On 08/10/2015 05:06 AM, Sergey Mironov wrote:
As for urweb-callback, the mailing-list archives will show my skepticism
of the basic abstraction, at the time it was being designed. I remain
unconvinced that it's necessary, given the periodic-tasks functionality
built into Ur/Web.
I understand that point of view, it just seems like a useful library in some
situations.
Regarding the urweb-callback issue - I think it is definitely useful
for OS communication, where periodic tasks may not meet latency
constraints. From the other hand, the low-level structure of the
library seems to me not very good now. I would like to re-design it.
Replacing queues and thread manager with simpler constructs is
probably a good idea . I hope I'll do this some day.
This question of latency is precisely where we disagreed last time. I
remain unconvinced that it matters for enough cases to justify building
an FFI callback library. I don't think anyone ran any performance
experiments with alternative implementations last time, to gather
evidence for one perspective or another.
_______________________________________________
Ur mailing list
[email protected]
http://www.impredicative.com/cgi-bin/mailman/listinfo/ur
