Ugh @oli how did we not land this PR yet. (I'm going through review queue)
I remember we looked at it together...
should I try to look at it again today and see if there's anything that I could help with?
my memory is that the PR itself was fairly harmless but that we needed (separately) to do a review of how normalization/canonicalization played into constants
I'm lokoing over again to try and jog my memory
ah I see
I had left some comments that oli didn't resolve
we need to fail gracefully everywhere, we can't ICE in any of this
(well, we can ICE in codegen, I guess. but not in inference)
@nikomatsakis should I wait for @oli or try to rebase the PR myself and make the changes I suggested?
I don't know. I'm having trouble remembering the details of this now.
@eddyb hmm I wonder if it makes sense to try and do an impromptu discussion this morning (perhaps in ~3 hours?) to talk about the overall strategy around normalization, constants, etc?
Failing to lift would intern?
Or we should never lift something that isn't already interned?
lift_to_global doesn't even exist anymore, now you have to check it some other way. but it was about inference variables, all along
if you have infer vars, you can't pass them to a query (we should maybe enforce this in the query engine, if we can do it cheaply, idk)