[swift-dev] Implementation of Type Erasers

[Dave Abrahams]

on Tue Sep 06 2016, Chris Eidhof <swift-dev-AT-swift.org> wrote:

> Thanks! That explains a lot.
>
> In my version, performance only is improved by 10%, so I'm not sure if the
> tradeoff is worth it:
>
> In the isolated case of AnyIterator, there's only one method, so the sizeof
> is the same (and the performance is worth it). In the case of AnySequence,
> it would be a lot more work to write the implementation (and the size would
> change a lot). Having two different mechanisms for type erasure seems like
> it might confuse things even more.

We'll happily take a litte more implementation complexity in exchange
for performance gains.  10% is nothing to sneeze at.

> On Mon, Sep 5, 2016 at 9:50 PM, Dmitri Gribenko <gribozavr at gmail.com> wrote:
>
>> On Mon, Sep 5, 2016 at 11:13 AM, Chris Eidhof via swift-dev
>> <swift-dev at swift.org> wrote:
>> > Hello swift-dev,
>> >
>> > I was wondering why type erasers (e.g. for AnyIterator) are implemented
>> the
>> > way they are. I would implement AnyIterator like this:
>> >
>> > final class MyAnyIterator<A>: IteratorProtocol {
>> >     var nextImpl: () -> A?
>> >
>> >     init<I: IteratorProtocol>(iterator: I) where I.Element == A {
>> >         var copy = iterator
>> >         nextImpl = { copy.next() }
>> >     }
>> >
>> >     func next() -> A? {
>> >         return nextImpl()
>> >     }
>> > }
>> >
>> > Instead, it is implemented in a class which contains a box.
>>
>> Your approach requires adding a new closure property for every method
>> that is forwarded.  Subclassing does not require that.
>>
>> If you see a case where performance is not good, we would appreciate a
>> bugreport at https://bugs.swift.org/ .  Please attach a self-contained
>> reproducer.  Thank you.
>>
>> Dmitri
>>
>> --
>> main(i,j){for(i=2;;i++){for(j=2;j<i;j++){if(!(i%j)){j=0;break;}}if
>> (j){printf("%d\n",i);}}} /*Dmitri Gribenko <gribozavr at gmail.com>*/
>>

-- 
-Dave