Implement move by word, fix rendering of Arabic shape joining

ui/gfx/render_text_linux.cc

Index: ui/gfx/render_text_linux.cc
===================================================================
--- ui/gfx/render_text_linux.cc	(revision 100777)
+++ ui/gfx/render_text_linux.cc	(working copy)
@@ -8,6 +8,7 @@
 
 #include <algorithm>
 
+#include "base/i18n/break_iterator.h"
 #include "base/logging.h"
 #include "ui/gfx/canvas_skia.h"
 #include "ui/gfx/pango_util.h"
@@ -184,8 +185,7 @@
     return LeftEndSelectionModel();
   if (break_type == CHARACTER_BREAK)
     return LeftSelectionModel(current);
-  // TODO(xji): not implemented yet.
-  return RenderText::GetLeftSelectionModel(current, break_type);
+  return LeftSelectionModelByWord(current);

[msw, 2011/09/15 01:26:00]

DCHECK(break_type == WORD_BREAK); before this to ensure someone doesn't
introduce a different break type that gets inadvertantly handled as a word here.
Same for GetRightSelectionModel below.

[xji, 2011/09/15 22:58:09]

Done.

 }
 
 SelectionModel RenderTextLinux::GetRightSelectionModel(
@@ -197,8 +197,7 @@
     return RightEndSelectionModel();
   if (break_type == CHARACTER_BREAK)
     return RightSelectionModel(current);
-  // TODO(xji): not implemented yet.
-  return RenderText::GetRightSelectionModel(current, break_type);
+  return RightSelectionModelByWord(current);
 }
 
 SelectionModel RenderTextLinux::LeftEndSelectionModel() {
@@ -428,6 +427,64 @@
                                       FirstSelectionModelInsideRun(item);
 }
 
+SelectionModel RenderTextLinux::LeftSelectionModelByWord(
+    const SelectionModel& selection) {
+  base::i18n::BreakIterator iter(text(), base::i18n::BreakIterator::BREAK_WORD);
+  bool success = iter.Init();
+  DCHECK(success);
+  if (!success)
+    return selection;
+
+  SelectionModel left_end = LeftEndSelectionModel();
+  SelectionModel left(selection);
+  while (!left.Equals(left_end)) {
+    left = LeftSelectionModel(left);
+    size_t caret = left.caret_pos();
+    GSList* run = GetRunContainingPosition(caret);
+    DCHECK(run);
+    PangoItem* item = reinterpret_cast<PangoLayoutRun*>(run->data)->item;
+    size_t cursor = left.selection_end();
+    if (item->analysis.level % 2 == 0) {  // LTR run.
+      if (iter.IsStartOfWord(cursor))
+        return left;
+    } else {  // RTL run.
+      if (iter.IsEndOfWord(cursor))
+        return left;
+    }
+  }
+
+  return left_end;
+}
+
+SelectionModel RenderTextLinux::RightSelectionModelByWord(
+    const SelectionModel& selection) {
+  base::i18n::BreakIterator iter(text(), base::i18n::BreakIterator::BREAK_WORD);
+  bool success = iter.Init();
+  DCHECK(success);
+  if (!success)
+    return selection;
+
+  SelectionModel right_end = RightEndSelectionModel();
+  SelectionModel right(selection);
+  while (!right.Equals(right_end)) {
+    right = RightSelectionModel(right);
+    size_t caret = right.caret_pos();
+    GSList* run = GetRunContainingPosition(caret);
+    DCHECK(run);
+    PangoItem* item = reinterpret_cast<PangoLayoutRun*>(run->data)->item;
+    size_t cursor = right.selection_end();
+    if (item->analysis.level % 2 == 0) {  // LTR run.
+      if (iter.IsEndOfWord(cursor))
+        return right;
+    } else {  // RTL run.
+      if (iter.IsStartOfWord(cursor))
+        return right;
+    }
+  }
+
+  return right_end;
+}
+
 PangoLayout* RenderTextLinux::EnsureLayout() {
   if (layout_ == NULL) {
     CanvasSkia canvas(display_rect().width(), display_rect().height(), false);
@@ -486,6 +543,15 @@
 void RenderTextLinux::SetupPangoAttributes(PangoLayout* layout) {
   PangoAttrList* attrs = pango_attr_list_new();
   // Set selection background color.
+  // TODO(xji): There's a bug in pango that it can't use two colors in one
+  // glyph. Please refer to https://bugzilla.gnome.org/show_bug.cgi?id=648157
+  // for detail. So for example, if a font has "ffi" ligature, but you select
+  // half of that glyph, you either get the entire "ffi" ligature
+  // selection-colored, or none of it.
+  // We could use clipping to render selection.
+  // Use pango_glyph_item_get_logical_widths to find the exact boundaries of
+  // selection, then cairo_clip that, paint background, set color to white and
+  // redraw the layout.
   SkColor selection_color =
       focused() ? kFocusedSelectionColor : kUnfocusedSelectionColor;
   size_t start = std::min(MinOfSelection(), text().length());
@@ -502,6 +568,8 @@
   StyleRanges ranges_of_style(style_ranges());
   ApplyCompositionAndSelectionStyles(&ranges_of_style);
 
+  PlatformFont* default_platform_font = default_style().font.platform_font();
+
   for (StyleRanges::const_iterator i = ranges_of_style.begin();
        i < ranges_of_style.end(); ++i) {
     start = std::min(i->range.start(), text().length());
@@ -509,13 +577,21 @@
     if (start >= end)
       continue;
 
-    const Font& font = !i->underline ? i->font :
-        i->font.DeriveFont(0, i->font.GetStyle() | Font::UNDERLINED);
-    PangoFontDescription* desc = font.GetNativeFont();
-    pango_attr = pango_attr_font_desc_new(desc);
-    AppendPangoAttribute(start, end, pango_attr, attrs);
-    pango_font_description_free(desc);
+    const Font& font = i->font;
+    if (font.platform_font() != default_platform_font) {

[msw, 2011/09/15 01:26:00]

I want to make sure this is correct. If you don't AppendPangoAttribute, it just
uses the default font right? (If it uses the previous run's font, then this is
incorrect.)

Still, this will break Arabic shaping if the two runs hosting that text have the
same non-default font. It may not occur in practice now, but this would
AppendPangoAttribute and split up the text for each half with a custom font.

I think we'll eventually need to split up StyleRange into components
(ColorRange, FontRange, etc.) so that we can combine adjacent ranges with the
same Fonts (to avoid unnecessarily splitting up runs).

[xji, 2011/09/15 22:58:09]

Yes.
You are right.
You are right. I added a TODO.

+      // In pango, different fonts means different runs, and it breaks Arabic

[msw, 2011/09/15 01:26:00]

Capitalize Pango?, also this comment probably belongs just above and outside
this conditional block.

[xji, 2011/09/15 22:58:09]

Done.

+      // shaping acorss run boundaries. So, set font to the largest range.
+      PangoFontDescription* desc = font.GetNativeFont();
+      pango_attr = pango_attr_font_desc_new(desc);
+      AppendPangoAttribute(start, end, pango_attr, attrs);
+      pango_font_description_free(desc);
+    }
 
+    if (i->underline) {
+      pango_attr = pango_attr_underline_new(PANGO_UNDERLINE_SINGLE);
+      AppendPangoAttribute(start, end, pango_attr, attrs);
+    }
+
     SkColor foreground = i->foreground;
     pango_attr = pango_attr_foreground_new(
         ConvertColorFrom8BitTo16Bit(SkColorGetR(foreground)),
@@ -542,6 +618,7 @@
   pango_attr_list_insert(attrs, pango_attr);
 }
 
+// TODO(xji): keep a vector of runs to avoid use single-linked list.

[msw, 2011/09/15 01:26:00]

Nits: "*K*eep", "us*ing*", "singl*y*" and "list*s*"

[xji, 2011/09/15 22:58:09]

done except "lists", it is one list.

[msw, 2011/09/17 00:27:10]

Technically that's grammatically incorrect, you'll need a preposition with a
singular noun in that case:
"... to avoid using *a* singly-linked list."

 PangoLayoutRun* RenderTextLinux::GetPreviousRun(PangoLayoutRun* run) const {
   GSList* current = current_line_->runs;
   GSList* prev = NULL;