oh yeah sure, but all i'm saying there's a reason we moved away from the rigid 1.0 structure primari

oh yeah sure, but all i'm saying there's a reason we moved away from the rigid 1.0 structure primarily because we got complaints about the overloads kinda being naff to use

Kai•8/4/23, 9:02 PM

and exposing all of them is problematic because then you get no intellisense and maybe your IDE just dies

Kai•8/4/23, 9:03 PM

which... yeah that's super bad

KKai yeah this is great that we talk about this tbh

PerkseyOP•8/4/23, 9:03 PM

for sure, this is also why we refused to include overloads in the original 3.0 proposals haha

Kai•8/4/23, 9:04 PM

f(int***, void**, uint**, res*, float**)f(int***, void**, uint**, res*, float**)
here there is a very good argument that we should support passing
T[]T[] in place of T*T* everywhere, I think everyone agrees?
f(int**[], void*[], uint*[], res[], float*[])f(int**[], void*[], uint*[], res[], float*[]) should be a valid way to pass the args

Maxine•8/4/23, 9:04 PM

implicit casts to ptr might genuinely be a good alternative to having every permutation of argument as an overload

Maxine•8/4/23, 9:04 PM

but also doesn't work for every possible case

Kai•8/4/23, 9:04 PM

but then what about
f(int*[][], void[][], uint[][], res[], float[][])f(int*[][], void[][], uint[][], res[], float[][]) (now next to void[][] not being valid, so it should stay void*[]void*[])
these seem fine?

AAqua i don't see why it's something that couldn't be returned to, just because it was...

Lyris the Kitori•8/4/23, 9:04 PM

i feel like this is also an IDE problem

Lyris the Kitori•8/4/23, 9:04 PM

intellisense should pick up on implicit casts and give you nice completions

PPerksey oh yeah sure, but all i'm saying there's a reason we moved away from the rigid 1...

Aqua•8/4/23, 9:05 PM

I'm not familliar with 1.0 so I can only say it from my own thoughts
one way to do it could be to simply stick to the spec, and don't provide any helpers etc
raw pointer overloads just have the exact same signature as the C API
and the safe overloads are similar, just any time there is a lengthlength and a pointer, remove that and convert it to an array
or something along those lines
in this way, you only need 2 or 3 overloads (if you include spans)

Kai•8/4/23, 9:05 PM

now we want to pass different versions of this to different args, so you take those 3 permutations I've shown and get 15 overloads (5 * 3 versions = 15) (minus the one void*[] that's not valid = 12 overloads)

AAqua I'm not familliar with 1.0 so I can only say it from my own thoughts one way to ...

Kai•8/4/23, 9:06 PM

This is true, but very much not ergonomic. C APIs will most likely not have a length or have a shared length for several of the arguments. It's hard to figure out what's what and even harder to really put this in an API

AAqua i don't see why it's something that couldn't be returned to, just because it was...

Lyris the Kitori•8/4/23, 9:06 PM

while i agree with the "its less visually obvious to newcomers", once you learn the way its structured its a lot better ergonomically, its just teaching people that way, we'd need excellent documentation for this to really be viable

Kai•8/4/23, 9:06 PM

and it also doesn't really offer a good solution for passing int*[][]int*[][] or int*[,]int*[,]

AAqua I'm not familliar with 1.0 so I can only say it from my own thoughts one way to ...

PerkseyOP•8/4/23, 9:07 PM

tbh i'm content with having a wrapped overload and an unwrapped overload i.e. if you used only the unwrapped functions, Silk.NET is literally just TerraFX but with pretty names.

Aqua•8/4/23, 9:07 PM

if it's anything, these are the overloads I provide for creating a buffer in pie
https://github.com/piegfx/Pie/blob/main/src/Pie/GraphicsDevice.cs#L83-L122
they are not perfect and could do with a little refining, but they encompass pretty much everything one may want to do with them
the unsafe overload is there for those who want it, and there are safe options as well

PPerksey tbh i'm content with having a wrapped overload and an unwrapped overload i.e. if...

Maxine•8/4/23, 9:07 PM

yeah fuck it why not

PPerksey tbh i'm content with having a wrapped overload and an unwrapped overload i.e. if...

Kai•8/4/23, 9:07 PM

this I would prefer personally

KKai now we want to pass different versions of this to different args, so you take th...

Lyris the Kitori•8/4/23, 9:07 PM

yeah if your methods have a lot of thing* and you do array overloading you start getting to 3**11 overload territory

PPerksey tbh i'm content with having a wrapped overload and an unwrapped overload i.e. if...

Aqua•8/4/23, 9:07 PM

yeah, that could work quite well

PerkseyOP•8/4/23, 9:07 PM

most of our users don't want that though, so we will encounter the education problem still.

PerkseyOP•8/4/23, 9:08 PM

but again, my intention is to have a "Silk DSL 101" page that just tells you everything you need to know in using these wrappers

Maxine•8/4/23, 9:08 PM

silk documentation already isn't any good, it can't get worse

Kai•8/4/23, 9:08 PM

It's the same problem Span introduces though. These are advanced APIs, you spent hours digging through docs. Read the 2 lines the intellisense will give you once you hover over the Ptr type

PerkseyOP•8/4/23, 9:08 PM

again we literally can't ship 3.0 until we have docs

PerkseyOP•8/4/23, 9:08 PM

the Working Group has decided that

Aqua•8/4/23, 9:08 PM

if you need docs written you know who to ask heh

Maxine•8/4/23, 9:08 PM

what a daunting name

Maxine•8/4/23, 9:09 PM

the Working Group

Aqua•8/4/23, 9:09 PM

i've been meaning to write new tutorials I've just been juggling around 6 projects at once

MMaxine the Working Group

PerkseyOP•8/4/23, 9:09 PM

it used to be the ARB - API Review Board

AAqua yeah, that could work quite well

Lyris the Kitori•8/4/23, 9:09 PM

i feel like this line from zig zen shows some of my reasonings for being against that

 * Only one obvious way to do things.

 * Only one obvious way to do things.

Lyris the Kitori•8/4/23, 9:09 PM

we want people to use these safer types because they would be less error prone

Lyris the Kitori•8/4/23, 9:09 PM

and more readable

AAqua if it's anything, these are the overloads I provide for creating a buffer in pie...

Kai•8/4/23, 9:09 PM

yeah this makes sense. Overloading void* and IntPtr is a bit tricky (and in modern code you'd prefer nint but whatever)

Lyris the Kitori•8/4/23, 9:09 PM

we dont want newcomers to just fallback to the unsafe API when they would probably be happier with the safer one, if they learned how to use it

LLyris the Kitori i feel like this line from zig zen shows some of my reasonings for being against...

Kai•8/4/23, 9:10 PM

Yes, but this is out the window given all we're doing is providing one way for each way C# provides us

KKai yeah this makes sense. Overloading void* and IntPtr is a bit tricky (and in mode...

Aqua•8/4/23, 9:10 PM

yeah i have been moving towards nint, I wrote this code before I was aware that IntPtr == nintIntPtr == nintwhich is why it is like it is, i should change it

Kai•8/4/23, 9:10 PM

like there's multiple obvious ways to represent a data structure that does this in C#, and so we provide one obvious way to go from each possible obvious data structure design to one obvious call pattern in Silk

Kai•8/4/23, 9:10 PM

it's not on us to reduce the bullshit C# has layed upon itself for years.

AAqua yeah i have been moving towards nint, I wrote this code before I was aware that ...

Kai•8/4/23, 9:11 PM

I think (might be wrong) void* implicitly converts to nint so

LLyris the Kitori we want people to use these safer types because they would be less error prone

PerkseyOP•8/4/23, 9:11 PM

yeah this is my reservation in providing both. like yeah i'm content in doing so, but also

please use the better types
you can still use the better types in exactly the same way as if you weren't

Aqua•8/4/23, 9:11 PM

hmm you might be right

Maxine•8/4/23, 9:11 PM

think of it this way, if we were the BCL, we'd have two overloads, like they have overloads for byte[] and Span<byte>

Kai•8/4/23, 9:11 PM

not really

Kai•8/4/23, 9:11 PM

byte[] and Span<byte> are for legacy reasons

Maxine•8/4/23, 9:12 PM

i suppose that's true

PerkseyOP•8/4/23, 9:12 PM

then again it is my intention to have things like the ReSharper/Rider plugin help out a lot in saying "you're using unsafe, wanna use the better, more concise way?"

Kai•8/4/23, 9:12 PM

BCL has mostly decided that Span<byte> is a strict improvement to byte[] and done that everywhere.
I might be wrong / outdated here but I don't think new APIs are introduced with byte[]byte[] where SpanSpan/ROSpanROSpan make more sense

Kai•8/4/23, 9:12 PM

(sometimes maybe for equality with old legacy APIs)

oh yeah sure, but all i'm saying there's a reason we moved away from the rigid 1.0 structure primari

Similar Threads

Similar Threads

Similar Threads