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

ui/gfx/render_text_harfbuzz.cc

Index: ui/gfx/render_text_harfbuzz.cc
diff --git a/ui/gfx/render_text_harfbuzz.cc b/ui/gfx/render_text_harfbuzz.cc
index 32a9e554f60b47127cef9d426aef71a9f410adfa..0db4394cd66ea3c7b3beb883e5a16d499d9254e7 100644
--- a/ui/gfx/render_text_harfbuzz.cc
+++ b/ui/gfx/render_text_harfbuzz.cc
@@ -90,15 +90,12 @@ size_t FindRunBreakingCharacter(const base::string16& text,
   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;
 }
@@ -167,6 +164,11 @@ int ScriptInterval(const base::string16& text,
 
   while (char_iterator.Advance()) {

[Jun Mukai, 2015/04/29 21:09:07]

Hm, why not just doing
if (u_isUWhiteSpace(char_iterator.get()))
  continue;
?
You don't have to compute ScriptSetIntersect for whitespaces, anyways it
shouldn't separate the text runs.

[xdai1, 2015/05/05 18:27:49]

As I said in the previous comment to msw@, this is depending on how we want to
break the text into runs. For example, suppose we have text "abc defg". If we
use  "continue;" here, then we'll get two runs: "abc " and "defg", but if we
fake out |scripts| and |scripts_size|, then we'll get one single run: "abc
defg". I think these two approaches are both fine. But since the previous one
makes more sense, I'll accept your suggestion and use "continue;" here.

     ScriptSetIntersect(char_iterator.get(), scripts, &scripts_size);
+    // Special handling to merge white space into the previous run.
+    if (u_isUWhiteSpace(char_iterator.get())) {
+      scripts[0] = *script;
+      scripts_size = 1U;
+    }
     if (scripts_size == 0U)
       return char_iterator.array_pos();
     *script = scripts[0];
@@ -239,7 +241,9 @@ class HarfBuzzLineBreaker {
         text_x_(0),
         line_x_(0),
         max_descent_(0),
-        max_ascent_(0) {
+        max_ascent_(0),
+        next_word_start_pos_(0),
+        word_width_(0) {
     DCHECK_EQ(multiline_, (words_ != nullptr));
     AdvanceLine();
   }
@@ -248,6 +252,15 @@ class HarfBuzzLineBreaker {
   void AddRun(int run_index) {
     const internal::TextRunHarfBuzz* run = run_list_.runs()[run_index];
     base::char16 first_char = text_[run->range.start()];
+
+    if (word_wrap_behavior_ == TRUNCATE_LONG_WORDS && words_) {
+      // If the run has been processed, skip it.
+      size_t current_word_start_pos =
+          words_->GetBreak(run->range.start())->first;
+      if (current_word_start_pos < next_word_start_pos_)
+        return;
+    }
+
     if (multiline_ && first_char == '\n') {
       AdvanceLine();
     } else if (multiline_ && (line_x_ + SkFloatToScalar(run->width)) >
@@ -317,8 +330,8 @@ class HarfBuzzLineBreaker {
     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;
+    next_word_start_pos_ =

[Jun Mukai, 2015/04/29 21:09:07]

Probably word_end_pos_ would be a better name?

[xdai1, 2015/05/05 18:27:49]

Done.

+        (next_word == words_->breaks().end()) ? text_.size() : next_word->first;
     *width = 0;
 
     Range char_range;
@@ -328,17 +341,11 @@ class HarfBuzzLineBreaker {
       // 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) {

[Jun Mukai, 2015/04/29 21:09:07]

I am not sure if removing this block works correctly.  Consider the input is
"aaaaa-b ccc" and the given width is for 3 characters, behavior is
IGNORE_LONG_WORDS.

In that case:
- first run "aaaaa" runs and no problem
- second run "-" has no problem
- third run "b ccc". It first scan "b ", it's *width is already bigger than
available_width, it needs to return true here, otherwise it continues to show
"ccc" in the same line.

[xdai1, 2015/05/05 18:27:49]

In third run it won't come inside this inner if because it can't pass the outer
if (next_word == words_->breaks().end()).

I also tested it by writing a test case and it works as expected. It creates
three lines (because it contains three words: [aaaaa-], [b ], [ccc]):
aaaaa-
b
ccc
Actually it just postpones returning at line 387.

[Jun Mukai, 2015/05/06 17:38:52]

For line 387, the behavior has to be TRUNCATE_LONG_WORDS. Consider the behavior
is IGNORE_LONG_WORDS.  In the case I commented above,
- *width > available_width, but
- line_x_ > 0 and word_width_ > *width
- word_wrap_behavior_ is neither WRAP_LONG_WORDS nor TRUNCATE_LONG_WORDS

Therefore, it doesn't match any cases in the if-clause of line 365-388.  It
eventually arrives here.

[xdai1, 2015/05/06 20:23:44]

In this case when it comes to "b", it will return in line 375. Because:
1) "b " is a single run, it doesn't combined into the previous "aaaaa-" run.
2) line_x_ != 0 && word_width_ == *width
I also added this test case to RenderTextTest.Multiline_LineBreakerBehavior.

[Jun Mukai, 2015/05/06 20:47:55]

No, "b " and "ccc" must be in a single run, they are only separated by a
whitespace.
word_width_ shouldn't equal to *width.  *width starts from 0 in each iteration
of this method, but word_width_ should start from 'aaaaa-' (oh, if hyphen "-"
splits into words, think that "aaaaa'b ccc").
My point is, think that a run contains multiple words (like "b ccc"), and its
first word ("b ") is actually a part of a larger word including previous runs
("aaaaa'"), and it's involved by linebreaks.
I don't get how your code deals with such cases.

-          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;
+        next_word_start_pos_ = (next_word == words_->breaks().end())
+                                   ? text_.size()
+                                   : next_word->first;
+        word_width_ = 0;
       }
 
       Range glyph_range;
@@ -351,15 +358,17 @@ class HarfBuzzLineBreaker {
                             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) {
+        if ((line_x_ != 0 && word_width_ <= *width) ||
+            (line_x_ == 0 && word_width_ < *width)) {
           // Roll back one word.
-          *width -= word_width;
+          *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) {
@@ -372,6 +381,10 @@ class HarfBuzzLineBreaker {
           }
           *end_char = *next_char;
           return true;
+        } else if (word_wrap_behavior_ == TRUNCATE_LONG_WORDS) {
+          *width = truncated_width;
+          *next_char = next_word_start_pos_;
+          return true;
         }
       } else {
         *end_char = char_range.end();
@@ -423,6 +436,7 @@ class HarfBuzzLineBreaker {
     max_descent_ = 0;
     max_ascent_ = 0;
     line_x_ = 0;
+    word_width_ = 0;
     lines_.push_back(internal::Line());
   }
 
@@ -468,6 +482,31 @@ class HarfBuzzLineBreaker {
     }
     text_x_ += SkFloatToScalar(width);
     line_x_ += SkFloatToScalar(width);
+
+    // Calculate |word_width_|.
+    if (words_) {
+      BreakList<size_t>::const_iterator current_word =
+          (char_range.start() == 0) ? words_->breaks().begin()
+                                    : words_->GetBreak(char_range.end() - 1);
+      BreakList<size_t>::const_iterator next_word =
+          (char_range.end() == text_.size())
+              ? words_->breaks().end()
+              : words_->GetBreak(char_range.end());
+      if (word_width_ == 0 && current_word == next_word) {
+        Range tmp_range;
+        Range glyph_range;
+        for (size_t pos = current_word->first; pos < char_range.end(); ++pos) {
+          run.GetClusterAt(pos, &tmp_range, &glyph_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 if (current_word != next_word) {
+        word_width_ = 0;
+      }
+    }
   }
 
   const SkScalar max_width_;
@@ -489,6 +528,13 @@ class HarfBuzzLineBreaker {
   float max_descent_;
   float max_ascent_;
 
+  // Stores the start position of next word that need to be processed.
+  size_t next_word_start_pos_;
+  // 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_;