Stream: t-libs/wg-allocators

Topic: Next steps?


Tim Diekmann (Jan 23 2020 at 00:02, on Zulip):

After our first merge has landed (I assume it will get through the FCP), what do you think are the logical next steps? One possibility would be to merge alloc-wg step by step as far as possible (this excludes Box, Vec etc.), starting with the separation of AllocRef into DeallocRef, AllocRef: DeallocRef, and ReallocRef: AllocRef. Another option could be the associated error type. Both shouldn't change much, but It would be helpful to get feedback from the outside of this WG regarding both.
If we want to experiment even more, BuildAllocRef could be interesting, but only paired with changing the signature to self instead of &mut self.

Opinions?

Lokathor (Jan 23 2020 at 00:20, on Zulip):

One step, not sure where it goes in the process, is to announce very loudly that this is moving into Nightly and for people to please try it. Be as weird as possible with it. Push the design to its absolute limits and see what breaks.

We don't want to have a lack of feedback on this one. Allocators are really going to have a lot of influence on Rust at least for this entire edition, if not forever.

Tim Diekmann (Jan 23 2020 at 00:24, on Zulip):

not sure where it goes in the process

https://users.rust-lang.org/ is a must-have I guess. Also reddit is a suitable place IMO.

Lokathor (Jan 23 2020 at 00:28, on Zulip):

i meant I'm not sure how soon to shout about it

Tim Diekmann (Jan 23 2020 at 00:28, on Zulip):

Ah

Tim Diekmann (Jan 23 2020 at 00:31, on Zulip):

I think we should first implement some things that differ strongly enough from the current implementation. I think BuildAllocRef is the biggest change so far and also needs most fine-tuning and testing.

Error handling is also a big point I think.

As soon as we can touch Box, Vec, etc., it becomes important again.

Mike Hommey (Jan 23 2020 at 00:59, on Zulip):

It feels to me the shouting and the getting people to test could be done with the alloc-wg crate

Tim Diekmann (Jan 23 2020 at 01:41, on Zulip):

What about a hybrid approach? Pushing some things upstream first, make an announcement and link the alloc-wg crate for more features? I think if some features are merged into master, more people will test these out.

Lokathor (Jan 23 2020 at 08:00, on Zulip):

Having it be in Nightly and "automatically available" is a big deal, worlds better than telling someone to try the alloc-wg crate.

Anyone, at any time, can go make some random crate with whatever insane ideas about allocation they want to cook up.

But, if you've put something into Nightly, there's at least like 3 or 4 other people in the world (maybe more?) who think it's a good enough idea to let the experimentation into the official compiler.

It's not the final step in the path at all, but it's a landmark that the wider public will easily recognize as being a significant point of progress for this project. That's good PR. That's a positive step towards fixing one of the biggest areas where Rust still lags behind C/C++.

There's such a thing as over selling it, but this is a pretty important step if you stop and think about it much.

Simon Sapin (Jan 23 2020 at 09:21, on Zulip):

Allocators are really going to have a lot of influence on Rust at least for this entire edition, if not forever.

Simon Sapin (Jan 23 2020 at 09:21, on Zulip):

Not disputing this, but what sort of influence do you have in mind?

Tim Diekmann (Jan 28 2020 at 13:16, on Zulip):

We did it, the first change has landed on master!

Last update: Feb 25 2020 at 04:05UTC