how to recover from a crash? about gpio HOT 5 CLOSED

The gpio library does not allow you to force use a pin. If it sees a pin locked it assumes another app already has it. If your app has left it locked then you need to release the pin yourself. That is done using

echo pin > /sys/class/gpio/unexport

where pin is the pin number.

That is using the sysfs GPIO interface which the gpio library uses for watches and which is reporting the pin in use.

In general you can can catch signals that cause your app to exit and call the Unwatch in the signal handler. Refer to the watcher example which does just that. It doesn't catch a SIGSEGV though - and I'm not sure that approach works for a SIGSEGV. You can always try it by adding syscall.SIGSEGV to the list of signals.

And most importantly, fix your app so it doesn't SIGSEGV.

from gpio.

I'm already catching signals to clean up everything. however, if there is for example a panic() inside the watch handler, the app will crash, leaving pins locked.

I don't think a go app can SIGSEGV unless you are using the unsafe package maybe. afaik panic() is just like an exception. maybe I'm missing something?

would it be possible to add something like gpio.Unexport() or maybe the watcher can defer Unwatch() inside the internal handler go routine so that it would be cleaned up even when some code in the handler from the lib consumer panics? this would be similar to how the http server is handling panics inside handlers: https://github.com/golang/go/blob/master/src/net/http/server.go#L81.

from gpio.

this shows that no real segmentation fault is happening in go: https://play.golang.org/p/SzaWnRlE5-k

from gpio.

Yeah, I realised that as I was going to sleep. Sorry, I've been working in C and the context switch took a while. In Go you get a panic, not a SIGSEGV. So shoot me.

I wont be adding anything to this library, and certainly not an unexport. When I wrote this lib I decided not to expose an unexport as it just allows users to easily override locks. And then locks become pointless. It is intentionally hard to override a lock - you need to write to that file yourself. Which isn't really that hard. You are welcome to add something yourself if you think that is the best way to solve your problem.

I don't think the lib should be trying to recover from a panic - you can easily wrap your handler in something that recovers from panics if you want. The HTTP server case is different. They chose to assume the panic is confined to the handler so they can keep the app, and the potentially thousands of connections, running. This library plays with hardware and I chose to play it safe and assume that if you have a panic then it is time to leave - hopefully before the smoke starts coming out. But there is nothing stopping you wrapping your existing handler with some recovery if you disagree. Personally, I'd fix the panic.

from gpio.

yes that makes sense. thanks for the input!

from gpio.