1697579 - CompositorBridgeParent destruction semantics aren't enforced properly

Status: RESOLVED FIXED

(Core :: IPC, defect)

Description

[Brad Werth [:bradwerth]]

It is important that CompositorBridgeParent::ActorDestroy must be called before a CompositorBridgeParent is destroyed. Failure to do so can lead to IPC problem, possibly crashes.

As noted by sotaro elsewhere:

Current implementation expects that CompositorBridgeParent is alive when WebRenderBridgeParent is alive. When CompositorBridgeParent is destroyed, CompositorBridgeParent::StopAndClearResources() should be called. And it seems not safe to call DestroyCompositor() from nsChildView::~nsChildView().

And also:

I wonder if we need to handle also ActorDealloc() in addition to ActorDestroy().

This bug is intended to cover enforcement of these semantics.

Comment 1

[Brad Werth [:bradwerth]]

Attachment: Bug 1697579 Part 1: Add diagnostic asserts to ~CompositorBridgeParent().

Comment 2

[Brad Werth [:bradwerth]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=ae5fad16f97ef6c017c8731f5a0aa215e4fe0aa4

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/1cf8864ed737
Part 1: Add diagnostic asserts to ~CompositorBridgeParent(). r=sotaro

Comment 4

[Sotaro Ikeda [:sotaro]]

I also added gfxCriticalNote to CompositorBridgeParent::~CompositorBridgeParent() by Bug 1697090. It might help for crash report.

Comment 5

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/1cf8864ed737

Comment 6

[Brad Werth [:bradwerth]]

Several bugs have landed that have improved these semantics, including Bug 1710347 and Bug 1700848, which addresses the core issue of this bug.