path: root/gfx/thebes/gfxCoreTextShaper.cpp

/* -*- Mode: C++; tab-width: 20; indent-tabs-mode: nil; c-basic-offset: 2 -*-
 * 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/. */

#include "mozilla/ArrayUtils.h"
#include "gfxCoreTextShaper.h"
#include "gfxMacFont.h"
#include "gfxFontUtils.h"
#include "gfxTextRun.h"
#include "mozilla/gfx/2D.h"
#include "mozilla/UniquePtrExtensions.h"

#include <algorithm>

#include <dlfcn.h>

using namespace mozilla;

// standard font descriptors that we construct the first time they're needed
CTFontDescriptorRef gfxCoreTextShaper::sFeaturesDescriptor[kMaxFontInstances];

// Helper to create a CFDictionary with the right attributes for shaping our
// text, including imposing the given directionality.
CFDictionaryRef gfxCoreTextShaper::CreateAttrDict(bool aRightToLeft) {
  // Because we always shape unidirectional runs, and may have applied
  // directional overrides, we want to force a direction rather than
  // allowing CoreText to do its own unicode-based bidi processing.
  SInt16 dirOverride = kCTWritingDirectionOverride |
                       (aRightToLeft ? kCTWritingDirectionRightToLeft
                                     : kCTWritingDirectionLeftToRight);
  CFNumberRef dirNumber =
      ::CFNumberCreate(kCFAllocatorDefault, kCFNumberSInt16Type, &dirOverride);
  CFArrayRef dirArray = ::CFArrayCreate(
      kCFAllocatorDefault, (const void**)&dirNumber, 1, &kCFTypeArrayCallBacks);
  ::CFRelease(dirNumber);
  CFTypeRef attrs[] = {kCTFontAttributeName, kCTWritingDirectionAttributeName};
  CFTypeRef values[] = {mCTFont[0], dirArray};
  CFDictionaryRef attrDict = ::CFDictionaryCreate(
      kCFAllocatorDefault, attrs, values, ArrayLength(attrs),
      &kCFTypeDictionaryKeyCallBacks, &kCFTypeDictionaryValueCallBacks);
  ::CFRelease(dirArray);
  return attrDict;
}

gfxCoreTextShaper::gfxCoreTextShaper(gfxMacFont* aFont)
    : gfxFontShaper(aFont),
      mAttributesDictLTR(nullptr),
      mAttributesDictRTL(nullptr) {
  for (size_t i = 0; i < kMaxFontInstances; i++) {
    mCTFont[i] = nullptr;
  }
  // Create our default CTFontRef
  mCTFont[0] = CreateCTFontWithFeatures(
      aFont->GetAdjustedSize(), GetFeaturesDescriptor(kDefaultFeatures));
}

gfxCoreTextShaper::~gfxCoreTextShaper() {
  if (mAttributesDictLTR) {
    ::CFRelease(mAttributesDictLTR);
  }
  if (mAttributesDictRTL) {
    ::CFRelease(mAttributesDictRTL);
  }
  for (size_t i = 0; i < kMaxFontInstances; i++) {
    if (mCTFont[i]) {
      ::CFRelease(mCTFont[i]);
    }
  }
}

static bool IsBuggyIndicScript(unicode::Script aScript) {
  return aScript == unicode::Script::BENGALI ||
         aScript == unicode::Script::KANNADA ||
         aScript == unicode::Script::ORIYA || aScript == unicode::Script::KHMER;
}

bool gfxCoreTextShaper::ShapeText(DrawTarget* aDrawTarget,
                                  const char16_t* aText, uint32_t aOffset,
                                  uint32_t aLength, Script aScript,
                                  nsAtom* aLanguage, bool aVertical,
                                  RoundingFlags aRounding,
                                  gfxShapedText* aShapedText) {
  // Create a CFAttributedString with text and style info, so we can use
  // CoreText to lay it out.
  bool isRightToLeft = aShapedText->IsRightToLeft();
  const UniChar* text = reinterpret_cast<const UniChar*>(aText);

  CFStringRef stringObj = ::CFStringCreateWithCharactersNoCopy(
      kCFAllocatorDefault, text, aLength, kCFAllocatorNull);

  // Figure out whether we should try to set the AAT small-caps feature:
  // examine OpenType tags for the requested style, and see if 'smcp' is
  // among them.
  const gfxFontStyle* style = mFont->GetStyle();
  gfxFontEntry* entry = mFont->GetFontEntry();
  auto handleFeatureTag = [](const uint32_t& aTag, uint32_t& aValue,
                             void* aUserArg) -> void {
    if (aTag == HB_TAG('s', 'm', 'c', 'p') && aValue) {
      *static_cast<bool*>(aUserArg) = true;
    }
  };
  bool addSmallCaps = false;
  MergeFontFeatures(style, entry->mFeatureSettings, false, entry->FamilyName(),
                    false, handleFeatureTag, &addSmallCaps);

  // Get an attributes dictionary suitable for shaping text in the
  // current direction, creating it if necessary.
  CFDictionaryRef attrObj =
      isRightToLeft ? mAttributesDictRTL : mAttributesDictLTR;
  if (!attrObj) {
    attrObj = CreateAttrDict(isRightToLeft);
    (isRightToLeft ? mAttributesDictRTL : mAttributesDictLTR) = attrObj;
  }

  FeatureFlags featureFlags = kDefaultFeatures;
  if (IsBuggyIndicScript(aScript)) {
    // To work around buggy Indic AAT fonts shipped with OS X,
    // we re-enable the Line Initial Smart Swashes feature that is needed
    // for "split vowels" to work in at least Bengali and Kannada fonts.
    // Affected fonts include Bangla MN, Bangla Sangam MN, Kannada MN,
    // Kannada Sangam MN. See bugs 686225, 728557, 953231, 1145515.
    // Also applies to Oriya and Khmer, see bug 1370927 and bug 1403166.
    featureFlags |= kIndicFeatures;
  }
  if (aShapedText->DisableLigatures()) {
    // For letterspacing (or maybe other situations) we need to make
    // a copy of the CTFont with the ligature feature disabled.
    featureFlags |= kDisableLigatures;
  }
  if (addSmallCaps) {
    featureFlags |= kAddSmallCaps;
  }

  // For the disabled-ligature, buggy-indic-font or small-caps case, replace
  // the default CTFont in the attribute dictionary with a tweaked version.
  CFMutableDictionaryRef mutableAttr = nullptr;
  if (featureFlags != 0) {
    if (!mCTFont[featureFlags]) {
      mCTFont[featureFlags] = CreateCTFontWithFeatures(
          mFont->GetAdjustedSize(), GetFeaturesDescriptor(featureFlags));
    }
    mutableAttr =
        ::CFDictionaryCreateMutableCopy(kCFAllocatorDefault, 2, attrObj);
    ::CFDictionaryReplaceValue(mutableAttr, kCTFontAttributeName,
                               mCTFont[featureFlags]);
    attrObj = mutableAttr;
  }

  // Now we can create an attributed string
  CFAttributedStringRef attrStringObj =
      ::CFAttributedStringCreate(kCFAllocatorDefault, stringObj, attrObj);
  ::CFRelease(stringObj);

  // Create the CoreText line from our string, then we're done with it
  CTLineRef line = ::CTLineCreateWithAttributedString(attrStringObj);
  ::CFRelease(attrStringObj);

  // and finally retrieve the glyph data and store into the gfxTextRun
  CFArrayRef glyphRuns = ::CTLineGetGlyphRuns(line);
  uint32_t numRuns = ::CFArrayGetCount(glyphRuns);

  // Iterate through the glyph runs.
  bool success = true;
  for (uint32_t runIndex = 0; runIndex < numRuns; runIndex++) {
    CTRunRef aCTRun = (CTRunRef)::CFArrayGetValueAtIndex(glyphRuns, runIndex);
    CFRange range = ::CTRunGetStringRange(aCTRun);
    CFDictionaryRef runAttr = ::CTRunGetAttributes(aCTRun);
    if (runAttr != attrObj) {
      // If Core Text manufactured a new dictionary, this may indicate
      // unexpected font substitution. In that case, we fail (and fall
      // back to harfbuzz shaping)...
      const void* font1 = ::CFDictionaryGetValue(attrObj, kCTFontAttributeName);
      const void* font2 = ::CFDictionaryGetValue(runAttr, kCTFontAttributeName);
      if (font1 != font2) {
        // ...except that if the fallback was only for a variation
        // selector or join control that is otherwise unsupported,
        // we just ignore it.
        if (range.length == 1) {
          char16_t ch = aText[range.location];
          if (gfxFontUtils::IsJoinControl(ch) ||
              gfxFontUtils::IsVarSelector(ch)) {
            continue;
          }
        }
        NS_WARNING("unexpected font fallback in Core Text");
        success = false;
        break;
      }
    }
    if (SetGlyphsFromRun(aShapedText, aOffset, aLength, aCTRun) != NS_OK) {
      success = false;
      break;
    }
  }

  if (mutableAttr) {
    ::CFRelease(mutableAttr);
  }
  ::CFRelease(line);

  return success;
}

#define SMALL_GLYPH_RUN \
  128  // preallocated size of our auto arrays for per-glyph data;
       // some testing indicates that 90%+ of glyph runs will fit
       // without requiring a separate allocation

nsresult gfxCoreTextShaper::SetGlyphsFromRun(gfxShapedText* aShapedText,
                                             uint32_t aOffset, uint32_t aLength,
                                             CTRunRef aCTRun) {
  typedef gfxShapedText::CompressedGlyph CompressedGlyph;

  int32_t direction = aShapedText->IsRightToLeft() ? -1 : 1;

  int32_t numGlyphs = ::CTRunGetGlyphCount(aCTRun);
  if (numGlyphs == 0) {
    return NS_OK;
  }

  int32_t wordLength = aLength;

  // character offsets get really confusing here, as we have to keep track of
  // (a) the text in the actual textRun we're constructing
  // (c) the string that was handed to CoreText, which contains the text of
  // the font run
  // (d) the CTRun currently being processed, which may be a sub-run of the
  // CoreText line

  // get the source string range within the CTLine's text
  CFRange stringRange = ::CTRunGetStringRange(aCTRun);
  // skip the run if it is entirely outside the actual range of the font run
  if (stringRange.location + stringRange.length <= 0 ||
      stringRange.location >= wordLength) {
    return NS_OK;
  }

  // retrieve the laid-out glyph data from the CTRun
  UniquePtr<CGGlyph[]> glyphsArray;
  UniquePtr<CGPoint[]> positionsArray;
  UniquePtr<CFIndex[]> glyphToCharArray;
  const CGGlyph* glyphs = nullptr;
  const CGPoint* positions = nullptr;
  const CFIndex* glyphToChar = nullptr;

  // Testing indicates that CTRunGetGlyphsPtr (almost?) always succeeds,
  // and so allocating a new array and copying data with CTRunGetGlyphs
  // will be extremely rare.
  // If this were not the case, we could use an AutoTArray<> to
  // try and avoid the heap allocation for small runs.
  // It's possible that some future change to CoreText will mean that
  // CTRunGetGlyphsPtr fails more often; if this happens, AutoTArray<>
  // may become an attractive option.
  glyphs = ::CTRunGetGlyphsPtr(aCTRun);
  if (!glyphs) {
    glyphsArray = MakeUniqueFallible<CGGlyph[]>(numGlyphs);
    if (!glyphsArray) {
      return NS_ERROR_OUT_OF_MEMORY;
    }
    ::CTRunGetGlyphs(aCTRun, ::CFRangeMake(0, 0), glyphsArray.get());
    glyphs = glyphsArray.get();
  }

  positions = ::CTRunGetPositionsPtr(aCTRun);
  if (!positions) {
    positionsArray = MakeUniqueFallible<CGPoint[]>(numGlyphs);
    if (!positionsArray) {
      return NS_ERROR_OUT_OF_MEMORY;
    }
    ::CTRunGetPositions(aCTRun, ::CFRangeMake(0, 0), positionsArray.get());
    positions = positionsArray.get();
  }

  // Remember that the glyphToChar indices relate to the CoreText line,
  // not to the beginning of the textRun, the font run,
  // or the stringRange of the glyph run
  glyphToChar = ::CTRunGetStringIndicesPtr(aCTRun);
  if (!glyphToChar) {
    glyphToCharArray = MakeUniqueFallible<CFIndex[]>(numGlyphs);
    if (!glyphToCharArray) {
      return NS_ERROR_OUT_OF_MEMORY;
    }
    ::CTRunGetStringIndices(aCTRun, ::CFRangeMake(0, 0),
                            glyphToCharArray.get());
    glyphToChar = glyphToCharArray.get();
  }

  double runWidth = ::CTRunGetTypographicBounds(aCTRun, ::CFRangeMake(0, 0),
                                                nullptr, nullptr, nullptr);

  AutoTArray<gfxShapedText::DetailedGlyph, 1> detailedGlyphs;
  CompressedGlyph* charGlyphs = aShapedText->GetCharacterGlyphs() + aOffset;

  // CoreText gives us the glyphindex-to-charindex mapping, which relates each
  // glyph to a source text character; we also need the charindex-to-glyphindex
  // mapping to find the glyph for a given char. Note that some chars may not
  // map to any glyph (ligature continuations), and some may map to several
  // glyphs (eg Indic split vowels). We set the glyph index to NO_GLYPH for
  // chars that have no associated glyph, and we record the last glyph index for
  // cases where the char maps to several glyphs, so that our clumping will
  // include all the glyph fragments for the character.

  // The charToGlyph array is indexed by char position within the stringRange of
  // the glyph run.

  static const int32_t NO_GLYPH = -1;
  AutoTArray<int32_t, SMALL_GLYPH_RUN> charToGlyphArray;
  if (!charToGlyphArray.SetLength(stringRange.length, fallible)) {
    return NS_ERROR_OUT_OF_MEMORY;
  }
  int32_t* charToGlyph = charToGlyphArray.Elements();
  for (int32_t offset = 0; offset < stringRange.length; ++offset) {
    charToGlyph[offset] = NO_GLYPH;
  }
  for (int32_t i = 0; i < numGlyphs; ++i) {
    int32_t loc = glyphToChar[i] - stringRange.location;
    if (loc >= 0 && loc < stringRange.length) {
      charToGlyph[loc] = i;
    }
  }

  // Find character and glyph clumps that correspond, allowing for ligatures,
  // indic reordering, split glyphs, etc.
  //
  // The idea is that we'll find a character sequence starting at the first char
  // of stringRange, and extend it until it includes the character associated
  // with the first glyph; we also extend it as long as there are "holes" in the
  // range of glyphs. So we will eventually have a contiguous sequence of
  // characters, starting at the beginning of the range, that map to a
  // contiguous sequence of glyphs, starting at the beginning of the glyph
  // array. That's a clump; then we update the starting positions and repeat.
  //
  // NB: In the case of RTL layouts, we iterate over the stringRange in reverse.
  //

  // This may find characters that fall outside the range 0:wordLength,
  // so we won't necessarily use everything we find here.

  bool isRightToLeft = aShapedText->IsRightToLeft();
  int32_t glyphStart =
      0;  // looking for a clump that starts at this glyph index
  int32_t charStart =
      isRightToLeft
          ? stringRange.length - 1
          : 0;  // and this char index (in the stringRange of the glyph run)

  while (glyphStart <
         numGlyphs) {  // keep finding groups until all glyphs are accounted for
    bool inOrder = true;
    int32_t charEnd = glyphToChar[glyphStart] - stringRange.location;
    NS_WARNING_ASSERTION(charEnd >= 0 && charEnd < stringRange.length,
                         "glyph-to-char mapping points outside string range");
    // clamp charEnd to the valid range of the string
    charEnd = std::max(charEnd, 0);
    charEnd = std::min(charEnd, int32_t(stringRange.length));

    int32_t glyphEnd = glyphStart;
    int32_t charLimit = isRightToLeft ? -1 : stringRange.length;
    do {
      // This is normally executed once for each iteration of the outer loop,
      // but in unusual cases where the character/glyph association is complex,
      // the initial character range might correspond to a non-contiguous
      // glyph range with "holes" in it. If so, we will repeat this loop to
      // extend the character range until we have a contiguous glyph sequence.
      NS_ASSERTION((direction > 0 && charEnd < charLimit) ||
                       (direction < 0 && charEnd > charLimit),
                   "no characters left in range?");
      charEnd += direction;
      while (charEnd != charLimit && charToGlyph[charEnd] == NO_GLYPH) {
        charEnd += direction;
      }

      // find the maximum glyph index covered by the clump so far
      if (isRightToLeft) {
        for (int32_t i = charStart; i > charEnd; --i) {
          if (charToGlyph[i] != NO_GLYPH) {
            // update extent of glyph range
            glyphEnd = std::max(glyphEnd, charToGlyph[i] + 1);
          }
        }
      } else {
        for (int32_t i = charStart; i < charEnd; ++i) {
          if (charToGlyph[i] != NO_GLYPH) {
            // update extent of glyph range
            glyphEnd = std::max(glyphEnd, charToGlyph[i] + 1);
          }
        }
      }

      if (glyphEnd == glyphStart + 1) {
        // for the common case of a single-glyph clump, we can skip the
        // following checks
        break;
      }

      if (glyphEnd == glyphStart) {
        // no glyphs, try to extend the clump
        continue;
      }

      // check whether all glyphs in the range are associated with the
      // characters in our clump; if not, we have a discontinuous range, and
      // should extend it unless we've reached the end of the text
      bool allGlyphsAreWithinCluster = true;
      int32_t prevGlyphCharIndex = charStart;
      for (int32_t i = glyphStart; i < glyphEnd; ++i) {
        int32_t glyphCharIndex = glyphToChar[i] - stringRange.location;
        if (isRightToLeft) {
          if (glyphCharIndex > charStart || glyphCharIndex <= charEnd) {
            allGlyphsAreWithinCluster = false;
            break;
          }
          if (glyphCharIndex > prevGlyphCharIndex) {
            inOrder = false;
          }
          prevGlyphCharIndex = glyphCharIndex;
        } else {
          if (glyphCharIndex < charStart || glyphCharIndex >= charEnd) {
            allGlyphsAreWithinCluster = false;
            break;
          }
          if (glyphCharIndex < prevGlyphCharIndex) {
            inOrder = false;
          }
          prevGlyphCharIndex = glyphCharIndex;
        }
      }
      if (allGlyphsAreWithinCluster) {
        break;
      }
    } while (charEnd != charLimit);

    NS_WARNING_ASSERTION(glyphStart < glyphEnd,
                         "character/glyph clump contains no glyphs!");
    if (glyphStart == glyphEnd) {
      ++glyphStart;  // make progress - avoid potential infinite loop
      charStart = charEnd;
      continue;
    }

    NS_WARNING_ASSERTION(charStart != charEnd,
                         "character/glyph clump contains no characters!");
    if (charStart == charEnd) {
      glyphStart = glyphEnd;  // this is bad - we'll discard the glyph(s),
                              // as there's nowhere to attach them
      continue;
    }

    // Now charStart..charEnd is a ligature clump, corresponding to
    // glyphStart..glyphEnd; Set baseCharIndex to the char we'll actually attach
    // the glyphs to (1st of ligature), and endCharIndex to the limit (position
    // beyond the last char), adjusting for the offset of the stringRange
    // relative to the textRun.
    int32_t baseCharIndex, endCharIndex;
    if (isRightToLeft) {
      while (charEnd >= 0 && charToGlyph[charEnd] == NO_GLYPH) {
        charEnd--;
      }
      baseCharIndex = charEnd + stringRange.location + 1;
      endCharIndex = charStart + stringRange.location + 1;
    } else {
      while (charEnd < stringRange.length && charToGlyph[charEnd] == NO_GLYPH) {
        charEnd++;
      }
      baseCharIndex = charStart + stringRange.location;
      endCharIndex = charEnd + stringRange.location;
    }

    // Then we check if the clump falls outside our actual string range; if so,
    // just go to the next.
    if (endCharIndex <= 0 || baseCharIndex >= wordLength) {
      glyphStart = glyphEnd;
      charStart = charEnd;
      continue;
    }
    // Ensure we won't try to go beyond the valid length of the word's text
    baseCharIndex = std::max(baseCharIndex, 0);
    endCharIndex = std::min(endCharIndex, wordLength);

    // Now we're ready to set the glyph info in the textRun; measure the glyph
    // width of the first (perhaps only) glyph, to see if it is "Simple"
    int32_t appUnitsPerDevUnit = aShapedText->GetAppUnitsPerDevUnit();
    double toNextGlyph;
    if (glyphStart < numGlyphs - 1) {
      toNextGlyph = positions[glyphStart + 1].x - positions[glyphStart].x;
    } else {
      toNextGlyph = positions[0].x + runWidth - positions[glyphStart].x;
    }
    int32_t advance = int32_t(toNextGlyph * appUnitsPerDevUnit);

    // Check if it's a simple one-to-one mapping
    int32_t glyphsInClump = glyphEnd - glyphStart;
    if (glyphsInClump == 1 &&
        gfxTextRun::CompressedGlyph::IsSimpleGlyphID(glyphs[glyphStart]) &&
        gfxTextRun::CompressedGlyph::IsSimpleAdvance(advance) &&
        charGlyphs[baseCharIndex].IsClusterStart() &&
        positions[glyphStart].y == 0.0) {
      charGlyphs[baseCharIndex].SetSimpleGlyph(advance, glyphs[glyphStart]);
    } else {
      // collect all glyphs in a list to be assigned to the first char;
      // there must be at least one in the clump, and we already measured its
      // advance, hence the placement of the loop-exit test and the measurement
      // of the next glyph
      while (true) {
        gfxTextRun::DetailedGlyph* details = detailedGlyphs.AppendElement();
        details->mGlyphID = glyphs[glyphStart];
        details->mOffset.y = -positions[glyphStart].y * appUnitsPerDevUnit;
        details->mAdvance = advance;
        if (++glyphStart >= glyphEnd) {
          break;
        }
        if (glyphStart < numGlyphs - 1) {
          toNextGlyph = positions[glyphStart + 1].x - positions[glyphStart].x;
        } else {
          toNextGlyph = positions[0].x + runWidth - positions[glyphStart].x;
        }
        advance = int32_t(toNextGlyph * appUnitsPerDevUnit);
      }

      aShapedText->SetDetailedGlyphs(aOffset + baseCharIndex,
                                     detailedGlyphs.Length(),
                                     detailedGlyphs.Elements());

      detailedGlyphs.Clear();
    }

    // the rest of the chars in the group are ligature continuations, no
    // associated glyphs
    while (++baseCharIndex != endCharIndex && baseCharIndex < wordLength) {
      CompressedGlyph& shapedTextGlyph = charGlyphs[baseCharIndex];
      NS_ASSERTION(!shapedTextGlyph.IsSimpleGlyph(),
                   "overwriting a simple glyph");
      shapedTextGlyph.SetComplex(inOrder && shapedTextGlyph.IsClusterStart(),
                                 false);
    }

    glyphStart = glyphEnd;
    charStart = charEnd;
  }

  return NS_OK;
}

#undef SMALL_GLYPH_RUN

// Construct the font attribute descriptor that we'll apply by default when
// creating a CTFontRef. This will turn off line-edge swashes by default,
// because we don't know the actual line breaks when doing glyph shaping.

// We also cache feature descriptors for shaping with disabled ligatures, and
// for buggy Indic AAT font workarounds, created on an as-needed basis.

#define MAX_FEATURES 5  // max used by any of our Get*Descriptor functions

CTFontDescriptorRef gfxCoreTextShaper::CreateFontFeaturesDescriptor(
    const std::pair<SInt16, SInt16>* aFeatures, size_t aCount) {
  MOZ_ASSERT(aCount <= MAX_FEATURES);

  CFDictionaryRef featureSettings[MAX_FEATURES];

  for (size_t i = 0; i < aCount; i++) {
    CFNumberRef type = ::CFNumberCreate(
        kCFAllocatorDefault, kCFNumberSInt16Type, &aFeatures[i].first);
    CFNumberRef selector = ::CFNumberCreate(
        kCFAllocatorDefault, kCFNumberSInt16Type, &aFeatures[i].second);

    CFTypeRef keys[] = {kCTFontFeatureTypeIdentifierKey,
                        kCTFontFeatureSelectorIdentifierKey};
    CFTypeRef values[] = {type, selector};
    featureSettings[i] = ::CFDictionaryCreate(
        kCFAllocatorDefault, (const void**)keys, (const void**)values,
        ArrayLength(keys), &kCFTypeDictionaryKeyCallBacks,
        &kCFTypeDictionaryValueCallBacks);

    ::CFRelease(selector);
    ::CFRelease(type);
  }

  CFArrayRef featuresArray =
      ::CFArrayCreate(kCFAllocatorDefault, (const void**)featureSettings,
                      aCount,  // not ArrayLength(featureSettings), as we
                               // may not have used all the allocated slots
                      &kCFTypeArrayCallBacks);

  for (size_t i = 0; i < aCount; i++) {
    ::CFRelease(featureSettings[i]);
  }

  const CFTypeRef attrKeys[] = {kCTFontFeatureSettingsAttribute};
  const CFTypeRef attrValues[] = {featuresArray};
  CFDictionaryRef attributesDict = ::CFDictionaryCreate(
      kCFAllocatorDefault, (const void**)attrKeys, (const void**)attrValues,
      ArrayLength(attrKeys), &kCFTypeDictionaryKeyCallBacks,
      &kCFTypeDictionaryValueCallBacks);
  ::CFRelease(featuresArray);

  CTFontDescriptorRef descriptor =
      ::CTFontDescriptorCreateWithAttributes(attributesDict);
  ::CFRelease(attributesDict);

  return descriptor;
}

CTFontDescriptorRef gfxCoreTextShaper::GetFeaturesDescriptor(
    FeatureFlags aFeatureFlags) {
  MOZ_ASSERT(aFeatureFlags < kMaxFontInstances);
  if (!sFeaturesDescriptor[aFeatureFlags]) {
    typedef std::pair<SInt16, SInt16> FeatT;
    AutoTArray<FeatT, MAX_FEATURES> features;
    features.AppendElement(
        FeatT(kSmartSwashType, kLineFinalSwashesOffSelector));
    if ((aFeatureFlags & kIndicFeatures) == 0) {
      features.AppendElement(
          FeatT(kSmartSwashType, kLineInitialSwashesOffSelector));
    }
    if (aFeatureFlags & kAddSmallCaps) {
      features.AppendElement(FeatT(kLetterCaseType, kSmallCapsSelector));
      features.AppendElement(
          FeatT(kLowerCaseType, kLowerCaseSmallCapsSelector));
    }
    if (aFeatureFlags & kDisableLigatures) {
      features.AppendElement(
          FeatT(kLigaturesType, kCommonLigaturesOffSelector));
    }
    MOZ_ASSERT(features.Length() <= MAX_FEATURES);
    sFeaturesDescriptor[aFeatureFlags] =
        CreateFontFeaturesDescriptor(features.Elements(), features.Length());
  }
  return sFeaturesDescriptor[aFeatureFlags];
}

CTFontRef gfxCoreTextShaper::CreateCTFontWithFeatures(
    CGFloat aSize, CTFontDescriptorRef aDescriptor) {
  const gfxFontEntry* fe = mFont->GetFontEntry();
  bool isInstalledFont = !fe->IsUserFont() || fe->IsLocalUserFont();
  CGFontRef cgFont = static_cast<gfxMacFont*>(mFont)->GetCGFontRef();
  return gfxMacFont::CreateCTFontFromCGFontWithVariations(
      cgFont, aSize, isInstalledFont, aDescriptor);
}

void gfxCoreTextShaper::Shutdown()  // [static]
{
  for (size_t i = 0; i < kMaxFontInstances; i++) {
    if (sFeaturesDescriptor[i] != nullptr) {
      ::CFRelease(sFeaturesDescriptor[i]);
      sFeaturesDescriptor[i] = nullptr;
    }
  }
}