Deadlock in [rest device] when called from activation state machine

[tcard]

A customer has reported a crash due to this. The call stack shows that [rest device] was called from the state machine while already holding the lock. We need to ensure this doesn't happen; we probably need a [rest device_nosync] method like the others we have around, and use that from the activation machine, and put an dispatch_async on entry points of the state machine.

[tcard]

Here's a screenshot of the crash:

[itolisto]

Hello @tcard and @ricardopereira

If I can help you, please let me know, I would gladly collaborate to solve this issue.

[itolisto]

Hello @tcard and @ricardopereira,

Just wondering if you have been able to take a look on this? And again, if there's anything I can do, I'll be happy to do it.

[ricardopereira]

@itolisto Not yet, sorry. But I'll look at it today. You're using the push branch, right?

[itolisto]

@ricardopereira Yes, I’m using the push branch.

[itolisto]

Hi @ricardopereira,

Just checking again if you have had the chance to take a look on this. Sorry if I'm being annoying (: And again, if I can help please tell me. 👍

[ricardopereira]

@itolisto Sorry. I had to deal with an other issue the other day. I'll reach you later today to test this out please 😊

[ricardopereira]

I reproduced it successfully:

Basically, I'm calling activate() in a deactivate event.

[ricardopereira]

A test should be added with this issue at #662

[ricardopereira]

@itolisto Can you test the fix using the push-fix-654 branch please? Thanks 👍

[itolisto]

@ricardopereira Testing this right now.

[itolisto]

@ricardopereira I can't reproduce the issue anymore, I believe this is fixed. 👍

Thank you for your help

[ricardopereira]

Good to hear! Thank you