Go to http://www.linuxgames.com/ -- upon fully loading, the main news table is pushed leftwards to overlap the links table on the left side of the page (I enclose a screenshot of the problem). Interestingly, adjusting the browser window size in any direction, ie. causing another document reflow, fixes the problem. The page is affected regardless of whether it is coming over the network or straight out of cache. This problem only started to occur some time in the last 14 hours (n.b. we've had nsViewManager2 removal and dbaron's big bug 86947 checkin in this period, as well as a few less-likely culprits).

mine

*** Bug 106810 has been marked as a duplicate of this bug. ***

the bug is visible in trunk 2001-102421 linux

Resizing after the bad layout brings it back to what it should have been.

Well, I'm a little puzzled by this bug. The problems seems to be that nsBlockReflowState::RecoverStateFrom doesn't descend into child blocks to fill in the space manager state with floaters whose placeholders (I think that's the criterion for floaters being "in" a line) are in lines *within* child blocks. What I don't understand is how this ever worked before. Did it work because we always reflowed enough that this wasn't a problem, or did it work because the space manager got filled in some other way?

/me shrugs. As a quick fix, during state recovery you could _always_ desecend into each line that has a block frame (which doesn't have a space manager itself). Longer term, this would be fixed by persisting the space manager's state across reflows, right?

confirming on win2k using build 2001102708

*** Bug 107592 has been marked as a duplicate of this bug. ***

*** Bug 107968 has been marked as a duplicate of this bug. ***

*** Bug 109168 has been marked as a duplicate of this bug. ***

*** Bug 109850 has been marked as a duplicate of this bug. ***

I still need to test the performance effects of this patch, both on typical pages (jrgm's tests) and on long pages.

Performance effects seem minimal. No significant change on jrgm's tests. Added about 10-15 seconds to a testcase with 100,000 paragraphs (from 4:00 to 4:10 or so).

Comment on attachment 57740 [details] [diff] [review] patch r=kin@netscape.com

Comment on attachment 57740 [details] [diff] [review] patch sr=waterson

Checked in to trunk 2001-11-14 23:28/30 PDT. Awaiting 0.9.6 branch approval.

Comment on attachment 57740 [details] [diff] [review] patch a=blizzard on behalf of drivers for 0.9.6

Checked in to 0.9.6 2001-11-15 15:56 PDT.

I think the reason my patch in bug 86947 caused this was because it partially disabled the working branch of the duplicated code described in bug 117374 that I'm now planning to remove.

Verified fixed.