1433787 - hit-test: short hang when (un)focusing Nightly

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect, P1)

Description

[Darkspirit]

Attachment: 2018-01-28_18-39-11.mp4

Nightly 60 x64 20180128100650 de_DE @ Debian Testing (KDE, Radeon RX480)
fresh profile: gfx.webrender.all, gfx.webrender.hit-test

1. open the testcase: https://bug1401665.bmoattachments.org/attachment.cgi?id=8910415
2. (you may zoom to 300%)
3. repeatedly click (or hold your mouse down) on Nightly, then beside it, then on Nightly again etc.

Comment 1

[Milan Sreckovic [:milan] (needinfo for best results)]

Linux only?

Comment 2

[Kartikaya Gupta (email:[e-mail])]

I can repro on OS X

Comment 3

[Kartikaya Gupta (email:[e-mail])]

Based on a profile [1] it looks like the compositor thread is stuck in FlushRendering() at [2]. This in turn because the Renderer thread in WR is busy doing lots of composites, each of which gets stuck in SwapBuffers for a vsync. The composites it's doing are presumably from a bunch of hit-tests. It's not really clear to me why we end up doing a lot of hit-tests when focusing/unfocusing the browser, and why each of those hit-tests triggers a render (it implies we're getting display list updates in between the hit-tests). But the fix I'm doing for the debug assertion failure in bug 1421380 fixes this by ensuring the Renderer thread doesn't try to composite these frames after rendering them.

[1] https://perfht.ml/2DMCVcT
[2] https://searchfox.org/mozilla-central/rev/11d0ff9f36465ce19b0c43d1ecc3025791eeb808/gfx/layers/wr/WebRenderBridgeParent.cpp#1358

Comment 4

[Darkspirit]

Nightly 60 x64 20180130223236 de_DE @ Debian Testing (KDE, Radeon RX480).
It seems to be fixed. Thank you!
Your try build from bug 1421380 comment 16 was already good.

Comment 5

[Kartikaya Gupta (email:[e-mail])]

Thanks for verifying!