Stop combining text runs which are connected by 'COMMON' blocks.

ui/gfx/render_text_harfbuzz.cc

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "ui/gfx/render_text_harfbuzz.h"
 
 #include <limits>
 #include <set>
 
 #include "base/i18n/bidi_line_iterator.h"
@@ skipping 27 matching lines @@
 
 // Text length limit. Longer strings are slow and not fully tested.
 const size_t kMaxTextLength = 10000;
 
 // The maximum number of scripts a Unicode character can belong to. This value
 // is arbitrarily chosen to be a good limit because it is unlikely for a single
 // character to belong to more scripts.
 const size_t kMaxScripts = 5;
 
 // Returns true if characters of |block_code| may trigger font fallback.
-// Dingbats and emoticons can be rendered through the color emoji font file,
-// therefore it needs to be trigerred as fallbacks. See crbug.com/448909
 bool IsUnusualBlockCode(UBlockCode block_code) {
   return block_code == UBLOCK_GEOMETRIC_SHAPES ||
-         block_code == UBLOCK_MISCELLANEOUS_SYMBOLS ||
-         block_code == UBLOCK_DINGBATS ||
-         block_code == UBLOCK_EMOTICONS;
+         block_code == UBLOCK_MISCELLANEOUS_SYMBOLS;

[Jun Mukai, 2015/05/11 23:21:40]

ckocagil@ or msw@ -- is this function still necessary?

 }
 
 bool IsBracket(UChar32 character) {
   static const char kBrackets[] = { '(', ')', '{', '}', '<', '>', };
   static const char* kBracketsEnd = kBrackets + arraysize(kBrackets);
   return std::find(kBrackets, kBracketsEnd, character) != kBracketsEnd;
 }
 
 // Returns the boundary between a special and a regular character. Special
 // characters are brackets or characters that satisfy |IsUnusualBlockCode|.
@@ skipping 18 matching lines @@
     const bool block_break = current_block != first_block &&
         (first_block_unusual || IsUnusualBlockCode(current_block));
     if (block_break || current_char == '\n' ||
         first_bracket != IsBracket(current_char)) {
       return run_start + iter.array_pos();
     }
   }
   return run_break;
 }
 
-// If the given scripts match, returns the one that isn't USCRIPT_COMMON or
-// USCRIPT_INHERITED, i.e. the more specific one. Otherwise returns
-// USCRIPT_INVALID_CODE.
+// If the given scripts match, returns the one that isn't USCRIPT_INHERITED,
+// i.e. the more specific one. Otherwise returns USCRIPT_INVALID_CODE.
 UScriptCode ScriptIntersect(UScriptCode first, UScriptCode second) {
-  if (first == second ||
-      (second > USCRIPT_INVALID_CODE && second <= USCRIPT_INHERITED)) {
+  if (first == second || second == USCRIPT_INHERITED)
     return first;
-  }
-  if (first > USCRIPT_INVALID_CODE && first <= USCRIPT_INHERITED)
+  if (first == USCRIPT_INHERITED)
     return second;
   return USCRIPT_INVALID_CODE;
 }
 
 // Writes the script and the script extensions of the character with the
 // Unicode |codepoint|. Returns the number of written scripts.
 int GetScriptExtensions(UChar32 codepoint, UScriptCode* scripts) {
   UErrorCode icu_error = U_ZERO_ERROR;
   // ICU documentation incorrectly states that the result of
   // |uscript_getScriptExtensions| will contain the regular script property.
@@ skipping 47 matching lines @@
                    UScriptCode* script) {
   DCHECK_GT(length, 0U);
 
   UScriptCode scripts[kMaxScripts] = { USCRIPT_INVALID_CODE };
 
   base::i18n::UTF16CharIterator char_iterator(text.c_str() + start, length);
   size_t scripts_size = GetScriptExtensions(char_iterator.get(), scripts);
   *script = scripts[0];
 
   while (char_iterator.Advance()) {
+    // Special handling to merge white space into the previous run.
+    if (u_isUWhiteSpace(char_iterator.get()))
+      continue;
     ScriptSetIntersect(char_iterator.get(), scripts, &scripts_size);
     if (scripts_size == 0U)
       return char_iterator.array_pos();
     *script = scripts[0];
   }
 
   return length;
 }
 
 // A port of hb_icu_script_to_script because harfbuzz on CrOS is built without
@@ skipping 53 matching lines @@
         min_baseline_(min_baseline),
         min_height_(min_height),
         multiline_(multiline),
         word_wrap_behavior_(word_wrap_behavior),
         text_(text),
         words_(words),
         run_list_(run_list),
         text_x_(0),
         line_x_(0),
         max_descent_(0),
-        max_ascent_(0) {
+        max_ascent_(0),
+        word_width_(0) {
     DCHECK_EQ(multiline_, (words_ != nullptr));
     AdvanceLine();
   }
 
   // Breaks the run at given |run_index| into Line structs.
   void AddRun(int run_index) {
     const internal::TextRunHarfBuzz* run = run_list_.runs()[run_index];
     base::char16 first_char = text_[run->range.start()];
     if (multiline_ && first_char == '\n') {
       AdvanceLine();
-    } else if (multiline_ && (line_x_ + SkFloatToScalar(run->width)) >
-               max_width_) {
+      return;
+    }
+
+    if (multiline_ && line_x_ != 0) {

[Jun Mukai, 2015/05/11 23:21:40]

Why does this have to be here?
This break a run and call AdvanceLine(), which sounds like a role of BreakRun().

That way, a benefit is that you don't have to implement GetWordWidth() by
yourself, it's a part of BreakRunAtWidth().

+      // If the following word is not the first word in the current line and
+      // can't be fit into the current line, advance a new line.
+      // See kTestScenarios[2], kTestScenarios[7] and kTestScenarios[10] in
+      // RenderTextTest.Multiline_LineBreakerBehavior for example.
+      BreakList<size_t>::const_iterator word =
+          words_->GetBreak(run->range.start());
+      if (run->range.start() == word->first &&
+          line_x_ + GetWordWidth(word) > max_width_) {
+        AdvanceLine();
+      }
+    }
+
+    if (multiline_ && (line_x_ + SkFloatToScalar(run->width)) > max_width_) {
       BreakRun(run_index);
     } else {
       AddSegment(run_index, run->range, run->width);
     }
   }
 
   // Finishes line breaking and outputs the results. Can be called at most once.
   void Finalize(std::vector<internal::Line>* lines, SizeF* size) {
     DCHECK(!lines_.empty());
     // Add an empty line to finish the line size calculation and remove it.
     AdvanceLine();
     lines_.pop_back();
     *size = total_size_;
     lines->swap(lines_);
   }
 
  private:
   // A (line index, segment index) pair that specifies a segment in |lines_|.
   typedef std::pair<size_t, size_t> SegmentHandle;
 
   internal::LineSegment* SegmentFromHandle(const SegmentHandle& handle) {
     return &lines_[handle.first].segments[handle.second];
   }
 
+  // Gets the glyph width for |word|.
+  SkScalar GetWordWidth(BreakList<size_t>::const_iterator word) {
+    if (word == words_->breaks().end())
+      return 0;
+    size_t word_st = word->first;
+    size_t word_en =
+        (word + 1 == words_->breaks().end()) ? text_.size() : (word + 1)->first;
+    internal::TextRunList::TextRunHarfBuzzIter run_st =
+        run_list_.GetRunAt(word_st);
+    internal::TextRunList::TextRunHarfBuzzIter run_en =
+        run_list_.GetRunAt(word_en);
+    SkScalar width = 0;
+    for (auto iter = run_st; iter != run_en; iter++) {
+      Range char_range = (*iter)->range.Intersect(Range(word_st, word_en));
+      Range glyph_range = (*iter)->CharRangeToGlyphRange(char_range);
+      SkScalar char_width = ((glyph_range.end() >= (*iter)->glyph_count)
+                                 ? SkFloatToScalar((*iter)->width)
+                                 : (*iter)->positions[glyph_range.end()].x()) -
+                            (*iter)->positions[glyph_range.start()].x();
+      width += char_width;
+    }
+    return width;
+  }
+
   // Breaks a run into segments that fit in the last line in |lines_| and adds
   // them. Adds a new Line to the back of |lines_| whenever a new segment can't
   // be added without the Line's width exceeding |max_width_|.
   void BreakRun(int run_index) {
     const internal::TextRunHarfBuzz& run = *(run_list_.runs()[run_index]);
     SkScalar width = 0;
     size_t next_char = run.range.start();
 
     // Break the run until it fits the current line.
     while (next_char < run.range.end()) {
@@ skipping 21 matching lines @@
   bool BreakRunAtWidth(const internal::TextRunHarfBuzz& run,
                        size_t start_char,
                        SkScalar* width,
                        size_t* end_char,
                        size_t* next_char) {
     DCHECK(words_);
     DCHECK(run.range.Contains(Range(start_char, start_char + 1)));
     SkScalar available_width = max_width_ - line_x_;
     BreakList<size_t>::const_iterator word = words_->GetBreak(start_char);
     BreakList<size_t>::const_iterator next_word = word + 1;
-    // Width from |std::max(word->first, start_char)| to the current character.
-    SkScalar word_width = 0;
+    size_t word_end_pos =
+        (next_word == words_->breaks().end()) ? text_.size() : next_word->first;
+    word_width_ = (start_char == word->first) ? 0 : word_width_;
     *width = 0;
 
     Range char_range;
     SkScalar truncated_width = 0;
     for (size_t i = start_char; i < run.range.end(); i += char_range.length()) {
       // |word| holds the word boundary at or before |i|, and |next_word| holds
       // the word boundary right after |i|. Advance both |word| and |next_word|
       // when |i| reaches |next_word|.
       if (next_word != words_->breaks().end() && i >= next_word->first) {
-        if (*width > available_width) {
-          DCHECK_NE(WRAP_LONG_WORDS, word_wrap_behavior_);
-          *next_char = i;
-          if (word_wrap_behavior_ != TRUNCATE_LONG_WORDS)
-            *end_char = *next_char;
-          else
-            *width = truncated_width;
-          return true;
-        }
         word = next_word++;
-        word_width = 0;
+        word_end_pos = (next_word == words_->breaks().end()) ? text_.size()
+                                                             : next_word->first;
+        word_width_ = 0;
       }
 
       Range glyph_range;
       run.GetClusterAt(i, &char_range, &glyph_range);
       DCHECK_LT(0U, char_range.length());
 
       SkScalar char_width = ((glyph_range.end() >= run.glyph_count)
                                  ? SkFloatToScalar(run.width)
                                  : run.positions[glyph_range.end()].x()) -
                             run.positions[glyph_range.start()].x();
 
       *width += char_width;
-      word_width += char_width;
+      word_width_ += char_width;
 
       // TODO(mukai): implement ELIDE_LONG_WORDS.
       if (*width > available_width) {
-        if (line_x_ != 0 || word_width < *width) {
-          // Roll back one word.
-          *width -= word_width;
+        if ((line_x_ != 0 && word_width_ <= *width) ||
+            (line_x_ == 0 && word_width_ < *width)) {
+          // If the run is not the first run in the current line (line_x_ != 0),
+          // rool back one word and advance a new line if the current word can't
+          // be added into the current line without exceeding |available_width|.
+          // If the run is the first run in the current line (line_x_ == 0),
+          // don't roll back if the current word width is larger than the line
+          // width to avoid constructing empty lines.
+          // See kTestScenarios[0-1] in
+          // RenderTextTest.Multiline_LineBreakerBehavior.
+          *width -= word_width_;
           *next_char = std::max(word->first, start_char);
           *end_char = *next_char;
+          word_width_ = 0;
           return true;
         } else if (word_wrap_behavior_ == WRAP_LONG_WORDS) {
-          if (char_width < *width) {
+          if (char_width < *width ||
+              (word_width_ > *width && char_width <= *width)) {
             // Roll back one character.
+            // See kTestScenarios[8] for the first clause and kTestScenarios[9]
+            // for the second in RenderTextTest.Multiline_LineBreakerBehavior.
             *width -= char_width;
             *next_char = i;
           } else {
             // Continue from the next character.
             *next_char = i + char_range.length();
           }
           *end_char = *next_char;
+          word_width_ = 0;
           return true;
+        } else if (word_wrap_behavior_ == TRUNCATE_LONG_WORDS) {
+          *width = truncated_width;
+          *next_char = word_end_pos;
+          if (i < word_end_pos - 1) {
+            // Ignore all characters that belongs to the current word which has
+            // been truncated. Use kTestScenarios[5] as example. "that's good"
+            // has three runs: "that", "'", "s good". Suppose each line could
+            // fit into 4 characters, then the expected 2 lines should be:
+            // "that", "good", which means run "'" should be totally ignored and
+            // run "s good" should be partially ignored.
+            return false;
+          } else {
+            // Only advance a new line when the current word has been processed.
+            word_width_ = 0;
+            return true;
+          }
         }
       } else {
         *end_char = char_range.end();
         truncated_width = *width;
       }
     }
 
     if (word_wrap_behavior_ == TRUNCATE_LONG_WORDS)
       *width = truncated_width;
     *end_char = *next_char = run.range.end();
@@ skipping 31 matching lines @@
       line->size.set_height(std::max(min_height_, max_descent_ + max_ascent_));
       line->baseline =
           std::max(min_baseline_, SkScalarRoundToInt(max_ascent_));
       line->preceding_heights = std::ceil(total_size_.height());
       total_size_.set_height(total_size_.height() + line->size.height());
       total_size_.set_width(std::max(total_size_.width(), line->size.width()));
     }
     max_descent_ = 0;
     max_ascent_ = 0;
     line_x_ = 0;
+    word_width_ = 0;
     lines_.push_back(internal::Line());
   }
 
   // Adds a new segment with the given properties to |lines_.back()|.
   void AddSegment(int run_index, Range char_range, float width) {
     if (char_range.is_empty()) {
       DCHECK_EQ(0, width);
       return;
     }
     const internal::TextRunHarfBuzz& run = *(run_list_.runs()[run_index]);
@@ skipping 25 matching lines @@
     if (run.is_rtl) {
       rtl_segments_.push_back(
           SegmentHandle(lines_.size() - 1, line->segments.size() - 1));
       // If this is the last segment of an RTL run, reprocess the text-space x
       // ranges of all segments from the run.
       if (char_range.end() == run.range.end())
         UpdateRTLSegmentRanges();
     }
     text_x_ += SkFloatToScalar(width);
     line_x_ += SkFloatToScalar(width);
+
+    // Calculate |word_width_|.

[Jun Mukai, 2015/05/11 23:21:40]

If you modify AddRun() to invoke always BreakRun() if |multiline_| (regardless
of run's width), then BreakRunAtWidth() can compute the word width. Then you
don't have to compute here.

Right?

+    if (multiline_) {
+      BreakList<size_t>::const_iterator word_for_range_start =
+          words_->GetBreak(char_range.start());
+      BreakList<size_t>::const_iterator word_for_range_end =
+          (char_range.end() == 0) ? words_->breaks().begin()
+                                  : words_->GetBreak(char_range.end() - 1);
+
+      if (word_for_range_start == word_for_range_end) {
+        // Current word covers the entire char_range.
+        if (word_width_ == 0) {
+          Range glyph_range = run.CharRangeToGlyphRange(char_range);
+          SkScalar char_width = ((glyph_range.end() >= run.glyph_count)
+                                     ? SkFloatToScalar(run.width)
+                                     : run.positions[glyph_range.end()].x()) -
+                                run.positions[glyph_range.start()].x();
+          word_width_ += char_width;
+        }
+      } else {
+        // Current char_range intersects with more than one word.
+        word_width_ = char_range.end() - word_for_range_end->first;
+      }
+    }
   }
 
   const SkScalar max_width_;
   const int min_baseline_;
   const float min_height_;
   const bool multiline_;
   const WordWrapBehavior word_wrap_behavior_;
   const base::string16& text_;
   const BreakList<size_t>* const words_;
   const internal::TextRunList& run_list_;
 
   // Stores the resulting lines.
   std::vector<internal::Line> lines_;
 
   // Text space and line space x coordinates of the next segment to be added.
   SkScalar text_x_;
   SkScalar line_x_;
 
   float max_descent_;
   float max_ascent_;
 
+  // Stores the glyph width between the current position and the start position
+  // of the current word. If the current position is pointing to the first
+  // character of the current word, |word_width_| is set to 0.
+  SkScalar word_width_;
+
   // Size of the multiline text, not including the currently processed line.
   SizeF total_size_;
 
   // The current RTL run segments, to be applied by |UpdateRTLSegmentRanges()|.
   std::vector<SegmentHandle> rtl_segments_;
 
   DISALLOW_COPY_AND_ASSIGN(HarfBuzzLineBreaker);
 };
 
 // Function object for case insensitive string comparison.
@@ skipping 151 matching lines @@
 void TextRunList::ComputePrecedingRunWidths() {
   // Precalculate run width information.
   width_ = 0.0f;
   for (size_t i = 0; i < runs_.size(); ++i) {
     TextRunHarfBuzz* run = runs_[visual_to_logical_[i]];
     run->preceding_run_widths = width_;
     width_ += run->width;
   }
 }
 
+TextRunList::TextRunHarfBuzzIter TextRunList::GetRunAt(size_t position) const {
+  for (TextRunList::TextRunHarfBuzzIter iter = runs_.begin();
+       iter != runs_.end(); iter++) {
+    if ((*iter)->range.start() <= position && (*iter)->range.end() > position)
+      return iter;
+  }
+  return runs_.end();
+}
+
 }  // namespace internal
 
 RenderTextHarfBuzz::RenderTextHarfBuzz()
     : RenderText(),
       update_layout_run_list_(false),
       update_display_run_list_(false),
       update_grapheme_iterator_(false),
       update_display_text_(false),
       glyph_width_for_test_(0u) {
   set_truncate_length(kMaxTextLength);
@@ skipping 806 matching lines @@
   DCHECK(!update_layout_run_list_);
   DCHECK(!update_display_run_list_);
   return text_elided() ? display_run_list_.get() : &layout_run_list_;
 }
 
 const internal::TextRunList* RenderTextHarfBuzz::GetRunList() const {
   return const_cast<RenderTextHarfBuzz*>(this)->GetRunList();
 }
 
 }  // namespace gfx