I don't really get why there's two? Why is the one without a return arg restricted to floating point

I don't really get why there's two?
Why is the one without a return arg restricted to floating point? Seems like extra typing for user for nothing?

KKai I don't really get why there's two? Why is the one without a return arg restrict...

Maxine•10/21/23, 9:03 PM

because there is no use of calling Length<T> if the result is not a decimal

khitiara•10/21/23, 9:04 PM

the one without a result is a shortcut if T is already floating

khitiara•10/21/23, 9:04 PM

and you just want to use that

Maxine•10/21/23, 9:04 PM

and because there is no square root for integers

MMaxine what is that comment about, though?

Perksey•10/21/23, 9:04 PM

basically no rules were defined for generic maths when Silk.NET.Maths when it was created, and we made no attempt to define them, instead allowing everything in the absence of rules set by an ecosystem authority (a la .NET API Review).
this led to us defining our own Add, Subtract, etc in Scalar.BaseOps, and implementing every algorithm under the sun in terms of those fully generalised BaseOps regardless of whether its correct, makes sense, or would yield the correct results for a given type.

KaiOP•10/21/23, 9:04 PM

On length I understand the roots interface, but on that and on square I don't get the floating point arg

khitiara•10/21/23, 9:04 PM

the one with a result could be like Vector3<int>.Length<float>()Vector3<int>.Length<float>()

khitiara•10/21/23, 9:04 PM

oh yeah the static lengthsquared has the float bit

khitiara•10/21/23, 9:04 PM

idk why

khitiara•10/21/23, 9:04 PM

might be a mistake

PPerksey basically no rules were defined for generic maths when Silk.NET.Maths when it wa...

Maxine•10/21/23, 9:04 PM

alright
so are you for or against Vector3<int>?

Perksey•10/21/23, 9:05 PM

i am against not looking at it at an individual operation level

KaiOP•10/21/23, 9:05 PM

Like imo we need a really good argument for restricting T beyond what's required for the implementation

KKai On length I understand the roots interface, but on that and on square I don't ge...

Maxine•10/21/23, 9:05 PM

on square, it's because it's a likely overflow

KaiOP•10/21/23, 9:05 PM

But so is add...

Maxine•10/21/23, 9:05 PM

not really?

khitiara•10/21/23, 9:05 PM

that shouldnt be a generic restriction though

khitiara•10/21/23, 9:05 PM

like thats an at your own risk thing

Perksey•10/21/23, 9:05 PM

dot product can be implemented for int, so LengthSquared can be implemented for int, but the BCL has decided we should not be able to square root an int, so we should not implement length

KaiOP•10/21/23, 9:06 PM

Or multiply? The argument is that squaring a byte is likely to overflow?

Maxine•10/21/23, 9:06 PM

ok, so pick one

both LengthSquared<T> and LengthSquared<T, TReturn>
only LengthSquared<T>

khitiara•10/21/23, 9:06 PM

both but no type restriction, because the treturn version can be good to implement length with

khitiara•10/21/23, 9:07 PM

is my vote

Maxine•10/21/23, 9:07 PM

okay

Maxine•10/21/23, 9:07 PM

i'm just asking because i don't want the proposal to have 8000 combinations

PPerksey dot product can be implemented for int, so LengthSquared can be implemented for ...

Perksey•10/21/23, 9:07 PM

now, i'm not against figuring out an API that allows the user to specify that they want an implicit upcast/downcast for a given operation, as this is what we did for some ops in some cases without the user knowing, but this way it codifies intent.

Maxine•10/21/23, 9:07 PM

noted

KaiOP•10/21/23, 9:07 PM

I like that I can specify the return type, but I don't understand restricting to floating point

khitiara•10/21/23, 9:07 PM

yeah

khitiara•10/21/23, 9:07 PM

that restriction makes no sense to me either

khitiara•10/21/23, 9:07 PM

for the squared

khitiara•10/21/23, 9:08 PM

because i could just as easily do a Vector3<short> and the do .LengthSquared<long>.LengthSquared<long>

khitiara•10/21/23, 9:08 PM

or whatever

Perksey•10/21/23, 9:08 PM

for LengthSquared no restriction makes sense for the return, you can implement Dot on any INumber and therefore any INumber is a valid return.

khitiara•10/21/23, 9:08 PM

yeah

KaiOP•10/21/23, 9:08 PM

for normal not either - the root restriction makes sense.
If BCL adds root capabilities to sbyte we should get it implicitly

khitiara•10/21/23, 9:09 PM

yeah

khitiara•10/21/23, 9:09 PM

requiring IRootFunctions for Length makes sense instead of specifically requiring IFloatingPoint because it lets other non-float types that have sqrt well defined can use it

khitiara•10/21/23, 9:10 PM

like complex probably could for instance

khitiara•10/21/23, 9:10 PM

if that ever gets that

khitiara•10/21/23, 9:10 PM

in general my opinion is we want the actual generic constraints to be as loose as possible

Perksey•10/21/23, 9:10 PM

^^^^^^^

Perksey•10/21/23, 9:10 PM

but still while respecting what the BCL have decided (i.e. "oh well i'm just going to require INumber for everything and reimplement everything that's not allowed myself because we want it to be as loose as possible" is not a valid take)

khitiara•10/21/23, 9:11 PM

yeah

khitiara•10/21/23, 9:12 PM

if the natural implementation of a (probably extension) function requires an operation imo we should have the generic constraint be the loosest interface for that, so just INumber and IRootFunctions for length for example

khitiara•10/21/23, 9:12 PM

im fine with requiring either inumber or inumberbase on everything but beyond that keep as loose as possible

khitiara•10/21/23, 9:14 PM

if we had true generic specialization or the extension objects thing id want to go all-out on the looseness bit

Perksey•10/21/23, 9:14 PM

but then you have Silk.NET.Maths 2.X haha

Maxine•10/21/23, 9:14 PM

detroit become C++

KaiOP•10/21/23, 9:15 PM

Kind of fundamental - why have we decided to restrict base Vector to INumber?

I don't really get why there's two? Why is the one without a return arg restricted to floating point

Similar Threads