Bug 15238 – Odd horizontal scrolling amount corrupts gif rendering in JavaScript context

• Status: UNCONFIRMED
• Component: WebCore Misc.

[Norman Hardy]

I have seen this only with ephemeral ads which seldom last long enough to last until report is read.
See two files at (http://cap-lore.com/bug/b2/b1.tiff) and (http://cap-lore.com/bug/b2/b2.tiff)
They are made by the grab Utility.
One can get from either to the other buy scrolling the image vertically out of the window, scrolling an odd number of pixels horizontally, and then scrolling the image back in vertically.

I think that the following html is what caused the image:
It came from the page at <http://www.nytimes.com/> at about 23:00 on 2007, Sept 18, PDT.

<div class="columnGroup" id="HPMiddle">
<!-- ADXINFO classification="bigad" campaign="Amex-STS-659605-nyt1"--><SCRIPT language='JavaScript1.1' SRC="http://ad.doubleclick.net/adj/N553.newyorktimes.com/B2348401.13;sz=300x250;ord=2007.09.19.05.42.55?">
</SCRIPT>
<NOSCRIPT>
<A HREF="http://ad.doubleclick.net/jump/N553.newyorktimes.com/B2348401.13;sz=300x250;ord=2007.09.19.05.42.55?">
<IMG SRC="http://ad.doubleclick.net/ad/N553.newyorktimes.com/B2348401.13;sz=300x250;ord=2007.09.19.05.42.55?" BORDER=0 WIDTH=300 HEIGHT=250 ALT="Click Here"></A>
</NOSCRIPT>
</div>

10 minutes later the same page shows a different ad in the same position.
The same fault happens with that ad too.

A similar but different bug is in Firefox.
I am running my screen with 16 bit pixels.
(about 5 in 10 tries are bad)
With 32 bit pixels the problem goes away (10 in 10 tries are good).
The same thing with Firefox.

I quit the browser, downloaded the nightly build of Web kit, and then went back to the same page; the problem persists.
I rebooted the Mac and relaunched Safari; the problem persists.
I launch Firefox 2.0.0.6; a similar but different problem occurs.
I cannot make it fail with Opera 9.20 build 3669 or MS Explorer 5.2.

In Safari scrolling horizontally, while keeping the pixels in the window, neither causes the corruption nor repairs it.
In Firefox, however, such horizontal scrolling causes the problem (50% of the time) or fixes it (50% of the time).
I think I have confirmed that it depends on whether scrolling amount, in pixels, is even or odd.
It is as if the pixel copying due to scrolling were at a different level of rendering.
The bug may not be shared between Firefox and Safari.

I think I can scroll horizontally one pixel at a time.
If I am right then I deduce that odd horizontal scroll positions have the problem and even ones do not, or vice versa.
It looks like a byte ordering problem, or really an ordering of 16 bit shorts.
Perhaps pixels are copied 32 bits at a time between areas with different ordering conventions.

[Gavin Barraclough]

This doesn't look like a JSC bug.