Clean asynchronous stop_callback implementation

[gpotter2]

This PR:

• adds an asynchronous way of stopping sniff
• WILL BREAK started_callback usage. However, it only exists in master (was not on 2.4.0), so we can allow such a thing
• fixes Provide stop for sniff even when no packets received #1505

I will add the unit tests if we agree on the PR

[gpotter2]

Note to self: separate

• WINDOWS and started_callback ==> use select_objects
• WINDOWS and not started_callback ==> previous behavior: dont use select_objects

[guedou]

@gpotter2 what is the status of this PR?

[gpotter2]

Quite unknown :/ I’m still not sure wether this was a good idea at all

I’ll run some benchmarks and try to make it work with the recent select overhaul, to see what it becomes :/

[voi98306]

This is a VERY useful feature, implemented in a way consistent with how it has been done in C since at least the mid 90's, and thus well battle proven. I would REALLY like to see this coming out in a release.

The only comment I have is that if you return something more resembling an object, rather than a mere stop_... function, you can easily in the future add additional reasons for communicating with the sniff function without breaking backwards compatibility.

So instead of sending the "stop_callback" function to "started_callback", send a communication structure allowing the user to call "commstruct.stop_sniffing()". That way, in the future you can add other methods, like "commstruct.extend_timeout()" or whatever. Other changes needed to accomodate new features (such as not merely clearing the select list when objpipe is written to) will be well hidden behind the interface.

I am a huge fan and really appreciate your work. I hope to see this feature in a not too distant release.

[gpotter2]

The only comment I have is that if you return something more resembling an object, rather than a mere stop_... function, you can easily in the future add additional reasons for communicating with the sniff function without breaking backwards compatibility.

Good point. I think the best option would be to implement a sniff_async function (or a threaded argument to sniff) that would pop a sniff in a thread and return such an object.

[gpotter2]

@voi98306 It's not ready yet (some tests still fail), but there's a working demo over #1999