add support for #[global_allocator]

[beepster4096]

This PR adds support for custom global allocators. Unfortunately, the code given in #1207 still causes errors when used with box. I believe this is because Box is special-cased in miri and stacked borrows.

[oli-obk]

Unfortunately, the code given in #1207 still causes errors when used with box

What is the difference between your test and the code in the issue that causes the failure? What kind of failure are we talking about? A stacked borrows failure?

[RalfJung]

Yeah, Box is a magic type. But I think the problem might be a different one -- that example code uses the "header before object base address" pattern, which simply does not work in Stacked Borrows (rust-lang/unsafe-code-guidelines#256). So this has little to do with Box.

[beepster4096]

Right, rustfmt refuses to work correctly on shims/foreign_items.rs. I'll remove the whitespace causing the format check to internal error, but we should find some workaround to make rustfmt work. (or fix rustfmt itself)

[RalfJung]

rustfmt refuses to work correctly on shims/foreign_items.rs

Interesting, I did not know this. Yeah might be good to report that with rustfmt.

[RalfJung]

This looks great. :) Could you squash the commits a bit?

[beepster4096]

I squashed the commits up.

[RalfJung]

Awesome, thanks a lot. :)

@bors r+

[bors]

📌 Commit e6a27a6 has been approved by RalfJung

[bors]

⌛ Testing commit e6a27a6 with merge adc26a3...

[bors]

☀️ Test successful - checks-actions
Approved by: RalfJung
Pushing adc26a3 to master...