User Agent: Mozilla/5.0 (Windows NT 6.1; Win64; x64; rv:18.0) Gecko/20130119 Firefox/18.0 Build ID: 20130119203203 Steps to reproduce: On this page: http://larvoltrials.com/ohm.html There are headers inside DIVs with position:relative that have their top property changed via javascript to follow the viewport on one axis only. Actual results: In Firefox 21 (and still in Nightly 24.0a1), while scrolling vertically, these headers follow the viewport as expected, until the viewport is scrolled enough so that the original (pre-positioning) area is completely outside the viewport, then they disappear. The Inspector tool still shows their boundaries as following the viewport, they just aren't visible. Expected results: The DIVs should continue to be visible regardless of vertical scroll position. (Worked in Firefox 18 and chrome/IE9)

The similar problem happens in Firefox4...17. So it is difficult to determine regression window. Fixed window(m-aurora) Bad: http://hg.mozilla.org/releases/mozilla-aurora/rev/fa9a4d400949 Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/18.0 Firefox/18.0 ID:20121010042011 Fixed: http://hg.mozilla.org/releases/mozilla-aurora/rev/e3f7fdf150a4 Mozilla/5.0 (Windows NT 6.1; WOW64; rv:18.0) Gecko/18.0 Firefox/18.0 ID:20121011042010 Fized pushlog: http://hg.mozilla.org/releases/mozilla-aurora/pushloghtml?fromchange=fa9a4d400949&tochange=e3f7fdf150a4 Fixed by cf470cfd18ca Ehsan Akhgari — Backout bug 157681 in order to fix bug 775350 on Aurora; a=lsblakk Unfortunately, The patch of bug 775350 does not fix this problem in Firfox19 and later.

Correct >>The similar problem happens in Firefox4...17. The similar problem happens in Firefox4...17 (when they were still in m-c).

I get a different regression range: Last good nightly: 2012-06-06 First bad nightly: 2012-06-07 Pushlog: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=6338a8988917&tochange=7e4c2abb9fc9 It appears to be caused by the number of rows in the table. If there are <= 20 rows, it works. If there are more than 20 rows, the bug happens.

The red div shouldn't disappear when the page is scrolled. If you remove one of the empty rows, it works.

(In reply to mjh563 from comment #3) > I get a different regression range: > > Last good nightly: 2012-06-06 > First bad nightly: 2012-06-07 > > Pushlog: > http://hg.mozilla.org/mozilla-central/ > pushloghtml?fromchange=6338a8988917&tochange=7e4c2abb9fc9 > > It appears to be caused by the number of rows in the table. If there are <= > 20 rows, it works. If there are more than 20 rows, the bug happens. Confirmed, it has regressed since FF16.

The comment 3 range would be bug 157681. The comment 1 range was for the problem being _fixed_ between 17 and 18, which was done by backing out bug 157681.

It looks to me like we're failing to update an overflow area somewhere, probably table-related since this seems to depend on tables.

Ah, this bug is because the table row cursor isn't being invalidated when UpdateOverflow happens.

Interesting that no equivalent bug was reported for blocks.

Comment on attachment 750279 [details] [diff] [review] fix >+ virtual bool UpdateOverflow(); MOZ_OVERRIDE, please. r=me