Note: this bug sounds sorta similar to bug 333822, but the information available there is so inconsistent and unclear I could not be sure if this is the same issue or not, so I'm filing this a separate bug. This is a relatively recent trunk regression, but I don't have a date (or time to find the regression range), so help with that would be appreciated. The reflow branch landing is a natural suspect. When a table cell with COLSPAN set also has a width specified, the table will expand all the columns in the span to the correct total width. This still works. However, if the table has a width set that stretches the table (the usual being WIDTH=100%), the width on the COLSPAN cell is not honored and it is stretched wider. I have a page showing this with two tables here: http://twpol.dyndns.org:82/temp/silver/html/Table%20COLSPAN%20bug.html And a testcase version here (to be attached): http://twpol.dyndns.org:82/temp/silver/html/Table%20COLSPAN%20bug%20(testcase).html As noted on the first page, this bug does not exist anywhere prior to current trunk builds.

this is regression from the reflow branch landing which heavily tweaked the layout strategy

The patch in bug 363874 fixes this.

Fixed by checkin of bug 363874.