Bugzilla

Here's one more testcase, generated by LQC with light edits by me, which I suspect is a variant of this same issue, since it's also got a writing-mode dependency.

(I'm not entirely sure it's precisely the same hence, I'm attaching it here, and we can spin off separate bugs as-needed for any testcases that remain broken when we're ready to land a patch here.)

Here's one more testcase, generated by LQC with light edits by me, which I suspect is a variant of this same issue, since it's also got a writing-mode dependency.

(I'm not entirely sure it's precisely the same hence, I'm attaching it here, and we can spin off separate bugs as-needed for any testcases that remain broken when we're ready to land a patch here.)

As with the others, the STR are to run `simpleRecreate()` in your console and note the discrepancy in logged values -- e.g. here's what I see:
```
Height after style changes, before reload
#three 1952
Height after reload
#three 968
```

If I disable the optimization pref, then I get 968 for both measurements.