Welcome to the main channel on the development of MoarVM, a virtual machine for NQP and Rakudo (moarvm.org). This channel is being logged for historical purposes.
Set by lizmat on 24 May 2021.
Nicholas good *, * 07:16
jnthn nine: Wow, that's some progress :-) 11:18
timo i wonder when the right time would be to start building something on top of rakuast that's not "make tests work" 11:35
nine My guess at this point: just a couple of months after you joined building something that's "make tests work" ;) 11:36
lizmat an intermediate project could be rewriting the "assuming" logic in RakuAST ? 11:37
timo :D 11:59
jnthn I think most ideally with assuming we'd have some role mixed in wrap-style that marks out the argument transformation, and use new-disp to do the argument insertion (because that's exactly what new-disp is good at :)) and only lazily produce a signature if it's ever introspected. 12:05
That way, the calls to such functions would spesh into pulling the assumed arguments out of a constants table and then a normal boring call 12:06
lizmat jnthn: are you saying this should be independent of RakuAST ? 12:07
jnthn lizmat: RakuAST would be a decent way to produce the transformed signature 12:09
lizmat ah, ok, *phew* :-)
jnthn But at least on MoarVM it's a pity to not make use of new-disp to do the argument insertion
lizmat so I guess we would keep the old code for the other backends? 12:11
jnthn I think for other backends RakuAST could be used to straight up produce the wrapper 12:12
(including signature)
Hm, wrapper is a bad name for it. The delegator perhaps. :)
Most of the work *is* producing the signature anyway 12:13
lizmat that makes it sound like new-disp is a disadvantage / more work ?
jnthn More work, but producing the signature lazily will be way faster for cases that never introspect, just call, and the calls will be somewhat faster if new-disp can just have it as an argument capture rewrite 12:14
As with many things, one doesn't *have* to teach the VM about them via new-disp, but doing so can bring significant speedups :) 12:15
lizmat and yet another Rakudo Weekly News hits the Net: rakudoweekly.blog/2022/07/04/2022-...re-videos/ 12:32
nine lizmat: brilliant image choice :) 12:35
lizmat thanks... :-)
taken about 25k from here 12:36
*km
jnthn lizmat++
nine Did you have this use case in mind when taking it or did it occur to you later? 12:37
lizmat I had it in mind
other than cropping and a little colour correction, no other tampering was performed :-) 12:38
timo i wonder how many (if any) signatures we have serialized in the core setting that would be saved if assuing created signatures lazily 21:12