Bug 52987 – REGRESSION (r73369-r73405): Overflowing columns not repainted correctly

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Joseph Pearson]

The first nightly this bug appeared in was r73429. r73340 exhibits the expected behaviour.

The Monocle ebook reading library utilises the following technique:

* All the content is in a fixed width/height iframe.
* In response to UI events, the document body inside the frame is translated horizontally.

You can see it in effect here: http://things-we-didnt-c2f954.reading.booki.sh. Turning past the first page in recent nightlies shows blank pages. In Safari 5.0.3 it behaves as expected.

The same bug exhibits if the body's scrollLeft is changed rather than its X translation.

My suspicion is that it is caused by the patch to this bug: https://bugs.webkit.org/show_bug.cgi?id=50593

I'm not familiar with WebKit development, but I'm happy to refine this bug report and do further investigation if required — the effect of this bug on us is huge!

[Alexey Proskuryakov]

Confirmed with r76456 on Mac OS X.

[Alexey Proskuryakov]

<rdar://problem/8905571>

[Simon Fraser (smfr)]

Can you create a simple testcase that shows the problem?

[Joseph Pearson]

Thanks for the prompt response, guys. Here's a test case that works as expected in Saf 5.0.3, broken in latest nightly:

http://pockets.inventivelabs.com.au/webkit-out-of-bounds/index.html

Looks like this was caused by <http://trac.webkit.org/changeset/73385>.

[Joseph Pearson]

I've updated the title of this bug, in line with what the testcase demonstrates — this is a more fundamental rendering problem than just Monocle compat. An element with columns and a height will disappear if translated or even if the document is scrolled.

This is all you need to see the problem:

body {
  height: 300px;
  -webkit-column-width: 200px;
}

It doesn't need to be in an iframe, it doesn't need to be translated. Just add a bunch of lorem ipsum and attempt to scroll horizontally.

Since this is now in Chrome betas, and a public release can therefore be expected any day, I wonder if the priority of this should be upped and whether it's worth assigning it to the presumably quite busy Dave Hyatt (who authored Changeset 73385)? A number of deployed applications depend on this, and I haven't yet found a workaround.

[Joseph Pearson]

Created attachment 83902 [details]
Testcase demonstrating disappearing columns when scrolling horizontally

This is a testcase for the problem as described in my comment #6. When scrolling horizontally, additional columns should be visible all the way to the end of the scroll region.

[Dave Hyatt]

Created attachment 85527 [details]
Patch

Found the issue.

[Dave Hyatt]

Fixed in r80883.