1539508 - [meta] Simplify WorkerPrivate's end-of-life flow

Status: NEW

(Core :: DOM: Workers, task, P2)

Description

[Yaron Tausky [:ytausky]]

Currently the circumstances that lead to WorkerPrivate getting destructed are quite complicated: it can happen on more than one thread, it involves a self-smart pointer, etc. We're also aware of several bugs that lurk there. We should rethink how it happens and modify accordingly.

Comment 1

[Jens Stutte [:jstutte]]

I think, this should have higher priority than P3, at least for the removal of raw pointers as of bug 1546331. But I agree that we should have a very close look at the lifetime of WorkerPrivate in general, too, and besides the destruction paths even the usage from multiple threads might cause issues. I assume though, that we are not talking about massive parallel access on these data structures from more than two threads but that it is kind of serialized during run-time through the event loop and that we have at most the spawning thread and the worker thread itself accessing it. But this is just a wild guess/hope and should be verified - I did not even try to analyze the code, yet.

Maybe :perry can structure better the single tasks that should be done for this, probably promoting this to a meta-bug?

Comment 2

[Jens Stutte [:jstutte]]

We start with a patch for bug 1546331 first.

Comment 3

[Perry Jiang [:perry] [no longer employee, use ni?]]

This will probably become a meta bug (there's agreement that there should/will be work done for this). In particular, worker "busy counts" should be factored into WorkerRefs.

Comment 4

[Jens Stutte [:jstutte]]

bug 1493591 might be related, too (crash during deconstructor call of WorkerPrivate). And it still gets fresh crash reports from time to time, though closed.

Comment 5

[Jens Stutte [:jstutte]]

Start to make this a meta-bug.

Comment 6

[Jens Stutte [:jstutte]]

Should we limit this to end-of-life? What about bugs like bug 1505409 ?

Comment 7

[Jens Stutte [:jstutte]]

Or bug 1552663 ?