Implement Pango RenderText for Linux.

ui/gfx/render_text_linux.cc

 // Copyright (c) 2011 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_linux.h"
 
+#include <pango/pangocairo.h>
+#include <algorithm>
+
+#include "base/logging.h"
+#include "ui/gfx/canvas_skia.h"
+#include "ui/gfx/gtk_util.h"
+#include "unicode/uchar.h"
+#include "unicode/ustring.h"
+
+// TODO(xji): instead of converting each R or G or B from 16-bite to 32-bit,
+// it should also massage A in the conversion.
+#define COLOR_16_TO_32_BIT(c) ((c)/0xff * 0xffff)

[msw, 2011/08/23 08:01:01]

Can you make this a file local function instead of a pre-processor directive?
Also, I'd appreciate if someone else could double-check this logic.

[xji, 2011/08/23 23:52:52]

changed to function in unnamed namespace.
checked with Behdad. changed to "c * (0xffff / 0xff)".

+
+// TODO(xji): index saved in upper layer is utf16 index. Pango uses utf8 index.
+// Since caret_pos is used internally, we could save utf8 index for caret_pos
+// to avoid conversion.
+
 namespace gfx {
 
 RenderTextLinux::RenderTextLinux()
-    : RenderText() {
+    : layout_(NULL),
+      layout_line_(NULL) {
 }
 
 RenderTextLinux::~RenderTextLinux() {
+  ResetLayout();
 }
 
 RenderText* RenderText::CreateRenderText() {
   return new RenderTextLinux;
 }
 
+void RenderTextLinux::SetText(const string16& text) {
+  RenderText::SetText(text);
+  ResetLayout();
+}
+
+void RenderTextLinux::SetDisplayRect(const Rect&r) {
+  RenderText::SetDisplayRect(r);
+  ResetLayout();
+}
+
+void RenderTextLinux::ApplyStyleRange(StyleRange style_range) {
+  RenderText::ApplyStyleRange(style_range);
+  ResetLayout();
+}
+
+void RenderTextLinux::ApplyDefaultStyle() {
+  RenderText::ApplyDefaultStyle();
+  ResetLayout();
+}
+
+base::i18n::TextDirection RenderTextLinux::GetTextDirection() {
+  EnsureLayout();
+
+  const char* layout_text = pango_layout_get_text(layout_);
+  PangoDirection base_dir =
+        pango_find_base_dir(layout_text, strlen(layout_text));
+  if (base_dir == PANGO_DIRECTION_RTL || base_dir == PANGO_DIRECTION_WEAK_RTL)
+    return base::i18n::RIGHT_TO_LEFT;
+  return base::i18n::LEFT_TO_RIGHT;
+}
+
+int RenderTextLinux::GetStringWidth() {
+  PangoLayout* layout = EnsureLayout();
+  int width, height;
+  pango_layout_get_pixel_size(layout, &width, &height);
+  return width;
+}
+
+void RenderTextLinux::Draw(Canvas* canvas) {
+  PangoLayout* layout = EnsureLayout();
+  Rect bounds(display_rect());
+
+  // Clip the canvas to the text display area.
+  CanvasSkia* canvas_skia = canvas->AsCanvasSkia();
+  canvas_skia->ClipRectInt(
+      bounds.x(), bounds.y(), bounds.width(), bounds.height());
+
+  // TODO(xji): use ScopedPlatformPaing.

[msw, 2011/08/23 08:01:01]

ScopedPlatformPain*t*

[xji, 2011/08/23 23:52:52]

Done.

+  cairo_t* cr = skia::BeginPlatformPaint(canvas_skia);
+  cairo_save(cr);
+  cairo_rectangle(cr, bounds.x(), bounds.y(), bounds.width(), bounds.height());
+  cairo_clip(cr);
+
+  int text_width, text_height;
+  pango_layout_get_pixel_size(layout, &text_width, &text_height);
+  Point offset(ToViewPoint(Point()));
+  // Vertically centered.
+  int text_y = offset.y() + ((bounds.height() - text_height) / 2);
+  cairo_move_to(cr, offset.x(), text_y);
+  pango_cairo_show_layout(cr, layout);
+
+  // Destructor.
+  cairo_restore(cr);
+  skia::EndPlatformPaint(canvas_skia);
+
+  // Paint cursor.
+  bounds = GetUpdatedCursorBounds();
+  if (cursor_visible() && focused() && !bounds.IsEmpty()) {
+    if (!bounds.IsEmpty())
+       canvas->DrawRectInt(kCursorColor,
+                          bounds.x(),
+                          bounds.y(),
+                          bounds.width(),
+                          bounds.height());
+  }
+}
+
+SelectionModel RenderTextLinux::FindCursorPosition(const Point& point) {
+  // TODO(xji): when points outside of text, return HOME/END position.
+  PangoLayout* layout = EnsureLayout();
+
+  if (text().empty())
+    return SelectionModel(0, 0, SelectionModel::LEADING);
+
+  Point p(ToTextPoint(point));
+  int caret_pos, trailing;
+  pango_layout_xy_to_index(layout, p.x() * PANGO_SCALE, p.y() * PANGO_SCALE,
+                           &caret_pos, &trailing);
+
+  size_t selection_end = caret_pos;
+  if (trailing > 0) {
+    const char* pango_text = pango_layout_get_text(layout);
+    const char* ch = g_utf8_offset_to_pointer(pango_text + caret_pos, trailing);
+    selection_end = ch - pango_text;
+  }
+
+  return SelectionModel(
+      Utf8IndexToUtf16Index(selection_end),
+      Utf8IndexToUtf16Index(caret_pos),
+      trailing > 0 ? SelectionModel::TRAILING : SelectionModel::LEADING);
+}
+
+Rect RenderTextLinux::GetCursorBounds(const SelectionModel& selection,
+                                      bool insert_mode) {
+  PangoLayout* layout = EnsureLayout();
+
+  size_t caret_pos = insert_mode ? selection.caret_pos() :
+                                   selection.selection_end();
+  PangoRectangle pos;
+  pango_layout_index_to_pos(layout, Utf16IndexToUtf8Index(caret_pos), &pos);
+
+  SelectionModel::CaretPlacement caret_placement = selection.caret_placement();
+  int x = pos.x;
+  if ((insert_mode && caret_placement == SelectionModel::TRAILING) ||
+      (!insert_mode && pos.width < 0))
+    x += pos.width;
+  x = x / PANGO_SCALE;
+
+  int h = std::min(display_rect().height(), pos.height/PANGO_SCALE);

[msw, 2011/08/23 08:01:01]

spaces around '/' operator.

[xji, 2011/08/23 23:52:52]

Done.

+  Rect bounds(x, (display_rect().height() - h) / 2, 0, h);
+  bounds.set_origin(ToViewPoint(bounds.origin()));
+
+  if (!insert_mode)
+    bounds.set_width(std::abs(pos.width));
+
+  return bounds;
+}
+
+SelectionModel RenderTextLinux::GetLeftSelectionModel(
+    const SelectionModel& current,
+    BreakType break_type) {
+  EnsureLayout();
+
+  if (break_type == LINE_BREAK || text().empty())
+    return LeftEndSelectionModel();
+  if (break_type == CHARACTER_BREAK)
+    return LeftSelectionModel(current);
+  // TODO(xji): not implemented yet.
+  return RenderText::GetLeftSelectionModel(current, break_type);
+}
+
+SelectionModel RenderTextLinux::GetRightSelectionModel(
+    const SelectionModel& current,
+    BreakType break_type) {
+  EnsureLayout();
+
+  if (break_type == LINE_BREAK || text().empty())
+    return RightEndSelectionModel();
+  if (break_type == CHARACTER_BREAK)
+    return RightSelectionModel(current);
+  // TODO(xji): not implemented yet.
+  return RenderText::GetRightSelectionModel(current, break_type);
+}
+
+size_t RenderTextLinux::GetIndexOfPreviousGrapheme(size_t position) {
+  EnsureLayout();
+  return Utf16IndexOfAdjacentGrapheme(position, PREVIOUS);
+}
+
+SelectionModel RenderTextLinux::LeftEndSelectionModel() {
+  if (GetTextDirection() == base::i18n::RIGHT_TO_LEFT) {
+    GSList* first_visual_run = layout_line_->runs;
+    if (first_visual_run) {
+      PangoItem* item =
+          reinterpret_cast<PangoLayoutRun*>(first_visual_run->data)->item;
+      if (item->analysis.level % 2 == 0) {  // LTR.
+        size_t caret = Utf8IndexToUtf16Index(item->offset);
+        return SelectionModel(text().length(), caret, SelectionModel::LEADING);
+      } else {  // RTL.
+        size_t caret = Utf8IndexToUtf16Index(item->offset + item->length);
+        caret = Utf16IndexOfAdjacentGrapheme(caret, PREVIOUS);
+        return SelectionModel(text().length(), caret, SelectionModel::TRAILING);
+      }
+    }
+  }
+  return SelectionModel(0, 0, SelectionModel::LEADING);
+}
+
+SelectionModel RenderTextLinux::RightEndSelectionModel() {
+  if (GetTextDirection() == base::i18n::LEFT_TO_RIGHT) {
+    GSList* last_visual_run = GetLastRun();
+    if (last_visual_run) {
+      PangoItem* item =
+          reinterpret_cast<PangoLayoutRun*>(last_visual_run->data)->item;
+      if (item->analysis.level % 2 == 0) {  // LTR.
+        size_t caret = Utf8IndexToUtf16Index(item->offset + item->length);
+        caret = Utf16IndexOfAdjacentGrapheme(caret, PREVIOUS);
+        return SelectionModel(text().length(), caret, SelectionModel::TRAILING);
+      } else {  // RTL.
+        size_t caret = Utf8IndexToUtf16Index(item->offset);
+        return SelectionModel(text().length(), caret, SelectionModel::LEADING);
+      }
+    }
+  }
+  return SelectionModel(0, 0, SelectionModel::LEADING);
+}
+
+GSList* RenderTextLinux::GetRunContainingPosition(size_t position) const {
+  GSList* run = layout_line_->runs;
+  while (run) {
+    PangoItem* box = reinterpret_cast<PangoLayoutRun*>(run->data)->item;
+    size_t run_start = Utf8IndexToUtf16Index(box->offset);
+    size_t run_end = Utf8IndexToUtf16Index(box->offset + box->length);
+
+    if (position >= run_start && position < run_end)
+      return run;
+    run = run->next;
+  }
+  return NULL;
+}
+
+size_t RenderTextLinux::Utf8IndexOfAdjacentGrapheme(
+    size_t utf16_index_of_current_grapheme,
+    RelativeLogicalPosition pos) const {
+  PangoLogAttr* log_attrs;
+  gint n_attrs;
+  pango_layout_get_log_attrs(layout_, &log_attrs, &n_attrs);
+
+  const char* layout_text = pango_layout_get_text(layout_);
+  const char* ch = layout_text +
+      Utf16IndexToUtf8Index(utf16_index_of_current_grapheme);
+  int char_offset = static_cast<int>(g_utf8_pointer_to_offset(layout_text, ch));
+
+  if (pos == PREVIOUS) {
+    if (ch > layout_text) {
+      do {
+        ch = g_utf8_find_prev_char(layout_text, ch);
+        --char_offset;
+      } while (ch && *ch && !log_attrs[char_offset].is_cursor_position);
+      if (!ch)
+        ch = layout_text;
+    }
+  } else {
+    if (ch < layout_text + strlen(layout_text)) {
+      do {
+        ch = g_utf8_find_next_char(ch, NULL);
+        ++char_offset;
+      } while (ch && *ch && !log_attrs[char_offset].is_cursor_position);
+      if (!ch || ch >= layout_text + strlen(layout_text))
+        ch = layout_text + strlen(layout_text);
+    }
+  }
+
+  size_t utf8_index = static_cast<size_t>(ch - layout_text);
+  g_free(log_attrs);
+  return utf8_index;
+}
+
+size_t RenderTextLinux::Utf16IndexOfAdjacentGrapheme(
+    size_t utf16_index_of_current_grapheme,
+    RelativeLogicalPosition pos) const {
+  size_t utf8_index = Utf8IndexOfAdjacentGrapheme(
+      utf16_index_of_current_grapheme, pos);
+  return Utf8IndexToUtf16Index(utf8_index);
+}
+
+SelectionModel RenderTextLinux::FirstSelectionModelInsideRun(
+    const PangoItem* run) const {
+  size_t caret = Utf8IndexToUtf16Index(run->offset);
+  size_t cursor = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+  return SelectionModel(cursor, caret, SelectionModel::TRAILING);
+}
+
+SelectionModel RenderTextLinux::LastSelectionModelInsideRun(
+    const PangoItem* run) const {
+  size_t caret = Utf8IndexToUtf16Index(run->offset + run->length);
+  caret = Utf16IndexOfAdjacentGrapheme(caret, PREVIOUS);
+  return SelectionModel(caret, caret, SelectionModel::LEADING);
+}
+
+// Assume caret_pos in |current| is n, 'l' represents leading in
+// caret_placement and 't' represents trailing in caret_placement. Following
+// is the calculation from (caret_pos, caret_placement) in |current| to
+// (selection_end, caret_pos, caret_placement) when moving cursor left by
+// one grapheme (for simplilcity, assume each grapheme is one character).
+// If n is in LTR run,
+// (n, t) ---> (n, n, l).
+// (n, l) ---> (n-1, n-1, l) if n is inside run (not at boundary).
+// (n, l) ---> goto across run case if n is at run boundary.
+// If n is in RTL run,
+// (n, l) --> (n+1, n, t).
+// (n, t) --> (n+2, n+1, t) if n is inside run.
+// (n, t) --> goto across run case if n is at run boundary.
+// If n is at run boundary, get its visually left run,
+// if left run is LTR run,

[msw, 2011/08/23 08:01:01]

Capitalize "If"

[xji, 2011/08/23 23:52:52]

Done.

+// (n, t) --> (left run's end, left run's end, l).
+// If rght run is RTL run,

[msw, 2011/08/23 08:01:01]

Don't you mean left run?

[xji, 2011/08/23 23:52:52]

yes. changed.

+// (n, t) --> (left run's begin + 1, left run's begin, t).
+SelectionModel RenderTextLinux::LeftSelectionModel(
+    const SelectionModel& selection) {
+  size_t caret = selection.caret_pos();
+  SelectionModel::CaretPlacement caret_placement = selection.caret_placement();
+  GSList* run = GetRunContainingPosition(caret);
+  DCHECK(run);
+
+  PangoItem* box = reinterpret_cast<PangoLayoutRun*>(run->data)->item;
+  size_t run_start = Utf8IndexToUtf16Index(box->offset);
+  size_t run_end = Utf8IndexToUtf16Index(box->offset + box->length);
+
+  if (box->analysis.level % 2 == 0) {  // LTR run.
+    if (caret_placement == SelectionModel::TRAILING)
+      return SelectionModel(caret, caret, SelectionModel::LEADING);
+    else if (caret > run_start) {
+      caret = Utf16IndexOfAdjacentGrapheme(caret, PREVIOUS);
+      return SelectionModel(caret, caret, SelectionModel::LEADING);
+    }
+  } else {  // RTL run.
+    if (caret_placement == SelectionModel::LEADING) {
+      size_t cursor = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+      return SelectionModel(cursor, caret, SelectionModel::TRAILING);
+    } else if (selection.selection_end() < run_end) {
+      caret = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+      size_t cursor = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+      return SelectionModel(cursor, caret, SelectionModel::TRAILING);
+    }
+  }
+
+  // The character is at the begin of its run; advance to the previous visual
+  // run.
+  GSList* prev_run = GetPreviousRun(run);
+  if (!prev_run)  // TODO(xji): check whether this is expected.
+    return LeftEndSelectionModel();
+
+  box = reinterpret_cast<PangoLayoutRun*>(prev_run->data)->item;
+  return (box->analysis.level % 2) ? FirstSelectionModelInsideRun(box) :
+                                     LastSelectionModelInsideRun(box);
+}
+
+// Assume caret_pos in |current| is n, 'l' represents leading in
+// caret_placement and 't' represents trailing in caret_placement. Following
+// is the calculation from (caret_pos, caret_placement) in |current| to
+// (selection_end, caret_pos, caret_placement) when moving cursor right by
+// one grapheme (for simplilcity, assume each grapheme is one character).
+// If n is in LTR run,
+// (n, l) ---> (n+1, n, t).
+// (n, t) ---> (n+2, n+1, t) if n is inside run (not at boundary).
+// (n, t) ---> goto across run case if n is at run boundary.
+// If n is in RTL run,
+// (n, t) --> (n, n, l).
+// (n, l) --> (n-1, n-1, l) if n is inside run.
+// (n, l) --> goto across run case if n is at run boundary.
+// If n is at run boundary, get its visually right run,
+// if right run is LTR run,

[msw, 2011/08/23 08:01:01]

Capitalize "If"

[xji, 2011/08/23 23:52:52]

Done.

+// (n, t) --> (right run's begin + 1, right run's begin, t).
+// If rght run is RTL run,

[msw, 2011/08/23 08:01:01]

r*i*ght

[xji, 2011/08/23 23:52:52]

Done.

+// (n, t) --> (right run's end, right run's end, l).
+SelectionModel RenderTextLinux::RightSelectionModel(
+    const SelectionModel& selection) {
+  size_t caret = selection.caret_pos();
+  SelectionModel::CaretPlacement caret_placement = selection.caret_placement();
+  GSList* run = GetRunContainingPosition(caret);
+  DCHECK(run);
+
+  PangoItem* box = reinterpret_cast<PangoLayoutRun*>(run->data)->item;
+  size_t run_start = Utf8IndexToUtf16Index(box->offset);
+  size_t run_end = Utf8IndexToUtf16Index(box->offset + box->length);
+
+  if (box->analysis.level % 2 == 0) {  // LTR run.
+    if (caret_placement == SelectionModel::LEADING) {
+      size_t cursor = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+      return SelectionModel(cursor, caret, SelectionModel::TRAILING);
+    } else if (selection.selection_end() < run_end) {
+      caret = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+      size_t cursor = Utf16IndexOfAdjacentGrapheme(caret, NEXT);
+      return SelectionModel(cursor, caret, SelectionModel::TRAILING);
+    }
+  } else {  // RTL run.
+    if (caret_placement == SelectionModel::TRAILING)
+      return SelectionModel(caret, caret, SelectionModel::LEADING);
+    else if (caret > run_start)
+      return SelectionModel(caret - 1, caret - 1, SelectionModel::LEADING);

[msw, 2011/08/23 08:01:01]

Do you mean to get Utf16IndexOfAdjacentGrapheme(caret, PREVIOUS) here? Also,
here and everywhere else, I'm curious what happens if there is no PREVIOUS or
NEXT grapheme.

[xji, 2011/08/23 23:52:52]

YES! thanks for catching it.

Also,
When there is no previous grapheme, the previous grapheme call returns 0.

when there is no next grapheme, the next grapheme returns the string length.

[msw, 2011/08/24 08:55:30]

Okay, then if the callers are trying to get the previous or next grapheme and
setting trailing/leading as opposed to leading/trailing, you should catch those
errors. For example if you're trying to set the caret position to trail the
cursor's previous grapheme, but the cursor is already at 0, then you're going to
set the caret to (0, trailing) but it really ought to be (0, leading).

We ought to audit all the callers of such functions to ensure that they
implement the correct behavior if the returned 'adjacent' grapheme is the same
as the input value.

[xji, 2011/08/25 03:15:18]

To match the restrictions you added to selection model,
I checked the usage of previous/next grapheme.
Looks like the usages in RenderTextLinux is fine.
I changed RenderText::GetRightSelectionModel(), and changed
RenderText::SelectWord() to use MoveCursorTo() as yours.

+  }
+
+  // The character is at the end of its run; advance to the next visual run.
+  GSList* next_run = run->next;
+  if (!next_run)  // TODO(xji): what is the expected behavior?
+    return RightEndSelectionModel();
+
+  box = reinterpret_cast<PangoLayoutRun*>(next_run->data)->item;
+  return (box->analysis.level % 2) ? LastSelectionModelInsideRun(box) :
+                                     FirstSelectionModelInsideRun(box);
+}
+
+PangoLayout* RenderTextLinux::EnsureLayout() {
+  if (layout_ == NULL) {
+    CanvasSkia canvas(display_rect().width(), display_rect().height(), false);
+    // TODO(xji): can use ScopedPlatformPaint ?
+    cairo_t* cr = skia::BeginPlatformPaint(&canvas);
+
+    layout_ = pango_cairo_create_layout(cr);
+    SetupPangoLayout(
+        layout_,
+        text(),
+        default_style().font,
+        display_rect().width(),
+        base::i18n::GetFirstStrongCharacterDirection(text()),
+        CanvasSkia::DefaultCanvasTextAlignment());
+
+    pango_layout_set_height(layout_, display_rect().height() * PANGO_SCALE);
+    SetupPangoAttributes(layout_);
+
+    skia::EndPlatformPaint(&canvas);
+
+    layout_line_ = pango_layout_get_line_readonly(layout_, 0);
+    pango_layout_line_ref(layout_line_);
+  }
+  return layout_;
+}
+
+void RenderTextLinux::ResetLayout() {
+  // set_cached_bounds_and_offset_valid(false) is done in RenderText for every
+  // operation that trigger ResetLayout().
+  if (layout_) {
+    g_object_unref(layout_);
+    layout_ = NULL;
+  }
+  if (layout_line_) {
+    pango_layout_line_unref(layout_line_);
+    layout_line_ = NULL;
+  }
+}
+
+void RenderTextLinux::SetupPangoAttributes(PangoLayout* layout) {
+  PangoAttrList* attrs = pango_attr_list_new();
+  // Set selection background color.
+  SkColor selection_color =
+      focused() ? kFocusedSelectionColor : kUnfocusedSelectionColor;
+  PangoAttribute* pango_attr = pango_attr_background_new(
+      COLOR_16_TO_32_BIT(SkColorGetR(selection_color)),
+      COLOR_16_TO_32_BIT(SkColorGetG(selection_color)),
+      COLOR_16_TO_32_BIT(SkColorGetB(selection_color)));
+  AppendPangoAttribute(MinOfSelection(), MaxOfSelection(),
+                       pango_attr, attrs);
+
+  StyleRanges ranges_of_style(style_ranges());
+  ApplyCompositionAndSelectionStyles(&ranges_of_style);
+
+  for (StyleRanges::const_iterator i = ranges_of_style.begin();
+       i < ranges_of_style.end(); ++i) {
+    const Font& font = !i->underline ? i->font :
+        i->font.DeriveFont(0, i->font.GetStyle() | Font::UNDERLINED);
+    pango_attr = pango_attr_font_desc_new(
+        reinterpret_cast<PangoFontDescription*>(font.GetNativeFont()));
+    AppendPangoAttribute(i->range.start(), i->range.end(), pango_attr, attrs);
+
+    SkColor foreground = i->foreground;
+    pango_attr = pango_attr_foreground_new(
+        COLOR_16_TO_32_BIT(SkColorGetR(foreground)),
+        COLOR_16_TO_32_BIT(SkColorGetG(foreground)),
+        COLOR_16_TO_32_BIT(SkColorGetB(foreground)));
+    AppendPangoAttribute(i->range.start(), i->range.end(), pango_attr, attrs);
+
+    if (i->strike) {
+      pango_attr = pango_attr_strikethrough_new(true);
+      AppendPangoAttribute(i->range.start(), i->range.end(), pango_attr, attrs);
+    }
+  }
+
+  pango_layout_set_attributes(layout, attrs);
+  pango_attr_list_unref(attrs);
+}
+
+void RenderTextLinux::AppendPangoAttribute(size_t start,
+                                           size_t end,
+                                           PangoAttribute* pango_attr,
+                                           PangoAttrList* attrs) {
+  pango_attr->start_index = Utf16IndexToUtf8Index(start);
+  pango_attr->end_index = Utf16IndexToUtf8Index(end);
+  pango_attr_list_insert(attrs, pango_attr);
+}
+
+GSList* RenderTextLinux::GetPreviousRun(GSList* run) const {
+  GSList* current = layout_line_->runs;
+  GSList* prev = NULL;
+  while (current) {
+    if (current == run)
+      return prev;
+    prev = current;
+    current = current->next;
+  }
+  return NULL;
+}
+
+GSList* RenderTextLinux::GetLastRun() const {
+  GSList* current = layout_line_->runs;
+  while (current && current->next) {
+    current = current->next;
+  }
+  return current;
+}
+
+size_t RenderTextLinux::Utf16IndexToUtf8Index(size_t index) const {
+  int32_t utf8_index = 0;
+  int32_t num_of_substitions = 0;
+  UErrorCode ec = U_ZERO_ERROR;
+  u_strToUTF8WithSub(NULL, 0, &utf8_index, text().data(), index,

[msw, 2011/08/23 08:01:01]

I'm worried by the potential performance losses incurred by calling this
function and its corresponding 8->16 frequently. RenderTextLinux and
RenderTextWin will both need some performance testing and tuning, but these
calls might warrant explicit TODOs, and I'd like to see official review from
someone that knows this code better.

[xji, 2011/08/23 23:52:52]

I have the same concern. I checked with Markus who is very familiar with the
above function, and he said that when destination buffer is NULL and destination
length is 0, the code has a quick path to get the index without doing real
conversion.

Since pango use utf8 internally, we have to do the conversion on
selection_start/end. But I have a TODO at the begin of file saying that to
remove the convresion on caret_pos.

I will ask Markus to take a look.

[markus.icu, 2011/08/25 19:09:03]

Well, I wrote and rewrote these functions :-)
They start with some extra tests that would not be necessary in a function which
never writes any output, but starting with the second input character they fall
into a tight loop that just counts the result length. These functions do need to
read the string up to the input index/length, but there is no way around that
for the index conversion.

+                     0xFFFD, &num_of_substitions, &ec);
+  return utf8_index;

[msw, 2011/08/23 08:01:01]

I think you're supposed to check ec for U_FAILURE (same below).

[xji, 2011/08/23 23:52:52]

From doc
"If the input string is not well-formed, then the U_INVALID_CHAR_FOUND error
code is set.".

I think our string should be well-formed. If it is well-formed, we could use
u_strToUTF8, which exits when found a not well-formed character.

To make it safe, I uses u_strToUTF8WithSub(), which convert the not well-formed
character to substitute character.

The return value means whether the input string contains not well-formed
character or not. I do not think we need to care about that. Or if there is one,
what should we do?

[msw, 2011/08/24 08:55:30]

Sounds like you should add a DCHECK_EQ(ec, U_ZERO_ERROR); or similar.

[xji, 2011/08/25 03:15:18]

Done.

+}
+
+size_t RenderTextLinux::Utf8IndexToUtf16Index(size_t index) const {
+  int32_t utf16_index = 0;
+  int32_t num_of_substitions = 0;
+  UErrorCode ec = U_ZERO_ERROR;
+  const char* layout_text = pango_layout_get_text(layout_);
+  u_strFromUTF8WithSub(NULL, 0, &utf16_index, layout_text, index,
+                     0xFFFD, &num_of_substitions, &ec);
+  return utf16_index;
+}
+
 }  // namespace gfx