1384025 - Two color children disappear after toggling to transform-style:flat.

Status: NEW

(Core :: Layout, defect, P3)

Description

[u459114]

Steps to reproduce:
1. Visit https://codepen.io/impressivewebs/pen/pIgBf
2. Click toggle transform-style button.

Actual results:
goldenrod and green divs disappear.

Expected results:
According to 2D rendering order, goldenrod and green divs should cover on the aquamarine one.

Comment 1

[Daniel Holbert [:dholbert]]

Attachment: testcase 1

Here's the testcase, taken from the codepen source.

Comment 2

[Daniel Holbert [:dholbert]]

I think this might be an invalidation issue of some sort.
 - When I click the button on the attached testcase, the goldenrod and green divs don't even disappear (i.e. the rendering doesn't change at all) until I alt-tab away (triggering a repaint).
 - From that point, I can make them reappear in the correct spot (flattened on top of the aquamarine one) if I right-click the aquamarine div and choose "inspect".

Comment 3

[Daniel Holbert [:dholbert]]

(In reply to Daniel Holbert [:dholbert] from comment #2)
> I think this might be an invalidation issue of some sort.
>  - When I click the button on the attached testcase, the goldenrod and green
> divs don't even disappear (i.e. the rendering doesn't change at all) until I
> alt-tab away (triggering a repaint).

The invalidation issue seems to have started with this range:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=11dc79e232110ba6de5179e46dfbda77b52a88c3&tochange=4313752f69956ae248bd4e7ff3913c8dd4252698

 --> bug 1097464 ("Move preserve-3d handling into the compositor")

Before that range, our behavior still isn't quite correct (i.e. we draw the overlayed divs too skinny at the end). But at least we draw them that way in a stable/consistent manner, and there aren't any rendering artifacts. (The rendering doesn't change if I alt-tab away or right-click and pick "inspect".)

So, it makes some sense to call this a regression from that bug.  Marking as such.  Given that that's quite a while ago (Firefox 46), I don't think there's any strong need to take this for 57 (and to take the risk of such a hypothetical uplift) --> wontfix for that release.

Comment 4

[Mike Taylor [:miketaylr]]

Let's remove the regression keywords since we've been shipping this for so long.