980886 - Assertion failure: "MIR instruction returned value with unexpected type" when starting Firefox with soundcloud.com as pinned

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine: JIT, defect, P3)

Description

[alessandro]

Attachment: Schermata 2014-03-07 alle 13.50.08.png

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:30.0) Gecko/20100101 Firefox/30.0 (Beta/Release)
Build ID: 20140306030201

Steps to reproduce:

when starting up firefox with soundcloud pinned tab i get a strange behaviour see pic in order to use soundcloud i have to refresh page every time not  abig deal but very annoying!!!!!!!!!!!!!!!
screenshot : https://copy.com/xAmbcrMp09z4
i tried in safe mode same odd behaviour!!!!!!!!!!!!!

Comment 1

[Paul Silaghi, QA [:pauly]]

Confirmed in nightly 30.0a1 (2014-03-09), Win 7 x64 if starting Firefox with soundcloud.com as pinned.

Comment 2

[Tim Taubert [:ttaubert] (inactive)]

Having https://soundcloud.com/explore open in an app tab, I can get Nightly to hang and crash shortly after startup. Will retry with a debug build and crash reporter.

Comment 3

[Tim Taubert [:ttaubert] (inactive)]

Can't reproduce in debug builds, maybe they're too slow. Got a crash report with an opt/crashreporter build that doesn't help at all. Rebuilding with debug symbols.

Comment 4

[Tim Taubert [:ttaubert] (inactive)]

Shortly before crashing I see:

Assertion failure: MIR instruction returned value with unexpected type, at js/src/jit/IonMacroAssembler.cpp:1274

Maybe that's the reason I can't get a proper stack trace.

Comment 5

[Till Schneidereit [:till]]

Hannes, can you look at this? It's not like soundcloud is only used by a handful of people, after all ...

Comment 6

[Hannes Verschore [:h4writer]]

Created bug 992535 for the GGC crash that started to show up at in Firefox 31 since two weeks ago.

Now I'll try to reproduce the "Assertion failure: MIR instruction returned value with unexpected type" one. Which is apparently already present since Firefox 30.

Comment 7

[Hannes Verschore [:h4writer]]

Over the weekend I tried to reproduce this issue but failed.

I got the builds from 9 March (since comment 0 and comment 1):
http://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/2014-03-09-mozilla-central-debug/

I tried:
- linux 32bit (noflash)
- linux 64bit (with flash installed)

I just pinned the soundcloud.com/explore. Close the browser and started again. But no crash?

@Pauly, can you give me more information about the crash. Which version/platform you tested and what the steps are to reproduce. (Please also try if disabling flash interferes with this issue). Thanks

Comment 8

[Tim Taubert [:ttaubert] (inactive)]

(In reply to Hannes Verschore [:h4writer] from comment #7)
> I got the builds from 9 March (since comment 0 and comment 1):
> http://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/2014-03-09-mozilla-
> central-debug/

I could reproduce this on OS X opt builds, I couldn't with debug builds.

Comment 9

[Hannes Verschore [:h4writer]]

(In reply to Tim Taubert [:ttaubert] from comment #8)
> (In reply to Hannes Verschore [:h4writer] from comment #7)
> > I got the builds from 9 March (since comment 0 and comment 1):
> > http://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/2014-03-09-mozilla-
> > central-debug/
> 
> I could reproduce this on OS X opt builds, I couldn't with debug builds.

So you saw "Assertion failure: MIR instruction returned value with unexpected type" with a opt build? That would be very strange, since these assertions are only enabled on debug/debug+optimize builds. Can you still reproduce with a nightly from 2014-03-09 ?

Comment 10

[Tim Taubert [:ttaubert] (inactive)]

Oh, it might have also been a debug+optimize build, I played around with a couple of build configs trying to get a stack trace. I can try doing the same again locally.

Comment 11

[Hannes Verschore [:h4writer]]

(In reply to Tim Taubert [:ttaubert] from comment #10)
> Oh, it might have also been a debug+optimize build, I played around with a
> couple of build configs trying to get a stack trace. I can try doing the
> same again locally.

A stacktrace won't give more information in this case. The assertion is throw in JIT-code, so nothing comprehensible will be on the stack to help to fix this.

I'm still trying to get to reproduce this.
1) Can you test if "disabling flash" makes a difference?
2) Is this a 32bit or a 64bit build
3) What are the steps to reproduce? Any more steps than pin http://soundcloud.com/explore. Close and re-open the browser? Any small step can make a difference. If you could just explain the exact steps you did?

Comment 12

[Tim Taubert [:ttaubert] (inactive)]

(In reply to Hannes Verschore [:h4writer] from comment #11)
> A stacktrace won't give more information in this case. The assertion is
> throw in JIT-code, so nothing comprehensible will be on the stack to help to
> fix this.

Yeah, but I didn't know that it's a JIT-code assertion before I was digging deeper :)

> I'm still trying to get to reproduce this.
> 1) Can you test if "disabling flash" makes a difference?
> 2) Is this a 32bit or a 64bit build
> 3) What are the steps to reproduce? Any more steps than pin
> http://soundcloud.com/explore. Close and re-open the browser? Any small step
> can make a difference. If you could just explain the exact steps you did?

I will re-test all of that and report back.

Comment 13

[Tim Taubert [:ttaubert] (inactive)]

I could just reproduce it again with a debug+opt build on OS X 64bit. Disabling Flash made no difference. I have a single window with three tabs:

1) soundcloud.com/explore (pinned)
2) golem.de (normal)
3) some image from imgur (normal, selected)

The session is automatically restored, i.e. "When Nightly starts: Show my windows and tabs from last time". I quit Firefox and then start it again and it crashes about every second time.

If there is anything I can do from here to debug this we can also team up on IRC if you like!

Comment 14

[Tim Taubert [:ttaubert] (inactive)]

I can also reproduce it with only the pinned tab and about:home, so it doesn't seem to be caused by the non-soundcloud tabs.

Comment 15

[Paul Silaghi, QA [:pauly]]

(In reply to Hannes Verschore [:h4writer] from comment #7)
> @Pauly, can you give me more information about the crash. Which
> version/platform you tested and what the steps are to reproduce.
It seems I can no longer reproduce the issue, both on 30.0a1 (2014-03-09) and 31.0a1 (2014-04-06) normal builds, Win 7 x64.

Comment 16

[Hannes Verschore [:h4writer]]

Ok I can reproduce also \0/ Thanks for the help and details about when it crashes!

The only strange part is that I can only let it crash with GGC builds. Which is strange, since the first report is on a non-ggc build.

Comment 17

[Till Schneidereit [:till]]

I'd be surprised if this still reproduces, but we should check.

Comment 18

[Naveed Ihsanullah [:naveed]]

This does not reproduce for me. It is likely fixed in the years since it was reported.