Minor incompatibility with bytes

[AdamGS]

First of all, this is an awesome crate! really cool to see people improving on some of the most popular crates in the ecosystem.
I tried arc-slice in Vortex, where we make very heavy use of Bytes. While it does seem to make a difference in some benchmarks, it also surfaced an issue where spare_capacity_mut is market unsafe here but isn't in bytes. I'm not if compatibility is a major goal here but I figured its worth reporting.

[wyfo]

You're right, spare_capacity_mut is unsafe in the port of bytes using arc-slice, and it was in fact done on purpose, as I suspect BytesMut::spare_capacity_mut to be unsound, and not having it unsafe for ArcSliceMut is definitely unsound.
Why is it unsound? Because spare_capacity_mut, combined with set_len, allows to write uninitialized memory to the buffer.
I'm currently asking if it is an issue for a Vec buffer (that BytesMut is using), but that is clearly an issue for arbitrary buffers, which ArcSliceMut supports. I may also open an issue on bytes repository, depending on the answer I get.

That being said, while in previous drafts I was also storing the Vec buffer directly in the Arc, I made a recent change to store only the pointer + capacity in case of Vec (if needs_drop::<T>() returns false). So technically, I'm protected about the unsoundness of writing uninitialized memory in my BytesMut reimplementation, and I can make BytesMut::spare_capacity_mut safe. I will try to push a new release tonight with this fix, making the API 100% compatible.

In any case, thank you a lot for your feedbacks, and for having tried arc-slice.

[wyfo]

0.1.0-alpha.2 has been published.