Stream: wg-traits

Topic: caching cycles #60010


nikomatsakis (May 01 2019 at 16:00, on Zulip):

so over in #t-compiler, I noted a problem with the way that we are caching things in our current trait evaluator

nikomatsakis (May 01 2019 at 16:00, on Zulip):

the problem has to do with cycles

nikomatsakis (May 01 2019 at 16:01, on Zulip):

I'm kind of chatting on here about possible solutions :)

nikomatsakis (May 01 2019 at 16:04, on Zulip):

anyway I mean the obvious thing is to do the usual algorithm for detecting SCCs, I think

nikomatsakis (May 01 2019 at 16:05, on Zulip):

basically I think we would not cache anything that depended on a cycle until we reach the "head" of the cycle in the stack, and then we'd cache everything at once (as appropriate)

nikomatsakis (May 01 2019 at 16:19, on Zulip):

implementing this though is gonna' be a bit annoying

nikomatsakis (May 01 2019 at 16:19, on Zulip):

I'm also worrying a bit about the SLG solver in chalk-engine; want to be sure it doesn't fall into the same trap... I feel like it may.

Last update: Nov 12 2019 at 17:00UTC