Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

BREAKING CHANGE: Unify Math & Mathf. Make every method in Math polymorphic #2335

Open
wants to merge 14 commits into
base: main
Choose a base branch
from

Conversation

MaxGraey
Copy link
Member

@MaxGraey MaxGraey commented Jun 24, 2022

In this PR we're doing several breaking changes:

  • move all detailed implementations to std/util/math
  • make all Math's methods polymorphic
  • remove non-standard NativeMath#scalebn
  • remove non-standard NativeMath#rem
  • remove non-standard NativeMath#mod (but not tests)
  • remove NativeMathf completely
  • for now, we have only single Math.random (for f64 type)

Caveats:

  1. I have observed quite often the script that users from JS for example:
function foo(x: i32, y: i32): i32 {
   let res = Math.floor(x / y) as i32;
   return res;
}

Before this PR this will create unnecessary cast to f64, Math.floor and downcast to i32. With this PR this code now equivalent to x / y without all this unnecessary work due to Math.floor, Math.trunc and etc accept integers and pass through it "as is".

  1. Now following code will generate a compilation error:
const cos3 = Math.cos(3);
// -> ERROR: Math.cos accept only f32 or f64 types
// Solution: it may be fixed as Math.cos(3.0)

const x: i32 = 123;
const sqr = Math.sqrt(x);
// -> ERROR: Math.sqrt accept only f32 or f64 types
// Solution: Math.sqrt(x as f64)

This is the most painful caveat, which can lead to many breaking changes in some code.
Perhaps we could add new generic builtin helper which forces parameter type to float. Something like:

// builtin will be equivalent to this:
type OnlyFloat<T> = T extends f32 | f64 ? T : f64

function sqrt<T extends i32 | i64 | f32 | f64>(x: T): OnlyFloat<T> {
   if (isInteger<T>()) {
     return builtin_sqrt<f64>(x as f64) as OnlyFloat<T>;
   }
   if (isFloat<T>()) {
     return builtin_sqrt<T>(x) as OnlyFloat<T>;
   }
}

// now we can use integer inputs which will forced to f64 for outputs but safely preserve f32
const sqrt3 = sqrt(3); // infer as sqrt<i32>(x) -> f64

Thoughts?

  • I've read the contributing guidelines
  • I've added my name and email to the NOTICE file

@MaxGraey MaxGraey marked this pull request as ready for review June 25, 2022 09:01
@MaxGraey MaxGraey requested a review from dcodeIO June 25, 2022 09:01
@dcodeIO
Copy link
Member

dcodeIO commented Jun 25, 2022

So far, Math is quite intentionally identical to JS Math, also because the implementations can be switched with --use. Making one generic and the other not seems problematic in this regard. Similarly, this might break portable code relying on both being identical. The mentioned caveat that integer literals will be inferred as i32, which sometimes might either not work or produce unexpected results in the absence of a type parameter, seems quite unfortunate as well. Are you sure this is actually viable?

@MaxGraey
Copy link
Member Author

MaxGraey commented Jun 25, 2022

abs, floor, ceil, round etc will be work properly with integers even in JSMath due to i32 -> f64 -> i32 is invariant. But u32 -> f64 -> u32 or i64 -> f64 -> i64 is not invariant, so this may be a problem. However, I don't think ppl will use that functions for such types. Also, I'm going to add some compiler warnings for integer paths for according methos.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants