STR: Open the testcase. Expected: rgb(0, 255, 0), rgb(255, 0, 0), consistently. Actual: Expected result when shift-reloaded, rgb(255, 0, 0), rgb(255, 0, 0) otherwise. Running a regression test. This blocks my work on bug 1381071. I suspect the regressor is bug 1203766, but not sure yet.

This is not stylo-only, btw.

Err, I meant bug 1366964 above, of course.

But it can't be, of course, since I can repro it in 54.0.1...

I suspect I know why this happens for stylo, but digging now.

So I think this is happening because we don't increment the restyle generation when we find a style change that doesn't touch any frame... And thus we assume we can keep the style context around. I'm going to fix this for stylo in bug 1381071, first conservatively, then reworking a few things (I don't think after that bug we need all the generation mess anyway).

So, I retract my words, I don't think that's the root cause...

Oh, actually it's not due display: none at all, and it's because the script-set value is preserved across reloads... So this is invalid I fear, sorry all for the noise :(