1517896 - PersistentRooted<GCHashSet> does not work in ShellContext

Status: RESOLVED FIXED

(Core :: JavaScript: GC, enhancement, P3)

Description

[Steve Fink [:sfink] [:s:]]

arai attempted to use PersistentRooted<GCHashSet> and got a !Empty() assertion during shutdown.

Comment 1

[Steve Fink [:sfink] [:s:]]

Attachment: Fix shell context shutdown ordering

The problem seems to be that we don't actually delete the main ShellContext. (We do delete worker ones).

This is not the minimal necessary change, I don't think -- when tracing through, it seemed wrong to be deleting the runtime before the context, so I swapped the order, then fixed up everything that broke.

If that causes too many issues, I can undo that part.

Comment 2

[Jon Coppeard (:jonco)]

Comment on attachment 9034520 [details] [diff] [review]
Fix shell context shutdown ordering

Review of attachment 9034520 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/shell/js.cpp
@@ +11102,5 @@
>    DestructSharedObjectMailbox();
>  
>    CancelOffThreadJobsForRuntime(cx);
>  
> +  JS_SetContextPrivate(cx, nullptr);

Do we not need to do this in the worker case?

::: js/src/threading/ProtectedData.cpp
@@ +55,2 @@
>    JSContext* cx = TlsContext.get();
> +  MOZ_ASSERT_IF(cx, CurrentThreadCanAccessRuntime(cx->runtime()));

Does this allow the check to pass if this is called on an unrelated (non-SM) thread?  That seems bad.

Comment 3

[Steve Fink [:sfink] [:s:]]

Attachment: Fix shell context shutdown ordering

Clearing out the context's private data isn't really necessary anywhere, it just felt dangerous to be running a bunch of code during shutdown with a dangling pointer. But you're right, I should clear it consistently. Fixed.

I switched to using AutoNoteSingleThreadedRegion to suppress the checks, and it appears to shut down successfully.

Do you have an opinion on whether a cx or rt should be shut down first, btw? Because if I didn't change that, then this patch would be simpler. Maybe it's just because my head is still stuck in a multiple-cx world that it seems wrong to me to blow away the runtime first and have a period where there is a cx without an rt. An rt without a cx didn't seem as bad.

Comment 4

[Jon Coppeard (:jonco)]

Comment on attachment 9034825 [details] [diff] [review]
Fix shell context shutdown ordering

Review of attachment 9034825 [details] [diff] [review]:
-----------------------------------------------------------------

Ah, I see.

I agree, it feels like we should destroy the context before the runtime.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/fdd74f1bbff3
Fix shell context shutdown ordering, r=jonco

Comment 6

[Cristina Coroiu [:ccoroiu]]

https://hg.mozilla.org/mozilla-central/rev/fdd74f1bbff3