36811 - CSS background does not tile properly

Status: VERIFIED FIXED

(Core Graveyard :: GFX, defect, P3)

Description

[tarnar]

From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux 2.2.14 i686; en-US; m16) Gecko/20000421
BuildID:    2000042113

The page is supposed to show a tiled image with a semi-transparent image above
it.  Mozilla renders the tiled image once, but doesn't actually tile it out.

Also, if you scroll down so the tiled pace is out of sight, then back, it
renders correctly.

Page again:  http://www.w3.org/Graphics/PNG/inline-alpha.html


Reproducible: Always
Steps to Reproduce:
1.Go to the page in question
2.Observe the non-tiled image
3.Scroll the page down, then back up so the grey area goes out of sight, then
back into sight

Comment 1

[Richard Zach]

Confirmed on Linux build 2000.04.21.13.  Background image tiles erratically,
expose event fixes it. -> Rendering

Comment 2

[Kevin McCluskey (gone)]

Don, I think this is a dup of one your existing tile bugs.

Comment 3

[dcone (gone)]

This is because for the initial update event.. the image is not loaded.  The 
paint system thinks the background is validated..but it has not been loaded yet.  
When the event comes through that the image is loaded.. only the rect of the 
image is invalidated.. and then painted.  If you invalidate the entire window 
after that..everything will be draw.  Mark.. do you have a bug to dup this to.. 
do we know who will fix this...

Comment 4

[Marc Attinasi]

I have not entered this bug yet, so we might as well use this one to track it.

I'm not sure either whot this goes to... It is probably a CSS problem because it 
is in CSSRendering, however the rendering in there is definitely shared. I 
suggest we leave it in my queue and fix it post-haste - this is a nasty problem 
for a large number of pages.

Marking M16 beta+

Comment 5

[Marc Attinasi]

Two problems here: 

1) the nsFrameImageLoader has a method (Notify) that takes care of invalidating 
the correct area when an asynch image load completes. The problem is that the 
method only invalidates the size of the image (if the size is known), so when 
the image is *tiled* the invalidated region is only the size of a single tile.
2) nsCSSRendering::PaintBackground is passing the frame being rendered to the 
image loader. This is not sufficient when the frame being rendered is the BODY 
or HTML frame, because we need to invalidate the CANVAS frame in those cases.

To fix problem 1 we change the nsFrameImageLoader::DamageRepairFrames to 
invalidate the frame's bounds, not just the image size. For problem 2 we need to 
work harder: the frame passed to the image loader needs to be swapped to the 
canvas frame when the frame being rendered is the HTML or BODY frame - this we 
can lookup by tags.

I have straight-forward fixes for these now, and it works fine. The proper 
solution is more involved still and involves an interface change on nsIFrame. 
Basically, the method nsFrameImageLoader::DamageRepairFrames should call into 
the frame and let it invalidate itself as it sees fit. It should pass the size 
of the image if it knows it, so in many cases that is all that gets invalidated, 
however if the frame wants to do something different it can encapsulate that 
behavior. This approach will come post-release; the code will be so documented.

Comment 6

[Marc Attinasi]

Fix checked in. 

Files Changed: nsCSSRendering.cpp, nsFrameImageLoader.cpp

Comment 7

[Chris Petersen]

Fixed in the July 14th build.