Jan, is this still valid?

(In reply to Robert Mader [:rmader] from comment #3)

Jan, is this still valid?

KDE with disabled compositor, Ubuntu 21.04, GTX 1060, Nvidia driver 470.63.01
Yes, still reproducible.
mozregression --launch 2021-09-21 --pref gfx.webrender.all:true gfx.x11-egl.force-disabled:true -a https://codepen.io/teerapuch/pen/vLJXeR

From the regressing pull request:
https://github.com/servo/webrender/pull/3195/files (bug 1496168)

Invalidate render targets when returning them to the pool.
This allows the driver to avoid preserving the old contents we don't
care about.

// Note: The invalidate extension may not be supported, in which
// case this is a no-op. That's ok though, because it's just a
// hint.

/// Transfers ownership of a render target back to the pool.

In bug 1723323 the window can also become partially black.
Maybe the rendered content is thrown away too soon?

In bug 1678804 the top-left quarter of an extension panel shows junk. Could it be related?

Back then, bug 1496168 caused another regression:
bug 1499785 == https://github.com/servo/webrender/pull/3195#issuecomment-430746224

@bholley @kvark any ideas off the top of your head why this PR would result in all the glyphs and images turning into black rectangles on Android? https://bugzilla.mozilla.org/show_bug.cgi?id=1499785

That Android bug was fixed by https://github.com/servo/webrender/pull/3239.

We recently changed to use the glTexStorage* family of functions,
along with sized internal format types, such as BGRA8. Unfortunately,
the GL_EXT_texture_format_BGRA8888 extension does not provide BGRA8 as
a valid format unless the GL_EXT_texture_storage extension is also
available, which is not usually the case on GLES3 as glTexStorage*
are built in.

Unfortunately this means we must fall back to using glTexImage* rather
than glTexStorage* when dealing with BGRA data. Longer term it will
make sense to ensure the provided data is in RGBA format, as the
benifits of immutable storage are desirable.