00:35 <companion_cube> I think it might be the case for some purely functional use cases, but it's unfit for a ML until it has a decent answer to mutability

07:11 <discocaml> <omentic> i caught this talk irl

07:12 <discocaml> <omentic> i was very impressed by the benchmarks being comparable (even if universally worse). there was much emphasis on it being a prototype

07:15 <discocaml> <omentic> mildly disappointed it doesn't seem to have been extended. i'm a big fan of reference counting. (mostly for simplicity + real-time systems guarantees)

07:19 <discocaml> <omentic> interestingly tidbit on the whole perceus algo, an essentially identical algorithm was implemented in nim some years prior. iirc nim's impl freed memory at end of last scope instead of point of last use, so they couldn't do the mutability in place stuff koka did, and nim is also a jank ass one-person systems language project so it was never written up comprehensively

07:19 <discocaml> <omentic> interesting tidbit on the whole perceus algo, an essentially identical algorithm was implemented in nim some years prior. iirc nim's impl freed memory at end of last scope instead of point of last use, so they couldn't do the mutability in place stuff koka did, and nim is also a jank ass one-person systems language project so it was never written up comprehensively

10:45 <companion_cube> If you don't do reuse there's no interesting comparison to perceus, it's just refcounting then

15:25 <discocaml> <omentic> there's the ownership analysis to elide counts

15:27 <discocaml> <omentic> i consider that the most interesting part of perceus personally

21:13 <discocaml> <diligentclerk> companion-cube - "good luck to them, perceus doesn't like mutability" - This is interesting, can you elaborate on this? Koka should support mutation right

21:20 <companion_cube> I think it does mutation like a state monad, by passing state around?

21:20 <companion_cube> but that's my recall of perceus, it's not compatible with mutation or cycles