983511 - With some css:device pixel ratios, interpolated style snaps to device pixels for some properties

Status: NEW

(Core :: CSS Parsing and Computation, defect)

Description

[Brian Birtles (:birtles)]

In the following test:

  http://dxr.mozilla.org/mozilla-central/source/layout/style/test/test_transitions_per_property.html

running on a machine whose resolution is not 96dpi (?) I get test failures for a handful of properties including -moz-column-rule-width and border-bottom-width.

On further inspection I notice it seems to correspond to properties that are represented as nscoord inside an nsStyleAnimation::Value.

Instead of getting 6px for the 1/4 value in test_length_transition I get 5.6px.

After a bit of digging I notice we calculate the interpolated value correctly as 360 app units (/ 60 = 6px), but when we go to store it in nsStyleColumn we round it to device pixels:

  mColumnRuleWidth = NS_ROUND_BORDER_TO_PIXELS(aWidth, mTwipsPerPixel);

which gives us 336 app units which produces 5.6px when calling getComputedStyle.

Comment 1

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

It's quite intentional that we're rounding to device pixels.  We want to ensure that after snapping *edges* to the nearest device pixel, we end up with a border of a given width having the same displayed width on all sides.

That said, it's not necessarily required that this snapping happen as part of what we consider to be the computed style, or even as part of style data computation.  But it (or at least the latter) does make things quite a bit easier.