Add #![no_std] support

[whitequark]

Fixes #138.

This was surprisingly easy. There are essentially no meaningful changes required.

[coveralls]

Coverage remained the same at 94.063% when pulling 8c295a7 on whitequark:master into 7025c09 on gimli-rs:master.

[coveralls]

Coverage remained the same at 94.063% when pulling c3194fd on whitequark:master into 7025c09 on gimli-rs:master.

[philipc]

Is there a reason for this to use a nightly feature, instead of not(feature = "std") like we do elsewhere? It'd be nice if cargo build --no-default-features could work.

[whitequark]

Yes. See Cargo.toml:

nightly = ["fallible-iterator/alloc"]

[philipc]

Ok, this is probably nitpicky, but I just want to make sure I understand if you don't mind:

• do we need to use fallible-iterator/alloc in gimli?
• why don't the uses of the alloc crate depend on the nightly feature?

[whitequark]

do we need to use fallible-iterator/alloc in gimli?

Yes. Otherwise, it cannot be tested without the std feature, I believe.

[philipc]

Thanks!

[fitzgen]

Thanks for adding no_std support @whitequark ! Was the btreemap vs hashmap related to no_std or was it for performance? And if the latter, what kind of difference did you observe?

[whitequark]

@fitzgen Just #![no_std]. There is no hash-anything without std, because they implicitly rely on the global state (the default RNG). This is actually very annoying, but oh well!

[fitzgen]

Ah, yeah that makes sense. Thanks again!