gecko-dev/layout/generic/nsTextRunTransformations.cpp

/* -*- Mode: C++; tab-width: 2; 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 "nsTextRunTransformations.h"

#include "nsTextFrameUtils.h"
#include "gfxSkipChars.h"
#include "nsGkAtoms.h"

#include "nsStyleConsts.h"
#include "nsStyleContext.h"
#include "gfxContext.h"
#include "nsUnicodeProperties.h"
#include "nsSpecialCasingData.h"

// Unicode characters needing special casing treatment in tr/az languages
#define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE  0x0130
#define LATIN_SMALL_LETTER_DOTLESS_I           0x0131

// Greek sigma needs custom handling for the lowercase transform; for details
// see comments under "case NS_STYLE_TEXT_TRANSFORM_LOWERCASE" within
// nsCaseTransformTextRunFactory::RebuildTextRun(), and bug 740120.
#define GREEK_CAPITAL_LETTER_SIGMA             0x03A3
#define GREEK_SMALL_LETTER_FINAL_SIGMA         0x03C2
#define GREEK_SMALL_LETTER_SIGMA               0x03C3

// Custom uppercase mapping for Greek; see bug 307039 for details
#define GREEK_LOWER_ALPHA                      0x03B1
#define GREEK_LOWER_ALPHA_TONOS                0x03AC
#define GREEK_LOWER_ALPHA_OXIA                 0x1F71
#define GREEK_LOWER_EPSILON                    0x03B5
#define GREEK_LOWER_EPSILON_TONOS              0x03AD
#define GREEK_LOWER_EPSILON_OXIA               0x1F73
#define GREEK_LOWER_ETA                        0x03B7
#define GREEK_LOWER_ETA_TONOS                  0x03AE
#define GREEK_LOWER_ETA_OXIA                   0x1F75
#define GREEK_LOWER_IOTA                       0x03B9
#define GREEK_LOWER_IOTA_TONOS                 0x03AF
#define GREEK_LOWER_IOTA_OXIA                  0x1F77
#define GREEK_LOWER_IOTA_DIALYTIKA             0x03CA
#define GREEK_LOWER_IOTA_DIALYTIKA_TONOS       0x0390
#define GREEK_LOWER_IOTA_DIALYTIKA_OXIA        0x1FD3
#define GREEK_LOWER_OMICRON                    0x03BF
#define GREEK_LOWER_OMICRON_TONOS              0x03CC
#define GREEK_LOWER_OMICRON_OXIA               0x1F79
#define GREEK_LOWER_UPSILON                    0x03C5
#define GREEK_LOWER_UPSILON_TONOS              0x03CD
#define GREEK_LOWER_UPSILON_OXIA               0x1F7B
#define GREEK_LOWER_UPSILON_DIALYTIKA          0x03CB
#define GREEK_LOWER_UPSILON_DIALYTIKA_TONOS    0x03B0
#define GREEK_LOWER_UPSILON_DIALYTIKA_OXIA     0x1FE3
#define GREEK_LOWER_OMEGA                      0x03C9
#define GREEK_LOWER_OMEGA_TONOS                0x03CE
#define GREEK_LOWER_OMEGA_OXIA                 0x1F7D
#define GREEK_UPPER_ALPHA                      0x0391
#define GREEK_UPPER_EPSILON                    0x0395
#define GREEK_UPPER_ETA                        0x0397
#define GREEK_UPPER_IOTA                       0x0399
#define GREEK_UPPER_IOTA_DIALYTIKA             0x03AA
#define GREEK_UPPER_OMICRON                    0x039F
#define GREEK_UPPER_UPSILON                    0x03A5
#define GREEK_UPPER_UPSILON_DIALYTIKA          0x03AB
#define GREEK_UPPER_OMEGA                      0x03A9
#define GREEK_UPPER_ALPHA_TONOS                0x0386
#define GREEK_UPPER_ALPHA_OXIA                 0x1FBB
#define GREEK_UPPER_EPSILON_TONOS              0x0388
#define GREEK_UPPER_EPSILON_OXIA               0x1FC9
#define GREEK_UPPER_ETA_TONOS                  0x0389
#define GREEK_UPPER_ETA_OXIA                   0x1FCB
#define GREEK_UPPER_IOTA_TONOS                 0x038A
#define GREEK_UPPER_IOTA_OXIA                  0x1FDB
#define GREEK_UPPER_OMICRON_TONOS              0x038C
#define GREEK_UPPER_OMICRON_OXIA               0x1FF9
#define GREEK_UPPER_UPSILON_TONOS              0x038E
#define GREEK_UPPER_UPSILON_OXIA               0x1FEB
#define GREEK_UPPER_OMEGA_TONOS                0x038F
#define GREEK_UPPER_OMEGA_OXIA                 0x1FFB
#define COMBINING_ACUTE_ACCENT                 0x0301
#define COMBINING_DIAERESIS                    0x0308
#define COMBINING_ACUTE_TONE_MARK              0x0341
#define COMBINING_GREEK_DIALYTIKA_TONOS        0x0344

// When doing an Uppercase transform in Greek, we need to keep track of the
// current state while iterating through the string, to recognize and process
// diphthongs correctly. For clarity, we define a state for each vowel and
// each vowel with accent, although a few of these do not actually need any
// special treatment and could be folded into kStart.
enum GreekCasingState {
  kStart,
  kAlpha,
  kEpsilon,
  kEta,
  kIota,
  kOmicron,
  kUpsilon,
  kOmega,
  kAlphaAcc,
  kEpsilonAcc,
  kEtaAcc,
  kIotaAcc,
  kOmicronAcc,
  kUpsilonAcc,
  kOmegaAcc,
  kOmicronUpsilon,
  kDiaeresis
};

static PRUint32
GreekUpperCase(PRUint32 aCh, GreekCasingState* aState)
{
  switch (aCh) {
  case GREEK_UPPER_ALPHA:
  case GREEK_LOWER_ALPHA:
    *aState = kAlpha;
    return GREEK_UPPER_ALPHA;

  case GREEK_UPPER_EPSILON:
  case GREEK_LOWER_EPSILON:
    *aState = kEpsilon;
    return GREEK_UPPER_EPSILON;

  case GREEK_UPPER_ETA:
  case GREEK_LOWER_ETA:
    *aState = kEta;
    return GREEK_UPPER_ETA;

  case GREEK_UPPER_IOTA:
    *aState = kIota;
    return GREEK_UPPER_IOTA;

  case GREEK_UPPER_OMICRON:
  case GREEK_LOWER_OMICRON:
    *aState = kOmicron;
    return GREEK_UPPER_OMICRON;

  case GREEK_UPPER_UPSILON:
    switch (*aState) {
    case kOmicron:
      *aState = kOmicronUpsilon;
      break;
    default:
      *aState = kUpsilon;
      break;
    }
    return GREEK_UPPER_UPSILON;

  case GREEK_UPPER_OMEGA:
  case GREEK_LOWER_OMEGA:
    *aState = kOmega;
    return GREEK_UPPER_OMEGA;

  // iota and upsilon may be the second vowel of a diphthong
  case GREEK_LOWER_IOTA:
    switch (*aState) {
    case kAlphaAcc:
    case kEpsilonAcc:
    case kOmicronAcc:
    case kUpsilonAcc:
      *aState = kStart;
      return GREEK_UPPER_IOTA_DIALYTIKA;
    default:
      break;
    }
    *aState = kIota;
    return GREEK_UPPER_IOTA;

  case GREEK_LOWER_UPSILON:
    switch (*aState) {
    case kAlphaAcc:
    case kEpsilonAcc:
    case kEtaAcc:
    case kOmicronAcc:
      *aState = kStart;
      return GREEK_UPPER_UPSILON_DIALYTIKA;
    case kOmicron:
      *aState = kOmicronUpsilon;
      break;
    default:
      *aState = kUpsilon;
      break;
    }
    return GREEK_UPPER_UPSILON;

  case GREEK_UPPER_IOTA_DIALYTIKA:
  case GREEK_LOWER_IOTA_DIALYTIKA:
  case GREEK_UPPER_UPSILON_DIALYTIKA:
  case GREEK_LOWER_UPSILON_DIALYTIKA:
  case COMBINING_DIAERESIS:
    *aState = kDiaeresis;
    return ToUpperCase(aCh);

  // remove accent if it follows a vowel or diaeresis,
  // and set appropriate state for diphthong detection
  case COMBINING_ACUTE_ACCENT:
  case COMBINING_ACUTE_TONE_MARK:
    switch (*aState) {
    case kAlpha:
      *aState = kAlphaAcc;
      return PRUint32(-1); // omit this char from result string
    case kEpsilon:
      *aState = kEpsilonAcc;
      return PRUint32(-1);
    case kEta:
      *aState = kEtaAcc;
      return PRUint32(-1);
    case kIota:
      *aState = kIotaAcc;
      return PRUint32(-1);
    case kOmicron:
      *aState = kOmicronAcc;
      return PRUint32(-1);
    case kUpsilon:
      *aState = kUpsilonAcc;
      return PRUint32(-1);
    case kOmicronUpsilon:
      *aState = kStart; // this completed a diphthong
      return PRUint32(-1);
    case kOmega:
      *aState = kOmegaAcc;
      return PRUint32(-1);
    case kDiaeresis:
      *aState = kStart;
      return PRUint32(-1);
    default:
      break;
    }
    break;

  // combinations with dieresis+accent just strip the accent,
  // and reset to start state (don't form diphthong with following vowel)
  case GREEK_LOWER_IOTA_DIALYTIKA_TONOS:
  case GREEK_LOWER_IOTA_DIALYTIKA_OXIA:
    *aState = kStart;
    return GREEK_UPPER_IOTA_DIALYTIKA;

  case GREEK_LOWER_UPSILON_DIALYTIKA_TONOS:
  case GREEK_LOWER_UPSILON_DIALYTIKA_OXIA:
    *aState = kStart;
    return GREEK_UPPER_UPSILON_DIALYTIKA;

  case COMBINING_GREEK_DIALYTIKA_TONOS:
    *aState = kStart;
    return COMBINING_DIAERESIS;

  // strip accents from vowels, and note the vowel seen so that we can detect
  // diphthongs where diaeresis needs to be added
  case GREEK_LOWER_ALPHA_TONOS:
  case GREEK_LOWER_ALPHA_OXIA:
  case GREEK_UPPER_ALPHA_TONOS:
  case GREEK_UPPER_ALPHA_OXIA:
    *aState = kAlphaAcc;
    return GREEK_UPPER_ALPHA;

  case GREEK_LOWER_EPSILON_TONOS:
  case GREEK_LOWER_EPSILON_OXIA:
  case GREEK_UPPER_EPSILON_TONOS:
  case GREEK_UPPER_EPSILON_OXIA:
    *aState = kEpsilonAcc;
    return GREEK_UPPER_EPSILON;

  case GREEK_LOWER_ETA_TONOS:
  case GREEK_LOWER_ETA_OXIA:
  case GREEK_UPPER_ETA_TONOS:
  case GREEK_UPPER_ETA_OXIA:
    *aState = kEtaAcc;
    return GREEK_UPPER_ETA;

  case GREEK_LOWER_IOTA_TONOS:
  case GREEK_LOWER_IOTA_OXIA:
  case GREEK_UPPER_IOTA_TONOS:
  case GREEK_UPPER_IOTA_OXIA:
    *aState = kIotaAcc;
    return GREEK_UPPER_IOTA;

  case GREEK_LOWER_OMICRON_TONOS:
  case GREEK_LOWER_OMICRON_OXIA:
  case GREEK_UPPER_OMICRON_TONOS:
  case GREEK_UPPER_OMICRON_OXIA:
    *aState = kOmicronAcc;
    return GREEK_UPPER_OMICRON;

  case GREEK_LOWER_UPSILON_TONOS:
  case GREEK_LOWER_UPSILON_OXIA:
  case GREEK_UPPER_UPSILON_TONOS:
  case GREEK_UPPER_UPSILON_OXIA:
    switch (*aState) {
    case kOmicron:
      *aState = kStart; // this completed a diphthong
      break;
    default:
      *aState = kUpsilonAcc;
      break;
    }
    return GREEK_UPPER_UPSILON;

  case GREEK_LOWER_OMEGA_TONOS:
  case GREEK_LOWER_OMEGA_OXIA:
  case GREEK_UPPER_OMEGA_TONOS:
  case GREEK_UPPER_OMEGA_OXIA:
    *aState = kOmegaAcc;
    return GREEK_UPPER_OMEGA;
  }

  // all other characters just reset the state, and use standard mappings
  *aState = kStart;
  return ToUpperCase(aCh);
}

nsTransformedTextRun *
nsTransformedTextRun::Create(const gfxTextRunFactory::Parameters* aParams,
                             nsTransformingTextRunFactory* aFactory,
                             gfxFontGroup* aFontGroup,
                             const PRUnichar* aString, PRUint32 aLength,
                             const PRUint32 aFlags, nsStyleContext** aStyles,
                             bool aOwnsFactory)
{
  NS_ASSERTION(!(aFlags & gfxTextRunFactory::TEXT_IS_8BIT),
               "didn't expect text to be marked as 8-bit here");

  void *storage = AllocateStorageForTextRun(sizeof(nsTransformedTextRun), aLength);
  if (!storage) {
    return nullptr;
  }

  return new (storage) nsTransformedTextRun(aParams, aFactory, aFontGroup,
                                            aString, aLength,
                                            aFlags, aStyles, aOwnsFactory);
}

void
nsTransformedTextRun::SetCapitalization(PRUint32 aStart, PRUint32 aLength,
                                        bool* aCapitalization,
                                        gfxContext* aRefContext)
{
  if (mCapitalize.IsEmpty()) {
    if (!mCapitalize.AppendElements(GetLength()))
      return;
    memset(mCapitalize.Elements(), 0, GetLength()*sizeof(bool));
  }
  memcpy(mCapitalize.Elements() + aStart, aCapitalization, aLength*sizeof(bool));
  mNeedsRebuild = true;
}

bool
nsTransformedTextRun::SetPotentialLineBreaks(PRUint32 aStart, PRUint32 aLength,
                                             PRUint8* aBreakBefore,
                                             gfxContext* aRefContext)
{
  bool changed = gfxTextRun::SetPotentialLineBreaks(aStart, aLength,
      aBreakBefore, aRefContext);
  if (changed) {
    mNeedsRebuild = true;
  }
  return changed;
}

size_t
nsTransformedTextRun::SizeOfExcludingThis(nsMallocSizeOfFun aMallocSizeOf)
{
  size_t total = gfxTextRun::SizeOfExcludingThis(aMallocSizeOf);
  total += mStyles.SizeOfExcludingThis(aMallocSizeOf);
  total += mCapitalize.SizeOfExcludingThis(aMallocSizeOf);
  if (mOwnsFactory) {
    total += aMallocSizeOf(mFactory);
  }
  return total;
}

size_t
nsTransformedTextRun::SizeOfIncludingThis(nsMallocSizeOfFun aMallocSizeOf)
{
  return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf);
}

nsTransformedTextRun*
nsTransformingTextRunFactory::MakeTextRun(const PRUnichar* aString, PRUint32 aLength,
                                          const gfxTextRunFactory::Parameters* aParams,
                                          gfxFontGroup* aFontGroup, PRUint32 aFlags,
                                          nsStyleContext** aStyles, bool aOwnsFactory)
{
  return nsTransformedTextRun::Create(aParams, this, aFontGroup,
                                      aString, aLength, aFlags, aStyles, aOwnsFactory);
}

nsTransformedTextRun*
nsTransformingTextRunFactory::MakeTextRun(const PRUint8* aString, PRUint32 aLength,
                                          const gfxTextRunFactory::Parameters* aParams,
                                          gfxFontGroup* aFontGroup, PRUint32 aFlags,
                                          nsStyleContext** aStyles, bool aOwnsFactory)
{
  // We'll only have a Unicode code path to minimize the amount of code needed
  // for these rarely used features
  NS_ConvertASCIItoUTF16 unicodeString(reinterpret_cast<const char*>(aString), aLength);
  return MakeTextRun(unicodeString.get(), aLength, aParams, aFontGroup,
                     aFlags & ~(gfxFontGroup::TEXT_IS_PERSISTENT | gfxFontGroup::TEXT_IS_8BIT),
                     aStyles, aOwnsFactory);
}

/**
 * Copy a given textrun, but merge certain characters into a single logical
 * character. Glyphs for a character are added to the glyph list for the previous
 * character and then the merged character is eliminated. Visually the results
 * are identical.
 * 
 * This is used for text-transform:uppercase when we encounter a SZLIG,
 * whose uppercase form is "SS", or other ligature or precomposed form
 * that expands to multiple codepoints during case transformation,
 * and for Greek text when combining diacritics have been deleted.
 * 
 * This function is unable to merge characters when they occur in different
 * glyph runs. This only happens in tricky edge cases where a character was
 * decomposed by case-mapping (e.g. there's no precomposed uppercase version
 * of an accented lowercase letter), and then font-matching caused the
 * diacritics to be assigned to a different font than the base character.
 * In this situation, the diacritic(s) get discarded, which is less than
 * ideal, but they probably weren't going to render very well anyway.
 * Bug 543200 will improve this by making font-matching operate on entire
 * clusters instead of individual codepoints.
 * 
 * For simplicity, this produces a textrun containing all DetailedGlyphs,
 * no simple glyphs. So don't call it unless you really have merging to do.
 * 
 * @param aCharsToMerge when aCharsToMerge[i] is true, this character in aSrc
 * is merged into the previous character
 *
 * @param aDeletedChars when aDeletedChars[i] is true, the character at this
 * position in aDest was deleted (has no corresponding char in aSrc)
 */
static void
MergeCharactersInTextRun(gfxTextRun* aDest, gfxTextRun* aSrc,
                         const bool* aCharsToMerge, const bool* aDeletedChars)
{
  aDest->ResetGlyphRuns();

  gfxTextRun::GlyphRunIterator iter(aSrc, 0, aSrc->GetLength());
  PRUint32 offset = 0;
  nsAutoTArray<gfxTextRun::DetailedGlyph,2> glyphs;
  while (iter.NextRun()) {
    gfxTextRun::GlyphRun* run = iter.GetGlyphRun();
    nsresult rv = aDest->AddGlyphRun(run->mFont, run->mMatchType,
                                     offset, false);
    if (NS_FAILED(rv))
      return;

    bool anyMissing = false;
    PRUint32 mergeRunStart = iter.GetStringStart();
    const gfxTextRun::CompressedGlyph *srcGlyphs = aSrc->GetCharacterGlyphs();
    gfxTextRun::CompressedGlyph mergedGlyph = srcGlyphs[mergeRunStart];
    PRUint32 stringEnd = iter.GetStringEnd();
    for (PRUint32 k = iter.GetStringStart(); k < stringEnd; ++k) {
      const gfxTextRun::CompressedGlyph g = srcGlyphs[k];
      if (g.IsSimpleGlyph()) {
        if (!anyMissing) {
          gfxTextRun::DetailedGlyph details;
          details.mGlyphID = g.GetSimpleGlyph();
          details.mAdvance = g.GetSimpleAdvance();
          details.mXOffset = 0;
          details.mYOffset = 0;
          glyphs.AppendElement(details);
        }
      } else {
        if (g.IsMissing()) {
          anyMissing = true;
          glyphs.Clear();
        }
        if (g.GetGlyphCount() > 0) {
          glyphs.AppendElements(aSrc->GetDetailedGlyphs(k), g.GetGlyphCount());
        }
      }

      if (k + 1 < iter.GetStringEnd() && aCharsToMerge[k + 1]) {
        // next char is supposed to merge with current, so loop without
        // writing current merged glyph to the destination
        continue;
      }

      // If the start of the merge run is actually a character that should
      // have been merged with the previous character (this can happen
      // if there's a font change in the middle of a case-mapped character,
      // that decomposed into a sequence of base+diacritics, for example),
      // just discard the entire merge run. See comment at start of this
      // function.
      NS_WARN_IF_FALSE(!aCharsToMerge[mergeRunStart],
                       "unable to merge across a glyph run boundary, "
                       "glyph(s) discarded");
      if (!aCharsToMerge[mergeRunStart]) {
        if (anyMissing) {
          mergedGlyph.SetMissing(glyphs.Length());
        } else {
          mergedGlyph.SetComplex(mergedGlyph.IsClusterStart(),
                                 mergedGlyph.IsLigatureGroupStart(),
                                 glyphs.Length());
        }
        aDest->SetGlyphs(offset, mergedGlyph, glyphs.Elements());
        ++offset;

        while (offset < aDest->GetLength() && aDeletedChars[offset]) {
          aDest->SetGlyphs(offset++, gfxTextRun::CompressedGlyph(), nullptr);
        }
      }

      glyphs.Clear();
      anyMissing = false;
      mergeRunStart = k + 1;
      if (mergeRunStart < stringEnd) {
        mergedGlyph = srcGlyphs[mergeRunStart];
      }
    }
    NS_ASSERTION(glyphs.Length() == 0,
                 "Leftover glyphs, don't request merging of the last character with its next!");  
  }
  NS_ASSERTION(offset == aDest->GetLength(), "Bad offset calculations");
}

static gfxTextRunFactory::Parameters
GetParametersForInner(nsTransformedTextRun* aTextRun, PRUint32* aFlags,
    gfxContext* aRefContext)
{
  gfxTextRunFactory::Parameters params =
    { aRefContext, nullptr, nullptr,
      nullptr, 0, aTextRun->GetAppUnitsPerDevUnit()
    };
  *aFlags = aTextRun->GetFlags() & ~gfxFontGroup::TEXT_IS_PERSISTENT;
  return params;
}

void
nsFontVariantTextRunFactory::RebuildTextRun(nsTransformedTextRun* aTextRun,
    gfxContext* aRefContext)
{
  gfxFontGroup* fontGroup = aTextRun->GetFontGroup();
  gfxFontStyle fontStyle = *fontGroup->GetStyle();
  fontStyle.size *= 0.8;
  nsRefPtr<gfxFontGroup> smallFont = fontGroup->Copy(&fontStyle);
  if (!smallFont)
    return;

  PRUint32 flags;
  gfxTextRunFactory::Parameters innerParams =
      GetParametersForInner(aTextRun, &flags, aRefContext);

  PRUint32 length = aTextRun->GetLength();
  const PRUnichar* str = aTextRun->mString.BeginReading();
  nsRefPtr<nsStyleContext>* styles = aTextRun->mStyles.Elements();
  // Create a textrun so we can check cluster-start properties
  nsAutoPtr<gfxTextRun> inner(fontGroup->MakeTextRun(str, length, &innerParams, flags));
  if (!inner.get())
    return;

  nsCaseTransformTextRunFactory uppercaseFactory(nullptr, true);

  aTextRun->ResetGlyphRuns();

  PRUint32 runStart = 0;
  nsAutoTArray<nsStyleContext*,50> styleArray;
  nsAutoTArray<PRUint8,50> canBreakBeforeArray;

  enum RunCaseState {
    kUpperOrCaseless, // will be untouched by font-variant:small-caps
    kLowercase,       // will be uppercased and reduced
    kSpecialUpper     // specials: don't shrink, but apply uppercase mapping
  };
  RunCaseState runCase = kUpperOrCaseless;

  // Note that this loop runs from 0 to length *inclusive*, so the last
  // iteration is in effect beyond the end of the input text, to give a
  // chance to finish the last casing run we've found.
  // The last iteration, when i==length, must not attempt to look at the
  // character position [i] or the style data for styles[i], as this would
  // be beyond the valid length of the textrun or its style array.
  for (PRUint32 i = 0; i <= length; ++i) {
    RunCaseState chCase = kUpperOrCaseless;
    // Unless we're at the end, figure out what treatment the current
    // character will need.
    if (i < length) {
      nsStyleContext* styleContext = styles[i];
      // Characters that aren't the start of a cluster are ignored here. They
      // get added to whatever lowercase/non-lowercase run we're in.
      if (!inner->IsClusterStart(i)) {
        chCase = runCase;
      } else {
        if (styleContext->GetStyleFont()->mFont.variant == NS_STYLE_FONT_VARIANT_SMALL_CAPS) {
          PRUint32 ch = str[i];
          if (NS_IS_HIGH_SURROGATE(ch) && i < length - 1 && NS_IS_LOW_SURROGATE(str[i + 1])) {
            ch = SURROGATE_TO_UCS4(ch, str[i + 1]);
          }
          PRUint32 ch2 = ToUpperCase(ch);
          if (ch != ch2 || mozilla::unicode::SpecialUpper(ch)) {
            chCase = kLowercase;
          } else if (styleContext->GetStyleFont()->mLanguage == nsGkAtoms::el) {
            // In Greek, check for characters that will be modified by the
            // GreekUpperCase mapping - this catches accented capitals where
            // the accent is to be removed (bug 307039). These are handled by
            // a transformed child run using the full-size font.
            GreekCasingState state = kStart; // don't need exact context here
            ch2 = GreekUpperCase(ch, &state);
            if (ch != ch2) {
              chCase = kSpecialUpper;
            }
          }
        } else {
          // Don't transform the character! I.e., pretend that it's not lowercase
        }
      }
    }

    // At the end of the text, or when the current character needs different
    // casing treatment from the current run, finish the run-in-progress
    // and prepare to accumulate a new run.
    // Note that we do not look at any source data for offset [i] here,
    // as that would be invalid in the case where i==length.
    if ((i == length || runCase != chCase) && runStart < i) {
      nsAutoPtr<nsTransformedTextRun> transformedChild;
      nsAutoPtr<gfxTextRun> cachedChild;
      gfxTextRun* child;

      switch (runCase) {
      case kUpperOrCaseless:
        cachedChild =
          fontGroup->MakeTextRun(str + runStart, i - runStart, &innerParams,
                                 flags);
        child = cachedChild.get();
        break;
      case kLowercase:
        transformedChild =
          uppercaseFactory.MakeTextRun(str + runStart, i - runStart,
                                       &innerParams, smallFont, flags,
                                       styleArray.Elements(), false);
        child = transformedChild;
        break;
      case kSpecialUpper:
        transformedChild =
          uppercaseFactory.MakeTextRun(str + runStart, i - runStart,
                                       &innerParams, fontGroup, flags,
                                       styleArray.Elements(), false);
        child = transformedChild;
        break;
      }
      if (!child)
        return;
      // Copy potential linebreaks into child so they're preserved
      // (and also child will be shaped appropriately)
      NS_ASSERTION(canBreakBeforeArray.Length() == i - runStart,
                   "lost some break-before values?");
      child->SetPotentialLineBreaks(0, canBreakBeforeArray.Length(),
          canBreakBeforeArray.Elements(), aRefContext);
      if (transformedChild) {
        transformedChild->FinishSettingProperties(aRefContext);
      }
      aTextRun->CopyGlyphDataFrom(child, 0, child->GetLength(), runStart);

      runStart = i;
      styleArray.Clear();
      canBreakBeforeArray.Clear();
    }

    if (i < length) {
      runCase = chCase;
      styleArray.AppendElement(styles[i]);
      canBreakBeforeArray.AppendElement(aTextRun->CanBreakLineBefore(i));
    }
  }
}

void
nsCaseTransformTextRunFactory::RebuildTextRun(nsTransformedTextRun* aTextRun,
    gfxContext* aRefContext)
{
  PRUint32 length = aTextRun->GetLength();
  const PRUnichar* str = aTextRun->mString.BeginReading();
  nsRefPtr<nsStyleContext>* styles = aTextRun->mStyles.Elements();

  nsAutoString convertedString;
  nsAutoTArray<bool,50> charsToMergeArray;
  nsAutoTArray<bool,50> deletedCharsArray;
  nsAutoTArray<nsStyleContext*,50> styleArray;
  nsAutoTArray<PRUint8,50> canBreakBeforeArray;
  bool mergeNeeded = false;

  // Some languages have special casing conventions that differ from the
  // default Unicode mappings.
  // The enum values here are named for well-known exemplar languages that
  // exhibit the behavior in question; multiple lang tags may map to the
  // same setting here, if the behavior is shared by other languages.
  enum {
    eNone,    // default non-lang-specific behavior
    eTurkish, // preserve dotted/dotless-i distinction in uppercase
    eDutch,   // treat "ij" digraph as a unit for capitalization
    eGreek    // strip accent when uppercasing Greek vowels
  } languageSpecificCasing = eNone;

  const nsIAtom* lang = nullptr;
  bool capitalizeDutchIJ = false;
  bool prevIsLetter = false;
  PRUint32 sigmaIndex = PRUint32(-1);
  nsIUGenCategory::nsUGenCategory cat;
  GreekCasingState greekState = kStart;
  PRUint32 i;
  for (i = 0; i < length; ++i) {
    PRUint32 ch = str[i];
    nsStyleContext* styleContext = styles[i];

    PRUint8 style = mAllUppercase ? NS_STYLE_TEXT_TRANSFORM_UPPERCASE
      : styleContext->GetStyleText()->mTextTransform;
    int extraChars = 0;
    const mozilla::unicode::MultiCharMapping *mcm;

    if (NS_IS_HIGH_SURROGATE(ch) && i < length - 1 && NS_IS_LOW_SURROGATE(str[i + 1])) {
      ch = SURROGATE_TO_UCS4(ch, str[i + 1]);
    }

    if (lang != styleContext->GetStyleFont()->mLanguage) {
      lang = styleContext->GetStyleFont()->mLanguage;
      if (lang == nsGkAtoms::tr || lang == nsGkAtoms::az ||
          lang == nsGkAtoms::ba || lang == nsGkAtoms::crh ||
          lang == nsGkAtoms::tt) {
        languageSpecificCasing = eTurkish;
      } else if (lang == nsGkAtoms::nl) {
        languageSpecificCasing = eDutch;
      } else if (lang == nsGkAtoms::el) {
        languageSpecificCasing = eGreek;
        greekState = kStart;
      } else {
        languageSpecificCasing = eNone;
      }
    }

    switch (style) {
    case NS_STYLE_TEXT_TRANSFORM_LOWERCASE:
      if (languageSpecificCasing == eTurkish) {
        if (ch == 'I') {
          ch = LATIN_SMALL_LETTER_DOTLESS_I;
          prevIsLetter = true;
          sigmaIndex = PRUint32(-1);
          break;
        }
        if (ch == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
          ch = 'i';
          prevIsLetter = true;
          sigmaIndex = PRUint32(-1);
          break;
        }
      }

      // Special lowercasing behavior for Greek Sigma: note that this is listed
      // as context-sensitive in Unicode's SpecialCasing.txt, but is *not* a
      // language-specific mapping; it applies regardless of the language of
      // the element.
      //
      // The lowercase mapping for CAPITAL SIGMA should be to SMALL SIGMA (i.e.
      // the non-final form) whenever there is a following letter, or when the
      // CAPITAL SIGMA occurs in isolation (neither preceded nor followed by a
      // LETTER); and to FINAL SIGMA when it is preceded by another letter but
      // not followed by one.
      //
      // To implement the context-sensitive nature of this mapping, we keep
      // track of whether the previous character was a letter. If not, CAPITAL
      // SIGMA will map directly to SMALL SIGMA. If the previous character
      // was a letter, CAPITAL SIGMA maps to FINAL SIGMA and we record the
      // position in the converted string; if we then encounter another letter,
      // that FINAL SIGMA is replaced with a standard SMALL SIGMA.

      cat = mozilla::unicode::GetGenCategory(ch);

      // If sigmaIndex is not -1, it marks where we have provisionally mapped
      // a CAPITAL SIGMA to FINAL SIGMA; if we now find another letter, we
      // need to change it to SMALL SIGMA.
      if (sigmaIndex != PRUint32(-1)) {
        if (cat == nsIUGenCategory::kLetter) {
          convertedString.SetCharAt(GREEK_SMALL_LETTER_SIGMA, sigmaIndex);
        }
      }

      if (ch == GREEK_CAPITAL_LETTER_SIGMA) {
        // If preceding char was a letter, map to FINAL instead of SMALL,
        // and note where it occurred by setting sigmaIndex; we'll change it
        // to standard SMALL SIGMA later if another letter follows
        if (prevIsLetter) {
          ch = GREEK_SMALL_LETTER_FINAL_SIGMA;
          sigmaIndex = convertedString.Length();
        } else {
          // CAPITAL SIGMA not preceded by a letter is unconditionally mapped
          // to SMALL SIGMA
          ch = GREEK_SMALL_LETTER_SIGMA;
          sigmaIndex = PRUint32(-1);
        }
        prevIsLetter = true;
        break;
      }

      // ignore diacritics for the purpose of contextual sigma mapping;
      // otherwise, reset prevIsLetter appropriately and clear the
      // sigmaIndex marker
      if (cat != nsIUGenCategory::kMark) {
        prevIsLetter = (cat == nsIUGenCategory::kLetter);
        sigmaIndex = PRUint32(-1);
      }

      mcm = mozilla::unicode::SpecialLower(ch);
      if (mcm) {
        int j = 0;
        while (j < 2 && mcm->mMappedChars[j + 1]) {
          convertedString.Append(mcm->mMappedChars[j]);
          ++extraChars;
          ++j;
        }
        ch = mcm->mMappedChars[j];
        break;
      }

      ch = ToLowerCase(ch);
      break;

    case NS_STYLE_TEXT_TRANSFORM_UPPERCASE:
      if (languageSpecificCasing == eTurkish && ch == 'i') {
        ch = LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
        break;
      }

      if (languageSpecificCasing == eGreek) {
        ch = GreekUpperCase(ch, &greekState);
        break;
      }

      mcm = mozilla::unicode::SpecialUpper(ch);
      if (mcm) {
        int j = 0;
        while (j < 2 && mcm->mMappedChars[j + 1]) {
          convertedString.Append(mcm->mMappedChars[j]);
          ++extraChars;
          ++j;
        }
        ch = mcm->mMappedChars[j];
        break;
      }

      ch = ToUpperCase(ch);
      break;

    case NS_STYLE_TEXT_TRANSFORM_CAPITALIZE:
      if (capitalizeDutchIJ && ch == 'j') {
        ch = 'J';
        capitalizeDutchIJ = false;
        break;
      }
      capitalizeDutchIJ = false;
      if (i < aTextRun->mCapitalize.Length() && aTextRun->mCapitalize[i]) {
        if (languageSpecificCasing == eTurkish && ch == 'i') {
          ch = LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
          break;
        }
        if (languageSpecificCasing == eDutch && ch == 'i') {
          ch = 'I';
          capitalizeDutchIJ = true;
          break;
        }

        mcm = mozilla::unicode::SpecialTitle(ch);
        if (mcm) {
          int j = 0;
          while (j < 2 && mcm->mMappedChars[j + 1]) {
            convertedString.Append(mcm->mMappedChars[j]);
            ++extraChars;
            ++j;
          }
          ch = mcm->mMappedChars[j];
          break;
        }

        ch = ToTitleCase(ch);
      }
      break;

    default:
      break;
    }

    if (ch == PRUint32(-1)) {
      deletedCharsArray.AppendElement(true);
      mergeNeeded = true;
    } else {
      deletedCharsArray.AppendElement(false);
      charsToMergeArray.AppendElement(false);
      styleArray.AppendElement(styleContext);
      canBreakBeforeArray.AppendElement(aTextRun->CanBreakLineBefore(i));

      if (IS_IN_BMP(ch)) {
        convertedString.Append(ch);
      } else {
        convertedString.Append(H_SURROGATE(ch));
        convertedString.Append(L_SURROGATE(ch));
        ++i;
        deletedCharsArray.AppendElement(true); // not exactly deleted, but the
                                               // trailing surrogate is skipped
        ++extraChars;
      }

      while (extraChars-- > 0) {
        mergeNeeded = true;
        charsToMergeArray.AppendElement(true);
        styleArray.AppendElement(styleContext);
        canBreakBeforeArray.AppendElement(false);
      }
    }
  }

  PRUint32 flags;
  gfxTextRunFactory::Parameters innerParams =
      GetParametersForInner(aTextRun, &flags, aRefContext);
  gfxFontGroup* fontGroup = aTextRun->GetFontGroup();

  nsAutoPtr<nsTransformedTextRun> transformedChild;
  nsAutoPtr<gfxTextRun> cachedChild;
  gfxTextRun* child;

  if (mInnerTransformingTextRunFactory) {
    transformedChild = mInnerTransformingTextRunFactory->MakeTextRun(
        convertedString.BeginReading(), convertedString.Length(),
        &innerParams, fontGroup, flags, styleArray.Elements(), false);
    child = transformedChild.get();
  } else {
    cachedChild = fontGroup->MakeTextRun(
        convertedString.BeginReading(), convertedString.Length(),
        &innerParams, flags);
    child = cachedChild.get();
  }
  if (!child)
    return;
  // Copy potential linebreaks into child so they're preserved
  // (and also child will be shaped appropriately)
  NS_ASSERTION(convertedString.Length() == canBreakBeforeArray.Length(),
               "Dropped characters or break-before values somewhere!");
  child->SetPotentialLineBreaks(0, canBreakBeforeArray.Length(),
      canBreakBeforeArray.Elements(), aRefContext);
  if (transformedChild) {
    transformedChild->FinishSettingProperties(aRefContext);
  }

  if (mergeNeeded) {
    // Now merge multiple characters into one multi-glyph character as required
    // and deal with skipping deleted accent chars
    NS_ASSERTION(charsToMergeArray.Length() == child->GetLength(),
                 "source length mismatch");
    NS_ASSERTION(deletedCharsArray.Length() == aTextRun->GetLength(),
                 "destination length mismatch");
    MergeCharactersInTextRun(aTextRun, child, charsToMergeArray.Elements(),
                             deletedCharsArray.Elements());
  } else {
    // No merging to do, so just copy; this produces a more optimized textrun.
    // We can't steal the data because the child may be cached and stealing
    // the data would break the cache.
    aTextRun->ResetGlyphRuns();
    aTextRun->CopyGlyphDataFrom(child, 0, child->GetLength(), 0);
  }
}