This is spun off of bug 424236, to cover the part that is a regression from bug 416168. What's happening in this testcase is that (based on the output of NOISY_VERTICAL_ALIGN), in standards mode, we make the inlines that are alone on a line not be empty, but we do make the line that contains them empty (because we enter the code that prints "zapping min/maxY" when we're aligning the root span). One possibility for how to fix this is to make these inlines hit the emptyContinuation logic, but I'm not sure that will fix all the cases we want to fix. I think the invariant that we want to preserve is that if we're collapsing a line to zero, all the inlines inside it should also be collapsed to zero. The end result should be that in this testcase, the inlines before and after the block should have at most a dot's worth of outline, not a vertical line's worth of outline.

We could special-case the outline drawing code to just not outline zero-width or zero-height rects.