When a transform is applied to an element with a CSS drop-shadow filter applied, the shadow can be clipped.

This only happens when the blur radius is relatively small, but this likely means there are some issues around inflating a picture's surface rect when a drop-shadow filter is applied.

mozregression --good 2018-08-15 --bad 2018-09-15 --pref gfx.webrender.all:true -a https://bug1561447.bmoattachments.org/attachment.cgi?id=9074000

8:54.99 INFO: Last good revision: febddb5a4dc21c7b09b02efc39568a03588d893b
8:54.99 INFO: First bad revision: 0192efea1b664db8d20a691401ec5d3f9d3a2122
8:54.99 INFO: Pushlog:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=febddb5a4dc21c7b09b02efc39568a03588d893b&tochange=0192efea1b664db8d20a691401ec5d3f9d3a2122

The same statement might apply here:

(Emilio Cobos Álvarez (:emilio) from bug 1514384 comment 2)

Well, sure, before that we wouldn't clip it at all :)

This problem only occurs when the square is partly off the page, in this case a clip task is created which gets applied to the square's shadow. If the square is entirely in the page, no clip task is created and the shadow is properly rendered.

As far as I can tell, this is an issue inside WebRender as there isn't a clip in the display list itself that would cause this.

Also as the blur radius increases, the clip rect seems to increase as well, so we are probably taking the blur radius into account without considering the offset.

Pushed by rmaries@mozilla.com:
https://hg.mozilla.org/integration/autoland/rev/7c1def596e46
Include picture shadow in local rect when building clip chain r=nical

https://hg.mozilla.org/mozilla-central/rev/7c1def596e46

Given that we're less than a week from Fx69 RC, I think we can let this fix ride Fx70 to release. Feel free to nominate for Beta approval if you feel strongly otherwise.