Feature socketcan bcm owned by bus

[hardbyte]

Competes with #409 because I forgot to look if you were already doing this @christiansandberg !

• adds task management responsibility to the Bus. By default tasks created with bus.send_periodic will have a reference held by the bus - this means in many cases the user doesn't need to keep the task in scope for their periodic messages to continue being sent. If this behavior isn't desired pass store_task=False to the send_periodic method.
• the bus now has a stop_all_periodic_tasks method
• The socketcan tasks now reuse a bcm socket

[christiansandberg]

We’ve discussed this before but this can potentially become a memory leak. There is not way for the user to clean out stopped tasks. I don't know what to do about it. Maybe it could be possible to remove them if they are stopped, because then we know that the application has its own reference to the task.

[christiansandberg]

...or shutdown() or stop_all_periodic_tasks() is called?

[felixdivo]

Note: We might also consider calling shutdown() in the __del__ method of the bus.

[codecov]

Codecov Report

Merging #412 into develop will increase coverage by 0.91%.
The diff coverage is 68.57%.

@@             Coverage Diff             @@
##           develop     #412      +/-   ##
===========================================
+ Coverage    61.51%   62.43%   +0.91%     
===========================================
  Files           55       55              
  Lines         4584     4600      +16     
===========================================
+ Hits          2820     2872      +52     
+ Misses        1764     1728      -36

[hardbyte]

Hmm okay what about this slightly hack approach of wrapping the task.stop() method inside bus.send_periodic. There would still be issues with resumable tasks and expired tasks though.

[hardbyte]

Regarding the memory leak issue at the least we should clearly document what the user can do to prevent it, but I'd rather have an api that allows the user to take ownership for the task management if they want to. Perhaps a flag in the bus initializer or the send_periodic method?

[christiansandberg]

Not all implementations may be able to handle shutdown being called twice.

[hardbyte]

Fair point - I'd rather remove the __del__ from the bus instead of adding shutdown state tracking. Do you agree?

[christiansandberg]

I think so too.

[hardbyte]

I think this is pretty much ready - I've just rebased.

The last change I made - added a remove_tasks parameter to the stop_all_periodic_tasks method which calls the wrapped stop method.

[felixdivo]

I just merged this. Someone else (@christiansandberg ?) needs to have a look at #292 and #404.