#rust-lang

aturon, woboats: very little is needed in const eval now. we don't even need to switch to miri, I don't think

and not supporting expressions means banning them, artificially (instead of giving them the projection semantics)

eddyb, we dont need to improve const eval to support expressions?

thats great

woboats: well, we did. it happened already :P

most of it was type-checking/inference before evaluation, and on-demand in general

that's when associated constants that don't depend on type parameters started working in array lengths

got it

how much of an effort would you say it is to make const params work?

woboats: well, so, have I shown you https://public.etherpad-mozilla.org/p/rust-comp...

I don't think there's anything difficult about the most conservative approach which does no HIR/MIR introspection of expressions

nrc: still up?

just the usual tedium of adding new core features to the compiler

but, e.g. ty::Kind is a tagged pointer with the lowest 2 bits used as a tag and only 2 of the 4 possible tags are used up (lifetimes and types)

if something I had to change recently makes implementing const generics harder, I screwed up :P

I don't know what chalk will look like in rustc though, that might be interesting

although I'm not sure chalk even needs to be able to tell between kinds