User Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:15.0) Gecko/15.0 Firefox/15.0a1 Build ID: 20120507171005 Steps to reproduce: 1. Loaded http://www.reddit.com/r/Android/comments/ta97g/what_dont_you_like_about_android/ via a hyperlink on reddit.com/r/Android 2. Scrolled about half way down the page Actual results: The page's background was replaced with a solid black color making everything unreadable. Expected results: The page's background stays the same

This happens on my Samsung Galaxy S2 Skyrocket which has an Adreno 220 GPU

I was able to reproduce this issue using the latest Nightly (2012-05-08) on Samsung Galaxy SII (2.3.4) and HTC Desire (2.2)

Reproduced, and quite unpleasant. -- Tested via: Nightly (05/08) * Galaxy Nexus (Android 4.0.4), Sony Ericsson Xperia Play (Android 2.3.4)

Unfortunately this has been a long-standing issue for a while now :(

I guess something made it easier to reproduce now; I haven't seen this before at all.

It should only happen on really long pages with a background, I think. If you feel like there is a regression here that makes this more apparent or obvious then we should bisect to find out when it happened.

This issue doesn't occur on: Maple 13.0a1 (2012-03-12) http://hg.mozilla.org/projects/maple/rev/e88141f7f64c but it occurs on: Maple 13.0a1 (2012-03-13) http://hg.mozilla.org/projects/maple/rev/16e04eaadc14 Possible range: http://hg.mozilla.org/projects/maple/pushloghtml?fromchange=e88141f7f64c&tochange=16e04eaadc14

Let's see if this is something we can make better (not use a black background), even if we can't fix the real bug.

Well, it doesn't make much sense, but... The first bad revision is: changeset: 89088:a5980ce21d40 user: Kartikaya Gupta <kgupta@mozilla.com> date: Mon Mar 12 12:03:38 2012 -0400 summary: Bug 732564 - Distinguish between changes where we should sync-update the Java viewport and where the compositor has to update Java. r=Cwiiis

qawanted for two reasons: one is for a reduced testcase; alternately, a different example of this on the internet, as I can't reproduce it reliably on Reddit any more :(

I can, however, reliably reproduce this on Aurora, strangely enough.

Ah, but I can reproduce this on Nightly on http://nomnompaleo.com

testcase wanted, but we need to reliable reproduce this first.

Tap on the link (or just scroll down a while) to get the black background. The background that is used in the testcase is pointing to http://static.tumblr.com/fftf9xi/27ll9xup0/texture_main.jpg which is a rather large image.

pixman's analyze_extent is returning FALSE here, which means that we're just trying to draw beyond the range pixman supports. This will need a fix in layout. The coordinates need to not be too big. I can reproduce very similar problems on GDI (i.e., bug 671302) due to the same underlying limitation in pixman.

I'm not sure how we'd fix this in Layout as there isn't a single bottleneck to check for extents prior to a paint. I think our best bet here is to implement comment 8 and paint the default HTML background color and not black when we overflow the extents.

I'll try something.

Comment on attachment 627348 [details] [diff] [review] This seems to do the trick Review of attachment 627348 [details] [diff] [review]: ----------------------------------------------------------------- ::: gfx/thebes/gfxUtils.cpp @@ +513,5 @@ > > + gfxMatrix m = aUserSpaceToImageSpace; > + if (doTile) { > + m.y0 = fmod(m.y0, aImageRect.height); > + m.x0 = fmod(m.x0, aImageRect.height); Should this one be aImageRect.width?

Comment on attachment 627605 [details] [diff] [review] Move the translation within pixman's range Review of attachment 627605 [details] [diff] [review]: ----------------------------------------------------------------- ::: gfx/thebes/gfxUtils.cpp @@ +518,1 @@ > #endif Use #ifdef MOZ_OPTIMIZE_MOBILE ... #else ... #endif instead of #ifndef

Comment on attachment 627605 [details] [diff] [review] Move the translation within pixman's range [Approval Request Comment] Bug caused by (feature/regressing bug #): pixman coordinate limits User impact if declined: black backgrounds instead of images on some pages Testing completed (on m-c, etc.): been on m-c for quite a while Risk to taking this patch (and alternatives if risky): mobile only, perhaps some image drawing wrongness. String or UUID changes made by this patch: None

Comment on attachment 627605 [details] [diff] [review] Move the translation within pixman's range This is already on Aurora. Please land on beta ASAP.