Bug 1521723 - Apply hyphenate-limit-chars settings to the potential breaks found by the hyphenator. r=layout-reviewers,emilio

Differential Revision: https://phabricator.services.mozilla.com/D234960

dom/base/nsLineBreaker.cpp

@@ -64,15 +64,7 @@ static constexpr bool IsNonBreakableChar(T aChar, bool aLegacyBehavior) {
 }
 
 nsLineBreaker::nsLineBreaker()
-    : mCurrentWordLanguage(nullptr),
+    : mLegacyBehavior(!mozilla::StaticPrefs::intl_icu4x_segmenter_enabled()) {}
-      mCurrentWordContainsMixedLang(false),
-      mScriptIsChineseOrJapanese(false),
-      mAfterBreakableSpace(false),
-      mBreakHere(false),
-      mWordBreak(WordBreakRule::Normal),
-      mLineBreak(LineBreakRule::Auto),
-      mWordContinuation(false),
-      mLegacyBehavior(!mozilla::StaticPrefs::intl_icu4x_segmenter_enabled()) {}
 
 nsLineBreaker::~nsLineBreaker() {
   NS_ASSERTION(mCurrentWord.Length() == 0,
@@ -422,14 +414,88 @@ void nsLineBreaker::FindHyphenationPoints(nsHyphenator* aHyphenator,
                                           const char16_t* aTextStart,
                                           const char16_t* aTextLimit,
                                           uint8_t* aBreakState) {
+  // Early-return for words that are definitely too short to hyphenate.
+  if (aTextLimit - aTextStart < mHyphenateLimitWord) {
+    return;
+  }
+
   nsDependentSubstring string(aTextStart, aTextLimit);
   AutoTArray<bool, 200> hyphens;
-  if (NS_SUCCEEDED(aHyphenator->Hyphenate(string, hyphens))) {
+  if (NS_FAILED(aHyphenator->Hyphenate(string, hyphens))) {
-    for (uint32_t i = 0; i + 1 < string.Length(); ++i) {
+    return;
-      if (hyphens[i]) {
+  }
-        aBreakState[i + 1] =
+
-            gfxTextRun::CompressedGlyph::FLAG_BREAK_TYPE_HYPHEN;
+  // Keep track of the length seen so far, in terms of characters that are
+  // countable for hyphenate-limit-chars purposes.
+  uint32_t length = 0;
+  AutoTArray<std::pair<uint32_t, uint32_t>, 16> positionAndLength;
+  for (uint32_t i = 0; i + 1 < string.Length(); ++i) {
+    // Get current character, converting surrogate pairs to UCS4 for char
+    // category lookup.
+    uint32_t ch = string[i];
+    if (NS_IS_HIGH_SURROGATE(ch) && i + 1 < string.Length() &&
+        NS_IS_LOW_SURROGATE(string[i + 1])) {
+      ch = SURROGATE_TO_UCS4(ch, string[i + 1]);
+    }
+
+    // According to CSS Text, "Nonspacing combining marks (Unicode General
+    // Category Mn) and intra-word punctuation (Unicode General Category P*)
+    // do not count towards the minimum."
+    // (https://drafts.csswg.org/css-text-4/#hyphenate-char-limits)
+    // We also don't count Control or Format categories.
+    using intl::GeneralCategory;
+    switch (UnicodeProperties::CharType(ch)) {
+      case GeneralCategory::Nonspacing_Mark:
+      case GeneralCategory::Dash_Punctuation:
+      case GeneralCategory::Open_Punctuation:
+      case GeneralCategory::Close_Punctuation:
+      case GeneralCategory::Connector_Punctuation:
+      case GeneralCategory::Other_Punctuation:
+      case GeneralCategory::Initial_Punctuation:
+      case GeneralCategory::Final_Punctuation:
+      case GeneralCategory::Control:
+      case GeneralCategory::Format:
+      case GeneralCategory::Surrogate:
+        break;
+      default:
+        ++length;
+        break;
+    }
+
+    // Don't accept any breaks until we're far enough into the word.
+    if (length >= mHyphenateLimitStart && hyphens[i]) {
+      MOZ_ASSERT(aBreakState[i + 1] ==
+                     gfxTextRun::CompressedGlyph::FLAG_BREAK_TYPE_NONE);
+      aBreakState[i + 1] = gfxTextRun::CompressedGlyph::FLAG_BREAK_TYPE_HYPHEN;
+      // Keep track of hyphen position and "countable" length of the word.
+      positionAndLength.AppendElement(
+          std::pair<uint32_t, uint32_t>(i + 1, length));
+    }
+
+    // If the character was outside the BMP, skip past the low surrogate.
+    if (!IS_IN_BMP(ch)) {
+      ++i;
+    }
+  }
+  ++length;  // Account for the last character (not counted by the loop above).
+
+  if (length < mHyphenateLimitWord) {
+    // After discounting combining marks, punctuation, controls, etc., the word
+    // was too short for hyphenate-limit-chars. If we've set any hyphen breaks,
+    // forget them.
+    while (!positionAndLength.IsEmpty()) {
+      auto [lastPos, lastLen] = positionAndLength.PopLastElement();
+      aBreakState[lastPos] = gfxTextRun::CompressedGlyph::FLAG_BREAK_TYPE_NONE;
+    }
+  } else {
+    // Check if trailing fragment is too short; if so, remove the last hyphen
+    // break(s) that we set, until the fragment will be long enough.
+    while (!positionAndLength.IsEmpty()) {
+      auto [lastPos, lastLen] = positionAndLength.PopLastElement();
+      if (length - lastLen >= mHyphenateLimitEnd) {
+        break;
       }
+      aBreakState[lastPos] = gfxTextRun::CompressedGlyph::FLAG_BREAK_TYPE_NONE;
     }
   }
 }

dom/base/nsLineBreaker.h

@@ -217,6 +217,16 @@ class nsLineBreaker {
     mWordContinuation = aContinuation;
   }
 
+  /**
+   * Set the hyphenate-limit-chars values. Values are clamped to be <= 255.
+   */
+  void SetHyphenateLimitChars(uint32_t aWordLength, uint32_t aStartLength,
+                              uint32_t aEndLength) {
+    mHyphenateLimitWord = std::min(255u, aWordLength);
+    mHyphenateLimitStart = std::min(255u, aStartLength);
+    mHyphenateLimitEnd = std::min(255u, aEndLength);
+  }
+
  private:
   // This is a list of text sources that make up the "current word" (i.e.,
   // run of text which does not contain any whitespace). All the mLengths
@@ -262,25 +272,38 @@ class nsLineBreaker {
   AutoTArray<char16_t, 100> mCurrentWord;
   // All the items that contribute to mCurrentWord
   AutoTArray<TextItem, 2> mTextItems;
-  nsAtom* mCurrentWordLanguage;
+  nsAtom* mCurrentWordLanguage = nullptr;
-  bool mCurrentWordContainsMixedLang;
+
+  // Constraints from CSS `hyphenate-limit-chars` property, to block the use of
+  // auto-hyphenation if the word is too short, or at positions too near the
+  // beginning/end of the word.
+  // (Note that per CSS Text spec, these counts ignore combining marks, etc.,
+  // so they are not purely codepoint or character counts.)
+  // (Zero values would have no effect; but text-frame code will update the
+  // values from CSS before calling the line-breaker.)
+  uint8_t mHyphenateLimitWord = 0;   // Min word length to auto-hyphenate
+  uint8_t mHyphenateLimitStart = 0;  // Min number of chars before the break
+  uint8_t mHyphenateLimitEnd = 0;    // Min number of chars after the break
+
+  bool mCurrentWordContainsMixedLang = false;
   bool mCurrentWordMightBeBreakable = false;
-  bool mScriptIsChineseOrJapanese;
+  bool mScriptIsChineseOrJapanese = false;
 
   // True if the previous character was breakable whitespace
-  bool mAfterBreakableSpace;
+  bool mAfterBreakableSpace = false;
   // True if a break must be allowed at the current position because
   // a run of breakable whitespace ends here
-  bool mBreakHere;
+  bool mBreakHere = false;
   // Break rules for letters from the "word-break" property.
-  mozilla::intl::WordBreakRule mWordBreak;
+  mozilla::intl::WordBreakRule mWordBreak =
+      mozilla::intl::WordBreakRule::Normal;
   // Line breaking strictness from the "line-break" property.
-  mozilla::intl::LineBreakRule mLineBreak;
+  mozilla::intl::LineBreakRule mLineBreak = mozilla::intl::LineBreakRule::Auto;
   // Should the text be treated as continuing a word-in-progress (for purposes
   // of initial capitalization)? Normally this is set to false whenever we
   // start using a linebreaker, but it may be set to true if the line-breaker
   // has been explicitly flushed mid-word.
-  bool mWordContinuation;
+  bool mWordContinuation = false;
   // True if using old line segmenter.
   const bool mLegacyBehavior;
 };

layout/generic/nsTextFrame.cpp

@@ -1733,6 +1733,30 @@ void BuildTextRunsScanner::AccumulateRunInfo(nsTextFrame* aFrame) {
     mLineBreakBeforeFrames.AppendElement(aFrame);
     mStartOfLine = false;
   }
+
+  // Default limits used by `hyphenate-limit-chars` for `auto` components, as
+  // suggested by the CSS Text spec.
+  // TODO: consider making these sensitive to the context, e.g. increasing the
+  // values for long line lengths to reduce the tendency to hyphenate too much.
+  const uint32_t kDefaultHyphenateTotalWordLength = 5;
+  const uint32_t kDefaultHyphenatePreBreakLength = 2;
+  const uint32_t kDefaultHyphenatePostBreakLength = 2;
+
+  const auto& hyphenateLimitChars = aFrame->StyleText()->mHyphenateLimitChars;
+  uint32_t pre =
+      hyphenateLimitChars.pre_hyphen_length.IsAuto()
+          ? kDefaultHyphenatePreBreakLength
+          : std::max(0, hyphenateLimitChars.pre_hyphen_length.AsNumber());
+  uint32_t post =
+      hyphenateLimitChars.post_hyphen_length.IsAuto()
+          ? kDefaultHyphenatePostBreakLength
+          : std::max(0, hyphenateLimitChars.post_hyphen_length.AsNumber());
+  uint32_t total =
+      hyphenateLimitChars.total_word_length.IsAuto()
+          ? kDefaultHyphenateTotalWordLength
+          : std::max(0, hyphenateLimitChars.total_word_length.AsNumber());
+  total = std::max(total, pre + post);
+  mLineBreaker.SetHyphenateLimitChars(total, pre, post);
 }
 
 static bool HasTerminalNewline(const nsTextFrame* aFrame) {