Context: test failures on https://codereview.chromium.org/2927953003/
A views_unittest fails on some versions of macOS due to a change in the font that RenderText is using.
The test, LabelSelectionTest.MouseDragSingleLineRT, renders בגדהוזא, then selects it with the mouse. The drag start point is chosen using GetSubstringBounds() but RenderText() decides whether to select to the left or the right of the *cursor* position, using GetCursorBounds().
Both are doing the right thing.
GetSubstringBounds() rounds using ceil(). This ensures the entire glyph is highlighted when selecting.
GetCursorBounds() rounds to nearest. This gives the cursor the best chance of fitting nicely "between" the glyphs.
RenderText::MoveCursor() is also doing the right thing by using GetCursorBounds() as the decider on which direction to go when selecting with the mouse.
What's "wrong" is the test using GetSubstringBounds() in its GetCursorPoint() helper method. Unfortunately, it can't (always) use GetCursorBounds, since it doesn't support multiline text. (you can't currently edit multiline, so there's no point drawing the insertion cursor).
I think to make the test happy, we can use GetCursorBounds() to select the start drag point when we know the text is multiline, and fall back to GetSubstringBounds otherwise.
Alternatively, we can force the test to use 'Lucida Grande' on Mac, but that feels more like it could bite us in future.
Context: test failures on https://codereview.chromium.org/2927953003/
A views_unittest fails on some versions of macOS due to a change in the font that RenderText is using.
The test, LabelSelectionTest.MouseDragSingleLineRT, renders בגדהוזא, then selects it with the mouse. The drag start point is chosen using GetSubstringBounds() but RenderText() decides whether to select to the left or the right of the *cursor* position, using GetCursorBounds().
Both are doing the right thing.
GetSubstringBounds() rounds using ceil(). This ensures the entire glyph is highlighted when selecting.
GetCursorBounds() rounds to nearest. This gives the cursor the best chance of fitting nicely "between" the glyphs.
RenderText::MoveCursor() is also doing the right thing by using GetCursorBounds() as the decider on which direction to go when selecting with the mouse.
What's "wrong" is the test using GetSubstringBounds() in its GetCursorPoint() helper method. Unfortunately, it can't (always) use GetCursorBounds, since it doesn't support multiline text. (you can't currently edit multiline, so there's no point drawing the insertion cursor).
I think to make the test happy, we can use GetCursorBounds() to select the start drag point when we know the text is single-line*, and fall back to GetSubstringBounds otherwise.
Alternatively, we can force the test to use 'Lucida Grande' on Mac, but that feels more like it could bite us in future.
*(Edit: multiline -> single-line)
Closing this out - nothing to do, but code references this issue.
Full multiline editing support in RenderTextHarfbuzz may make it less interesting - that's Issue 248597
Comment 1 by tapted@chromium.org
, Jun 21 2017