The last example in the above URL is only displaced 50px, but it should be 100px. I can't tell if this also happens for bottom and top because the viewer has started crashing. It doesn't happen for left. It feels like it should be my mistake, but it could also have something to do with the width of the div (i.e. try making div.ten have width 450px and see if it still happens - that was what I was doing when the viewer crashed (and now it crashes when I try to start it)).

It looks like my problem. 'right' is inheriting properly, but when I determine the computed value for 'margin-left' I have a check for whether the available margin space (per the constraint in 10.3.7) is < 0 and if it is I leave the computed value as 0. In your example both the P and its DIV containing block have specified 'width' values of 400 That seems wrong and what I should do instead is allow the value to be negative, after all the spec allows negative margins In the case where both 'margin-left' and 'margin-right' are 'auto' and the available space per the constraint in 10.3.7 is negative should we also compute a negative value for 'margin-left' in that case? The idea in that case is to center the element within its parent, so I guess it should extend an equal amount outside both the parent's left and right edges

Yes. auto values on things that can be negative (i.e., not width) should always be allowed to become negative from the calculations (not just for margin-right). Could this explain bug 2058 too??? (I think not, but I'm just checking...)

Okay, I just checked in my fix and now I allow the margins to be negative. Things won't look exactly right until the view manager changes happen so you'll see content getting clipped when a child sticks outside its parent

Verified fixed.