The reftest mathml/semantics-1.xhtml has just started failing for me - some things are being drawn slightly lower on the screen in the presence of <semantics> tags than they are without. I will attach screen shots, it's hard to describe. This worked a week ago, is (apparently) not failing on the build farm, and the only system change that appears to be relevant is that my Linux distributor has just picked up the 1.0.0 release of the STIX fonts, so I'm provisionally blaming that.

The most obvious difference between these screenshots is that the first left-justified row of black and white boxes is slightly farther away from its overline in the test than in the reference, but if you flip back and forth between them, other things move too. (I freehanded the area to grab, so the top and left margins may not match. That's not the problem I'm talking about.)

By the way, is this test _supposed_ to draw checkerboard rows, or does that indicate some other problem?

No, the checkerboard is not a problem. FYI, the purpose of these tests is to check that 1) semantics does not prevent displaystyle to be inherited (i.e. mtext children of munderover are drawn as under/over scripts not sub/sup scripts). 2) semantics does not prevent displaystyle the transmission of embellishement data (i.e. the bar stretches). 3) the semantics does not produce extra spaces around it. 4 & 5) the semantics does not interfere with maction (i.e. the selected child of maction should be 3 and 4 respectively) Apparently, the difference is only with tests 1) and 2) on your screenshot. I've only the problem with test 2). It seems that spaces are added around semantics frame but I can not explain how STIX 1.0 could be involved. Trying to force margin, border, padding to 0 for semantics does not seem to be helpful.

Replacing the <mtext> by an <mspace> seems to work better for me. Does that fix the problem for you too?

Comment on attachment 611174 [details] [diff] [review] Patch V1 Yes, that fixes it for me too.