On Wednesday, 5 April 2017 at 16:04, rintaro ishizaki wrote: > Just FYI > > Implementation of Type Erasers > https://lists.swift.org/pipermail/swift-dev/Week-of-Mon-20160905/002852.html
Thanks for that link! This looks like a followup to Chris’s question, so I’m pinging Dmitri to chime in on this. Bug report: https://bugs.swift.org/browse/SR-4499 For a reproducer, see the benchmark for `run_SuffixAnySequence` - that is utterly dominated by wending elements from the underlying sequence using generic witness table. There is no specialization occurring at all. As I’ve reasoned in my original e-mail, this is understandable - semantics of _AnyIteratorBox used in AnyIterator’s `next` method require virtual dispatch. Closure based approach to type erasure should’t have such a semantic barrier for optimization - the limit would probably just be currently conservative optimization? Best regards Pavol Vaskovic _______________________________________________ swift-dev mailing list swift-dev@swift.org https://lists.swift.org/mailman/listinfo/swift-dev
