3 ms·

Much of nalgebra's woes come from it trying to be n-dimensional without support for constants-in generics yet (this is in the works). To do this it does crazy w

by bjz_ 9y ago

Much of nalgebra's woes come from it trying to be n-dimensional without support for constants-in generics yet (this is in the works). To do this it does crazy wizardy using traits to model numbers at compile time. It's impressive for sure, but the compile errors are equally impressive. Which is why work is being done to add support for constants in generic parameters. But this is the case of one library. Another library to suffer from poor errors is the futures lib, but that will be solved with the `impl Trait` syntax that is in the works. Definitely agree it's a good idea to have a balance when it comes to generics, but likewise I'm glad folks are pushing the boundaries and finding room where things could be improved. It does take time for the language to catch up though, but they're working on it. Lots of good stuff is coming out of the current impl period.

seanmcdirmid 9y ago
Using classes or traits to model dimension numbers is something I did in C# when working on Bling 10 or so years ago. If you only have a few dimensions you need to handle anyways, it works out, and you can embed predecessor and successor dimensions in the types. Good days, though I guess I’m not that reckless anymore.
- bjz_ 9y ago
  Yeah, same in the C++ world with templates. They're just so handy, especially when you want performance! Thankfully C++'s `constexpr` has made this kind of thing much more tenable from a library consumer's perspective, and Rust will again be following in its footsteps, albeit with a less adhoc version based on dependent types. For what its worth, my cgmath library just has a bunch of hardcoded `Vector{2,3,4}`, `Point{2,3}`, and `Matrix{2,3,4}` types.
  - seanmcdirmid 9y ago
    Ya, that like is familiar. In C#, you can declare class D4 : Dim<D3,D4,D5> ...., and then use extension methods to access the D3 and D5 type parameters as previous and next dimensions. It all was nice and f-bounded as well. C# is an under appreciated language for these kinds of things.

by bjz_ 9y ago

seanmcdirmid 9y ago

Using classes or traits to model dimension numbers is something I did in C# when working on Bling 10 or so years ago. If you only have a few dimensions you need to handle anyways, it works out, and you can embed predecessor and successor dimensions in the types. Good days, though I guess I’m not that reckless anymore.

bjz_ 9y ago
Yeah, same in the C++ world with templates. They're just so handy, especially when you want performance! Thankfully C++'s `constexpr` has made this kind of thing much more tenable from a library consumer's perspective, and Rust will again be following in its footsteps, albeit with a less adhoc version based on dependent types. For what its worth, my cgmath library just has a bunch of hardcoded `Vector{2,3,4}`, `Point{2,3}`, and `Matrix{2,3,4}` types.
- seanmcdirmid 9y ago
  Ya, that like is familiar. In C#, you can declare class D4 : Dim<D3,D4,D5> ...., and then use extension methods to access the D3 and D5 type parameters as previous and next dimensions. It all was nice and f-bounded as well. C# is an under appreciated language for these kinds of things.