Stream: wg-traits

Topic: leak check, regionck


nikomatsakis (Jan 13 2020 at 14:49, on Zulip):

So @Matthew Jasper I've rebased the leak check PR. I think it's kind of a crucial PR to land for a bunch of other changes, but I'm debating what needs to be done in the PR.

I'd also (relatedly?) like to discuss the removal of regionck.

nikomatsakis (Jan 13 2020 at 15:33, on Zulip):

Two thoughts in this regard:

It occurred to me that we might be able to do a warning period, but I'm not sure just how hard it would be.

nikomatsakis (Jan 13 2020 at 15:33, on Zulip):

@Simon Sapin I have a question for you =) I believe servo is still relying on some kind of hack around fn(): PartialEq?

nikomatsakis (Jan 13 2020 at 15:36, on Zulip):

Or some such trait?

Matthew Jasper (Jan 13 2020 at 18:26, on Zulip):

I agree that we should have a write up of where we are going with regionck/lexical_region_resolve.

Matthew Jasper (Jan 13 2020 at 18:30, on Zulip):

I think that we probably need a warning. Maybe if it's really only servo...

nikomatsakis (Jan 13 2020 at 18:33, on Zulip):

I agree that we should have a write up of where we are going with regionck/lexical_region_resolve.

It seems like we need both a plan from a technical perspective but I would also like to have a write-up of how region handling is expected to work from a more "language" perspective; kind of like what we were discussing earlier, where I was proposing that chalk could (in theory) fully handle higher-ranked obligations.

I guess that in the meantime the existing PR is kind of a "maximally conservative" plan, in that it is putting all the logic into the region solver.

nikomatsakis (Jan 13 2020 at 18:33, on Zulip):

I think that we probably need a warning. Maybe if it's really only servo...

Yeah so i'm pondering how to implement that. One thing we could do is to keep the leak-check. We can use it to have coherence test things both ways.

nikomatsakis (Jan 13 2020 at 18:35, on Zulip):

But it seems like if we're going to have a warning period

nikomatsakis (Jan 13 2020 at 18:35, on Zulip):

We probably need to use leak-check in most of the code

nikomatsakis (Jan 13 2020 at 18:35, on Zulip):

Until we turn it off everywhere

nikomatsakis (Jan 13 2020 at 18:35, on Zulip):

That's not that bad in that it would give us time to fixup NLL region check too

Matthew Jasper (Jan 13 2020 at 22:44, on Zulip):

How do you want to work on the plan? Should we schedule a time to talk about this? Do you want to write up some more thoughts on this first?

nikomatsakis (Jan 14 2020 at 20:06, on Zulip):

Hey @Matthew Jasper so I have a little time (30min) now.. I think otherwise the best thing for scheduling a time to chat would be to schedule a time on Friday

nikomatsakis (Jan 14 2020 at 20:06, on Zulip):

er, actually, I would probably talk tomorrow

nikomatsakis (Jan 14 2020 at 20:06, on Zulip):

starting at 14:30 UTC-04:00 (Boston time)

Matthew Jasper (Jan 14 2020 at 20:07, on Zulip):

I'm available now

Matthew Jasper (Jan 14 2020 at 20:15, on Zulip):

Tomorrow it is then.

nikomatsakis (Jan 14 2020 at 21:17, on Zulip):

Hey, sorry, got pulled off again :P

nikomatsakis (Jan 14 2020 at 21:17, on Zulip):

Let me try to leave a few notes

nikomatsakis (Jan 14 2020 at 21:17, on Zulip):

First off, I think adding a leak-check warning period seems doable, as I noted, I guess I would have some sort of "flag" to skip leak-check that...can be set on the infcx? and cleared?

nikomatsakis (Jan 14 2020 at 21:18, on Zulip):

But in terms of moving away from regionck, the first thing is probably to enumerate all the places that lexical regionck is used, as they are going to have move towards the "NLL checker" (or whatever replaces it)

nikomatsakis (Jan 14 2020 at 21:18, on Zulip):

Although I suppose that we could conceivably keep the lexical checker around and just not use it from type-check?

nikomatsakis (Jan 18 2020 at 12:30, on Zulip):

So @Matthew Jasper I'm working on the "leak check removal warning" PR

Matthew Jasper (Jan 18 2020 at 12:31, on Zulip):

ok

nikomatsakis (Jan 18 2020 at 12:35, on Zulip):

feeling sad about it. I want to remove it entirely :P

nikomatsakis (Jan 18 2020 at 12:35, on Zulip):

but I think a short warning period is right

Last update: Jan 21 2020 at 08:25UTC