bubble up out-of-memory errors from liballoc

[thestinger]

This makes the low-level allocation API suitable for use cases where
out-of-memory conditions need to be handled.

Closes #18292

[breaking-change]

[rust-highfive]

Warning

• These commits modify unsafe code. Please review it carefully!

[cgaebel]

That seems bad

[huonw]

This needs a rebase, also, the documentation of alloc explicitly states:

This is the lowest level library through which allocation in Rust can be performed where the allocation is assumed to succeed. This library will abort the process when allocation fails.

So this needs to be updated, and maybe the design decision for this choice should be addressed. (git blame implies @alexcrichton wrote that piece of documentation.)

[thestinger]

@huonw: It's wrong because reallocate_inplace doesn't abort on failure. It wasn't a design decision but rather an existing assumption that I had to uphold when I wrote originally wrote this code.

[thestinger]

There's still not much you can do if you intend to handle out-of-memory conditions, but we can work towards moving the problem further up the stack to make Rust more usable in niches like kernels. For example, there's no reason that Box<T> and Rc<T> couldn't expose a way to handle out-of-memory.