Support for #![no_std]?

[mitchmindtree]

Hi there! I'm wondering if there is any interest in support for #![no_std] in noise?

I often come across use cases where I'd like to use noise functions in #![no_std] environments. So far these have included:

• embedded devices for art installations
• embedded audio projects
• rust-gpu shaders
• writing libraries which aim to support the above

I haven't taken a close look yet to get an idea of how much work would be involved, but I'm curious to get thoughts!

[Razaekel]

Thank you for the question! I had occasionally wondered about making the library no_std compatible, but never really had the motivation to go do it.

It didn't take too long to hammer it into shape. Check out the no-std branch. It still defaults to including std, so turn off the default features so that it doesn't include std.

[mitchmindtree]

Nice one, thanks for the quick branch!

I haven't tested the functionality, but depending on the feature/no-std in a downstream no_std crate with default-features = false appears to work nicely 👍

Any chance of seeing this land in 0.7?

[kpreid]

It looks like noise is still indirectly depending on std, through the default features of rand and num-traits. Disabling those default features and building on a std-less target (cargo check --target thumbv6m-none-eabi) reveals further constraints:

• The version of rand needs to be upgraded or it will depend on std via getrandom even with default features off.
• num_traits::Float and num_traits::Real don't exist without std; num_traits::float::FloatCore is mostly an adequate replacement.
• sqrt(), sin() and cos() are not available without std. This could be addressed by disabling certain noise functions and other items (but Worley noise can't be constructed without distance_functions::euclidean, which uses sqrt), or the libm crate offers software replacements that could be used.
• Floats don't implement num_traits::MulAdd. Some of the conditionals for target_os = "emscripten" already help with this if modified to also cover not(feature = "std").

As you might imagine from the length of this list, I gave fixing them a try, but there's it's not a strictly backwards compatible change, so I thought I'd provide this report and ask whether I should send PRs with the incomplete but straightforward parts.