Stream: wg-traits

Topic: leak-check removal


nikomatsakis (Feb 10 2020 at 19:24, on Zulip):

So @Matthew Jasper -- now that the first leak-check PR landed, one of the things I wanted to do was to enumerate the next steps. I think the things on my mind are:

nikomatsakis (Feb 10 2020 at 19:24, on Zulip):

On the latter point in particular, I was curious if you agreed with my reasoning on why we probably should be accepting those impls

Matthew Jasper (Feb 10 2020 at 19:54, on Zulip):

I agree that dyn for<'a> Fn(&'a A) and dyn Fn(&'r A) are not equal for any choice of 'r. I had two concerns:

nikomatsakis (Feb 10 2020 at 20:40, on Zulip):

@Matthew Jasper yes, I agree on the second point, and I remember that this specifically was a reason I wanted to be "maximally cosnervative" here. i.e., to give us room to get smarter about which regions outlive which other ones

nikomatsakis (Feb 10 2020 at 20:40, on Zulip):

or at least, I agree that's something I want to think carefully about

nikomatsakis (Feb 10 2020 at 20:41, on Zulip):

as to the first point, I guess one way to go about this might be to rework the leak-check to be much more conservative about what it rejects

nikomatsakis (Feb 10 2020 at 20:41, on Zulip):

that would apply equally

nikomatsakis (Feb 10 2020 at 20:41, on Zulip):

but it still requires us to wrestle with the second point, I have to look carefully at those types you gave as examples :)

Matthew Jasper (Feb 10 2020 at 20:52, on Zulip):

as to the first point, I guess one way to go about this might be to rework the leak-check to be much more conservative about what it rejects

It does seem like "a worse leak check" is roughly what this would have to be, for now.

nikomatsakis (Feb 10 2020 at 21:02, on Zulip):

"worse"?

nikomatsakis (Feb 10 2020 at 21:02, on Zulip):

I'd have said "better" -- i.e., not rejecting things that are valid :)

Last update: Feb 25 2020 at 04:25UTC