Rename OnceCell methods to be more consistent with std conventions

[matklad]

WDYT?

r? @ghost

[matklad]

r? @KodrAus
cc @m-ou-se
cc #74465

[m-ou-se]

For Option we use get_or_init (and get_or_init_with): https://doc.rust-lang.org/stable/std/option/enum.Option.html#method.get_or_insert_with

get_with (without or) sounds like the closure is always run.

[matklad]

Hm, I guess than it makes sense to use

get_or_insert_with
try_get_or_insert_with

for OnceCell?

The benefit is that it exactly matches the Option's method (and has the same semantics).

The drawback is that it is long -- that's OK for Option, as it is one of it's rarely used methods. For OnceCell, however, that's the main API...

[KodrAus]

I was just coming to say the same thing 🙂 I don’t personally think get_or_init_with is unreasonably long.

[matklad]

rerenamed!

[m-ou-se]

So would a future get_or_init take the init value directly, instead of from a closure, just like Option? Or only have an get_or_init_with and no by-value counterpart?

[matklad]

Yup, I think we can have get_or_insert(&self, value: T) -> &T, as a reference-returning counterpart of set. We might bikeshed the return type (ideally, it is Result<&T, (&T, T)>)

[matklad]

CI fails because, apparently, one does not simply deprecate unstable std API. I'll remove the commit than...

[m-ou-se]

I think #[rustc_deprecated] (instead of #[deprecated]) would work on an unstable api.

[KodrAus]

A by-value get-or-init API doesn’t seem very useful for these, have you ever come across use-cases that would want one? I think the _with suffix is a good change though!

I also didn’t think we needed to be consistent with Option and collections’ insert vs the once cells’ init, because in the context of once cells we are initializing the container. We have better terminology than insert to reach for which also has the nice benefit of being a little shorter.

[KodrAus]

Actually... Thinking about it you might want something like that if each caller might initialize with a different value. In that case you’re not worried about expensive initialization, but who got in first.

[matklad]

Ok, it seems there's no quick consensus here... Let me register this as an unresolved question in the tracking issue.