Add custom allocator support

[bolshoytoster]

Closes #55.

This doesn't add support in the following APIs:
From
from_elem
Default
from_buf
from_buf_and_len
from_buf_and_len_unchecked
from_vec
from_raw_parts
splice
from_slice_copy
FromIterator

new, try_with_capacity and with_capacity now have _in counterparts, to match the standard library.

PartialEq does not support comparing two SmallVecs with different allocators (this is different from std Vec). This shouldn't matter too much, since you can just compare them as slices instead.

You also can't append two SmallVecs with different allocators, this matches std Vec.

[bolshoytoster]

I'm not really sure how to read the fuzz failure. It looks like one of the tests times out, but I can't seem to install honggfuzz locally to test (TRUE being undefined in a c file)

[fereidani]

Try to rerun the CI with a force push: git commit --amend --no-edit && git push --force

[bolshoytoster]

Oh it might be due to a change I made to grow. It used to be able to shrink allocations if new_capacity < old_capacity, but since the allocator API has separate grow and shrink functions (no realloc), I split it into two matching functions, and grow now panics if you give it a smaller capacity.

Should I revert the behaviour and just have a if new_capacity < old_capacity { /*shrink*/ } else { /*grow*/ }?

[fereidani]

I'm a contributor like you, I think project members should comment on it.
But as your PR title is about custom allocator support, I would suggest to keep it as minimal as possible and direct to the point without no extra behavior changes.
Then check rust-lang/rust alloc behavior if smallvec does not behave the same, send another PR and fix that issue separately.

[alejandro-vaz]

yeah probably just revert grow behavior with that if statement, it's what realloc does somewhat internally anyway

I have absolutely no idea what's happening with fuzzing. I haven't touched nor taken a look at it since I'm here, and I don't believe anyone here now has

I can't understand the traceback neither

I'll give it a look tomorrow

why doesn't equality allow for different allocators though??

[bolshoytoster]

why doesn't equality allow for different allocators though??

It's just impl <T, U, A: Allocator, const N: usize, const M: usize> PartialEq<SmallVec<U, N, A>> for SmallVec<T, M, A>, I wasn't sure if adding an A2 would make it too much more complicated.

[alejandro-vaz]

as long as it's correct why not do it??

[alejandro-vaz]

did fuzzing solve itself or did I just miss something

[bolshoytoster]

did fuzzing solve itself or did I just miss something

The problem was the change in grow.

[alejandro-vaz]

I've taken a look at the diff, looks good

I'll review it properly tomorrow.