search

lf-lang / lingua-franca

Intuitive concurrent programming in any language
https://www.lf-lang.org
Other
240 stars 63 forks source link

Deprecate the keepalive target property #592

Open cmnrd opened 3 years ago

cmnrd commented 3 years ago

Given the recent development in #546, I wonder if we still need the keepalive target property. If there are no physical actions in the program, then keepalive=true does not make any sense because there is no way for new events to be created once the event queue is empty. If there are physical actions, then we automatically set keepalive=true to account for asynchronous events. We could also set keepalive=false explicitly in case we still want the program to terminate when there are no more events. However, I am wondering: is this a common use-case?

Couldn't we simply omit the keepalive property if we set it automatically to the expected value? In the presence of physical actions, keepalive=false is likely a bad idea in most cases, because it is unpredictable. Depending on the precise timing of physical events, the program could nondeterminsitcally terminate at different points each time we run it. I think a controlled shutdown using an API call like request_stop() when the program reaches a certain state is generally the preferred option for stopping programs with physical actions.

edwardalee commented 3 years ago

This is a good point. I think we could probably get rid of keepalive.

lhstrh commented 3 years ago

We've had this discussion before. See: https://github.com/lf-lang/lingua-franca/issues/131#issuecomment-611813536

That earlier discussion was closed with the conclusion that keepalive could not be automatic, but I don't quite understand the reasoning there either because we could distinguish between generated physical actions and programmed ones.

lhstrh commented 3 years ago

These were my 2 cents in that thread:

If the event queue is empty, and there are only logical actions, no new events can emerge -- unless we're scheduling logical events asynchronously (which would be an abuse of logical actions). In other words, there would be no point in enforcing keepalive in that scenario. I think we should eliminate the keepalive target property altogether (not just for the C target) and implement the behavior as you suggest; enforce keep alive if there are physical actions.

edwardalee commented 3 years ago

Actually, re-reading my comment from back then (about physical connections), I think it is reasonable to force keepalive to be true when using physical connections. The programmer still has the option of specifying a timeout or invoking stop_request. So I agree that we should eliminate the keepalive target property.