Running the regression tests in viewer will not flag any bbox (and probably other kinds) when there are differences. I haven't run the regression tests for at least 5 weeks, so this could have been broken that long. We've been plagued with false positives for quite a while, now we're seeing the flip side.
Created attachment 69425 [details] [diff] [review] patch to terminate xul style in the regression data
The bug only happens when style data is written in the regression data, which is the default for the table regression tests. nsStyleContext has had this bug for quite a while, so I'm a bit confused as to why I didn't see this before, since I think we have been writing style data for a long time.
Comment on attachment 69425 [details] [diff] [review] patch to terminate xul style in the regression data This was a typo from when the SVG branch landed.
The patch is in.