rethink the purpose of the `smallvec!` / `smallvec_inline!` macros

[alejandro-vaz]

ft. #424 @bolshoytoster

these macros are essentially wrappers around the associated functions SmallVec::from and SmallVec::from_buf.

rust policy with macros, as written on the rust book, is simple: don't make a macro out of which can be made as a function

thus they seem very redundant

their origins trace to Vec API parity and a hack that had to be placed in 2021 in order to please the CTFE

conceptually:

smallvec![1u8, 2, 3] ~= SmallVec::from([1u8, 2, 3]);

smallvec_inline![1u8, 2, 3] ~= SmallVec::from_buf([1u8, 2, 3]);

alternatives

• keep the macros as they are now
• redesign the smallvec_inline! macro to be a wrapper of ::from_buf as smallvec! is of ::from
• remove the macros and accept that their only reason of being was Vec API parity

[alejandro-vaz]

@bolshoytoster would love your thoughts

[bolshoytoster]

My main concern is that removing the macros might add extra work for people migrating from v1 -> v2, but smallvec![1, 2, 3] -> [1, 2, 3].into() shouldn't be too bad.

Another thing is that I'm not sure how well the compiler will optimize this for large arrays (e.g. SmallVec<u8, 256> as From<[u8; 65536]>) especially in debug builds.

It might be nice to have some documentation saying to use SmallVec::from_buf([1, 2, 3]) for arrays that fit inline, and SmallVec::from_vec(vec![1, 2, 3]) otherwise.

[alejandro-vaz]

@bolshoytoster yeah it's true that it might add extra work for the migration, but I don't think we should keep around a macro which has lost its purpose just to satisfy backwards-compatibility (we don't have to think about that now, it's v2, it's a rewrite)

if we commit that to v2 stable then that's basically unremovable

it looks to me we're keeping technical debt

I think the compiler would optimize it much better than we think. there's a reason why arrays are copy when the type is copy, it's very cheap for reasonable sizes. and there's a very real chance that the compiler is inlining it and not copying anything, though we'd have to inspect assembly

It might be nice to have some documentation saying to use SmallVec::from_buf([1, 2, 3]) for arrays that fit inline, and SmallVec::from_vec(vec![1, 2, 3]) otherwise.

yeah that'd be good