Loading the testcase gives me: ###!!! ASSERTION: Bogus availSize.width. Should be bigger: 'availSize.width >= mLegendRect.width', file /Users/jruderman/trunk/mozilla/layout/forms/nsFieldSetFrame.cpp, line 533

Yeah, the fieldset code I wrote on reflow branch really doesn't handle percentage stuff all that well. And this particular assertion is probably just wrong....

This removes the unenforceable assertion, as well as a lot of the complexity, and the bogus possible manipulation of NS_UNCONSTRAINEDSIZE. I'm also open to removing some of those !important rules this patch is adding, if we want to.

See also bug 402850.

Comment on attachment 290819 [details] [diff] [review] Possible fix In the forms.css changes, I think you should also add rules for min-height and max-height. Also, to be compatible with what you were doing before, I think you need to specify width:-moz-max-content, not width:auto. Otherwise the min-width computation for the fieldset will assume line breaking inside the legend. Or was that change intentional? (If you do this, fix the assertion in nsFieldSetFrame::Reflow to match.) (Could you at least add a testcase for min-width computation of a fieldset with a legend that has multiple words in it?) r+sr=dbaron with that (although it's also ok if your change to allow the contents of the legend to wrap was intentional -- assuming you've tested that it works reasonably)

> I think you should also add rules for min-height and max-height. Will do. > Also, to be compatible with what you were doing before, I think you need > to specify width:-moz-max-content, not width:auto. Otherwise the > min-width computation for the fieldset will assume line breaking inside > the legend. Or was that change intentional? This was intentional. The white-space:nowrap should keep the current behavior for normal cases, but if someone wants their legends to wrap, they can. That fixes bug 354502. > (Could you at least add a testcase for min-width computation of a > fieldset with a legend that has multiple words in it?) Yeah, I need to write some tests. I did test this stuff at the time, but didn't seem to get those data: URIs into reftest format. :(

David, see comment 5.

Drivers, this is a patch that makes sure that we don't do unconstrained math with fieldsets... Risk should be low-ish.

Comment on attachment 308982 [details] [diff] [review] Updated to comments a1.9=beltzner

Checked in.