385298 - Figure out why the checkin for Bug 382092 fixed a bunch of Mac rounding errors

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Eli Friedman]

The checkin for bug 382092 caused some unexpected passes on mac by fixing some rounding errors; my best guess is that it is due to using gfxASurface::CONTENT_COLOR as a parameter for aThebesContext->PushGroup instead of gfxASurface::CONTENT_COLOR_ALPHA.  Whatever it is, it really ought to get figured out and fixed, because it could affect other users of Thebes.

(Bugs affected: bug 382039, bug 382399, bug 382040, bug 381831, bug 381834, bug 383386.)

Comment 1

[Eli Friedman]

Oh, and I guess I should mention that I marked all the relevant tests as passing to fix the resulting orange.

Comment 2

[Jesse Ruderman]

Bug 382399 sounded like a real bug (off by one color), but the rest only appeared in reftests.  Are they real bugs?  Is this a real bug?

Comment 3

[Eli Friedman]

I think there's a real bug somewhere in all this; I'm not sure where it is, though.

It would be nice if someone with a mac could try changing nsPresShell::RenderDocument to unconditionally pass gfxASurface::CONTENT_COLOR_ALPHA to PushGroup and see if it affects reftests.  If it does, then we've found the bug.

Comment 4

[Joe Drew (not getting mail)]

I suspect that we're not going to be too concerned with this going forward, due to us moving away from Thebes, but this code does indeed still exist in nsPresShell::RenderDocument.