Change gen_statem:call(_, _, {clean_timeout,infinity}) to use proxy #1595
Conversation
|
In the docs:
Should be "So why not just let the calling process die by not catching the exception?" Otherwise I'm okay with things. If the default does not change, leaving some control to the caller makes sense at no requirements to change existing code. The complexity of the API is still quite large, but at that point, the local vs. proxy mechanism had already been exposed so this modification just makes it more consistent I think, by not messing with the explicit options given by the user. I'd give a tentative +1 to this. |
371848c
to
8efef4a
Compare
|
@ferd I made a mistake thinking messing with the explicit options given by the user would be an unobservable optimization. But I have learned and want to fix this in a 21.0. I approve your rephrasing suggestion. |
I would like community feedback on a changing a corner case for timeouts in
gen_statem:call(Server, Request, Timeout)for OTP 21.0.The
Timeout == {clean_timeout,infinity}today means{dirty_timeout,infinity}i.e does not use a proxy process. This was intended as an optimization.But since in the event of network problems such as the connection going down and up the monitor set before sending
Requestmay trigger causing the client call to fail, but theRequestmay actually have arrived at the server, which may reply so we get an unhandled late reply in the inbox. Very rarely.I think that giving the programmer the possibility to use a proxy process even for infinity timeout simplifies the semantics and may be found useful.
The behaviour for
Timeout == infinitywill still mean{dirty_timeout,infinity}that is: no proxy process.