856452 - mysteriously fragile PNG reftests

Status: NEW

(Core :: Graphics: ImageLib, defect)

Description

[Jonathan Kew [:jfkthame]]

Attachment: patch that triggers mysterious PNG test failures

The attached patch just adds a (pointless) call to gfxWindowsPlatform::GetPlatform() within the windows widget code (nsWindow::GetDefaultScaleInternal). I don't see why this should have any effect at all on behavior - but it causes two PNG reftests to consistently fail, across all four Windows test configurations on tryserver:

  https://tbpl.mozilla.org/?tree=Try&rev=068acae8c6c5

12:11:48 INFO - REFTEST TEST-UNEXPECTED-FAIL | file:///C:/talos-slave/test/build/tests/reftest/tests/image/test/reftest/pngsuite-ancillary/ccwn2c08.png | image comparison (==), max difference: 8, number of differing pixels: 569
12:11:49 INFO - REFTEST TEST-UNEXPECTED-FAIL | file:///C:/talos-slave/test/build/tests/reftest/tests/image/test/reftest/pngsuite-ancillary/ccwn3p08.png | image comparison (==), max difference: 8, number of differing pixels: 577 

The failures seem to be completely consistent, but I see no connection to the patch that's triggering them.

This turned up while testing the "part 2" patch in bug 818927, which uses a method from gfxWindowsPlatform to replace the body of GetDefaultScaleInternal. In that bug, I'm proposing to mark these two tests as fuzzy on windows, but it worries me that this patch, which should be a no-op, somehow perturbs them.