gecko/layout/generic/MathMLTextRunFactory.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 "MathMLTextRunFactory.h"

#include "mozilla/ArrayUtils.h"
 
#include "nsStyleConsts.h"
#include "nsStyleContext.h"
#include "nsTextFrameUtils.h"
#include "nsFontMetrics.h"
#include "nsDeviceContext.h"

using namespace mozilla;

/*
  Entries for the mathvariant lookup tables.  mKey represents the Unicode
  character to be transformed and is used for searching the tables.
  mReplacement represents the mapped mathvariant Unicode character.
*/
typedef struct
{
  uint32_t mKey;
  uint32_t mReplacement;
} MathVarMapping;

/*
 Lookup tables for use with mathvariant mappings to transform a unicode
 character point to another unicode character that indicates the proper output.
 mKey represents one of two concepts.
 1.  In the Latin table it represents a hole in the mathematical alphanumeric
     block, where the character that should occupy that position is located
     elsewhere.
 2.  It represents an Arabic letter.

  As a replacement, 0 is reserved to indicate no mapping was found.
*/
static const MathVarMapping gArabicInitialMapTable[] = {
  { 0x628, 0x1EE21 },
  { 0x62A, 0x1EE35 },
  { 0x62B, 0x1EE36 },
  { 0x62C, 0x1EE22 },
  { 0x62D, 0x1EE27 },
  { 0x62E, 0x1EE37 },
  { 0x633, 0x1EE2E },
  { 0x634, 0x1EE34 },
  { 0x635, 0x1EE31 },
  { 0x636, 0x1EE39 },
  { 0x639, 0x1EE2F },
  { 0x63A, 0x1EE3B },
  { 0x641, 0x1EE30 },
  { 0x642, 0x1EE32 },
  { 0x643, 0x1EE2A },
  { 0x644, 0x1EE2B },
  { 0x645, 0x1EE2C },
  { 0x646, 0x1EE2D },
  { 0x647, 0x1EE24 },
  { 0x64A, 0x1EE29 }
};

static const MathVarMapping gArabicTailedMapTable[] = {
  { 0x62C, 0x1EE42 },
  { 0x62D, 0x1EE47 },
  { 0x62E, 0x1EE57 },
  { 0x633, 0x1EE4E },
  { 0x634, 0x1EE54 },
  { 0x635, 0x1EE51 },
  { 0x636, 0x1EE59 },
  { 0x639, 0x1EE4F },
  { 0x63A, 0x1EE5B },
  { 0x642, 0x1EE52 },
  { 0x644, 0x1EE4B },
  { 0x646, 0x1EE4D },
  { 0x64A, 0x1EE49 },
  { 0x66F, 0x1EE5F },
  { 0x6BA, 0x1EE5D }
};

static const MathVarMapping gArabicStretchedMapTable[] = {
  { 0x628, 0x1EE61 },
  { 0x62A, 0x1EE75 },
  { 0x62B, 0x1EE76 },
  { 0x62C, 0x1EE62 },
  { 0x62D, 0x1EE67 },
  { 0x62E, 0x1EE77 },
  { 0x633, 0x1EE6E },
  { 0x634, 0x1EE74 },
  { 0x635, 0x1EE71 },
  { 0x636, 0x1EE79 },
  { 0x637, 0x1EE68 },
  { 0x638, 0x1EE7A },
  { 0x639, 0x1EE6F },
  { 0x63A, 0x1EE7B },
  { 0x641, 0x1EE70 },
  { 0x642, 0x1EE72 },
  { 0x643, 0x1EE6A },
  { 0x645, 0x1EE6C },
  { 0x646, 0x1EE6D },
  { 0x647, 0x1EE64 },
  { 0x64A, 0x1EE69 },
  { 0x66E, 0x1EE7C },
  { 0x6A1, 0x1EE7E }
};

static const MathVarMapping gArabicLoopedMapTable[] = {
  { 0x627, 0x1EE80 },
  { 0x628, 0x1EE81 },
  { 0x62A, 0x1EE95 },
  { 0x62B, 0x1EE96 },
  { 0x62C, 0x1EE82 },
  { 0x62D, 0x1EE87 },
  { 0x62E, 0x1EE97 },
  { 0x62F, 0x1EE83 },
  { 0x630, 0x1EE98 },
  { 0x631, 0x1EE93 },
  { 0x632, 0x1EE86 },
  { 0x633, 0x1EE8E },
  { 0x634, 0x1EE94 },
  { 0x635, 0x1EE91 },
  { 0x636, 0x1EE99 },
  { 0x637, 0x1EE88 },
  { 0x638, 0x1EE9A },
  { 0x639, 0x1EE8F },
  { 0x63A, 0x1EE9B },
  { 0x641, 0x1EE90 },
  { 0x642, 0x1EE92 },
  { 0x644, 0x1EE8B },
  { 0x645, 0x1EE8C },
  { 0x646, 0x1EE8D },
  { 0x647, 0x1EE84 },
  { 0x648, 0x1EE85 },
  { 0x64A, 0x1EE89 }
};

static const MathVarMapping gArabicDoubleMapTable[] = {
  { 0x628, 0x1EEA1 },
  { 0x62A, 0x1EEB5 },
  { 0x62B, 0x1EEB6 },
  { 0x62C, 0x1EEA2 },
  { 0x62D, 0x1EEA7 },
  { 0x62E, 0x1EEB7 },
  { 0x62F, 0x1EEA3 },
  { 0x630, 0x1EEB8 },
  { 0x631, 0x1EEB3 },
  { 0x632, 0x1EEA6 },
  { 0x633, 0x1EEAE },
  { 0x634, 0x1EEB4 },
  { 0x635, 0x1EEB1 },
  { 0x636, 0x1EEB9 },
  { 0x637, 0x1EEA8 },
  { 0x638, 0x1EEBA },
  { 0x639, 0x1EEAF },
  { 0x63A, 0x1EEBB },
  { 0x641, 0x1EEB0 },
  { 0x642, 0x1EEB2 },
  { 0x644, 0x1EEAB },
  { 0x645, 0x1EEAC },
  { 0x646, 0x1EEAD },
  { 0x648, 0x1EEA5 },
  { 0x64A, 0x1EEA9 }
};

static const MathVarMapping gLatinExceptionMapTable[] = {
  { 0x1D455, 0x210E },
  { 0x1D49D, 0x212C },
  { 0x1D4A0, 0x2130 },
  { 0x1D4A1, 0x2131 },
  { 0x1D4A3, 0x210B },
  { 0x1D4A4, 0x2110 },
  { 0x1D4A7, 0x2112 },
  { 0x1D4A8, 0x2133 },
  { 0x1D4AD, 0x211B },
  { 0x1D4BA, 0x212F },
  { 0x1D4BC, 0x210A },
  { 0x1D4C4, 0x2134 },
  { 0x1D506, 0x212D },
  { 0x1D50B, 0x210C },
  { 0x1D50C, 0x2111 },
  { 0x1D515, 0x211C },
  { 0x1D51D, 0x2128 },
  { 0x1D53A, 0x2102 },
  { 0x1D53F, 0x210D },
  { 0x1D545, 0x2115 },
  { 0x1D547, 0x2119 },
  { 0x1D548, 0x211A },
  { 0x1D549, 0x211D },
  { 0x1D551, 0x2124 }
};

// Finds a MathVarMapping struct with the specified key (aKey) within aTable.
// aTable must be an array, whose length is specified by aNumElements
static uint32_t
MathvarMappingSearch(uint32_t aKey, const MathVarMapping* aTable, uint32_t aNumElements)
{
  uint32_t low = 0;
  uint32_t high = aNumElements;
  while (high > low) {
    uint32_t midPoint = (low+high) >> 1;
    if (aKey == aTable[midPoint].mKey) {
      return aTable[midPoint].mReplacement;
    }
    if (aKey > aTable[midPoint].mKey) {
      low = midPoint + 1;
    } else {
      high = midPoint;
    }
  }
  return 0;
}

#define GREEK_UPPER_THETA               0x03F4
#define HOLE_GREEK_UPPER_THETA          0x03A2
#define NABLA                           0x2207
#define PARTIAL_DIFFERENTIAL            0x2202
#define GREEK_UPPER_ALPHA               0x0391
#define GREEK_UPPER_OMEGA               0x03A9
#define GREEK_LOWER_ALPHA               0x03B1
#define GREEK_LOWER_OMEGA               0x03C9
#define GREEK_LUNATE_EPSILON_SYMBOL     0x03F5
#define GREEK_THETA_SYMBOL              0x03D1
#define GREEK_KAPPA_SYMBOL              0x03F0
#define GREEK_PHI_SYMBOL                0x03D5
#define GREEK_RHO_SYMBOL                0x03F1
#define GREEK_PI_SYMBOL                 0x03D6
#define GREEK_LETTER_DIGAMMA            0x03DC
#define GREEK_SMALL_LETTER_DIGAMMA      0x03DD
#define MATH_BOLD_CAPITAL_DIGAMMA       0x1D7CA
#define MATH_BOLD_SMALL_DIGAMMA         0x1D7CB

#define LATIN_SMALL_LETTER_DOTLESS_I    0x0131
#define LATIN_SMALL_LETTER_DOTLESS_J    0x0237

#define MATH_ITALIC_SMALL_DOTLESS_I     0x1D6A4
#define MATH_ITALIC_SMALL_DOTLESS_J     0x1D6A5

#define MATH_BOLD_UPPER_A               0x1D400
#define MATH_ITALIC_UPPER_A             0x1D434
#define MATH_BOLD_SMALL_A               0x1D41A
#define MATH_BOLD_UPPER_ALPHA           0x1D6A8
#define MATH_BOLD_SMALL_ALPHA           0x1D6C2
#define MATH_ITALIC_UPPER_ALPHA         0x1D6E2
#define MATH_BOLD_DIGIT_ZERO            0x1D7CE
#define MATH_DOUBLE_STRUCK_ZERO         0x1D7D8

#define MATH_BOLD_UPPER_THETA           0x1D6B9
#define MATH_BOLD_NABLA                 0x1D6C1
#define MATH_BOLD_PARTIAL_DIFFERENTIAL  0x1D6DB
#define MATH_BOLD_EPSILON_SYMBOL        0x1D6DC
#define MATH_BOLD_THETA_SYMBOL          0x1D6DD
#define MATH_BOLD_KAPPA_SYMBOL          0x1D6DE
#define MATH_BOLD_PHI_SYMBOL            0x1D6DF
#define MATH_BOLD_RHO_SYMBOL            0x1D6E0
#define MATH_BOLD_PI_SYMBOL             0x1D6E1

/*
  Performs the character mapping needed to implement MathML's mathvariant
  attribute.  It takes a unicode character and maps it to its appropriate
  mathvariant counterpart specified by aMathVar.  The mapped character is
  typically located within Unicode's mathematical blocks (0x1D***, 0x1EE**) but
  there are exceptions which this function accounts for.
  Characters without a valid mapping or valid aMathvar value are returned
  unaltered.  Characters already in the mathematical blocks (or are one of the
  exceptions) are never transformed.
  Acceptable values for aMathVar are specified in layout/style/nsStyleConsts.h.
  The transformable characters can be found at:
  http://lists.w3.org/Archives/Public/www-math/2013Sep/0012.html and
  https://en.wikipedia.org/wiki/Mathematical_Alphanumeric_Symbols
*/
static uint32_t
MathVariant(uint32_t aCh, uint8_t aMathVar)
{
  uint32_t baseChar;
  enum CharacterType {
    kIsLatin,
    kIsGreekish,
    kIsNumber,
    kIsArabic,
  };
  CharacterType varType;

  int8_t multiplier;

  if (aMathVar <= NS_MATHML_MATHVARIANT_NORMAL) {
    // nothing to do here
    return aCh;
  }
  if (aMathVar > NS_MATHML_MATHVARIANT_STRETCHED) {
    NS_ASSERTION(false, "Illegal mathvariant value");
    return aCh;
  }

  // Exceptional characters with at most one possible transformation
  if (aCh == HOLE_GREEK_UPPER_THETA) {
    // Nothing at this code point is transformed
    return aCh;
  }
  if (aCh == GREEK_LETTER_DIGAMMA) {
    if (aMathVar == NS_MATHML_MATHVARIANT_BOLD) {
      return MATH_BOLD_CAPITAL_DIGAMMA;
    }
    return aCh;
  }
  if (aCh == GREEK_SMALL_LETTER_DIGAMMA) {
    if (aMathVar == NS_MATHML_MATHVARIANT_BOLD) {
      return MATH_BOLD_SMALL_DIGAMMA;
    }
    return aCh;
  }
  if (aCh == LATIN_SMALL_LETTER_DOTLESS_I) {
    if (aMathVar == NS_MATHML_MATHVARIANT_ITALIC) {
      return MATH_ITALIC_SMALL_DOTLESS_I;
    }
    return aCh;
  }
  if (aCh == LATIN_SMALL_LETTER_DOTLESS_J) {
    if (aMathVar == NS_MATHML_MATHVARIANT_ITALIC) {
      return MATH_ITALIC_SMALL_DOTLESS_J;
    }
    return aCh;
  }

  // The Unicode mathematical blocks are divided into four segments: Latin,
  // Greek, numbers and Arabic.  In the case of the first three
  // baseChar represents the relative order in which the characters are
  // encoded in the Unicode mathematical block, normalised to the first
  // character of that sequence.
  //
  if ('A' <= aCh && aCh <= 'Z') {
    baseChar = aCh - 'A';
    varType = kIsLatin;
  } else if ('a' <= aCh && aCh <= 'z') {
    // Lowercase characters are placed immediately after the uppercase
    // characters in the Unicode mathematical block.  The constant subtraction
    // represents the number of characters between the start of the sequence
    // (capital A) and the first lowercase letter.
    baseChar = MATH_BOLD_SMALL_A-MATH_BOLD_UPPER_A + aCh - 'a';
    varType = kIsLatin;
  } else if ('0' <= aCh && aCh <= '9') {
    baseChar = aCh - '0';
    varType = kIsNumber;
  } else if (GREEK_UPPER_ALPHA <= aCh && aCh <= GREEK_UPPER_OMEGA) {
    baseChar = aCh-GREEK_UPPER_ALPHA;
    varType = kIsGreekish;
  } else if (GREEK_LOWER_ALPHA <= aCh && aCh <= GREEK_LOWER_OMEGA) {
    // Lowercase Greek comes after uppercase Greek.
    // Note in this instance the presence of an additional character (Nabla)
    // between the end of the uppercase Greek characters and the lowercase
    // ones.
    baseChar =  MATH_BOLD_SMALL_ALPHA - MATH_BOLD_UPPER_ALPHA
                + aCh-GREEK_LOWER_ALPHA;
    varType = kIsGreekish;
  } else if (0x0600 <= aCh && aCh <= 0x06FF) {
    // Arabic characters are defined within this range
    varType = kIsArabic;
  } else {
    switch (aCh) {
      case GREEK_UPPER_THETA:
        baseChar = MATH_BOLD_UPPER_THETA-MATH_BOLD_UPPER_ALPHA;
        break;
      case NABLA:
        baseChar = MATH_BOLD_NABLA-MATH_BOLD_UPPER_ALPHA;
        break;
      case PARTIAL_DIFFERENTIAL:
        baseChar = MATH_BOLD_PARTIAL_DIFFERENTIAL - MATH_BOLD_UPPER_ALPHA;
        break;
      case GREEK_LUNATE_EPSILON_SYMBOL:
        baseChar = MATH_BOLD_EPSILON_SYMBOL - MATH_BOLD_UPPER_ALPHA;
        break;
      case GREEK_THETA_SYMBOL:
        baseChar = MATH_BOLD_THETA_SYMBOL - MATH_BOLD_UPPER_ALPHA;
        break;
      case GREEK_KAPPA_SYMBOL:
        baseChar = MATH_BOLD_KAPPA_SYMBOL - MATH_BOLD_UPPER_ALPHA;
        break;
      case GREEK_PHI_SYMBOL:
        baseChar = MATH_BOLD_PHI_SYMBOL - MATH_BOLD_UPPER_ALPHA;
        break;
      case GREEK_RHO_SYMBOL:
        baseChar = MATH_BOLD_RHO_SYMBOL - MATH_BOLD_UPPER_ALPHA;
        break;
      case GREEK_PI_SYMBOL:
        baseChar = MATH_BOLD_PI_SYMBOL - MATH_BOLD_UPPER_ALPHA;
        break;
      default:
        return aCh;
    }

    varType = kIsGreekish;
  }

  if (varType == kIsNumber) {
    switch (aMathVar) {
      // Each possible number mathvariant is encoded in a single, contiguous
      // block.  For example the beginning of the double struck number range
      // follows immediately after the end of the bold number range.
      // multiplier represents the order of the sequences relative to the first
      // one.
      case NS_MATHML_MATHVARIANT_BOLD:
        multiplier = 0;
        break;
      case NS_MATHML_MATHVARIANT_DOUBLE_STRUCK:
        multiplier = 1;
        break;
      case NS_MATHML_MATHVARIANT_SANS_SERIF:
        multiplier = 2;
        break;
      case NS_MATHML_MATHVARIANT_BOLD_SANS_SERIF:
        multiplier = 3;
        break;
      case NS_MATHML_MATHVARIANT_MONOSPACE:
        multiplier = 4;
        break;
      default:
        // This mathvariant isn't defined for numbers or is otherwise normal
        return aCh;
    }
    // As the ranges are contiguous, to find the desired mathvariant range it
    // is sufficient to multiply the position within the sequence order
    // (multiplier) with the period of the sequence (which is constant for all
    // number sequences) and to add the character point of the first character
    // within the number mathvariant range.
    // To this the baseChar calculated earlier is added to obtain the final
    // code point.
    return baseChar+multiplier*(MATH_DOUBLE_STRUCK_ZERO-MATH_BOLD_DIGIT_ZERO)
             +MATH_BOLD_DIGIT_ZERO;
  } else if (varType == kIsGreekish) {
    switch (aMathVar) {
      case NS_MATHML_MATHVARIANT_BOLD:
        multiplier = 0;
        break;
      case NS_MATHML_MATHVARIANT_ITALIC:
        multiplier = 1;
        break;
      case NS_MATHML_MATHVARIANT_BOLD_ITALIC:
        multiplier = 2;
        break;
      case NS_MATHML_MATHVARIANT_BOLD_SANS_SERIF:
        multiplier = 3;
        break;
      case NS_MATHML_MATHVARIANT_SANS_SERIF_BOLD_ITALIC:
        multiplier = 4;
        break;
      default:
        // This mathvariant isn't defined for Greek or is otherwise normal
        return aCh;
    }
    // See the kIsNumber case for an explanation of the following calculation
    return baseChar + MATH_BOLD_UPPER_ALPHA +
             multiplier*(MATH_ITALIC_UPPER_ALPHA - MATH_BOLD_UPPER_ALPHA);
  }

  uint32_t tempChar;
  uint32_t newChar;
  if (varType == kIsArabic) {
    const MathVarMapping* mapTable;
    uint32_t tableLength;
    switch (aMathVar) {
      /* The Arabic mathematical block is not continuous, nor does it have a
       * monotonic mapping to the unencoded characters, requiring the use of a
       * lookup table.
       */
      case NS_MATHML_MATHVARIANT_INITIAL:
        mapTable = gArabicInitialMapTable;
        tableLength = ArrayLength(gArabicInitialMapTable);
        break;
      case NS_MATHML_MATHVARIANT_TAILED:
        mapTable = gArabicTailedMapTable;
        tableLength = ArrayLength(gArabicTailedMapTable);
        break;
      case NS_MATHML_MATHVARIANT_STRETCHED:
        mapTable = gArabicStretchedMapTable;
        tableLength = ArrayLength(gArabicStretchedMapTable);
        break;
      case NS_MATHML_MATHVARIANT_LOOPED:
        mapTable = gArabicLoopedMapTable;
        tableLength = ArrayLength(gArabicLoopedMapTable);
        break;
      case NS_MATHML_MATHVARIANT_DOUBLE_STRUCK:
        mapTable = gArabicDoubleMapTable;
        tableLength = ArrayLength(gArabicDoubleMapTable);
        break;
      default:
        // No valid transformations exist
        return aCh;
    }
    newChar = MathvarMappingSearch(aCh, mapTable, tableLength);
  } else {
    // Must be Latin
    if (aMathVar > NS_MATHML_MATHVARIANT_MONOSPACE) {
      // Latin doesn't support the Arabic mathvariants
      return aCh;
    }
    multiplier = aMathVar - 2;
    // This is possible because the values for NS_MATHML_MATHVARIANT_* are
    // chosen to coincide with the order in which the encoded mathvariant
    // characters are located within their unicode block (less an offset to
    // avoid _NONE and _NORMAL variants)
    // See the kIsNumber case for an explanation of the following calculation
    tempChar =  baseChar + MATH_BOLD_UPPER_A +
                multiplier*(MATH_ITALIC_UPPER_A - MATH_BOLD_UPPER_A);
    // There are roughly twenty characters that are located outside of the
    // mathematical block, so the spaces where they ought to be are used
    // as keys for a lookup table containing the correct character mappings.
    newChar = MathvarMappingSearch(tempChar, gLatinExceptionMapTable,
                                   ArrayLength(gLatinExceptionMapTable));
  }

  if (newChar) {
    return newChar;
  } else if (varType == kIsLatin) {
    return tempChar;
  } else {
    // An Arabic character without a corresponding mapping
    return aCh;
  }

}

void
MathMLTextRunFactory::RebuildTextRun(nsTransformedTextRun* aTextRun,
                                     gfxContext* aRefContext)
{
  gfxFontGroup* fontGroup = aTextRun->GetFontGroup();

  nsAutoString convertedString;
  nsAutoTArray<bool,50> charsToMergeArray;
  nsAutoTArray<bool,50> deletedCharsArray;
  nsAutoTArray<nsStyleContext*,50> styleArray;
  nsAutoTArray<uint8_t,50> canBreakBeforeArray;
  bool mergeNeeded = false;

  bool singleCharMI =
    aTextRun->GetFlags() & nsTextFrameUtils::TEXT_IS_SINGLE_CHAR_MI;

  uint32_t length = aTextRun->GetLength();
  const char16_t* str = aTextRun->mString.BeginReading();
  nsRefPtr<nsStyleContext>* styles = aTextRun->mStyles.Elements();
  nsFont font;
  if (length) {
    font = styles[0]->StyleFont()->mFont;

    if (mSSTYScriptLevel) {
      bool found = false;
      // We respect ssty settings explicitly set by the user
      for (uint32_t i = 0; i < font.fontFeatureSettings.Length(); i++) {
        if (font.fontFeatureSettings[i].mTag == TRUETYPE_TAG('s', 's', 't', 'y')) {
          found = true;
          break;
        }
      }
      if (!found) {
        uint8_t sstyLevel = 0;
        float scriptScaling = pow(styles[0]->StyleFont()->mScriptSizeMultiplier,
                                  mSSTYScriptLevel);
        static_assert(NS_MATHML_DEFAULT_SCRIPT_SIZE_MULTIPLIER < 1,
                      "Shouldn't it make things smaller?");
        /*
          An SSTY level of 2 is set if the scaling factor is less than or equal
          to halfway between that for a scriptlevel of 1 (0.71) and that of a
          scriptlevel of 2 (0.71^2), assuming the default script size multiplier.
          An SSTY level of 1 is set if the script scaling factor is less than
          or equal that for a scriptlevel of 1 assuming the default script size
          multiplier.

          User specified values of script size multiplier will change the scaling
          factor which mSSTYScriptLevel values correspond to.

          In the event that the script size multiplier actually makes things
          larger, no change is made.

          To opt out of this change, add the following to the stylesheet:
          "font-feature-settings: 'ssty' 0"
        */
        if (scriptScaling <= (NS_MATHML_DEFAULT_SCRIPT_SIZE_MULTIPLIER +
                              (NS_MATHML_DEFAULT_SCRIPT_SIZE_MULTIPLIER *
                               NS_MATHML_DEFAULT_SCRIPT_SIZE_MULTIPLIER))/2) {
          // Currently only the first two ssty settings are used, so two is large
          // as we go
          sstyLevel = 2;
        } else if (scriptScaling <= NS_MATHML_DEFAULT_SCRIPT_SIZE_MULTIPLIER) {
          sstyLevel = 1;
        }
        if (sstyLevel) {
          gfxFontFeature settingSSTY;
          settingSSTY.mTag = TRUETYPE_TAG('s','s','t','y');
          settingSSTY.mValue = sstyLevel;
          font.fontFeatureSettings.AppendElement(settingSSTY);
        }
      }
    }
  }

  uint8_t mathVar = NS_MATHML_MATHVARIANT_NONE;
  bool doMathvariantStyling = true;

  for (uint32_t i = 0; i < length; ++i) {
    int extraChars = 0;
    nsStyleContext* styleContext = styles[i];
    mathVar = styleContext->StyleFont()->mMathVariant;

    if (singleCharMI && mathVar == NS_MATHML_MATHVARIANT_NONE) {
      // If the user has explicitly set a non-default value for fontstyle or
      // fontweight, the italic mathvariant behaviour of <mi> is disabled
      // This overrides the initial values specified in fontStyle, to avoid
      // inconsistencies in which attributes allow CSS changes and which do not.
      if (mFlags & MATH_FONT_WEIGHT_BOLD) {
        font.weight = NS_FONT_WEIGHT_BOLD;
        if (mFlags & MATH_FONT_STYLING_NORMAL) {
          font.style = NS_FONT_STYLE_NORMAL;
        } else {
          font.style = NS_FONT_STYLE_ITALIC;
        }
      } else if (mFlags & MATH_FONT_STYLING_NORMAL) {
        font.style = NS_FONT_STYLE_NORMAL;
        font.weight = NS_FONT_WEIGHT_NORMAL;
      } else {
        mathVar = NS_MATHML_MATHVARIANT_ITALIC;
      }
    }

    uint32_t 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]);
    }
    uint32_t ch2 = MathVariant(ch, mathVar);

    if (mathVar == NS_MATHML_MATHVARIANT_BOLD ||
        mathVar == NS_MATHML_MATHVARIANT_BOLD_ITALIC ||
        mathVar == NS_MATHML_MATHVARIANT_ITALIC) {
      if (ch == ch2  && ch != 0x20 && ch != 0xA0) {
        // Don't apply the CSS style if a character cannot be
        // transformed. There is an exception for whitespace as it is both
        // common and innocuous.
        doMathvariantStyling = false;
      }
      if (ch2 != ch) {
        // Bug 930504. Some platforms do not have fonts for Mathematical
        // Alphanumeric Symbols. Hence we check whether the transformed
        // character is actually available.
        uint8_t matchType;
        nsRefPtr<gfxFont> mathFont = fontGroup->
          FindFontForChar(ch2, 0, HB_SCRIPT_COMMON, nullptr, &matchType);
        if (mathFont) {
          // Don't apply the CSS style if there is a math font for at least one
          // of the transformed character in this text run.
          doMathvariantStyling = false;
        } else {
          // We fallback to the original character.
          ch2 = ch;
        }
      }
    }

    deletedCharsArray.AppendElement(false);
    charsToMergeArray.AppendElement(false);
    styleArray.AppendElement(styleContext);
    canBreakBeforeArray.AppendElement(aTextRun->CanBreakLineBefore(i));

    if (IS_IN_BMP(ch2)) {
      convertedString.Append(ch2);
    } else {
      convertedString.Append(H_SURROGATE(ch2));
      convertedString.Append(L_SURROGATE(ch2));
      ++extraChars;
      if (!IS_IN_BMP(ch)) {
        deletedCharsArray.AppendElement(true); // not exactly deleted, but
                                          // the trailing surrogate is skipped
        ++i;
      }
    }

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

  uint32_t flags;
  gfxTextRunFactory::Parameters innerParams =
      GetParametersForInner(aTextRun, &flags, aRefContext);

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

  if (mathVar == NS_MATHML_MATHVARIANT_BOLD && doMathvariantStyling) {
    font.style = NS_FONT_STYLE_NORMAL;
    font.weight = NS_FONT_WEIGHT_BOLD;
  } else if (mathVar == NS_MATHML_MATHVARIANT_ITALIC && doMathvariantStyling) {
    font.style = NS_FONT_STYLE_ITALIC;
    font.weight = NS_FONT_WEIGHT_NORMAL;
  } else if (mathVar == NS_MATHML_MATHVARIANT_BOLD_ITALIC &&
             doMathvariantStyling) {
    font.style = NS_FONT_STYLE_ITALIC;
    font.weight = NS_FONT_WEIGHT_BOLD;
  } else if (mathVar != NS_MATHML_MATHVARIANT_NONE) {
    // Mathvariant overrides fontstyle and fontweight
    // Need to check to see if mathvariant is actually applied as this function
    // is used for other purposes.
    font.style = NS_FONT_STYLE_NORMAL;
    font.weight = NS_FONT_WEIGHT_NORMAL;
  }
  gfxFontGroup* newFontGroup = nullptr;

  // Get the correct gfxFontGroup that corresponds to the earlier font changes.
  if (length) {
    nsPresContext* pc = styles[0]->PresContext();
    nsRefPtr<nsFontMetrics> metrics;
    pc->DeviceContext()->GetMetricsFor(font,
                                       styles[0]->StyleFont()->mLanguage,
                                       pc->GetUserFontSet(),
                                       pc->GetTextPerfMetrics(),
                                       *getter_AddRefs(metrics));
    if (metrics) {
      newFontGroup = metrics->GetThebesFontGroup();
    }
  }

  if (!newFontGroup) {
    // If we can't get a new font group, fall back to the old one.  Rendering
    // will be incorrect, but not significantly so.
    newFontGroup = fontGroup;
  }

  if (mInnerTransformingTextRunFactory) {
    transformedChild = mInnerTransformingTextRunFactory->MakeTextRun(
        convertedString.BeginReading(), convertedString.Length(),
        &innerParams, newFontGroup, flags, styleArray.Elements(), false);
    child = transformedChild.get();
  } else {
    cachedChild = newFontGroup->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
    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);
  }
}