Open Bug 885348 Opened 11 years ago Updated 2 years ago

Downscaled favicon re-scaled on mouse-out

Categories

(Core :: Graphics: Layers, defect)

x86
macOS
defect

Tracking

()

People

(Reporter: johnath, Unassigned)

References

Details

STR:

1) Open an image in a tab such that you get a downscaled favicon (e.g. http://i.imgur.com/E07UH4d.jpg )
2) Switch to another tab
3) Mouse over the image tab - the tab decoration highlights on mouseover, the favicon doesn't change, as expected
4) Mouse off the image tab

Expected: tab decoration highlight disappears, favicon doesn't change
Actual: favicon is re-downscaled blocky, and then reverts to its smoother downscaling again

Cc'ng Matt Woodrow since Jeff/Joe claim he was investigating this. This appears to be layers-related, and J/J originally suspected it has been around forever (at least since ca. FF3). Turns out no:

Last good nightly: 2012-09-28
First bad nightly: 2012-09-29

Pushlog:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=895f66c4eada&tochange=c09a0c022b2e
i herd u liek dlbi
Flags: needinfo?(matt.woodrow)
Blocks: dlbi
I wonder whether Australis styling changes might affect how/whether this bug occurs.
I'm able to reproduce this issue on a UX build on OSX, so I can say with 100% confidence that the Australis styling changes do not fix the problem.
(In reply to Joe Drew (:JOEDREW! \o/) from comment #1)
> i herd u liek dlbi

False.

This isn't really DLBI as such, it's a regression from bug 781053.

The problem is that the mouseover causes the opacity on the thumbnail image to change opacity, and we switch it to being an 'active' layer.

This moves the thumbnail image to being an ImageLayer, where the downscaling is done on the GPU. This is visibly lower quality than what Quartz does.

After half a second, we transition back to being inactive.

I'm not sure what we want to do here. Improving the scaling quality would be nice.

We could also change the heuristics so that this doesn't trigger active layers, or change the css so that the image isn't the only thing within the opacity.
Flags: needinfo?(matt.woodrow)
(In reply to Matt Woodrow (:mattwoodrow) from comment #4)
> (In reply to Joe Drew (:JOEDREW! \o/) from comment #1)
> > i herd u liek dlbi
> 
> False.
> 
> This isn't really DLBI as such, it's a regression from bug 781053.
> 
> The problem is that the mouseover causes the opacity on the thumbnail image
> to change opacity, and we switch it to being an 'active' layer.
> 
> This moves the thumbnail image to being an ImageLayer, where the downscaling
> is done on the GPU. This is visibly lower quality than what Quartz does.
> 
> After half a second, we transition back to being inactive.

What about bug 781053 caused the behaviour change? Was the layer not becoming active before?

> 
> I'm not sure what we want to do here. Improving the scaling quality would be
> nice.
> 
> We could also change the heuristics so that this doesn't trigger active
> layers, or change the css so that the image isn't the only thing within the
> opacity.

I think ideally we'd upload the downscaled version. This already cached somewhere, so it's sort of silly for us to upload the entire image only to downscale it on the gpu.
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.