1815480 - Investigate BrowserContext attach child -> parent flow vs. child shutdown

Status: ASSIGNED

(Core :: DOM: Navigation, task)

Description

[Jens Stutte [:jstutte]]

From bug 1811746 comment 3:

It seems we can find our process still subscribed to the given BrowsingContextGroup despite having started our content process' shutdown. IIUC, we always will ContentParent::MarkAsDead and thus ContentParent::RemoveFromList where we unsubscribe the process from all mGroups at the same time when we have done SignalImpendingShutdownToContentJS, so this is not really expected, unless there is some misalignment in our book-keeping for BrowsingContextGroups between what the ContentParent knows about it and who else might reference them.

In bug 1814603 we added a paper-over check that might result in process creation where it is not really expected nor useful.

Comment 1

[Jens Stutte [:jstutte]]

Attachment: WIP: Bug 1815480 - BrowserContextGroup::EnsureHostProcess should not accept dying processes.

Comment 2

[Jens Stutte [:jstutte]]

Attachment: WIP: Bug 1815480 - Ignore requests to create a BrowsingContext from a dying child process.

Comment 3

[Olli Pettay [:smaug][[e-mail]]]

Added some comments to phab.

Comment 4

[Artur Iunusov]

(I can take a look)

Comment 5

[Jens Stutte [:jstutte]]

Not sure if the patch from bug 1811746 wants to suggest that this is already fixed in the meantime, could you try to verify?

Comment 6

[Artur Iunusov]

looking right now

Comment 7

[Artur Iunusov]

Need to do a second look after finishing some patches regarding to workers throttling