i am slightly concerned that this is becoming too far away from the language's design remember - mos

i am slightly concerned that this is becoming too far away from the language's design
remember - most people (myself included) aren't gonna want to learn and use a bunch of custom types that aren't language constructs that do basically the same thing

MMaxine IT'S ALL INTEGERS PERKSEY

Perksey•10/17/23, 9:29 PM

AAqua i am slightly concerned that this is becoming too far away from the language's d...

Perksey•10/17/23, 9:29 PM

then just pretend they're all unmanaged pointers...

Perksey•10/17/23, 9:29 PM

the only place where it'll hurt is return typing

Perksey•10/17/23, 9:30 PM

everywhere else it's basically just designed to be equivalent to what we have in Silk.NET today where you can pass in whatever, just without millions of overloads

AquaOP•10/17/23, 9:31 PM

i'm trying to think from the perspective of a relative newb
where they will be used to arrays or maybe spans
you can teach them what a pointer is and how its worked, but give them PtrMutPtrPtrMutPtr or what have you? i'm not so sure
i can certainly say that I stopped being able to wrap my head around the pointer structs as soon as it got beyond Ptr<T>Ptr<T>, which on a personal level is all that I think is needed
where Ptr<T>Ptr<T> is a high level pointer to just stuff

Perksey•10/17/23, 9:32 PM

yes but you can't do that because it's a ref struct

Perksey•10/17/23, 9:32 PM

unless we want to discard strong typing entirely

Perksey•10/17/23, 9:32 PM

which... we can do but it feels wrong

Perksey•10/17/23, 9:32 PM

and prone to misuse

Perksey•10/17/23, 9:32 PM

more prone to misuse than if our users just read the docs

AquaOP•10/17/23, 9:33 PM

it's a library binding to opengl and other low level libraries
they're the misuse capitals of programming
i think trying to go this far to prevent misuse is a bit of a waste of time, people will misuse it regardless

AquaOP•10/17/23, 9:33 PM

i misuse the entirety of C# so

AquaOP•10/17/23, 9:33 PM

if there's a will there's a way

Curin•10/17/23, 9:34 PM

and what happens when they see byte**const (which I don't know if C# can even replicate with base keywords)?

Perksey•10/17/23, 9:35 PM

yeah C# can't replicate it, it's just all byte**byte** which the user can do whatever they please with

AAqua it's a library binding to opengl and other low level libraries they're the misus...

Perksey•10/17/23, 9:36 PM

i'll write some words on the proposals, but it's best to stick with this for now. we could go with an analyser based approach and go with stuff as you suggest, but this does still lose type safety a lot.

Perksey•10/17/23, 9:36 PM

i feel like generalising any pointer to a single type (or single generic type) is extremely dangerous and becomes even more so when you start thinking about structs, which we haven't tackled yet

AquaOP•10/17/23, 9:39 PM

unsafeunsafe in C# is by its very nature "extremely dangerous"
if the user fucks it up, it should be because of something they did
not because the library provided a billion custom pointer types that caused the user to use the wrong one
i'm assuming this whole thing is so that silk can be used in a "safe" environment, which is fair enough, but there's only so safe you can get with opengl and other lower level libraries
try to make it too safe and you enter rust territory, and i don't think anyone really wants that
if you take a look at opentk, its still in a safe context but you can very easily screw it up just like anything else

Perksey•10/17/23, 9:39 PM

you have a fence around your garden, right? people can just climb over that, stand on it, do unspeakable things to it, whatever. is this an argument for not having a fence at all?

AquaOP•10/17/23, 9:40 PM

no, but you could make the argument that it's pointless making the fence too tall. sure it makes it harder to climb over, but it also makes things worse for you as well

AquaOP•10/17/23, 9:40 PM

i think having some levels of type safety is good, just not a massive wall of custom types for the name of absolute type safety

Perksey•10/17/23, 9:42 PM

i think as i said on the coffee & code this needs to be discussed in a formal working group meeting once we have something to discuss

khitiara•10/17/23, 9:44 PM

also this is at least partially to help cut down on overload count separate from the safety issue

khitiara•10/17/23, 9:45 PM

provides implicit conversions you can use instead of having a silly number of overloads

Perksey•10/17/23, 9:45 PM

yeah that's the sole goal

Perksey•10/17/23, 9:46 PM

we could also just go fully unsafe but have extension methods generated on-the-fly to do marshalling and the IDE experience would be even more terrible than this

Curin•10/17/23, 9:46 PM

yes, D2D in specific has a 16 ptr overload that could take spans for each, 16^16 is stupid

Perksey•10/17/23, 9:50 PM

ultimately these days in the advent of TerraFX being an official DNF project Silk.NET has no reason to exist other than its overloading abilities, and the current implementation of this is known-problematic which is why we're trying to come up with another one which is less problematic. still problematic mind, but a less so in that you don't have to sift through hundreds of overloads.

AquaOP•10/17/23, 9:51 PM

if that's the case, could it be worth considering switching to a more safe/managed environment? instead of trying to get both safe and unsafe at once
there are no (sane) safe vulkan libraries for example, silk could fill that gap quite easily if tried
and by safe i don't mean arrays etc, i mean spans

Perksey•10/17/23, 9:52 PM

but then you can't mix and match which was the reason our overloading system was so great

Perksey•10/17/23, 9:52 PM

and we consistently got complaints that you couldn't do that when we tried it in 1.0

Curin•10/17/23, 9:53 PM

spans also don't work well for double ptr cases, which can be either lists of lists or a ptr to place a single object (DX has this a lot)

khitiara•10/17/23, 9:53 PM

yeah

PPerksey and we consistently got complaints that you couldn't do that when we tried it in...

AquaOP•10/17/23, 9:55 PM

you'll never please everyone - last time i checked opentk was completely safe and i've not seen too many complaints about that (i use opentk as an example as i used it for a good year or so before switching to silk)
personally i can't see a time when you'd want to mix and match - you either use unsafe or safe
so silk could be the safer alternative to terrafx for example

CCurin spans also don't work well for double ptr cases, which can be either lists of li...

AquaOP•10/17/23, 9:55 PM

fair point

Maxine•10/17/23, 9:55 PM

you'll never please everyone

but when both sides are really angry, it's annoying

khitiara•10/17/23, 9:56 PM

yeah having this as a safe way to handle jagged arrays and ref to pointer situations is why I'm most interested

AAqua you'll never please everyone - last time i checked opentk was completely safe an...

Perksey•10/17/23, 9:56 PM

https://discord.com/channels/521092042781229087/939645285896962088/1137442690687062089 ¯\_(ツ)_/¯

khitiara•10/17/23, 9:56 PM

that and the new implementation of silkmarshal since that rules for strings

Maxine•10/17/23, 9:57 PM

i'm only here because i didn't know what terrafx was and perks is a friend

khitiara•10/17/23, 9:58 PM

I'm here for the overloads and the general offtopic channel tbh

Maxine•10/17/23, 9:59 PM

i thought we settled on implicit casts for overloading and it worked reasonably well?

khitiara•10/17/23, 9:59 PM

yeah that's this

Perksey•10/17/23, 9:59 PM

we didn't settle, this is just us pursuing a prototype

dfkeenan🍍🍕•10/17/23, 9:59 PM

I don't know why I'm here.

PPerksey we didn't settle, this is just us pursuing a prototype

Perksey•10/17/23, 9:59 PM

none of this has been formally agreed

AquaOP•10/17/23, 9:59 PM

i'm here because i enjoyed that silk offered a lot more options than opentk
i didn't care for unsafe initially
and i'm currently trying to play devil's advocate, that is all
and i'm trying to come at this objectively, subjectively my opinions are somewhat different to what i am saying

Maxine•10/17/23, 9:59 PM

yeah but it's good enough

AAqua i'm here because i enjoyed that silk offered a lot more options than opentk i di...

Perksey•10/17/23, 10:00 PM

oh yeah no it's appreciated

Maxine•10/17/23, 10:00 PM

all bureaucracy aside, i don't see any way of doing this in practicality that isn't the cast system

i am slightly concerned that this is becoming too far away from the language's design remember - mos

Similar Threads