Merge lp://qastaging/~vanvugt/compiz/fix-1046661 into lp://qastaging/compiz/0.9.8

PASSED: Continuous integration, rev:3348
http://jenkins.qa.ubuntu.com/job/compiz-ci/32/
Executed test runs:
    SUCCESS: http://jenkins.qa.ubuntu.com/job/compiz-ci/./build=pbuilder,distribution=quantal,flavor=amd64/32/console

!(flags & (Desktop | Alpha)

That can be expressed as a temporary rather than in a compound if statement

const bool ignoreWindow = (flags & (Desktop | Alpha)

119 + /*
120 + * Alpha windows (status == false) can cover/cancel fullscreen
121 + * unredirected windows. But they can't be unredirected themselves.
122 + * I wonder if this is sensible as some games may use fullscreen
123 + * RGBA (alpha) windows that are fully opaque and need unredirect.
124 + */

This comment is confusing. We're checking whether or not the window has an rgba visual (eg, w->alpha () not whether or not it failed to paint (as status == false) would indicate to me.

Also, the default visual for most windows would be an RGB24 visual, so I don't think the concern is too much. (And we wouldn't be able to detect it anyways, as an RGBA visual just means a client is at liberty to have alpha values in its backing pixmap, which we must blend).

I'm also a bit confused as to why it matters that the window has an alpha channel as to whether or not a fullscreen window underneath it is unredirected. Shouldn't the fullscreen window underneath always be unredirected if there is some other window that is occluding it?

> Shouldn't the fullscreen window underneath always be
> unredirected if there is some other window that is occluding it?

Correction: That should read

"Shouldn't the fullscreen window underneath always be redirected if there is some other window that is occluding it"

!(flags & (Desktop | Alpha) is clearer, I think. Programmers think in logic, not words. We don't need to be translating clear logic into extraneous English. It complicates the thought process. Also, "ignoreWindow" is not accurate because it's not the only condition for ignoring a window.

Yeah, the comment is slightly out-dated still. It belongs next to some code that's not there any more. But technically it's still true. status is false for alpha windows (and other cases too).

Alpha "matters" for FullscreenRegion, only because we don't want to accidentally unredirect a translucent fullscreen window. You are right that alpha does not matter for the non-fullscreen windows.

Okay, I'm good with it now after discussing.

Noted that this proposal doesn't deal with the issue of "not counting" windows that are fully-transparent or not painted at all on top of fullscreen windows. This case should be dealt with later.