Support for const types

[ApoorvaJ]

Currently, constant types don't seem to be supported:

error[E0015]: calls in constants are limited to constant functions, tuple structs and tuple variants
  --> src/main.rs:
   |
17 | const INIT_POS: Vec3 = vec3(1.0, 1.0, 1.0);
   |                        ^^^^^^^^^^^^^^^^^^^

I can't construct the tuple manually either, since there is only crate-level visibility. Would it be possible to implement this?

[bitshifter]

I can't do this across the board because the intrinsics that glam uses are not const fn. I'm a bit reluctant to do it to some types and not others because it would be inconsistent which could be confusing. As soon as intrinsics get const fn I would be keen to use it.

In the interim you could use an array or tuple are truly in the into/from implementations.

[bitshifter]

I guess I could do some union hacks to avoid calling the intrinsic functions. Last time I played around with doing that it was a bit of a perf hit so it might not pan out, should be possible though.

[ApoorvaJ]

It's definitely not a pressing issue. I can easily work around it. I really like glam for its simplicity and uniformity. Just wanted to check whether this was an oversight or a limitation. :) Feel free to close this issue as you see fit.

[bitshifter]

No worries, it's something I would like to address. Will leave this open and try out the union cast thing some time.

[bitshifter]

I tried this out but unfortunately performing union casts and other unsafe casts is not allowed in const fn, tracking issue is here rust-lang/rust#51909.

It might be a while until Rust gets that feature but I'll leave this open for now anyway.

[FSMaxB]

Rather than making operations const fn, would it be possible to make just the constructors const fn?

[bitshifter]

That is what I was attempting to do. The simd types like __m128 are opaque and you need to call functions like _mm_set_ps to initialise them. Those functions are not const. I don't know if there's any reason those std lib intrinsic functions couldn't be made const though.

[FSMaxB]

That is what I was attempting to do.

Oh, ok.

[bitshifter]

I raised an issue with stdarch about this. Let's see where it goes. rust-lang/stdarch#906

[bitshifter]

It looks like it should be possible to make the constructors in std::arch const fn. That might take a while do though.

In the interim I've added a bunch of macros that can be used to create const values of all glam types:
const_mat2, const_mat3, const_mat4, const_quat, const_vec2, const_vec3, const_vec3a and const_vec4.

Internally these perform a union cast which is allowed for const values but is not allowed in a const fn in stable Rust.

This is just in git for now, at some point I'll make a new release to crates.io. I still need to write some docs and make some other changes. Let me know if you try them out.

[alice-i-cecile]

I would also be grateful to see some more utility functions for vectors made const where possible, such as .normalize <3

[bitshifter]

@alice-i-cecile I don't believe that's possible right now either. Floating point arithmetic is not allowed in a const fn context in stable Rust. See rust-lang/rust#57241. I guess if that issue is resolved, glam would also need most SSE2 functions in std::arch to also be made const fn. It could happen eventually but it might be a while.

[alice-i-cecile]

Darn, that's fair enough. I'll keep my eyes on that issue too then.

[SUPERCILEX]

Is it possible to make just the ::new functions const so the macros aren't needed?

[bitshifter]

Not at the moment, see #76 (comment)

[SUPERCILEX]

::new shouldn't be doing arithmetic though, right?

[bitshifter]

It does not, as the comment above says, the SSE init method is not const fun. This needs to be addressed in the std library

[SUPERCILEX]

Ohhhhh, I see, thanks.

[bitshifter]

@SkiFire13 pointed out unions can be used in a const eval context now so this issue is finally resolved as part of #294.