1034611 - Make worker JS runtime creation fallible

Status: RESOLVED DUPLICATE of bug 1247122

(Core :: DOM: Workers, defect)

Description

[Andrew McCreight [:mccr8]]

With GGC, workers apparently need a 16MB nursery, which is created when we create the JSRuntime.  That's huge, so we should make it fallible.  Fortunately, this only requires some minor code shuffling.

(We may separately want to consider reducing the nursery size for workers.)

Comment 1

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

That's crazy.  We absolutely cannot have a 16 MB nursery for workers.

Comment 2

[Andrew McCreight [:mccr8]]

Attachment: Make worker JS runtime allocation fallible.

This compiles and passes worker tests, though I'm getting one failure locally:

TEST-UNEXPECTED-FAIL | /tests/dom/workers/test/test_multi_sharedWorker_lifetimes.html | Still have data stored - got undefined, expected 0123456789abcdefghijklmnopqrstuvwxyz

Comment 3

[Ryan VanderMeulen [:RyanVM]]

That's bug 995688, no? Of course, bonus points if you can fix that while you're at it ;)

Comment 4

[Andrew McCreight [:mccr8]]

Hmm, yeah, looks like I commented back in April that it is permaorange for me. ;)

Comment 5

[Ben Turner (not reading bugmail, use the needinfo flag!)]

We know that we fail this allocation at least on tinderboxes so we'll need to do something more sophisticated than just bailing out of WorkerThreadPrimaryRunnable::Run I think.

Comment 6

[Andrew McCreight [:mccr8]]

I was hoping that returning an error would somehow magically turn into an exception in the parent thread.  Is that the sort of behavior you'd expect, Ben?

Comment 7

[Ben Turner (not reading bugmail, use the needinfo flag!)]

No, sadly :(

Comment 8

[Andrew McCreight [:mccr8]]

By the last sentence, I meant what is the desired behavior here? :)

Comment 9

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

Comment on attachment 8450985 [details] [diff] [review]
Make worker JS runtime allocation fallible.

So we established that this approach doesn't work, right?

Comment 10

[Andrew McCreight [:mccr8]]

I haven't checked yet, but I'm assuming it doesn't.  Sorry, I should have cleared the feedback flag after bent commented.