> On Feb 22, 2017, at 1:41 PM, Charles Srstka via swift-evolution
> <swift-evolution@swift.org> wrote:
>
>> * Updating completedUnitCount atomically
>>
>> The best practice here is to keep the progress object thread local. I think
>> that updating one progress from multiple threads could be a code smell.
>> Perhaps then you are doing several parts of the work and you should instead
>> form a tree. This also leads in a direction where completed unit count is
>> either 100% handed out to children or 100% controlled by your
>> work-update-progress loop. Mixing the two leads to easy confusion about who
>> is responsible for which portion. If you follow this rule, you never have to
>> get the completed unit count, which means the race you describe does not
>> exist.
>
> NSProgress is designed to be thread-safe by protecting access to its
> properties with an NSLock, but it is not in fact thread-safe. It is making a
> promise to its users that it is not keeping. In addition, the lack of an
> atomic increment means that if you are getting data from an API that only
> reports deltas, you either have to get the old completedUnitCount, or keep
> track of the running total yourself. An atomic increment operation would not
> only improve thread safety, but it would improve the experience, since it
> would reduce the work the user needs to do, while remaining safe and
> performant.
One thing to add to this is that when dealing with async APIs that you don’t
control, you don’t always know that things will be thread-local.
So if some library provides:
public func OpaqueAPI(input: SomeInput, dataHandler: (Data) -> ())
then in your client code:
let progress = Progress.discreteProgress(totalUnitCount: ...)
OpaqueAPI(input: blah) { data in
// Update my progress by data.count, but I don’t necessarily know what
thread this will happen on, or that it will be the same one every time
}
So, now I have to protect either the progress or some running total var that
I’m keeping with locks/mutexes/semaphores/sync queues/etc., adding a ton of
boilerplate to the method.
Charles
_______________________________________________
swift-evolution mailing list
swift-evolution@swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution
