font's space glyph, so we don't have
|
|
// to call FetchGlyphExtents here.
|
|
return textRun;
|
|
}
|
|
|
|
gfxTextRun *
|
|
gfxFontGroup::MakeBlankTextRun(uint32_t aLength,
|
|
const Parameters *aParams, uint32_t aFlags)
|
|
{
|
|
gfxTextRun *textRun =
|
|
gfxTextRun::Create(aParams, aLength, this, aFlags);
|
|
if (!textRun) {
|
|
return nullptr;
|
|
}
|
|
|
|
textRun->AddGlyphRun(GetFontAt(0), gfxTextRange::kFontGroup, 0, false);
|
|
return textRun;
|
|
}
|
|
|
|
gfxTextRun *
|
|
gfxFontGroup::MakeTextRun(const uint8_t *aString, uint32_t aLength,
|
|
const Parameters *aParams, uint32_t aFlags)
|
|
{
|
|
if (aLength == 0) {
|
|
return MakeEmptyTextRun(aParams, aFlags);
|
|
}
|
|
if (aLength == 1 && aString[0] == ' ') {
|
|
return MakeSpaceTextRun(aParams, aFlags);
|
|
}
|
|
|
|
aFlags |= TEXT_IS_8BIT;
|
|
|
|
if (GetStyle()->size == 0) {
|
|
// Short-circuit for size-0 fonts, as Windows and ATSUI can't handle
|
|
// them, and always create at least size 1 fonts, i.e. they still
|
|
// render something for size 0 fonts.
|
|
return MakeBlankTextRun(aLength, aParams, aFlags);
|
|
}
|
|
|
|
gfxTextRun *textRun = gfxTextRun::Create(aParams, aLength,
|
|
this, aFlags);
|
|
if (!textRun) {
|
|
return nullptr;
|
|
}
|
|
|
|
InitTextRun(aParams->mContext, textRun, aString, aLength);
|
|
|
|
textRun->FetchGlyphExtents(aParams->mContext);
|
|
|
|
return textRun;
|
|
}
|
|
|
|
gfxTextRun *
|
|
gfxFontGroup::MakeTextRun(const PRUnichar *aString, uint32_t aLength,
|
|
const Parameters *aParams, uint32_t aFlags)
|
|
{
|
|
if (aLength == 0) {
|
|
return MakeEmptyTextRun(aParams, aFlags);
|
|
}
|
|
if (aLength == 1 && aString[0] == ' ') {
|
|
return MakeSpaceTextRun(aParams, aFlags);
|
|
}
|
|
if (GetStyle()->size == 0) {
|
|
return MakeBlankTextRun(aLength, aParams, aFlags);
|
|
}
|
|
|
|
gfxTextRun *textRun = gfxTextRun::Create(aParams, aLength,
|
|
this, aFlags);
|
|
if (!textRun) {
|
|
return nullptr;
|
|
}
|
|
|
|
InitTextRun(aParams->mContext, textRun, aString, aLength);
|
|
|
|
textRun->FetchGlyphExtents(aParams->mContext);
|
|
|
|
return textRun;
|
|
}
|
|
|
|
template<typename T>
|
|
void
|
|
gfxFontGroup::InitTextRun(gfxContext *aContext,
|
|
gfxTextRun *aTextRun,
|
|
const T *aString,
|
|
uint32_t aLength)
|
|
{
|
|
NS_ASSERTION(aLength > 0, "don't call InitTextRun for a zero-length run");
|
|
|
|
// we need to do numeral processing even on 8-bit text,
|
|
// in case we're converting Western to Hindi/Arabic digits
|
|
int32_t numOption = gfxPlatform::GetPlatform()->GetBidiNumeralOption();
|
|
nsAutoArrayPtr<PRUnichar> transformedString;
|
|
if (numOption != IBMBIDI_NUMERAL_NOMINAL) {
|
|
// scan the string for numerals that may need to be transformed;
|
|
// if we find any, we'll make a local copy here and use that for
|
|
// font matching and glyph generation/shaping
|
|
bool prevIsArabic =
|
|
(aTextRun->GetFlags() & gfxTextRunFactory::TEXT_INCOMING_ARABICCHAR) != 0;
|
|
for (uint32_t i = 0; i < aLength; ++i) {
|
|
PRUnichar origCh = aString[i];
|
|
PRUnichar newCh = HandleNumberInChar(origCh, prevIsArabic, numOption);
|
|
if (newCh != origCh) {
|
|
if (!transformedString) {
|
|
transformedString = new PRUnichar[aLength];
|
|
if (sizeof(T) == sizeof(PRUnichar)) {
|
|
memcpy(transformedString.get(), aString, i * sizeof(PRUnichar));
|
|
} else {
|
|
for (uint32_t j = 0; j < i; ++j) {
|
|
transformedString[j] = aString[j];
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (transformedString) {
|
|
transformedString[i] = newCh;
|
|
}
|
|
prevIsArabic = IS_ARABIC_CHAR(newCh);
|
|
}
|
|
}
|
|
|
|
if (sizeof(T) == sizeof(uint8_t) && !transformedString) {
|
|
// the text is still purely 8-bit; bypass the script-run itemizer
|
|
// and treat it as a single Latin run
|
|
InitScriptRun(aContext, aTextRun, aString,
|
|
0, aLength, MOZ_SCRIPT_LATIN);
|
|
} else {
|
|
const PRUnichar *textPtr;
|
|
if (transformedString) {
|
|
textPtr = transformedString.get();
|
|
} else {
|
|
// typecast to avoid compilation error for the 8-bit version,
|
|
// even though this is dead code in that case
|
|
textPtr = reinterpret_cast<const PRUnichar*>(aString);
|
|
}
|
|
|
|
// split into script runs so that script can potentially influence
|
|
// the font matching process below
|
|
gfxScriptItemizer scriptRuns(textPtr, aLength);
|
|
|
|
#ifdef PR_LOGGING
|
|
PRLogModuleInfo *log = (mStyle.systemFont ?
|
|
gfxPlatform::GetLog(eGfxLog_textrunui) :
|
|
gfxPlatform::GetLog(eGfxLog_textrun));
|
|
#endif
|
|
|
|
uint32_t runStart = 0, runLimit = aLength;
|
|
int32_t runScript = MOZ_SCRIPT_LATIN;
|
|
while (scriptRuns.Next(runStart, runLimit, runScript)) {
|
|
|
|
#ifdef PR_LOGGING
|
|
if (NS_UNLIKELY(log)) {
|
|
nsAutoCString lang;
|
|
mStyle.language->ToUTF8String(lang);
|
|
uint32_t runLen = runLimit - runStart;
|
|
PR_LOG(log, PR_LOG_WARNING,\
|
|
("(%s) fontgroup: [%s] lang: %s script: %d len %d "
|
|
"weight: %d width: %d style: %s "
|
|
"TEXTRUN [%s] ENDTEXTRUN\n",
|
|
(mStyle.systemFont ? "textrunui" : "textrun"),
|
|
NS_ConvertUTF16toUTF8(mFamilies).get(),
|
|
lang.get(), runScript, runLen,
|
|
uint32_t(mStyle.weight), uint32_t(mStyle.stretch),
|
|
(mStyle.style & NS_FONT_STYLE_ITALIC ? "italic" :
|
|
(mStyle.style & NS_FONT_STYLE_OBLIQUE ? "oblique" :
|
|
"normal")),
|
|
NS_ConvertUTF16toUTF8(textPtr + runStart, runLen).get()));
|
|
}
|
|
#endif
|
|
|
|
InitScriptRun(aContext, aTextRun, textPtr,
|
|
runStart, runLimit, runScript);
|
|
}
|
|
}
|
|
|
|
if (sizeof(T) == sizeof(PRUnichar) && aLength > 0) {
|
|
gfxTextRun::CompressedGlyph *glyph = aTextRun->GetCharacterGlyphs();
|
|
if (!glyph->IsSimpleGlyph()) {
|
|
glyph->SetClusterStart(true);
|
|
}
|
|
}
|
|
|
|
// It's possible for CoreText to omit glyph runs if it decides they contain
|
|
// only invisibles (e.g., U+FEFF, see reftest 474417-1). In this case, we
|
|
// need to eliminate them from the glyph run array to avoid drawing "partial
|
|
// ligatures" with the wrong font.
|
|
// We don't do this during InitScriptRun (or gfxFont::InitTextRun) because
|
|
// it will iterate back over all glyphruns in the textrun, which leads to
|
|
// pathologically-bad perf in the case where a textrun contains many script
|
|
// changes (see bug 680402) - we'd end up re-sanitizing all the earlier runs
|
|
// every time a new script subrun is processed.
|
|
aTextRun->SanitizeGlyphRuns();
|
|
|
|
aTextRun->SortGlyphRuns();
|
|
}
|
|
|
|
template<typename T>
|
|
void
|
|
gfxFontGroup::InitScriptRun(gfxContext *aContext,
|
|
gfxTextRun *aTextRun,
|
|
const T *aString,
|
|
uint32_t aScriptRunStart,
|
|
uint32_t aScriptRunEnd,
|
|
int32_t aRunScript)
|
|
{
|
|
NS_ASSERTION(aScriptRunEnd > aScriptRunStart,
|
|
"don't call InitScriptRun for a zero-length run");
|
|
|
|
gfxFont *mainFont = GetFontAt(0);
|
|
|
|
uint32_t runStart = aScriptRunStart;
|
|
nsAutoTArray<gfxTextRange,3> fontRanges;
|
|
ComputeRanges(fontRanges, aString + aScriptRunStart,
|
|
aScriptRunEnd - aScriptRunStart, aRunScript);
|
|
uint32_t numRanges = fontRanges.Length();
|
|
|
|
for (uint32_t r = 0; r < numRanges; r++) {
|
|
const gfxTextRange& range = fontRanges[r];
|
|
uint32_t matchedLength = range.Length();
|
|
gfxFont *matchedFont = (range.font ? range.font.get() : nullptr);
|
|
|
|
// create the glyph run for this range
|
|
if (matchedFont) {
|
|
aTextRun->AddGlyphRun(matchedFont, range.matchType,
|
|
runStart, (matchedLength > 0));
|
|
// do glyph layout and record the resulting positioned glyphs
|
|
if (!matchedFont->SplitAndInitTextRun(aContext, aTextRun, aString,
|
|
runStart, matchedLength,
|
|
aRunScript)) {
|
|
// glyph layout failed! treat as missing glyphs
|
|
matchedFont = nullptr;
|
|
}
|
|
} else {
|
|
aTextRun->AddGlyphRun(mainFont, gfxTextRange::kFontGroup,
|
|
runStart, (matchedLength > 0));
|
|
}
|
|
|
|
if (!matchedFont) {
|
|
// for PRUnichar text, we need to set cluster boundaries so that
|
|
// surrogate pairs, combining characters, etc behave properly,
|
|
// even if we don't have glyphs for them
|
|
if (sizeof(T) == sizeof(PRUnichar)) {
|
|
gfxShapedWord::SetupClusterBoundaries(aTextRun->GetCharacterGlyphs() + runStart,
|
|
reinterpret_cast<const PRUnichar*>(aString) + runStart,
|
|
matchedLength);
|
|
}
|
|
|
|
// various "missing" characters may need special handling,
|
|
// so we check for them here
|
|
uint32_t runLimit = runStart + matchedLength;
|
|
for (uint32_t index = runStart; index < runLimit; index++) {
|
|
T ch = aString[index];
|
|
|
|
// tab and newline are not to be displayed as hexboxes,
|
|
// but do need to be recorded in the textrun
|
|
if (ch == '\n') {
|
|
aTextRun->SetIsNewline(index);
|
|
continue;
|
|
}
|
|
if (ch == '\t') {
|
|
aTextRun->SetIsTab(index);
|
|
continue;
|
|
}
|
|
|
|
// for 16-bit textruns only, check for surrogate pairs and
|
|
// special Unicode spaces; omit these checks in 8-bit runs
|
|
if (sizeof(T) == sizeof(PRUnichar)) {
|
|
if (NS_IS_HIGH_SURROGATE(ch) &&
|
|
index + 1 < aScriptRunEnd &&
|
|
NS_IS_LOW_SURROGATE(aString[index + 1]))
|
|
{
|
|
aTextRun->SetMissingGlyph(index,
|
|
SURROGATE_TO_UCS4(ch,
|
|
aString[index + 1]));
|
|
index++;
|
|
aTextRun->SetIsLowSurrogate(index);
|
|
continue;
|
|
}
|
|
|
|
// check if this is a known Unicode whitespace character that
|
|
// we can render using the space glyph with a custom width
|
|
gfxFloat wid = mainFont->SynthesizeSpaceWidth(ch);
|
|
if (wid >= 0.0) {
|
|
nscoord advance =
|
|
aTextRun->GetAppUnitsPerDevUnit() * floor(wid + 0.5);
|
|
gfxTextRun::CompressedGlyph g;
|
|
if (gfxTextRun::CompressedGlyph::IsSimpleAdvance(advance)) {
|
|
aTextRun->SetSimpleGlyph(index,
|
|
g.SetSimpleGlyph(advance,
|
|
mainFont->GetSpaceGlyph()));
|
|
} else {
|
|
gfxTextRun::DetailedGlyph detailedGlyph;
|
|
detailedGlyph.mGlyphID = mainFont->GetSpaceGlyph();
|
|
detailedGlyph.mAdvance = advance;
|
|
detailedGlyph.mXOffset = detailedGlyph.mYOffset = 0;
|
|
g.SetComplex(true, true, 1);
|
|
aTextRun->SetGlyphs(index,
|
|
g, &detailedGlyph);
|
|
}
|
|
continue;
|
|
}
|
|
}
|
|
|
|
if (IsInvalidChar(ch)) {
|
|
// invalid chars are left as zero-width/invisible
|
|
continue;
|
|
}
|
|
|
|
// record char code so we can draw a box with the Unicode value
|
|
aTextRun->SetMissingGlyph(index, ch);
|
|
}
|
|
}
|
|
|
|
runStart += matchedLength;
|
|
}
|
|
}
|
|
|
|
already_AddRefed<gfxFont>
|
|
gfxFontGroup::TryOtherFamilyMembers(gfxFont* aFont, uint32_t aCh)
|
|
{
|
|
gfxFontFamily *family = aFont->GetFontEntry()->Family();
|
|
if (family && !aFont->GetFontEntry()->mIsProxy &&
|
|
family->TestCharacterMap(aCh)) {
|
|
// Note that we don't need the actual runScript in matchData for
|
|
// gfxFontFamily::SearchAllFontsForChar, it's only used for the
|
|
// system-fallback case. So we can just set it to 0 here.
|
|
GlobalFontMatch matchData(aCh, 0, &mStyle);
|
|
family->SearchAllFontsForChar(&matchData);
|
|
gfxFontEntry *fe = matchData.mBestMatch;
|
|
if (fe) {
|
|
bool needsBold = aFont->GetStyle()->weight >= 600 && !fe->IsBold();
|
|
nsRefPtr<gfxFont> font = fe->FindOrMakeFont(&mStyle, needsBold);
|
|
if (font) {
|
|
return font.forget();
|
|
}
|
|
}
|
|
}
|
|
return nullptr;
|
|
}
|
|
|
|
already_AddRefed<gfxFont>
|
|
gfxFontGroup::FindFontForChar(uint32_t aCh, uint32_t aPrevCh,
|
|
int32_t aRunScript, gfxFont *aPrevMatchedFont,
|
|
uint8_t *aMatchType)
|
|
{
|
|
// To optimize common cases, try the first font in the font-group
|
|
// before going into the more detailed checks below
|
|
uint32_t nextIndex = 0;
|
|
bool isJoinControl = gfxFontUtils::IsJoinControl(aCh);
|
|
bool wasJoinCauser = gfxFontUtils::IsJoinCauser(aPrevCh);
|
|
bool isVarSelector = gfxFontUtils::IsVarSelector(aCh);
|
|
|
|
if (!isJoinControl && !wasJoinCauser && !isVarSelector) {
|
|
gfxFont *firstFont = mFonts[0];
|
|
if (firstFont->HasCharacter(aCh)) {
|
|
*aMatchType = gfxTextRange::kFontGroup;
|
|
firstFont->AddRef();
|
|
return firstFont;
|
|
}
|
|
// It's possible that another font in the family (e.g. regular face,
|
|
// where the requested style was italic) will support the character
|
|
nsRefPtr<gfxFont> font = TryOtherFamilyMembers(firstFont, aCh);
|
|
if (font) {
|
|
*aMatchType = gfxTextRange::kFontGroup;
|
|
return font.forget();
|
|
}
|
|
// we don't need to check the first font again below
|
|
++nextIndex;
|
|
}
|
|
|
|
if (aPrevMatchedFont) {
|
|
// Don't switch fonts for control characters, regardless of
|
|
// whether they are present in the current font, as they won't
|
|
// actually be rendered (see bug 716229)
|
|
uint8_t category = GetGeneralCategory(aCh);
|
|
if (category == HB_UNICODE_GENERAL_CATEGORY_CONTROL) {
|
|
aPrevMatchedFont->AddRef();
|
|
return aPrevMatchedFont;
|
|
}
|
|
|
|
// if this character is a join-control or the previous is a join-causer,
|
|
// use the same font as the previous range if we can
|
|
if (isJoinControl || wasJoinCauser) {
|
|
if (aPrevMatchedFont->HasCharacter(aCh)) {
|
|
aPrevMatchedFont->AddRef();
|
|
return aPrevMatchedFont;
|
|
}
|
|
}
|
|
}
|
|
|
|
// if this character is a variation selector,
|
|
// use the previous font regardless of whether it supports VS or not.
|
|
// otherwise the text run will be divided.
|
|
if (isVarSelector) {
|
|
if (aPrevMatchedFont) {
|
|
aPrevMatchedFont->AddRef();
|
|
return aPrevMatchedFont;
|
|
}
|
|
// VS alone. it's meaningless to search different fonts
|
|
return nullptr;
|
|
}
|
|
|
|
// 1. check remaining fonts in the font group
|
|
uint32_t fontListLength = FontListLength();
|
|
for (uint32_t i = nextIndex; i < fontListLength; i++) {
|
|
nsRefPtr<gfxFont> font = mFonts[i];
|
|
if (font->HasCharacter(aCh)) {
|
|
*aMatchType = gfxTextRange::kFontGroup;
|
|
return font.forget();
|
|
}
|
|
|
|
font = TryOtherFamilyMembers(font, aCh);
|
|
if (font) {
|
|
*aMatchType = gfxTextRange::kFontGroup;
|
|
return font.forget();
|
|
}
|
|
}
|
|
|
|
// if character is in Private Use Area, don't do matching against pref or system fonts
|
|
if ((aCh >= 0xE000 && aCh <= 0xF8FF) || (aCh >= 0xF0000 && aCh <= 0x10FFFD))
|
|
return nullptr;
|
|
|
|
// 2. search pref fonts
|
|
nsRefPtr<gfxFont> font = WhichPrefFontSupportsChar(aCh);
|
|
if (font) {
|
|
*aMatchType = gfxTextRange::kPrefsFallback;
|
|
return font.forget();
|
|
}
|
|
|
|
// 3. use fallback fonts
|
|
// -- before searching for something else check the font used for the previous character
|
|
if (aPrevMatchedFont && aPrevMatchedFont->HasCharacter(aCh)) {
|
|
*aMatchType = gfxTextRange::kSystemFallback;
|
|
aPrevMatchedFont->AddRef();
|
|
return aPrevMatchedFont;
|
|
}
|
|
|
|
// never fall back for characters from unknown scripts
|
|
if (aRunScript == HB_SCRIPT_UNKNOWN) {
|
|
return nullptr;
|
|
}
|
|
|
|
// for known "space" characters, don't do a full system-fallback search;
|
|
// we'll synthesize appropriate-width spaces instead of missing-glyph boxes
|
|
if (GetGeneralCategory(aCh) ==
|
|
HB_UNICODE_GENERAL_CATEGORY_SPACE_SEPARATOR &&
|
|
GetFontAt(0)->SynthesizeSpaceWidth(aCh) >= 0.0)
|
|
{
|
|
return nullptr;
|
|
}
|
|
|
|
// -- otherwise look for other stuff
|
|
*aMatchType = gfxTextRange::kSystemFallback;
|
|
font = WhichSystemFontSupportsChar(aCh, aRunScript);
|
|
return font.forget();
|
|
}
|
|
|
|
template<typename T>
|
|
void gfxFontGroup::ComputeRanges(nsTArray<gfxTextRange>& aRanges,
|
|
const T *aString, uint32_t aLength,
|
|
int32_t aRunScript)
|
|
{
|
|
NS_ASSERTION(aRanges.Length() == 0, "aRanges must be initially empty");
|
|
NS_ASSERTION(aLength > 0, "don't call ComputeRanges for zero-length text");
|
|
|
|
uint32_t prevCh = 0;
|
|
uint8_t matchType = 0;
|
|
int32_t lastRangeIndex = -1;
|
|
|
|
// initialize prevFont to the group's primary font, so that this will be
|
|
// used for string-initial control chars, etc rather than risk hitting font
|
|
// fallback for these (bug 716229)
|
|
gfxFont *prevFont = GetFontAt(0);
|
|
|
|
for (uint32_t i = 0; i < aLength; i++) {
|
|
|
|
const uint32_t origI = i; // save off in case we increase for surrogate
|
|
|
|
// set up current ch
|
|
uint32_t ch = aString[i];
|
|
|
|
// in 16-bit case only, check for surrogate pair
|
|
if (sizeof(T) == sizeof(PRUnichar)) {
|
|
if ((i + 1 < aLength) && NS_IS_HIGH_SURROGATE(ch) &&
|
|
NS_IS_LOW_SURROGATE(aString[i + 1])) {
|
|
i++;
|
|
ch = SURROGATE_TO_UCS4(ch, aString[i]);
|
|
}
|
|
}
|
|
|
|
if (ch == 0xa0) {
|
|
ch = ' ';
|
|
}
|
|
|
|
// find the font for this char
|
|
nsRefPtr<gfxFont> font =
|
|
FindFontForChar(ch, prevCh, aRunScript, prevFont, &matchType);
|
|
|
|
prevCh = ch;
|
|
|
|
if (lastRangeIndex == -1) {
|
|
// first char ==> make a new range
|
|
aRanges.AppendElement(gfxTextRange(0, 1, font, matchType));
|
|
lastRangeIndex++;
|
|
prevFont = font;
|
|
} else {
|
|
// if font has changed, make a new range
|
|
gfxTextRange& prevRange = aRanges[lastRangeIndex];
|
|
if (prevRange.font != font || prevRange.matchType != matchType) {
|
|
// close out the previous range
|
|
prevRange.end = origI;
|
|
aRanges.AppendElement(gfxTextRange(origI, i + 1,
|
|
font, matchType));
|
|
lastRangeIndex++;
|
|
|
|
// update prevFont for the next match, *unless* we switched
|
|
// fonts on a ZWJ, in which case propagating the changed font
|
|
// is probably not a good idea (see bug 619511)
|
|
if (sizeof(T) == sizeof(uint8_t) ||
|
|
!gfxFontUtils::IsJoinCauser(ch))
|
|
{
|
|
prevFont = font;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
aRanges[lastRangeIndex].end = aLength;
|
|
}
|
|
|
|
gfxUserFontSet*
|
|
gfxFontGroup::GetUserFontSet()
|
|
{
|
|
return mUserFontSet;
|
|
}
|
|
|
|
void
|
|
gfxFontGroup::SetUserFontSet(gfxUserFontSet *aUserFontSet)
|
|
{
|
|
NS_IF_RELEASE(mUserFontSet);
|
|
mUserFontSet = aUserFontSet;
|
|
NS_IF_ADDREF(mUserFontSet);
|
|
mCurrGeneration = GetGeneration();
|
|
}
|
|
|
|
uint64_t
|
|
gfxFontGroup::GetGeneration()
|
|
{
|
|
if (!mUserFontSet)
|
|
return 0;
|
|
return mUserFontSet->GetGeneration();
|
|
}
|
|
|
|
void
|
|
gfxFontGroup::UpdateFontList()
|
|
{
|
|
if (mUserFontSet && mCurrGeneration != GetGeneration()) {
|
|
// xxx - can probably improve this to detect when all fonts were found, so no need to update list
|
|
mFonts.Clear();
|
|
mUnderlineOffset = UNDERLINE_OFFSET_NOT_SET;
|
|
mSkipDrawing = false;
|
|
|
|
// bug 548184 - need to clean up FT2, OS/2 platform code to use BuildFontList
|
|
#if defined(XP_MACOSX) || defined(XP_WIN) || defined(ANDROID)
|
|
BuildFontList();
|
|
#else
|
|
ForEachFont(FindPlatformFont, this);
|
|
#endif
|
|
mCurrGeneration = GetGeneration();
|
|
}
|
|
}
|
|
|
|
struct PrefFontCallbackData {
|
|
PrefFontCallbackData(nsTArray<nsRefPtr<gfxFontFamily> >& aFamiliesArray)
|
|
: mPrefFamilies(aFamiliesArray)
|
|
{}
|
|
|
|
nsTArray<nsRefPtr<gfxFontFamily> >& mPrefFamilies;
|
|
|
|
static bool AddFontFamilyEntry(eFontPrefLang aLang, const nsAString& aName, void *aClosure)
|
|
{
|
|
PrefFontCallbackData *prefFontData = static_cast<PrefFontCallbackData*>(aClosure);
|
|
|
|
gfxFontFamily *family = gfxPlatformFontList::PlatformFontList()->FindFamily(aName);
|
|
if (family) {
|
|
prefFontData->mPrefFamilies.AppendElement(family);
|
|
}
|
|
return true;
|
|
}
|
|
};
|
|
|
|
already_AddRefed<gfxFont>
|
|
gfxFontGroup::WhichPrefFontSupportsChar(uint32_t aCh)
|
|
{
|
|
gfxFont *font;
|
|
|
|
// get the pref font list if it hasn't been set up already
|
|
uint32_t unicodeRange = FindCharUnicodeRange(aCh);
|
|
eFontPrefLang charLang = gfxPlatform::GetPlatform()->GetFontPrefLangFor(unicodeRange);
|
|
|
|
// if the last pref font was the first family in the pref list, no need to recheck through a list of families
|
|
if (mLastPrefFont && charLang == mLastPrefLang &&
|
|
mLastPrefFirstFont && mLastPrefFont->HasCharacter(aCh)) {
|
|
font = mLastPrefFont;
|
|
NS_ADDREF(font);
|
|
return font;
|
|
}
|
|
|
|
// based on char lang and page lang, set up list of pref lang fonts to check
|
|
eFontPrefLang prefLangs[kMaxLenPrefLangList];
|
|
uint32_t i, numLangs = 0;
|
|
|
|
gfxPlatform::GetPlatform()->GetLangPrefs(prefLangs, numLangs, charLang, mPageLang);
|
|
|
|
for (i = 0; i < numLangs; i++) {
|
|
nsAutoTArray<nsRefPtr<gfxFontFamily>, 5> families;
|
|
eFontPrefLang currentLang = prefLangs[i];
|
|
|
|
gfxPlatformFontList *fontList = gfxPlatformFontList::PlatformFontList();
|
|
|
|
// get the pref families for a single pref lang
|
|
if (!fontList->GetPrefFontFamilyEntries(currentLang, &families)) {
|
|
eFontPrefLang prefLangsToSearch[1] = { currentLang };
|
|
PrefFontCallbackData prefFontData(families);
|
|
gfxPlatform::ForEachPrefFont(prefLangsToSearch, 1, PrefFontCallbackData::AddFontFamilyEntry,
|
|
&prefFontData);
|
|
fontList->SetPrefFontFamilyEntries(currentLang, families);
|
|
}
|
|
|
|
// find the first pref font that includes the character
|
|
uint32_t j, numPrefs;
|
|
numPrefs = families.Length();
|
|
for (j = 0; j < numPrefs; j++) {
|
|
// look up the appropriate face
|
|
gfxFontFamily *family = families[j];
|
|
if (!family) continue;
|
|
|
|
// if a pref font is used, it's likely to be used again in the same text run.
|
|
// the style doesn't change so the face lookup can be cached rather than calling
|
|
// FindOrMakeFont repeatedly. speeds up FindFontForChar lookup times for subsequent
|
|
// pref font lookups
|
|
if (family == mLastPrefFamily && mLastPrefFont->HasCharacter(aCh)) {
|
|
font = mLastPrefFont;
|
|
NS_ADDREF(font);
|
|
return font;
|
|
}
|
|
|
|
bool needsBold;
|
|
gfxFontEntry *fe = family->FindFontForStyle(mStyle, needsBold);
|
|
// if ch in cmap, create and return a gfxFont
|
|
if (fe && fe->TestCharacterMap(aCh)) {
|
|
nsRefPtr<gfxFont> prefFont = fe->FindOrMakeFont(&mStyle, needsBold);
|
|
if (!prefFont) continue;
|
|
mLastPrefFamily = family;
|
|
mLastPrefFont = prefFont;
|
|
mLastPrefLang = charLang;
|
|
mLastPrefFirstFont = (i == 0 && j == 0);
|
|
return prefFont.forget();
|
|
}
|
|
|
|
}
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
already_AddRefed<gfxFont>
|
|
gfxFontGroup::WhichSystemFontSupportsChar(uint32_t aCh, int32_t aRunScript)
|
|
{
|
|
gfxFontEntry *fe =
|
|
gfxPlatformFontList::PlatformFontList()->
|
|
SystemFindFontForChar(aCh, aRunScript, &mStyle);
|
|
if (fe) {
|
|
bool wantBold = mStyle.ComputeWeight() >= 6;
|
|
nsRefPtr<gfxFont> font =
|
|
fe->FindOrMakeFont(&mStyle, wantBold && !fe->IsBold());
|
|
return font.forget();
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
/*static*/ void
|
|
gfxFontGroup::Shutdown()
|
|
{
|
|
NS_IF_RELEASE(gLangService);
|
|
}
|
|
|
|
nsILanguageAtomService* gfxFontGroup::gLangService = nullptr;
|
|
|
|
|
|
#define DEFAULT_PIXEL_FONT_SIZE 16.0f
|
|
|
|
/*static*/ uint32_t
|
|
gfxFontStyle::ParseFontLanguageOverride(const nsString& aLangTag)
|
|
{
|
|
if (!aLangTag.Length() || aLangTag.Length() > 4) {
|
|
return NO_FONT_LANGUAGE_OVERRIDE;
|
|
}
|
|
uint32_t index, result = 0;
|
|
for (index = 0; index < aLangTag.Length(); ++index) {
|
|
PRUnichar ch = aLangTag[index];
|
|
if (!nsCRT::IsAscii(ch)) { // valid tags are pure ASCII
|
|
return NO_FONT_LANGUAGE_OVERRIDE;
|
|
}
|
|
result = (result << 8) + ch;
|
|
}
|
|
while (index++ < 4) {
|
|
result = (result << 8) + 0x20;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
gfxFontStyle::gfxFontStyle() :
|
|
language(nsGkAtoms::x_western),
|
|
size(DEFAULT_PIXEL_FONT_SIZE), sizeAdjust(0.0f),
|
|
languageOverride(NO_FONT_LANGUAGE_OVERRIDE),
|
|
weight(NS_FONT_WEIGHT_NORMAL), stretch(NS_FONT_STRETCH_NORMAL),
|
|
systemFont(true), printerFont(false),
|
|
style(NS_FONT_STYLE_NORMAL)
|
|
{
|
|
}
|
|
|
|
gfxFontStyle::gfxFontStyle(uint8_t aStyle, uint16_t aWeight, int16_t aStretch,
|
|
gfxFloat aSize, nsIAtom *aLanguage,
|
|
float aSizeAdjust, bool aSystemFont,
|
|
bool aPrinterFont,
|
|
const nsString& aLanguageOverride):
|
|
language(aLanguage),
|
|
size(aSize), sizeAdjust(aSizeAdjust),
|
|
languageOverride(ParseFontLanguageOverride(aLanguageOverride)),
|
|
weight(aWeight), stretch(aStretch),
|
|
systemFont(aSystemFont), printerFont(aPrinterFont),
|
|
style(aStyle)
|
|
{
|
|
if (weight > 900)
|
|
weight = 900;
|
|
if (weight < 100)
|
|
weight = 100;
|
|
|
|
if (size >= FONT_MAX_SIZE) {
|
|
size = FONT_MAX_SIZE;
|
|
sizeAdjust = 0.0;
|
|
} else if (size < 0.0) {
|
|
NS_WARNING("negative font size");
|
|
size = 0.0;
|
|
}
|
|
|
|
if (!language) {
|
|
NS_WARNING("null language");
|
|
language = nsGkAtoms::x_western;
|
|
}
|
|
}
|
|
|
|
gfxFontStyle::gfxFontStyle(const gfxFontStyle& aStyle) :
|
|
language(aStyle.language),
|
|
size(aStyle.size), sizeAdjust(aStyle.sizeAdjust),
|
|
languageOverride(aStyle.languageOverride),
|
|
weight(aStyle.weight), stretch(aStyle.stretch),
|
|
systemFont(aStyle.systemFont), printerFont(aStyle.printerFont),
|
|
style(aStyle.style)
|
|
{
|
|
featureSettings.AppendElements(aStyle.featureSettings);
|
|
}
|
|
|
|
int8_t
|
|
gfxFontStyle::ComputeWeight() const
|
|
{
|
|
int8_t baseWeight = (weight + 50) / 100;
|
|
|
|
if (baseWeight < 0)
|
|
baseWeight = 0;
|
|
if (baseWeight > 9)
|
|
baseWeight = 9;
|
|
|
|
return baseWeight;
|
|
}
|
|
|
|
// This is not a member function of gfxShapedWord because it is also used
|
|
// by gfxFontGroup on missing-glyph runs, where we don't actually "shape"
|
|
// anything but still need to set cluster info.
|
|
/*static*/ void
|
|
gfxShapedWord::SetupClusterBoundaries(CompressedGlyph *aGlyphs,
|
|
const PRUnichar *aString, uint32_t aLength)
|
|
{
|
|
gfxTextRun::CompressedGlyph extendCluster;
|
|
extendCluster.SetComplex(false, true, 0);
|
|
|
|
ClusterIterator iter(aString, aLength);
|
|
|
|
// the ClusterIterator won't be able to tell us if the string
|
|
// _begins_ with a cluster-extender, so we handle that here
|
|
if (aLength && IsClusterExtender(*aString)) {
|
|
*aGlyphs = extendCluster;
|
|
}
|
|
|
|
while (!iter.AtEnd()) {
|
|
// advance iter to the next cluster-start (or end of text)
|
|
iter.Next();
|
|
// step past the first char of the cluster
|
|
aString++;
|
|
aGlyphs++;
|
|
// mark all the rest as cluster-continuations
|
|
while (aString < iter) {
|
|
*aGlyphs++ = extendCluster;
|
|
aString++;
|
|
}
|
|
}
|
|
}
|
|
|
|
gfxShapedWord::DetailedGlyph *
|
|
gfxShapedWord::AllocateDetailedGlyphs(uint32_t aIndex, uint32_t aCount)
|
|
{
|
|
NS_ASSERTION(aIndex < Length(), "Index out of range");
|
|
|
|
if (!mDetailedGlyphs) {
|
|
mDetailedGlyphs = new DetailedGlyphStore();
|
|
}
|
|
|
|
DetailedGlyph *details = mDetailedGlyphs->Allocate(aIndex, aCount);
|
|
if (!details) {
|
|
mCharacterGlyphs[aIndex].SetMissing(0);
|
|
return nullptr;
|
|
}
|
|
|
|
return details;
|
|
}
|
|
|
|
void
|
|
gfxShapedWord::SetGlyphs(uint32_t aIndex, CompressedGlyph aGlyph,
|
|
const DetailedGlyph *aGlyphs)
|
|
{
|
|
NS_ASSERTION(!aGlyph.IsSimpleGlyph(), "Simple glyphs not handled here");
|
|
NS_ASSERTION(aIndex > 0 || aGlyph.IsLigatureGroupStart(),
|
|
"First character can't be a ligature continuation!");
|
|
|
|
uint32_t glyphCount = aGlyph.GetGlyphCount();
|
|
if (glyphCount > 0) {
|
|
DetailedGlyph *details = AllocateDetailedGlyphs(aIndex, glyphCount);
|
|
if (!details) {
|
|
return;
|
|
}
|
|
memcpy(details, aGlyphs, sizeof(DetailedGlyph)*glyphCount);
|
|
}
|
|
mCharacterGlyphs[aIndex] = aGlyph;
|
|
}
|
|
|
|
#define ZWNJ 0x200C
|
|
#define ZWJ 0x200D
|
|
static inline bool
|
|
IsDefaultIgnorable(uint32_t aChar)
|
|
{
|
|
return GetIdentifierModification(aChar) == XIDMOD_DEFAULT_IGNORABLE ||
|
|
aChar == ZWNJ || aChar == ZWJ;
|
|
}
|
|
|
|
void
|
|
gfxShapedWord::SetMissingGlyph(uint32_t aIndex, uint32_t aChar, gfxFont *aFont)
|
|
{
|
|
DetailedGlyph *details = AllocateDetailedGlyphs(aIndex, 1);
|
|
if (!details) {
|
|
return;
|
|
}
|
|
|
|
details->mGlyphID = aChar;
|
|
if (IsDefaultIgnorable(aChar)) {
|
|
// Setting advance width to zero will prevent drawing the hexbox
|
|
details->mAdvance = 0;
|
|
} else {
|
|
gfxFloat width = NS_MAX(aFont->GetMetrics().aveCharWidth,
|
|
gfxFontMissingGlyphs::GetDesiredMinWidth(aChar));
|
|
details->mAdvance = uint32_t(width * mAppUnitsPerDevUnit);
|
|
}
|
|
details->mXOffset = 0;
|
|
details->mYOffset = 0;
|
|
mCharacterGlyphs[aIndex].SetMissing(1);
|
|
}
|
|
|
|
bool
|
|
gfxShapedWord::FilterIfIgnorable(uint32_t aIndex)
|
|
{
|
|
uint32_t ch = GetCharAt(aIndex);
|
|
if (IsDefaultIgnorable(ch)) {
|
|
DetailedGlyph *details = AllocateDetailedGlyphs(aIndex, 1);
|
|
if (details) {
|
|
details->mGlyphID = ch;
|
|
details->mAdvance = 0;
|
|
details->mXOffset = 0;
|
|
details->mYOffset = 0;
|
|
mCharacterGlyphs[aIndex].SetMissing(1);
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
void
|
|
gfxShapedWord::AdjustAdvancesForSyntheticBold(float aSynBoldOffset)
|
|
{
|
|
uint32_t synAppUnitOffset = aSynBoldOffset * mAppUnitsPerDevUnit;
|
|
for (uint32_t i = 0; i < Length(); ++i) {
|
|
CompressedGlyph *glyphData = &mCharacterGlyphs[i];
|
|
if (glyphData->IsSimpleGlyph()) {
|
|
// simple glyphs ==> just add the advance
|
|
int32_t advance = glyphData->GetSimpleAdvance() + synAppUnitOffset;
|
|
if (CompressedGlyph::IsSimpleAdvance(advance)) {
|
|
glyphData->SetSimpleGlyph(advance, glyphData->GetSimpleGlyph());
|
|
} else {
|
|
// rare case, tested by making this the default
|
|
uint32_t glyphIndex = glyphData->GetSimpleGlyph();
|
|
glyphData->SetComplex(true, true, 1);
|
|
DetailedGlyph detail = {glyphIndex, advance, 0, 0};
|
|
SetGlyphs(i, *glyphData, &detail);
|
|
}
|
|
} else {
|
|
// complex glyphs ==> add offset at cluster/ligature boundaries
|
|
uint32_t detailedLength = glyphData->GetGlyphCount();
|
|
if (detailedLength) {
|
|
DetailedGlyph *details = GetDetailedGlyphs(i);
|
|
if (!details) {
|
|
continue;
|
|
}
|
|
if (IsRightToLeft()) {
|
|
details[0].mAdvance += synAppUnitOffset;
|
|
} else {
|
|
details[detailedLength - 1].mAdvance += synAppUnitOffset;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bool
|
|
gfxTextRun::GlyphRunIterator::NextRun() {
|
|
if (mNextIndex >= mTextRun->mGlyphRuns.Length())
|
|
return false;
|
|
mGlyphRun = &mTextRun->mGlyphRuns[mNextIndex];
|
|
if (mGlyphRun->mCharacterOffset >= mEndOffset)
|
|
return false;
|
|
|
|
mStringStart = NS_MAX(mStartOffset, mGlyphRun->mCharacterOffset);
|
|
uint32_t last = mNextIndex + 1 < mTextRun->mGlyphRuns.Length()
|
|
? mTextRun->mGlyphRuns[mNextIndex + 1].mCharacterOffset : mTextRun->mCharacterCount;
|
|
mStringEnd = NS_MIN(mEndOffset, last);
|
|
|
|
++mNextIndex;
|
|
return true;
|
|
}
|
|
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
static void
|
|
AccountStorageForTextRun(gfxTextRun *aTextRun, int32_t aSign)
|
|
{
|
|
// Ignores detailed glyphs... we don't know when those have been constructed
|
|
// Also ignores gfxSkipChars dynamic storage (which won't be anything
|
|
// for preformatted text)
|
|
// Also ignores GlyphRun array, again because it hasn't been constructed
|
|
// by the time this gets called. If there's only one glyphrun that's stored
|
|
// directly in the textrun anyway so no additional overhead.
|
|
uint32_t length = aTextRun->GetLength();
|
|
int32_t bytes = length * sizeof(gfxTextRun::CompressedGlyph);
|
|
bytes += sizeof(gfxTextRun);
|
|
gTextRunStorage += bytes*aSign;
|
|
gTextRunStorageHighWaterMark = NS_MAX(gTextRunStorageHighWaterMark, gTextRunStorage);
|
|
}
|
|
#endif
|
|
|
|
// Helper for textRun creation to preallocate storage for glyph records;
|
|
// this function returns a pointer to the newly-allocated glyph storage.
|
|
// Returns nullptr if allocation fails.
|
|
void *
|
|
gfxTextRun::AllocateStorageForTextRun(size_t aSize, uint32_t aLength)
|
|
{
|
|
// Allocate the storage we need, returning nullptr on failure rather than
|
|
// throwing an exception (because web content can create huge runs).
|
|
void *storage = moz_malloc(aSize + aLength * sizeof(CompressedGlyph));
|
|
if (!storage) {
|
|
NS_WARNING("failed to allocate storage for text run!");
|
|
return nullptr;
|
|
}
|
|
|
|
// Initialize the glyph storage (beyond aSize) to zero
|
|
memset(reinterpret_cast<char*>(storage) + aSize, 0,
|
|
aLength * sizeof(CompressedGlyph));
|
|
|
|
return storage;
|
|
}
|
|
|
|
gfxTextRun *
|
|
gfxTextRun::Create(const gfxTextRunFactory::Parameters *aParams,
|
|
uint32_t aLength, gfxFontGroup *aFontGroup, uint32_t aFlags)
|
|
{
|
|
void *storage = AllocateStorageForTextRun(sizeof(gfxTextRun), aLength);
|
|
if (!storage) {
|
|
return nullptr;
|
|
}
|
|
|
|
return new (storage) gfxTextRun(aParams, aLength, aFontGroup, aFlags);
|
|
}
|
|
|
|
gfxTextRun::gfxTextRun(const gfxTextRunFactory::Parameters *aParams,
|
|
uint32_t aLength, gfxFontGroup *aFontGroup, uint32_t aFlags)
|
|
: mUserData(aParams->mUserData),
|
|
mFontGroup(aFontGroup),
|
|
mAppUnitsPerDevUnit(aParams->mAppUnitsPerDevUnit),
|
|
mFlags(aFlags), mCharacterCount(aLength)
|
|
{
|
|
NS_ASSERTION(mAppUnitsPerDevUnit != 0, "Invalid app unit scale");
|
|
MOZ_COUNT_CTOR(gfxTextRun);
|
|
NS_ADDREF(mFontGroup);
|
|
|
|
mCharacterGlyphs = reinterpret_cast<CompressedGlyph*>(this + 1);
|
|
|
|
if (aParams->mSkipChars) {
|
|
mSkipChars.TakeFrom(aParams->mSkipChars);
|
|
}
|
|
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
AccountStorageForTextRun(this, 1);
|
|
#endif
|
|
|
|
mSkipDrawing = mFontGroup->ShouldSkipDrawing();
|
|
}
|
|
|
|
gfxTextRun::~gfxTextRun()
|
|
{
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
AccountStorageForTextRun(this, -1);
|
|
#endif
|
|
#ifdef DEBUG
|
|
// Make it easy to detect a dead text run
|
|
mFlags = 0xFFFFFFFF;
|
|
#endif
|
|
|
|
NS_RELEASE(mFontGroup);
|
|
MOZ_COUNT_DTOR(gfxTextRun);
|
|
}
|
|
|
|
bool
|
|
gfxTextRun::SetPotentialLineBreaks(uint32_t aStart, uint32_t aLength,
|
|
uint8_t *aBreakBefore,
|
|
gfxContext *aRefContext)
|
|
{
|
|
NS_ASSERTION(aStart + aLength <= mCharacterCount, "Overflow");
|
|
|
|
uint32_t changed = 0;
|
|
uint32_t i;
|
|
CompressedGlyph *charGlyphs = mCharacterGlyphs + aStart;
|
|
for (i = 0; i < aLength; ++i) {
|
|
uint8_t canBreak = aBreakBefore[i];
|
|
if (canBreak && !charGlyphs[i].IsClusterStart()) {
|
|
// This can happen ... there is no guarantee that our linebreaking rules
|
|
// align with the platform's idea of what constitutes a cluster.
|
|
NS_WARNING("Break suggested inside cluster!");
|
|
canBreak = CompressedGlyph::FLAG_BREAK_TYPE_NONE;
|
|
}
|
|
changed |= charGlyphs[i].SetCanBreakBefore(canBreak);
|
|
}
|
|
return changed != 0;
|
|
}
|
|
|
|
gfxTextRun::LigatureData
|
|
gfxTextRun::ComputeLigatureData(uint32_t aPartStart, uint32_t aPartEnd,
|
|
PropertyProvider *aProvider)
|
|
{
|
|
NS_ASSERTION(aPartStart < aPartEnd, "Computing ligature data for empty range");
|
|
NS_ASSERTION(aPartEnd <= mCharacterCount, "Character length overflow");
|
|
|
|
LigatureData result;
|
|
CompressedGlyph *charGlyphs = mCharacterGlyphs;
|
|
|
|
uint32_t i;
|
|
for (i = aPartStart; !charGlyphs[i].IsLigatureGroupStart(); --i) {
|
|
NS_ASSERTION(i > 0, "Ligature at the start of the run??");
|
|
}
|
|
result.mLigatureStart = i;
|
|
for (i = aPartStart + 1; i < mCharacterCount && !charGlyphs[i].IsLigatureGroupStart(); ++i) {
|
|
}
|
|
result.mLigatureEnd = i;
|
|
|
|
int32_t ligatureWidth =
|
|
GetAdvanceForGlyphs(result.mLigatureStart, result.mLigatureEnd);
|
|
// Count the number of started clusters we have seen
|
|
uint32_t totalClusterCount = 0;
|
|
uint32_t partClusterIndex = 0;
|
|
uint32_t partClusterCount = 0;
|
|
for (i = result.mLigatureStart; i < result.mLigatureEnd; ++i) {
|
|
// Treat the first character of the ligature as the start of a
|
|
// cluster for our purposes of allocating ligature width to its
|
|
// characters.
|
|
if (i == result.mLigatureStart || charGlyphs[i].IsClusterStart()) {
|
|
++totalClusterCount;
|
|
if (i < aPartStart) {
|
|
++partClusterIndex;
|
|
} else if (i < aPartEnd) {
|
|
++partClusterCount;
|
|
}
|
|
}
|
|
}
|
|
NS_ASSERTION(totalClusterCount > 0, "Ligature involving no clusters??");
|
|
result.mPartAdvance = partClusterIndex * (ligatureWidth / totalClusterCount);
|
|
result.mPartWidth = partClusterCount * (ligatureWidth / totalClusterCount);
|
|
|
|
// Any rounding errors are apportioned to the final part of the ligature,
|
|
// so that measuring all parts of a ligature and summing them is equal to
|
|
// the ligature width.
|
|
if (aPartEnd == result.mLigatureEnd) {
|
|
gfxFloat allParts = totalClusterCount * (ligatureWidth / totalClusterCount);
|
|
result.mPartWidth += ligatureWidth - allParts;
|
|
}
|
|
|
|
if (partClusterCount == 0) {
|
|
// nothing to draw
|
|
result.mClipBeforePart = result.mClipAfterPart = true;
|
|
} else {
|
|
// Determine whether we should clip before or after this part when
|
|
// drawing its slice of the ligature.
|
|
// We need to clip before the part if any cluster is drawn before
|
|
// this part.
|
|
result.mClipBeforePart = partClusterIndex > 0;
|
|
// We need to clip after the part if any cluster is drawn after
|
|
// this part.
|
|
result.mClipAfterPart = partClusterIndex + partClusterCount < totalClusterCount;
|
|
}
|
|
|
|
if (aProvider && (mFlags & gfxTextRunFactory::TEXT_ENABLE_SPACING)) {
|
|
gfxFont::Spacing spacing;
|
|
if (aPartStart == result.mLigatureStart) {
|
|
aProvider->GetSpacing(aPartStart, 1, &spacing);
|
|
result.mPartWidth += spacing.mBefore;
|
|
}
|
|
if (aPartEnd == result.mLigatureEnd) {
|
|
aProvider->GetSpacing(aPartEnd - 1, 1, &spacing);
|
|
result.mPartWidth += spacing.mAfter;
|
|
}
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
gfxFloat
|
|
gfxTextRun::ComputePartialLigatureWidth(uint32_t aPartStart, uint32_t aPartEnd,
|
|
PropertyProvider *aProvider)
|
|
{
|
|
if (aPartStart >= aPartEnd)
|
|
return 0;
|
|
LigatureData data = ComputeLigatureData(aPartStart, aPartEnd, aProvider);
|
|
return data.mPartWidth;
|
|
}
|
|
|
|
int32_t
|
|
gfxTextRun::GetAdvanceForGlyphs(uint32_t aStart, uint32_t aEnd)
|
|
{
|
|
const CompressedGlyph *glyphData = mCharacterGlyphs + aStart;
|
|
int32_t advance = 0;
|
|
uint32_t i;
|
|
for (i = aStart; i < aEnd; ++i, ++glyphData) {
|
|
if (glyphData->IsSimpleGlyph()) {
|
|
advance += glyphData->GetSimpleAdvance();
|
|
} else {
|
|
uint32_t glyphCount = glyphData->GetGlyphCount();
|
|
if (glyphCount == 0) {
|
|
continue;
|
|
}
|
|
const DetailedGlyph *details = GetDetailedGlyphs(i);
|
|
if (details) {
|
|
uint32_t j;
|
|
for (j = 0; j < glyphCount; ++j, ++details) {
|
|
advance += details->mAdvance;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return advance;
|
|
}
|
|
|
|
static void
|
|
GetAdjustedSpacing(gfxTextRun *aTextRun, uint32_t aStart, uint32_t aEnd,
|
|
gfxTextRun::PropertyProvider *aProvider,
|
|
gfxTextRun::PropertyProvider::Spacing *aSpacing)
|
|
{
|
|
if (aStart >= aEnd)
|
|
return;
|
|
|
|
aProvider->GetSpacing(aStart, aEnd - aStart, aSpacing);
|
|
|
|
#ifdef DEBUG
|
|
// Check to see if we have spacing inside ligatures
|
|
|
|
const gfxTextRun::CompressedGlyph *charGlyphs = aTextRun->GetCharacterGlyphs();
|
|
uint32_t i;
|
|
|
|
for (i = aStart; i < aEnd; ++i) {
|
|
if (!charGlyphs[i].IsLigatureGroupStart()) {
|
|
NS_ASSERTION(i == aStart || aSpacing[i - aStart].mBefore == 0,
|
|
"Before-spacing inside a ligature!");
|
|
NS_ASSERTION(i - 1 <= aStart || aSpacing[i - 1 - aStart].mAfter == 0,
|
|
"After-spacing inside a ligature!");
|
|
}
|
|
}
|
|
#endif
|
|
}
|
|
|
|
bool
|
|
gfxTextRun::GetAdjustedSpacingArray(uint32_t aStart, uint32_t aEnd,
|
|
PropertyProvider *aProvider,
|
|
uint32_t aSpacingStart, uint32_t aSpacingEnd,
|
|
nsTArray<PropertyProvider::Spacing> *aSpacing)
|
|
{
|
|
if (!aProvider || !(mFlags & gfxTextRunFactory::TEXT_ENABLE_SPACING))
|
|
return false;
|
|
if (!aSpacing->AppendElements(aEnd - aStart))
|
|
return false;
|
|
memset(aSpacing->Elements(), 0, sizeof(gfxFont::Spacing)*(aSpacingStart - aStart));
|
|
GetAdjustedSpacing(this, aSpacingStart, aSpacingEnd, aProvider,
|
|
aSpacing->Elements() + aSpacingStart - aStart);
|
|
memset(aSpacing->Elements() + aSpacingEnd - aStart, 0, sizeof(gfxFont::Spacing)*(aEnd - aSpacingEnd));
|
|
return true;
|
|
}
|
|
|
|
void
|
|
gfxTextRun::ShrinkToLigatureBoundaries(uint32_t *aStart, uint32_t *aEnd)
|
|
{
|
|
if (*aStart >= *aEnd)
|
|
return;
|
|
|
|
CompressedGlyph *charGlyphs = mCharacterGlyphs;
|
|
|
|
while (*aStart < *aEnd && !charGlyphs[*aStart].IsLigatureGroupStart()) {
|
|
++(*aStart);
|
|
}
|
|
if (*aEnd < mCharacterCount) {
|
|
while (*aEnd > *aStart && !charGlyphs[*aEnd].IsLigatureGroupStart()) {
|
|
--(*aEnd);
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxTextRun::DrawGlyphs(gfxFont *aFont, gfxContext *aContext,
|
|
gfxFont::DrawMode aDrawMode, gfxPoint *aPt,
|
|
gfxTextObjectPaint *aObjectPaint,
|
|
uint32_t aStart, uint32_t aEnd,
|
|
PropertyProvider *aProvider,
|
|
uint32_t aSpacingStart, uint32_t aSpacingEnd)
|
|
{
|
|
nsAutoTArray<PropertyProvider::Spacing,200> spacingBuffer;
|
|
bool haveSpacing = GetAdjustedSpacingArray(aStart, aEnd, aProvider,
|
|
aSpacingStart, aSpacingEnd, &spacingBuffer);
|
|
aFont->Draw(this, aStart, aEnd, aContext, aDrawMode, aPt,
|
|
haveSpacing ? spacingBuffer.Elements() : nullptr, aObjectPaint);
|
|
}
|
|
|
|
static void
|
|
ClipPartialLigature(gfxTextRun *aTextRun, gfxFloat *aLeft, gfxFloat *aRight,
|
|
gfxFloat aXOrigin, gfxTextRun::LigatureData *aLigature)
|
|
{
|
|
if (aLigature->mClipBeforePart) {
|
|
if (aTextRun->IsRightToLeft()) {
|
|
*aRight = NS_MIN(*aRight, aXOrigin);
|
|
} else {
|
|
*aLeft = NS_MAX(*aLeft, aXOrigin);
|
|
}
|
|
}
|
|
if (aLigature->mClipAfterPart) {
|
|
gfxFloat endEdge = aXOrigin + aTextRun->GetDirection()*aLigature->mPartWidth;
|
|
if (aTextRun->IsRightToLeft()) {
|
|
*aLeft = NS_MAX(*aLeft, endEdge);
|
|
} else {
|
|
*aRight = NS_MIN(*aRight, endEdge);
|
|
}
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxTextRun::DrawPartialLigature(gfxFont *aFont, gfxContext *aCtx,
|
|
uint32_t aStart, uint32_t aEnd,
|
|
gfxPoint *aPt,
|
|
PropertyProvider *aProvider,
|
|
gfxTextRun::DrawCallbacks *aCallbacks)
|
|
{
|
|
if (aStart >= aEnd)
|
|
return;
|
|
|
|
// Draw partial ligature. We hack this by clipping the ligature.
|
|
LigatureData data = ComputeLigatureData(aStart, aEnd, aProvider);
|
|
gfxRect clipExtents = aCtx->GetClipExtents();
|
|
gfxFloat left = clipExtents.X()*mAppUnitsPerDevUnit;
|
|
gfxFloat right = clipExtents.XMost()*mAppUnitsPerDevUnit;
|
|
ClipPartialLigature(this, &left, &right, aPt->x, &data);
|
|
|
|
{
|
|
// Need to preserve the path, otherwise this can break canvas text-on-path;
|
|
// in general it seems like a good thing, as naive callers probably won't
|
|
// expect gfxTextRun::Draw to implicitly destroy the current path.
|
|
gfxContextPathAutoSaveRestore savePath(aCtx);
|
|
|
|
// use division here to ensure that when the rect is aligned on multiples
|
|
// of mAppUnitsPerDevUnit, we clip to true device unit boundaries.
|
|
// Also, make sure we snap the rectangle to device pixels.
|
|
aCtx->Save();
|
|
aCtx->NewPath();
|
|
aCtx->Rectangle(gfxRect(left / mAppUnitsPerDevUnit,
|
|
clipExtents.Y(),
|
|
(right - left) / mAppUnitsPerDevUnit,
|
|
clipExtents.Height()), true);
|
|
aCtx->Clip();
|
|
}
|
|
|
|
gfxFloat direction = GetDirection();
|
|
gfxPoint pt(aPt->x - direction*data.mPartAdvance, aPt->y);
|
|
DrawGlyphs(aFont, aCtx,
|
|
aCallbacks ? gfxFont::GLYPH_PATH : gfxFont::GLYPH_FILL, &pt,
|
|
nullptr, data.mLigatureStart, data.mLigatureEnd, aProvider,
|
|
aStart, aEnd);
|
|
if (aCallbacks) {
|
|
aCallbacks->NotifyGlyphPathEmitted();
|
|
}
|
|
aCtx->Restore();
|
|
|
|
aPt->x += direction*data.mPartWidth;
|
|
}
|
|
|
|
// returns true if a glyph run is using a font with synthetic bolding enabled, false otherwise
|
|
static bool
|
|
HasSyntheticBold(gfxTextRun *aRun, uint32_t aStart, uint32_t aLength)
|
|
{
|
|
gfxTextRun::GlyphRunIterator iter(aRun, aStart, aLength);
|
|
while (iter.NextRun()) {
|
|
gfxFont *font = iter.GetGlyphRun()->mFont;
|
|
if (font && font->IsSyntheticBold()) {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
// returns true if color is non-opaque (i.e. alpha != 1.0) or completely transparent, false otherwise
|
|
// if true, color is set on output
|
|
static bool
|
|
HasNonOpaqueColor(gfxContext *aContext, gfxRGBA& aCurrentColor)
|
|
{
|
|
if (aContext->GetDeviceColor(aCurrentColor)) {
|
|
if (aCurrentColor.a < 1.0 && aCurrentColor.a > 0.0) {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
// helper class for double-buffering drawing with non-opaque color
|
|
struct BufferAlphaColor {
|
|
BufferAlphaColor(gfxContext *aContext)
|
|
: mContext(aContext)
|
|
{
|
|
|
|
}
|
|
|
|
~BufferAlphaColor() {}
|
|
|
|
void PushSolidColor(const gfxRect& aBounds, const gfxRGBA& aAlphaColor, uint32_t appsPerDevUnit)
|
|
{
|
|
mContext->Save();
|
|
mContext->NewPath();
|
|
mContext->Rectangle(gfxRect(aBounds.X() / appsPerDevUnit,
|
|
aBounds.Y() / appsPerDevUnit,
|
|
aBounds.Width() / appsPerDevUnit,
|
|
aBounds.Height() / appsPerDevUnit), true);
|
|
mContext->Clip();
|
|
mContext->SetColor(gfxRGBA(aAlphaColor.r, aAlphaColor.g, aAlphaColor.b));
|
|
mContext->PushGroup(gfxASurface::CONTENT_COLOR_ALPHA);
|
|
mAlpha = aAlphaColor.a;
|
|
}
|
|
|
|
void PopAlpha()
|
|
{
|
|
// pop the text, using the color alpha as the opacity
|
|
mContext->PopGroupToSource();
|
|
mContext->SetOperator(gfxContext::OPERATOR_OVER);
|
|
mContext->Paint(mAlpha);
|
|
mContext->Restore();
|
|
}
|
|
|
|
gfxContext *mContext;
|
|
gfxFloat mAlpha;
|
|
};
|
|
|
|
void
|
|
gfxTextRun::Draw(gfxContext *aContext, gfxPoint aPt, gfxFont::DrawMode aDrawMode,
|
|
uint32_t aStart, uint32_t aLength,
|
|
PropertyProvider *aProvider, gfxFloat *aAdvanceWidth,
|
|
gfxTextObjectPaint *aObjectPaint,
|
|
gfxTextRun::DrawCallbacks *aCallbacks)
|
|
{
|
|
NS_ASSERTION(aStart + aLength <= mCharacterCount, "Substring out of range");
|
|
NS_ASSERTION(aDrawMode <= gfxFont::GLYPH_PATH, "GLYPH_PATH cannot be used with GLYPH_FILL or GLYPH_STROKE");
|
|
NS_ASSERTION(aDrawMode == gfxFont::GLYPH_PATH || !aCallbacks, "callback must not be specified unless using GLYPH_PATH");
|
|
|
|
gfxFloat direction = GetDirection();
|
|
|
|
if (mSkipDrawing) {
|
|
// We're waiting for a user font to finish downloading;
|
|
// but if the caller wants advance width, we need to compute it here
|
|
if (aAdvanceWidth) {
|
|
gfxTextRun::Metrics metrics = MeasureText(aStart, aLength,
|
|
gfxFont::LOOSE_INK_EXTENTS,
|
|
aContext, aProvider);
|
|
*aAdvanceWidth = metrics.mAdvanceWidth * direction;
|
|
}
|
|
|
|
// return without drawing
|
|
return;
|
|
}
|
|
|
|
gfxPoint pt = aPt;
|
|
|
|
// synthetic bolding draws glyphs twice ==> colors with opacity won't draw correctly unless first drawn without alpha
|
|
BufferAlphaColor syntheticBoldBuffer(aContext);
|
|
gfxRGBA currentColor;
|
|
bool needToRestore = false;
|
|
|
|
if (aDrawMode == gfxFont::GLYPH_FILL && HasNonOpaqueColor(aContext, currentColor)
|
|
&& HasSyntheticBold(this, aStart, aLength)) {
|
|
needToRestore = true;
|
|
// measure text, use the bounding box
|
|
gfxTextRun::Metrics metrics = MeasureText(aStart, aLength, gfxFont::LOOSE_INK_EXTENTS,
|
|
aContext, aProvider);
|
|
metrics.mBoundingBox.MoveBy(aPt);
|
|
syntheticBoldBuffer.PushSolidColor(metrics.mBoundingBox, currentColor, GetAppUnitsPerDevUnit());
|
|
}
|
|
|
|
GlyphRunIterator iter(this, aStart, aLength);
|
|
while (iter.NextRun()) {
|
|
gfxFont *font = iter.GetGlyphRun()->mFont;
|
|
uint32_t start = iter.GetStringStart();
|
|
uint32_t end = iter.GetStringEnd();
|
|
uint32_t ligatureRunStart = start;
|
|
uint32_t ligatureRunEnd = end;
|
|
ShrinkToLigatureBoundaries(&ligatureRunStart, &ligatureRunEnd);
|
|
|
|
bool drawPartial = aDrawMode == gfxFont::GLYPH_FILL ||
|
|
aDrawMode == gfxFont::GLYPH_PATH && aCallbacks;
|
|
|
|
if (drawPartial) {
|
|
DrawPartialLigature(font, aContext, start, ligatureRunStart, &pt,
|
|
aProvider, aCallbacks);
|
|
}
|
|
|
|
DrawGlyphs(font, aContext, aDrawMode, &pt, aObjectPaint, ligatureRunStart,
|
|
ligatureRunEnd, aProvider, ligatureRunStart, ligatureRunEnd);
|
|
|
|
if (aCallbacks) {
|
|
aCallbacks->NotifyGlyphPathEmitted();
|
|
}
|
|
|
|
if (drawPartial) {
|
|
DrawPartialLigature(font, aContext, ligatureRunEnd, end, &pt,
|
|
aProvider, aCallbacks);
|
|
}
|
|
}
|
|
|
|
// composite result when synthetic bolding used
|
|
if (needToRestore) {
|
|
syntheticBoldBuffer.PopAlpha();
|
|
}
|
|
|
|
if (aAdvanceWidth) {
|
|
*aAdvanceWidth = (pt.x - aPt.x)*direction;
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxTextRun::AccumulateMetricsForRun(gfxFont *aFont,
|
|
uint32_t aStart, uint32_t aEnd,
|
|
gfxFont::BoundingBoxType aBoundingBoxType,
|
|
gfxContext *aRefContext,
|
|
PropertyProvider *aProvider,
|
|
uint32_t aSpacingStart, uint32_t aSpacingEnd,
|
|
Metrics *aMetrics)
|
|
{
|
|
nsAutoTArray<PropertyProvider::Spacing,200> spacingBuffer;
|
|
bool haveSpacing = GetAdjustedSpacingArray(aStart, aEnd, aProvider,
|
|
aSpacingStart, aSpacingEnd, &spacingBuffer);
|
|
Metrics metrics = aFont->Measure(this, aStart, aEnd, aBoundingBoxType, aRefContext,
|
|
haveSpacing ? spacingBuffer.Elements() : nullptr);
|
|
aMetrics->CombineWith(metrics, IsRightToLeft());
|
|
}
|
|
|
|
void
|
|
gfxTextRun::AccumulatePartialLigatureMetrics(gfxFont *aFont,
|
|
uint32_t aStart, uint32_t aEnd,
|
|
gfxFont::BoundingBoxType aBoundingBoxType, gfxContext *aRefContext,
|
|
PropertyProvider *aProvider, Metrics *aMetrics)
|
|
{
|
|
if (aStart >= aEnd)
|
|
return;
|
|
|
|
// Measure partial ligature. We hack this by clipping the metrics in the
|
|
// same way we clip the drawing.
|
|
LigatureData data = ComputeLigatureData(aStart, aEnd, aProvider);
|
|
|
|
// First measure the complete ligature
|
|
Metrics metrics;
|
|
AccumulateMetricsForRun(aFont, data.mLigatureStart, data.mLigatureEnd,
|
|
aBoundingBoxType, aRefContext,
|
|
aProvider, aStart, aEnd, &metrics);
|
|
|
|
// Clip the bounding box to the ligature part
|
|
gfxFloat bboxLeft = metrics.mBoundingBox.X();
|
|
gfxFloat bboxRight = metrics.mBoundingBox.XMost();
|
|
// Where we are going to start "drawing" relative to our left baseline origin
|
|
gfxFloat origin = IsRightToLeft() ? metrics.mAdvanceWidth - data.mPartAdvance : 0;
|
|
ClipPartialLigature(this, &bboxLeft, &bboxRight, origin, &data);
|
|
metrics.mBoundingBox.x = bboxLeft;
|
|
metrics.mBoundingBox.width = bboxRight - bboxLeft;
|
|
|
|
// mBoundingBox is now relative to the left baseline origin for the entire
|
|
// ligature. Shift it left.
|
|
metrics.mBoundingBox.x -=
|
|
IsRightToLeft() ? metrics.mAdvanceWidth - (data.mPartAdvance + data.mPartWidth)
|
|
: data.mPartAdvance;
|
|
metrics.mAdvanceWidth = data.mPartWidth;
|
|
|
|
aMetrics->CombineWith(metrics, IsRightToLeft());
|
|
}
|
|
|
|
gfxTextRun::Metrics
|
|
gfxTextRun::MeasureText(uint32_t aStart, uint32_t aLength,
|
|
gfxFont::BoundingBoxType aBoundingBoxType,
|
|
gfxContext *aRefContext,
|
|
PropertyProvider *aProvider)
|
|
{
|
|
NS_ASSERTION(aStart + aLength <= mCharacterCount, "Substring out of range");
|
|
|
|
Metrics accumulatedMetrics;
|
|
GlyphRunIterator iter(this, aStart, aLength);
|
|
while (iter.NextRun()) {
|
|
gfxFont *font = iter.GetGlyphRun()->mFont;
|
|
uint32_t start = iter.GetStringStart();
|
|
uint32_t end = iter.GetStringEnd();
|
|
uint32_t ligatureRunStart = start;
|
|
uint32_t ligatureRunEnd = end;
|
|
ShrinkToLigatureBoundaries(&ligatureRunStart, &ligatureRunEnd);
|
|
|
|
AccumulatePartialLigatureMetrics(font, start, ligatureRunStart,
|
|
aBoundingBoxType, aRefContext, aProvider, &accumulatedMetrics);
|
|
|
|
// XXX This sucks. We have to get glyph extents just so we can detect
|
|
// glyphs outside the font box, even when aBoundingBoxType is LOOSE,
|
|
// even though in almost all cases we could get correct results just
|
|
// by getting some ascent/descent from the font and using our stored
|
|
// advance widths.
|
|
AccumulateMetricsForRun(font,
|
|
ligatureRunStart, ligatureRunEnd, aBoundingBoxType,
|
|
aRefContext, aProvider, ligatureRunStart, ligatureRunEnd,
|
|
&accumulatedMetrics);
|
|
|
|
AccumulatePartialLigatureMetrics(font, ligatureRunEnd, end,
|
|
aBoundingBoxType, aRefContext, aProvider, &accumulatedMetrics);
|
|
}
|
|
|
|
return accumulatedMetrics;
|
|
}
|
|
|
|
#define MEASUREMENT_BUFFER_SIZE 100
|
|
|
|
uint32_t
|
|
gfxTextRun::BreakAndMeasureText(uint32_t aStart, uint32_t aMaxLength,
|
|
bool aLineBreakBefore, gfxFloat aWidth,
|
|
PropertyProvider *aProvider,
|
|
bool aSuppressInitialBreak,
|
|
gfxFloat *aTrimWhitespace,
|
|
Metrics *aMetrics,
|
|
gfxFont::BoundingBoxType aBoundingBoxType,
|
|
gfxContext *aRefContext,
|
|
bool *aUsedHyphenation,
|
|
uint32_t *aLastBreak,
|
|
bool aCanWordWrap,
|
|
gfxBreakPriority *aBreakPriority)
|
|
{
|
|
aMaxLength = NS_MIN(aMaxLength, mCharacterCount - aStart);
|
|
|
|
NS_ASSERTION(aStart + aMaxLength <= mCharacterCount, "Substring out of range");
|
|
|
|
uint32_t bufferStart = aStart;
|
|
uint32_t bufferLength = NS_MIN<uint32_t>(aMaxLength, MEASUREMENT_BUFFER_SIZE);
|
|
PropertyProvider::Spacing spacingBuffer[MEASUREMENT_BUFFER_SIZE];
|
|
bool haveSpacing = aProvider && (mFlags & gfxTextRunFactory::TEXT_ENABLE_SPACING) != 0;
|
|
if (haveSpacing) {
|
|
GetAdjustedSpacing(this, bufferStart, bufferStart + bufferLength, aProvider,
|
|
spacingBuffer);
|
|
}
|
|
bool hyphenBuffer[MEASUREMENT_BUFFER_SIZE];
|
|
bool haveHyphenation = aProvider &&
|
|
(aProvider->GetHyphensOption() == NS_STYLE_HYPHENS_AUTO ||
|
|
(aProvider->GetHyphensOption() == NS_STYLE_HYPHENS_MANUAL &&
|
|
(mFlags & gfxTextRunFactory::TEXT_ENABLE_HYPHEN_BREAKS) != 0));
|
|
if (haveHyphenation) {
|
|
aProvider->GetHyphenationBreaks(bufferStart, bufferLength,
|
|
hyphenBuffer);
|
|
}
|
|
|
|
gfxFloat width = 0;
|
|
gfxFloat advance = 0;
|
|
// The number of space characters that can be trimmed
|
|
uint32_t trimmableChars = 0;
|
|
// The amount of space removed by ignoring trimmableChars
|
|
gfxFloat trimmableAdvance = 0;
|
|
int32_t lastBreak = -1;
|
|
int32_t lastBreakTrimmableChars = -1;
|
|
gfxFloat lastBreakTrimmableAdvance = -1;
|
|
bool aborted = false;
|
|
uint32_t end = aStart + aMaxLength;
|
|
bool lastBreakUsedHyphenation = false;
|
|
|
|
uint32_t ligatureRunStart = aStart;
|
|
uint32_t ligatureRunEnd = end;
|
|
ShrinkToLigatureBoundaries(&ligatureRunStart, &ligatureRunEnd);
|
|
|
|
uint32_t i;
|
|
for (i = aStart; i < end; ++i) {
|
|
if (i >= bufferStart + bufferLength) {
|
|
// Fetch more spacing and hyphenation data
|
|
bufferStart = i;
|
|
bufferLength = NS_MIN(aStart + aMaxLength, i + MEASUREMENT_BUFFER_SIZE) - i;
|
|
if (haveSpacing) {
|
|
GetAdjustedSpacing(this, bufferStart, bufferStart + bufferLength, aProvider,
|
|
spacingBuffer);
|
|
}
|
|
if (haveHyphenation) {
|
|
aProvider->GetHyphenationBreaks(bufferStart, bufferLength,
|
|
hyphenBuffer);
|
|
}
|
|
}
|
|
|
|
// There can't be a word-wrap break opportunity at the beginning of the
|
|
// line: if the width is too small for even one character to fit, it
|
|
// could be the first and last break opportunity on the line, and that
|
|
// would trigger an infinite loop.
|
|
if (!aSuppressInitialBreak || i > aStart) {
|
|
bool lineBreakHere = mCharacterGlyphs[i].CanBreakBefore() == 1;
|
|
bool hyphenation = haveHyphenation && hyphenBuffer[i - bufferStart];
|
|
bool wordWrapping =
|
|
aCanWordWrap && mCharacterGlyphs[i].IsClusterStart() &&
|
|
*aBreakPriority <= eWordWrapBreak;
|
|
|
|
if (lineBreakHere || hyphenation || wordWrapping) {
|
|
gfxFloat hyphenatedAdvance = advance;
|
|
if (!lineBreakHere && !wordWrapping) {
|
|
hyphenatedAdvance += aProvider->GetHyphenWidth();
|
|
}
|
|
|
|
if (lastBreak < 0 || width + hyphenatedAdvance - trimmableAdvance <= aWidth) {
|
|
// We can break here.
|
|
lastBreak = i;
|
|
lastBreakTrimmableChars = trimmableChars;
|
|
lastBreakTrimmableAdvance = trimmableAdvance;
|
|
lastBreakUsedHyphenation = !lineBreakHere && !wordWrapping;
|
|
*aBreakPriority = hyphenation || lineBreakHere ?
|
|
eNormalBreak : eWordWrapBreak;
|
|
}
|
|
|
|
width += advance;
|
|
advance = 0;
|
|
if (width - trimmableAdvance > aWidth) {
|
|
// No more text fits. Abort
|
|
aborted = true;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
gfxFloat charAdvance;
|
|
if (i >= ligatureRunStart && i < ligatureRunEnd) {
|
|
charAdvance = GetAdvanceForGlyphs(i, i + 1);
|
|
if (haveSpacing) {
|
|
PropertyProvider::Spacing *space = &spacingBuffer[i - bufferStart];
|
|
charAdvance += space->mBefore + space->mAfter;
|
|
}
|
|
} else {
|
|
charAdvance = ComputePartialLigatureWidth(i, i + 1, aProvider);
|
|
}
|
|
|
|
advance += charAdvance;
|
|
if (aTrimWhitespace) {
|
|
if (mCharacterGlyphs[i].CharIsSpace()) {
|
|
++trimmableChars;
|
|
trimmableAdvance += charAdvance;
|
|
} else {
|
|
trimmableAdvance = 0;
|
|
trimmableChars = 0;
|
|
}
|
|
}
|
|
}
|
|
|
|
if (!aborted) {
|
|
width += advance;
|
|
}
|
|
|
|
// There are three possibilities:
|
|
// 1) all the text fit (width <= aWidth)
|
|
// 2) some of the text fit up to a break opportunity (width > aWidth && lastBreak >= 0)
|
|
// 3) none of the text fits before a break opportunity (width > aWidth && lastBreak < 0)
|
|
uint32_t charsFit;
|
|
bool usedHyphenation = false;
|
|
if (width - trimmableAdvance <= aWidth) {
|
|
charsFit = aMaxLength;
|
|
} else if (lastBreak >= 0) {
|
|
charsFit = lastBreak - aStart;
|
|
trimmableChars = lastBreakTrimmableChars;
|
|
trimmableAdvance = lastBreakTrimmableAdvance;
|
|
usedHyphenation = lastBreakUsedHyphenation;
|
|
} else {
|
|
charsFit = aMaxLength;
|
|
}
|
|
|
|
if (aMetrics) {
|
|
*aMetrics = MeasureText(aStart, charsFit - trimmableChars,
|
|
aBoundingBoxType, aRefContext, aProvider);
|
|
}
|
|
if (aTrimWhitespace) {
|
|
*aTrimWhitespace = trimmableAdvance;
|
|
}
|
|
if (aUsedHyphenation) {
|
|
*aUsedHyphenation = usedHyphenation;
|
|
}
|
|
if (aLastBreak && charsFit == aMaxLength) {
|
|
if (lastBreak < 0) {
|
|
*aLastBreak = UINT32_MAX;
|
|
} else {
|
|
*aLastBreak = lastBreak - aStart;
|
|
}
|
|
}
|
|
|
|
return charsFit;
|
|
}
|
|
|
|
gfxFloat
|
|
gfxTextRun::GetAdvanceWidth(uint32_t aStart, uint32_t aLength,
|
|
PropertyProvider *aProvider)
|
|
{
|
|
NS_ASSERTION(aStart + aLength <= mCharacterCount, "Substring out of range");
|
|
|
|
uint32_t ligatureRunStart = aStart;
|
|
uint32_t ligatureRunEnd = aStart + aLength;
|
|
ShrinkToLigatureBoundaries(&ligatureRunStart, &ligatureRunEnd);
|
|
|
|
gfxFloat result = ComputePartialLigatureWidth(aStart, ligatureRunStart, aProvider) +
|
|
ComputePartialLigatureWidth(ligatureRunEnd, aStart + aLength, aProvider);
|
|
|
|
// Account for all remaining spacing here. This is more efficient than
|
|
// processing it along with the glyphs.
|
|
if (aProvider && (mFlags & gfxTextRunFactory::TEXT_ENABLE_SPACING)) {
|
|
uint32_t i;
|
|
nsAutoTArray<PropertyProvider::Spacing,200> spacingBuffer;
|
|
if (spacingBuffer.AppendElements(aLength)) {
|
|
GetAdjustedSpacing(this, ligatureRunStart, ligatureRunEnd, aProvider,
|
|
spacingBuffer.Elements());
|
|
for (i = 0; i < ligatureRunEnd - ligatureRunStart; ++i) {
|
|
PropertyProvider::Spacing *space = &spacingBuffer[i];
|
|
result += space->mBefore + space->mAfter;
|
|
}
|
|
}
|
|
}
|
|
|
|
return result + GetAdvanceForGlyphs(ligatureRunStart, ligatureRunEnd);
|
|
}
|
|
|
|
bool
|
|
gfxTextRun::SetLineBreaks(uint32_t aStart, uint32_t aLength,
|
|
bool aLineBreakBefore, bool aLineBreakAfter,
|
|
gfxFloat *aAdvanceWidthDelta,
|
|
gfxContext *aRefContext)
|
|
{
|
|
// Do nothing because our shaping does not currently take linebreaks into
|
|
// account. There is no change in advance width.
|
|
if (aAdvanceWidthDelta) {
|
|
*aAdvanceWidthDelta = 0;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
uint32_t
|
|
gfxTextRun::FindFirstGlyphRunContaining(uint32_t aOffset)
|
|
{
|
|
NS_ASSERTION(aOffset <= mCharacterCount, "Bad offset looking for glyphrun");
|
|
NS_ASSERTION(mCharacterCount == 0 || mGlyphRuns.Length() > 0,
|
|
"non-empty text but no glyph runs present!");
|
|
if (aOffset == mCharacterCount)
|
|
return mGlyphRuns.Length();
|
|
uint32_t start = 0;
|
|
uint32_t end = mGlyphRuns.Length();
|
|
while (end - start > 1) {
|
|
uint32_t mid = (start + end)/2;
|
|
if (mGlyphRuns[mid].mCharacterOffset <= aOffset) {
|
|
start = mid;
|
|
} else {
|
|
end = mid;
|
|
}
|
|
}
|
|
NS_ASSERTION(mGlyphRuns[start].mCharacterOffset <= aOffset,
|
|
"Hmm, something went wrong, aOffset should have been found");
|
|
return start;
|
|
}
|
|
|
|
nsresult
|
|
gfxTextRun::AddGlyphRun(gfxFont *aFont, uint8_t aMatchType,
|
|
uint32_t aUTF16Offset, bool aForceNewRun)
|
|
{
|
|
NS_ASSERTION(aFont, "adding glyph run for null font!");
|
|
if (!aFont) {
|
|
return NS_OK;
|
|
}
|
|
uint32_t numGlyphRuns = mGlyphRuns.Length();
|
|
if (!aForceNewRun && numGlyphRuns > 0) {
|
|
GlyphRun *lastGlyphRun = &mGlyphRuns[numGlyphRuns - 1];
|
|
|
|
NS_ASSERTION(lastGlyphRun->mCharacterOffset <= aUTF16Offset,
|
|
"Glyph runs out of order (and run not forced)");
|
|
|
|
// Don't append a run if the font is already the one we want
|
|
if (lastGlyphRun->mFont == aFont &&
|
|
lastGlyphRun->mMatchType == aMatchType)
|
|
{
|
|
return NS_OK;
|
|
}
|
|
|
|
// If the offset has not changed, avoid leaving a zero-length run
|
|
// by overwriting the last entry instead of appending...
|
|
if (lastGlyphRun->mCharacterOffset == aUTF16Offset) {
|
|
|
|
// ...except that if the run before the last entry had the same
|
|
// font as the new one wants, merge with it instead of creating
|
|
// adjacent runs with the same font
|
|
if (numGlyphRuns > 1 &&
|
|
mGlyphRuns[numGlyphRuns - 2].mFont == aFont &&
|
|
mGlyphRuns[numGlyphRuns - 2].mMatchType == aMatchType)
|
|
{
|
|
mGlyphRuns.TruncateLength(numGlyphRuns - 1);
|
|
return NS_OK;
|
|
}
|
|
|
|
lastGlyphRun->mFont = aFont;
|
|
lastGlyphRun->mMatchType = aMatchType;
|
|
return NS_OK;
|
|
}
|
|
}
|
|
|
|
NS_ASSERTION(aForceNewRun || numGlyphRuns > 0 || aUTF16Offset == 0,
|
|
"First run doesn't cover the first character (and run not forced)?");
|
|
|
|
GlyphRun *glyphRun = mGlyphRuns.AppendElement();
|
|
if (!glyphRun)
|
|
return NS_ERROR_OUT_OF_MEMORY;
|
|
glyphRun->mFont = aFont;
|
|
glyphRun->mCharacterOffset = aUTF16Offset;
|
|
glyphRun->mMatchType = aMatchType;
|
|
return NS_OK;
|
|
}
|
|
|
|
void
|
|
gfxTextRun::SortGlyphRuns()
|
|
{
|
|
if (mGlyphRuns.Length() <= 1)
|
|
return;
|
|
|
|
nsTArray<GlyphRun> runs(mGlyphRuns);
|
|
GlyphRunOffsetComparator comp;
|
|
runs.Sort(comp);
|
|
|
|
// Now copy back, coalescing adjacent glyph runs that have the same font
|
|
mGlyphRuns.Clear();
|
|
uint32_t i, count = runs.Length();
|
|
for (i = 0; i < count; ++i) {
|
|
// a GlyphRun with the same font as the previous GlyphRun can just
|
|
// be skipped; the last GlyphRun will cover its character range.
|
|
if (i == 0 || runs[i].mFont != runs[i - 1].mFont) {
|
|
mGlyphRuns.AppendElement(runs[i]);
|
|
// If two fonts have the same character offset, Sort() will have
|
|
// randomized the order.
|
|
NS_ASSERTION(i == 0 ||
|
|
runs[i].mCharacterOffset !=
|
|
runs[i - 1].mCharacterOffset,
|
|
"Two fonts for the same run, glyph indices may not match the font");
|
|
}
|
|
}
|
|
}
|
|
|
|
// Note that SanitizeGlyphRuns scans all glyph runs in the textrun;
|
|
// therefore we only call it once, at the end of textrun construction,
|
|
// NOT incrementally as each glyph run is added (bug 680402).
|
|
void
|
|
gfxTextRun::SanitizeGlyphRuns()
|
|
{
|
|
if (mGlyphRuns.Length() <= 1)
|
|
return;
|
|
|
|
// If any glyph run starts with ligature-continuation characters, we need to advance it
|
|
// to the first "real" character to avoid drawing partial ligature glyphs from wrong font
|
|
// (seen with U+FEFF in reftest 474417-1, as Core Text eliminates the glyph, which makes
|
|
// it appear as if a ligature has been formed)
|
|
int32_t i, lastRunIndex = mGlyphRuns.Length() - 1;
|
|
const CompressedGlyph *charGlyphs = mCharacterGlyphs;
|
|
for (i = lastRunIndex; i >= 0; --i) {
|
|
GlyphRun& run = mGlyphRuns[i];
|
|
while (charGlyphs[run.mCharacterOffset].IsLigatureContinuation() &&
|
|
run.mCharacterOffset < mCharacterCount) {
|
|
run.mCharacterOffset++;
|
|
}
|
|
// if the run has become empty, eliminate it
|
|
if ((i < lastRunIndex &&
|
|
run.mCharacterOffset >= mGlyphRuns[i+1].mCharacterOffset) ||
|
|
(i == lastRunIndex && run.mCharacterOffset == mCharacterCount)) {
|
|
mGlyphRuns.RemoveElementAt(i);
|
|
--lastRunIndex;
|
|
}
|
|
}
|
|
}
|
|
|
|
uint32_t
|
|
gfxTextRun::CountMissingGlyphs()
|
|
{
|
|
uint32_t i;
|
|
uint32_t count = 0;
|
|
for (i = 0; i < mCharacterCount; ++i) {
|
|
if (mCharacterGlyphs[i].IsMissing()) {
|
|
++count;
|
|
}
|
|
}
|
|
return count;
|
|
}
|
|
|
|
gfxTextRun::DetailedGlyph *
|
|
gfxTextRun::AllocateDetailedGlyphs(uint32_t aIndex, uint32_t aCount)
|
|
{
|
|
NS_ASSERTION(aIndex < mCharacterCount, "Index out of range");
|
|
|
|
if (!mDetailedGlyphs) {
|
|
mDetailedGlyphs = new DetailedGlyphStore();
|
|
}
|
|
|
|
DetailedGlyph *details = mDetailedGlyphs->Allocate(aIndex, aCount);
|
|
if (!details) {
|
|
mCharacterGlyphs[aIndex].SetMissing(0);
|
|
return nullptr;
|
|
}
|
|
|
|
return details;
|
|
}
|
|
|
|
void
|
|
gfxTextRun::SetGlyphs(uint32_t aIndex, CompressedGlyph aGlyph,
|
|
const DetailedGlyph *aGlyphs)
|
|
{
|
|
NS_ASSERTION(!aGlyph.IsSimpleGlyph(), "Simple glyphs not handled here");
|
|
NS_ASSERTION(aIndex > 0 || aGlyph.IsLigatureGroupStart(),
|
|
"First character can't be a ligature continuation!");
|
|
|
|
uint32_t glyphCount = aGlyph.GetGlyphCount();
|
|
if (glyphCount > 0) {
|
|
DetailedGlyph *details = AllocateDetailedGlyphs(aIndex, glyphCount);
|
|
if (!details)
|
|
return;
|
|
memcpy(details, aGlyphs, sizeof(DetailedGlyph)*glyphCount);
|
|
}
|
|
mCharacterGlyphs[aIndex] = aGlyph;
|
|
}
|
|
|
|
void
|
|
gfxTextRun::SetMissingGlyph(uint32_t aIndex, uint32_t aChar)
|
|
{
|
|
uint8_t category = GetGeneralCategory(aChar);
|
|
if (category >= HB_UNICODE_GENERAL_CATEGORY_SPACING_MARK &&
|
|
category <= HB_UNICODE_GENERAL_CATEGORY_NON_SPACING_MARK)
|
|
{
|
|
mCharacterGlyphs[aIndex].SetComplex(false, true, 0);
|
|
}
|
|
|
|
DetailedGlyph *details = AllocateDetailedGlyphs(aIndex, 1);
|
|
if (!details)
|
|
return;
|
|
|
|
details->mGlyphID = aChar;
|
|
GlyphRun *glyphRun = &mGlyphRuns[FindFirstGlyphRunContaining(aIndex)];
|
|
if (IsDefaultIgnorable(aChar)) {
|
|
// Setting advance width to zero will prevent drawing the hexbox
|
|
details->mAdvance = 0;
|
|
} else {
|
|
gfxFloat width = NS_MAX(glyphRun->mFont->GetMetrics().aveCharWidth,
|
|
gfxFontMissingGlyphs::GetDesiredMinWidth(aChar));
|
|
details->mAdvance = uint32_t(width*GetAppUnitsPerDevUnit());
|
|
}
|
|
details->mXOffset = 0;
|
|
details->mYOffset = 0;
|
|
mCharacterGlyphs[aIndex].SetMissing(1);
|
|
}
|
|
|
|
void
|
|
gfxTextRun::CopyGlyphDataFrom(const gfxShapedWord *aShapedWord, uint32_t aOffset)
|
|
{
|
|
uint32_t wordLen = aShapedWord->Length();
|
|
NS_ASSERTION(aOffset + wordLen <= GetLength(),
|
|
"word overruns end of textrun!");
|
|
|
|
const CompressedGlyph *wordGlyphs = aShapedWord->GetCharacterGlyphs();
|
|
if (aShapedWord->HasDetailedGlyphs()) {
|
|
for (uint32_t i = 0; i < wordLen; ++i, ++aOffset) {
|
|
const CompressedGlyph& g = wordGlyphs[i];
|
|
if (g.IsSimpleGlyph()) {
|
|
SetSimpleGlyph(aOffset, g);
|
|
} else {
|
|
const DetailedGlyph *details =
|
|
g.GetGlyphCount() > 0 ?
|
|
aShapedWord->GetDetailedGlyphs(i) : nullptr;
|
|
SetGlyphs(aOffset, g, details);
|
|
}
|
|
}
|
|
} else {
|
|
memcpy(GetCharacterGlyphs() + aOffset, wordGlyphs,
|
|
wordLen * sizeof(CompressedGlyph));
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxTextRun::CopyGlyphDataFrom(gfxTextRun *aSource, uint32_t aStart,
|
|
uint32_t aLength, uint32_t aDest)
|
|
{
|
|
NS_ASSERTION(aStart + aLength <= aSource->GetLength(),
|
|
"Source substring out of range");
|
|
NS_ASSERTION(aDest + aLength <= GetLength(),
|
|
"Destination substring out of range");
|
|
|
|
if (aSource->mSkipDrawing) {
|
|
mSkipDrawing = true;
|
|
}
|
|
|
|
// Copy base glyph data, and DetailedGlyph data where present
|
|
const CompressedGlyph *srcGlyphs = aSource->mCharacterGlyphs + aStart;
|
|
CompressedGlyph *dstGlyphs = mCharacterGlyphs + aDest;
|
|
for (uint32_t i = 0; i < aLength; ++i) {
|
|
CompressedGlyph g = srcGlyphs[i];
|
|
g.SetCanBreakBefore(!g.IsClusterStart() ?
|
|
CompressedGlyph::FLAG_BREAK_TYPE_NONE :
|
|
dstGlyphs[i].CanBreakBefore());
|
|
if (!g.IsSimpleGlyph()) {
|
|
uint32_t count = g.GetGlyphCount();
|
|
if (count > 0) {
|
|
DetailedGlyph *dst = AllocateDetailedGlyphs(i + aDest, count);
|
|
if (dst) {
|
|
DetailedGlyph *src = aSource->GetDetailedGlyphs(i + aStart);
|
|
if (src) {
|
|
::memcpy(dst, src, count * sizeof(DetailedGlyph));
|
|
} else {
|
|
g.SetMissing(0);
|
|
}
|
|
} else {
|
|
g.SetMissing(0);
|
|
}
|
|
}
|
|
}
|
|
dstGlyphs[i] = g;
|
|
}
|
|
|
|
// Copy glyph runs
|
|
GlyphRunIterator iter(aSource, aStart, aLength);
|
|
#ifdef DEBUG
|
|
gfxFont *lastFont = nullptr;
|
|
#endif
|
|
while (iter.NextRun()) {
|
|
gfxFont *font = iter.GetGlyphRun()->mFont;
|
|
NS_ASSERTION(font != lastFont, "Glyphruns not coalesced?");
|
|
#ifdef DEBUG
|
|
lastFont = font;
|
|
uint32_t end = iter.GetStringEnd();
|
|
#endif
|
|
uint32_t start = iter.GetStringStart();
|
|
|
|
// These used to be NS_ASSERTION()s, but WARNING is more appropriate.
|
|
// Although it's unusual (and not desirable), it's possible for us to assign
|
|
// different fonts to a base character and a following diacritic.
|
|
// Example on OSX 10.5/10.6 with default fonts installed:
|
|
// data:text/html,<p style="font-family:helvetica, arial, sans-serif;">
|
|
// &%23x043E;&%23x0486;&%23x20;&%23x043E;&%23x0486;
|
|
// This means the rendering of the cluster will probably not be very good,
|
|
// but it's the best we can do for now if the specified font only covered the
|
|
// initial base character and not its applied marks.
|
|
NS_WARN_IF_FALSE(aSource->IsClusterStart(start),
|
|
"Started font run in the middle of a cluster");
|
|
NS_WARN_IF_FALSE(end == aSource->GetLength() || aSource->IsClusterStart(end),
|
|
"Ended font run in the middle of a cluster");
|
|
|
|
nsresult rv = AddGlyphRun(font, iter.GetGlyphRun()->mMatchType,
|
|
start - aStart + aDest, false);
|
|
if (NS_FAILED(rv))
|
|
return;
|
|
}
|
|
}
|
|
|
|
void
|
|
gfxTextRun::SetSpaceGlyph(gfxFont *aFont, gfxContext *aContext,
|
|
uint32_t aCharIndex)
|
|
{
|
|
if (SetSpaceGlyphIfSimple(aFont, aContext, aCharIndex, ' ')) {
|
|
return;
|
|
}
|
|
|
|
aFont->InitWordCache();
|
|
static const uint8_t space = ' ';
|
|
gfxShapedWord *sw = aFont->GetShapedWord(aContext,
|
|
&space, 1,
|
|
HashMix(0, ' '),
|
|
MOZ_SCRIPT_LATIN,
|
|
mAppUnitsPerDevUnit,
|
|
gfxTextRunFactory::TEXT_IS_8BIT |
|
|
gfxTextRunFactory::TEXT_IS_ASCII |
|
|
gfxTextRunFactory::TEXT_IS_PERSISTENT);
|
|
if (sw) {
|
|
AddGlyphRun(aFont, gfxTextRange::kFontGroup, aCharIndex, false);
|
|
CopyGlyphDataFrom(sw, aCharIndex);
|
|
}
|
|
}
|
|
|
|
bool
|
|
gfxTextRun::SetSpaceGlyphIfSimple(gfxFont *aFont, gfxContext *aContext,
|
|
uint32_t aCharIndex, PRUnichar aSpaceChar)
|
|
{
|
|
uint32_t spaceGlyph = aFont->GetSpaceGlyph();
|
|
if (!spaceGlyph || !CompressedGlyph::IsSimpleGlyphID(spaceGlyph)) {
|
|
return false;
|
|
}
|
|
|
|
uint32_t spaceWidthAppUnits =
|
|
NS_lroundf(aFont->GetMetrics().spaceWidth * mAppUnitsPerDevUnit);
|
|
if (!CompressedGlyph::IsSimpleAdvance(spaceWidthAppUnits)) {
|
|
return false;
|
|
}
|
|
|
|
AddGlyphRun(aFont, gfxTextRange::kFontGroup, aCharIndex, false);
|
|
CompressedGlyph g;
|
|
g.SetSimpleGlyph(spaceWidthAppUnits, spaceGlyph);
|
|
if (aSpaceChar == ' ') {
|
|
g.SetIsSpace();
|
|
}
|
|
SetSimpleGlyph(aCharIndex, g);
|
|
return true;
|
|
}
|
|
|
|
void
|
|
gfxTextRun::FetchGlyphExtents(gfxContext *aRefContext)
|
|
{
|
|
bool needsGlyphExtents = NeedsGlyphExtents(this);
|
|
if (!needsGlyphExtents && !mDetailedGlyphs)
|
|
return;
|
|
|
|
uint32_t i, runCount = mGlyphRuns.Length();
|
|
CompressedGlyph *charGlyphs = mCharacterGlyphs;
|
|
for (i = 0; i < runCount; ++i) {
|
|
const GlyphRun& run = mGlyphRuns[i];
|
|
gfxFont *font = run.mFont;
|
|
uint32_t start = run.mCharacterOffset;
|
|
uint32_t end = i + 1 < runCount ?
|
|
mGlyphRuns[i + 1].mCharacterOffset : GetLength();
|
|
bool fontIsSetup = false;
|
|
uint32_t j;
|
|
gfxGlyphExtents *extents = font->GetOrCreateGlyphExtents(mAppUnitsPerDevUnit);
|
|
|
|
for (j = start; j < end; ++j) {
|
|
const gfxTextRun::CompressedGlyph *glyphData = &charGlyphs[j];
|
|
if (glyphData->IsSimpleGlyph()) {
|
|
// If we're in speed mode, don't set up glyph extents here; we'll
|
|
// just return "optimistic" glyph bounds later
|
|
if (needsGlyphExtents) {
|
|
uint32_t glyphIndex = glyphData->GetSimpleGlyph();
|
|
if (!extents->IsGlyphKnown(glyphIndex)) {
|
|
if (!fontIsSetup) {
|
|
if (!font->SetupCairoFont(aRefContext)) {
|
|
NS_WARNING("failed to set up font for glyph extents");
|
|
break;
|
|
}
|
|
fontIsSetup = true;
|
|
}
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
++gGlyphExtentsSetupEagerSimple;
|
|
#endif
|
|
font->SetupGlyphExtents(aRefContext, glyphIndex, false, extents);
|
|
}
|
|
}
|
|
} else if (!glyphData->IsMissing()) {
|
|
uint32_t glyphCount = glyphData->GetGlyphCount();
|
|
if (glyphCount == 0) {
|
|
continue;
|
|
}
|
|
const gfxTextRun::DetailedGlyph *details = GetDetailedGlyphs(j);
|
|
if (!details) {
|
|
continue;
|
|
}
|
|
for (uint32_t k = 0; k < glyphCount; ++k, ++details) {
|
|
uint32_t glyphIndex = details->mGlyphID;
|
|
if (!extents->IsGlyphKnownWithTightExtents(glyphIndex)) {
|
|
if (!fontIsSetup) {
|
|
if (!font->SetupCairoFont(aRefContext)) {
|
|
NS_WARNING("failed to set up font for glyph extents");
|
|
break;
|
|
}
|
|
fontIsSetup = true;
|
|
}
|
|
#ifdef DEBUG_TEXT_RUN_STORAGE_METRICS
|
|
++gGlyphExtentsSetupEagerTight;
|
|
#endif
|
|
font->SetupGlyphExtents(aRefContext, glyphIndex, true, extents);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
gfxTextRun::ClusterIterator::ClusterIterator(gfxTextRun *aTextRun)
|
|
: mTextRun(aTextRun), mCurrentChar(uint32_t(-1))
|
|
{
|
|
}
|
|
|
|
void
|
|
gfxTextRun::ClusterIterator::Reset()
|
|
{
|
|
mCurrentChar = uint32_t(-1);
|
|
}
|
|
|
|
bool
|
|
gfxTextRun::ClusterIterator::NextCluster()
|
|
{
|
|
uint32_t len = mTextRun->GetLength();
|
|
while (++mCurrentChar < len) {
|
|
if (mTextRun->IsClusterStart(mCurrentChar)) {
|
|
return true;
|
|
}
|
|
}
|
|
|
|
mCurrentChar = uint32_t(-1);
|
|
return false;
|
|
}
|
|
|
|
uint32_t
|
|
gfxTextRun::ClusterIterator::ClusterLength() const
|
|
{
|
|
if (mCurrentChar == uint32_t(-1)) {
|
|
return 0;
|
|
}
|
|
|
|
uint32_t i = mCurrentChar,
|
|
len = mTextRun->GetLength();
|
|
while (++i < len) {
|
|
if (mTextRun->IsClusterStart(i)) {
|
|
break;
|
|
}
|
|
}
|
|
|
|
return i - mCurrentChar;
|
|
}
|
|
|
|
gfxFloat
|
|
gfxTextRun::ClusterIterator::ClusterAdvance(PropertyProvider *aProvider) const
|
|
{
|
|
if (mCurrentChar == uint32_t(-1)) {
|
|
return 0;
|
|
}
|
|
|
|
return mTextRun->GetAdvanceWidth(mCurrentChar, ClusterLength(), aProvider);
|
|
}
|
|
|
|
size_t
|
|
gfxTextRun::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf)
|
|
{
|
|
// The second arg is how much gfxTextRun::AllocateStorage would have
|
|
// allocated.
|
|
size_t total = mGlyphRuns.SizeOfExcludingThis(aMallocSizeOf);
|
|
|
|
if (mDetailedGlyphs) {
|
|
total += mDetailedGlyphs->SizeOfIncludingThis(aMallocSizeOf);
|
|
}
|
|
|
|
return total;
|
|
}
|
|
|
|
size_t
|
|
gfxTextRun::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf)
|
|
{
|
|
return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf);
|
|
}
|
|
|
|
|
|
#ifdef DEBUG
|
|
void
|
|
gfxTextRun::Dump(FILE* aOutput) {
|
|
if (!aOutput) {
|
|
aOutput = stdout;
|
|
}
|
|
|
|
uint32_t i;
|
|
fputc('[', aOutput);
|
|
for (i = 0; i < mGlyphRuns.Length(); ++i) {
|
|
if (i > 0) {
|
|
fputc(',', aOutput);
|
|
}
|
|
gfxFont* font = mGlyphRuns[i].mFont;
|
|
const gfxFontStyle* style = font->GetStyle();
|
|
NS_ConvertUTF16toUTF8 fontName(font->GetName());
|
|
nsAutoCString lang;
|
|
style->language->ToUTF8String(lang);
|
|
fprintf(aOutput, "%d: %s %f/%d/%d/%s", mGlyphRuns[i].mCharacterOffset,
|
|
fontName.get(), style->size,
|
|
style->weight, style->style, lang.get());
|
|
}
|
|
fputc(']', aOutput);
|
|
}
|
|
#endif
|