In an RTL table, if a table cell (TD) with a width="n" attribute contains a table with width="m", and m>n, the cell does not expand to surround the nested table, but remains in its specified width (n). This is the root cause of the rendering problem in http://www.hondaclub-il.net/, originally reported by arik487 on the mozilla.co.il Firefox forum. Testcase coming up.

The outer (red) table should expand to surround the inner (blue) table. This happens for LTR but not for RTL.

Interestingly, repeated reflows (caused e.g. by changing the text size back and forth), causes the outer table to gradually grow until it reaches the correct width.

There is no need to second guess whats going on here, just run a reflow log... http://www.mozilla.org/newlayout/doc/frame_reflow_debug.html

the inner table gets a negative left margin computed and as consequence the outer table reports a bogus size. tblO 033A7CC8 r=0 a=UC,UC c=UC,UC cnt=149 tbl 033A7DC4 r=0 a=UC,UC c=2445,UC cnt=150 rowG 033A755C r=0 a=UC,UC c=UC,UC cnt=151 row 033A7F84 r=0 a=UC,UC c=UC,UC cnt=152 cell 033A81F4 r=0 a=UC,UC c=UC,UC cnt=153 block 033A8254 r=0 a=UC,UC c=UC,UC cnt=154 text 033A8348 r=0 a=UC,UC c=UC,UC cnt=155 text 033A8348 d=1425,285 me=795 block 033A8254 d=1425,300 me=795 cell 033A81F4 d=1485,360 me=855 row 033A7F84 d=UC,360 rowG 033A755C d=UC,360 colG 033A7744 r=0 a=UC,UC c=UC,UC cnt=156 col 033A53E4 r=0 a=0,0 c=-30,UC cnt=157 col 033A53E4 d=0,0 colG 033A7744 d=0,0 tbl 033A7DC4 d=2475,450 me=2475 tblO 033A7CC8 d=45,450 me=45 o=(-2430,0) 2475 x 450

Thanks, Brend. The bugus left margin originates from http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/layout/generic/nsHTMLReflowState.cpp&rev=1.233#2137 This code (availMarginSpace < 0) was originally introduced for bug 6155. The table case specifically was discussed much in bug 185411 comment #16 and onwards. Removing the |if (mStyleVisibility->mDirection == NS_STYLE_DIRECTION_RTL)| case fixes the symptoms of this bug, but causes RTL tables to overflow to the right when the viewport is too narrow to contain them. I'm not sure if that's really wrong (Safari does the same), and in any event we have more "content overflows to the right when it should overflow to the left in RTL blocks" issues (see bug 318116) - so maybe the fix should really be somewhere else. Perhaps someone who really understands the code can comment?

>who really understands the code can comment This should exclude me from commenting, but what I know for sure is that the table margin handling is a blatant hack (see bug 260980 comment 9). I have to admit that I took part in this.

This is an LTR-only testcase, demonstrating a related issue when a left or right margin is set on an inner table inside a cell with a specified small width. cases 2 and 4 should look like 1 and 3 respectively, and a single forced reflow (using ctrl-+) indeed causes them to look that way. See also bug 164457, which is similar (but not identical, as far as I can tell).

Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9a1) Gecko/20061208 Minefield/3.0a1 ID:2006120812 [cairo] seems fixed after reflow branch landing

Indeed, fixed by the reflow branch landing. Verified with: Mozilla/5.0 (Macintosh; U; PPC Mac OS X Mach-O; en-US; rv:1.9a1) Gecko/20061208 Minefield/3.0a1 ID:2006120806

Need regression test.

Mass-assigning the new rtl keyword to RTL-related (see bug 349193).