The tables seem to overlap. Build: 1999051808

I've added a reduced case (attachment). Please try to fix the table bug, and if there's also a layout bug, pass it to kipp.

Troy, please take a look at this.

This is triggered by the following sequence: * block level element with bottom margin nonzero (anything nonzero) * table, any size, floated left (including align=left) * table, any size attaching simpler test case. I imagine this has something to do with how you make tables an exception from the block-level flow-around-floats rules. See, perhaps, my proposal for float-displace, in which I think tables would have the value block (I think that's legacy behavior): http://lists.w3.org/Archives/Public/www-style/1998Dec/0082.html One question is how to handle this in standard mode...I would say do it the legacy way, since the spec doesn't really address this very well.

Kipp, this is a pretty bug change and with the table work I won't get to it for a while

*** Bug 12626 has been marked as a duplicate of this bug. ***

*** Bug 12626 has been marked as a duplicate of this bug. ***

http://bugzilla.mozilla.org/showattachment.cgi?attach_id=1416 is a slight variant from bug 12626 involving a form element.

*** Bug 15134 has been marked as a duplicate of this bug. ***

Dmose says that this bug makes calendar.yahoo.com utterly useless. He argues it should be fixed for dogfood. /be

Note for me: the bug has to do with the fact that the available space computed for the floater is computed at the wrong location - the block reflow context is computing the top margin and after that's computed we need to get the available space so that we know what the proper floater impacts are. To fix this I need to reconfigure block reflow context to break out the top margin computation in advance of the reflow. However, to do that is a royal pain because the top margin calculations depend on the containing block width (maybe - if they are percentages) and all of that logic is rolled up into nsHTMLReflowState's ctor and would require alot of work to fix. urk.

Fixed. I updated the block reflow logic to pre-compute the top-margin and use that to properly determine the available reflow space.

In the Oct 21 build (1999102109), this problem is fixed.