RenderTextHarfBuzz: Decide run direction by BiDi embedding level

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 <map>
 
 #include "base/i18n/bidi_line_iterator.h"
 #include "base/i18n/break_iterator.h"
@@ skipping 845 matching lines @@
   size_t position = LayoutIndexToTextIndex(run->range.end());
   position = IndexOfAdjacentGrapheme(position, CURSOR_BACKWARD);
   return SelectionModel(position, CURSOR_FORWARD);
 }
 
 void RenderTextHarfBuzz::ItemizeText() {
   const base::string16& text = GetLayoutText();
   const bool is_text_rtl = GetTextDirection() == base::i18n::RIGHT_TO_LEFT;
   DCHECK_NE(0U, text.length());
 
-  // If ICU fails to itemize the text, we set |fake_runs| and create a run that
-  // spans the entire text. This is needed because early returning and leaving
-  // the runs set empty causes some clients to crash/misbehave since they expect
-  // non-zero text metrics from a non-empty text.
+  // If ICU fails to itemize the text, we create a run that spans the entire
+  // text. This is needed because leaving the runs set empty causes some clients
+  // to crash/misbehave since they expect non-zero text metrics from a non-empty

[msw, 2014/06/10 17:42:52]

nit: remove "crash/" to shorten the comment by a line.

[ckocagil, 2014/06/10 19:12:52]

Done.

+  // text.
   base::i18n::BiDiLineIterator bidi_iterator;
-  bool fake_runs = !bidi_iterator.Open(text, is_text_rtl, false);
+  if (!bidi_iterator.Open(text, is_text_rtl, false)) {
+    internal::TextRunHarfBuzz* run = new internal::TextRunHarfBuzz;
+    run->range = Range(0, text.length());
+    runs_.push_back(run);
+    visual_to_logical_ = logical_to_visual_ = std::vector<int32_t>(1, 0);
+    return;
+  }
 
   // Temporarily apply composition underlines and selection colors.
   ApplyCompositionAndSelectionStyles();
 
   // Build the list of runs from the script items and ranged styles. Use an
   // empty color BreakList to avoid breaking runs at color boundaries.
   BreakList<SkColor> empty_colors;
   empty_colors.SetMax(text.length());
   internal::StyleIterator style(empty_colors, styles());
 
   for (size_t run_break = 0; run_break < text.length();) {
     internal::TextRunHarfBuzz* run = new internal::TextRunHarfBuzz;
     run->range.set_start(run_break);
     run->font_style = (style.style(BOLD) ? Font::BOLD : 0) |
                       (style.style(ITALIC) ? Font::ITALIC : 0);
     run->strike = style.style(STRIKE);
     run->diagonal_strike = style.style(DIAGONAL_STRIKE);
     run->underline = style.style(UNDERLINE);
 
-    if (fake_runs) {
-      run_break = text.length();
-    } else {
-      int32 script_item_break = 0;
-      bidi_iterator.GetLogicalRun(run_break, &script_item_break, &run->level);
-      // Find the length and script of this script run.
-      script_item_break = ScriptInterval(text, run_break,
-          script_item_break - run_break, &run->script) + run_break;
+    int32 script_item_break = 0;
+    bidi_iterator.GetLogicalRun(run_break, &script_item_break, &run->level);
+    // Odd BiDi embedding levels correspond to RTL runs.
+    run->is_rtl = (run->level % 2) == 1;

[msw, 2014/06/10 17:42:51]

Is this also true in RTL UI (--lang=he)?

[ckocagil, 2014/06/10 19:12:52]

AFAICT, yes.

[msw, 2014/06/10 20:47:56]

I wonder if that's right. Do we need to set a base direction, like
RenderTextWin::ItemizeLogicalText()?
// Set Uniscribe's base text direction.
  script_state_.uBidiLevel =
      (GetTextDirection() == base::i18n::RIGHT_TO_LEFT) ? 1 : 0;
https://code.google.com/p/chromium/codesearch#chromium/src/ui/gfx/render_text...

[ckocagil, 2014/06/11 07:18:08]

I think base directions are only relevant *before* establishing the run levels.
Once the levels are established, the directions are already decided. The levels
generated by i18n::BiDiLineIterator take into account the GetTextDirection()
value we pass it.

I tested this by making GetTextDirection() return RTL and typing a bunch of weak
characters (periods, commas) in an empty textfield. It was correctly displayed
as RTL.

So yeah, this seems to work right.

[msw, 2014/06/11 07:54:19]

You're right about the function ordering, but my question is still relevant: in
--lang=he RTL UI, does weak text still yield RTL runs? If not, we need to do
something as I suggest (setting a base direction before itemizing the runs).

+    // Find the length and script of this script run.
+    script_item_break = ScriptInterval(text, run_break,
+        script_item_break - run_break, &run->script) + run_break;
 
-      // Find the next break and advance the iterators as needed.
-      run_break = std::min(static_cast<size_t>(script_item_break),
-                           TextIndexToLayoutIndex(style.GetRange().end()));
+    // Find the next break and advance the iterators as needed.
+    run_break = std::min(static_cast<size_t>(script_item_break),
+                         TextIndexToLayoutIndex(style.GetRange().end()));
 
-      // Break runs adjacent to character substrings in certain code blocks.
-      // This avoids using their fallback fonts for more characters than needed,
-      // in cases like "\x25B6 Media Title", etc. http://crbug.com/278913
-      if (run_break > run->range.start()) {
-        const size_t run_start = run->range.start();
-        const int32 run_length = static_cast<int32>(run_break - run_start);
-        base::i18n::UTF16CharIterator iter(text.c_str() + run_start,
-                                           run_length);
-        const UBlockCode first_block_code = ublock_getCode(iter.get());
-        const bool first_block_unusual = IsUnusualBlockCode(first_block_code);
-        while (iter.Advance() && iter.array_pos() < run_length) {
-          const UBlockCode current_block_code = ublock_getCode(iter.get());
-          if (current_block_code != first_block_code &&
-              (first_block_unusual || IsUnusualBlockCode(current_block_code))) {
-            run_break = run_start + iter.array_pos();
-            break;
-          }
+    // Break runs adjacent to character substrings in certain code blocks.
+    // This avoids using their fallback fonts for more characters than needed,
+    // in cases like "\x25B6 Media Title", etc. http://crbug.com/278913
+    if (run_break > run->range.start()) {
+      const size_t run_start = run->range.start();
+      const int32 run_length = static_cast<int32>(run_break - run_start);
+      base::i18n::UTF16CharIterator iter(text.c_str() + run_start,
+                                         run_length);
+      const UBlockCode first_block_code = ublock_getCode(iter.get());
+      const bool first_block_unusual = IsUnusualBlockCode(first_block_code);
+      while (iter.Advance() && iter.array_pos() < run_length) {
+        const UBlockCode current_block_code = ublock_getCode(iter.get());
+        if (current_block_code != first_block_code &&
+            (first_block_unusual || IsUnusualBlockCode(current_block_code))) {
+          run_break = run_start + iter.array_pos();
+          break;
         }
       }
     }
 
     DCHECK(IsValidCodePointIndex(text, run_break));
     style.UpdatePosition(LayoutIndexToTextIndex(run_break));
     run->range.set_end(run_break);
-    UBiDiDirection direction = ubidi_getBaseDirection(
-        text.c_str() + run->range.start(), run->range.length());
-    if (direction == UBIDI_NEUTRAL)
-      run->is_rtl = is_text_rtl;
-    else
-      run->is_rtl = direction == UBIDI_RTL;
+
     runs_.push_back(run);
   }
 
   // Undo the temporarily applied composition underlines and selection colors.
   UndoCompositionAndSelectionStyles();
 
   const size_t num_runs = runs_.size();
   std::vector<UBiDiLevel> levels(num_runs);
   for (size_t i = 0; i < num_runs; ++i)
     levels[i] = runs_[i]->level;
@@ skipping 48 matching lines @@
     run->positions[i].set(run->width + x_offset, y_offset);
     run->width +=
         SkScalarRoundToInt(SkFixedToScalar(hb_positions[i].x_advance));
   }
 
   hb_buffer_destroy(buffer);
   hb_font_destroy(harfbuzz_font);
 }
 
 }  // namespace gfx