Silk.NET•3y ago

store a ref

PerkseyOP•1/6/23, 9:56 PM

a ref can be anything

PerkseyOP•1/6/23, 9:56 PM

managed, unmanaged

PerkseyOP•1/6/23, 9:56 PM

the only plus being if it's managed, it'll still be GC tracked

TechPizza•1/6/23, 9:56 PM

i bet rollie wants to store it in an ECS .)

Aqua•1/6/23, 9:56 PM

no, it's because i use the struct as the input parameter for my texture creation

PPerksey store a ref

Aqua•1/6/23, 9:57 PM

interesting, although i am not on .net 7 so i can't use ref fields
hm

PerkseyOP•1/6/23, 9:57 PM

yeah

PerkseyOP•1/6/23, 9:57 PM

that doesn't use .net 7 ref fields

PerkseyOP•1/6/23, 9:57 PM

i didn't know they existed until 5 seconds ago

PerkseyOP•1/6/23, 9:57 PM

so it'd be impressive if i managed to use ref fields before knowing about it!

AAqua no, it's because i use the struct as the input parameter for my texture creation

TechPizza•1/6/23, 9:58 PM

take a pointer to a ref struct and pass that when

PPerksey i didn't know they existed until 5 seconds ago

TechPizza•1/6/23, 9:58 PM

you can take pointers to object locals now as well

PerkseyOP•1/6/23, 9:58 PM

is that a GC hole

Aqua•1/6/23, 9:58 PM

so basically
if i am understanding this
storing it as a span is the correct way to do it
instead of storing a raw pointer?
i think i have confused myself

AAqua so basically if i am understanding this storing it as a span is the correct way ...

PerkseyOP•1/6/23, 9:58 PM

well storing it as a ref

PPerksey is that a GC hole

TechPizza•1/6/23, 9:58 PM

nope, pointers to locals with GC refs are now correctly tracked

PerkseyOP•1/6/23, 9:58 PM

so that could be ref fields, or if you're on anything but .net 7 span is the way to do that

Aqua•1/6/23, 9:58 PM

ah
gotcha

TechPizza•1/6/23, 9:59 PM

so you can do this:

object local = new();
object* ptr = &local;
PassToNative(ptr);

object local = new();
object* ptr = &local;
PassToNative(ptr);

TechPizza•1/6/23, 9:59 PM

it's a bit cursed but super useful

PerkseyOP•1/6/23, 9:59 PM

TechPizza•1/6/23, 10:03 PM

way better than GC handles. i'm already using it in my V8 bindings lol

Aqua•1/6/23, 10:03 PM

so i changed the struct from the pointer to the span and it doesn't seem to be complaining so i'm happy with this, thanks!

TTechPizza way better than GC handles. i'm already using it in my V8 bindings lol

TechPizza•1/6/23, 10:04 PM

although non-pinning GCHandle is still useful for long-term stuff in native

Aqua•1/6/23, 10:06 PM

this also helped me understand my original question of why you need to deallocate the memory with silkmarshal instead of just using fixed on it so that's good too

Aqua•1/6/23, 10:06 PM

how i have had 0 issues with what i've been doing i've got no idea, i wish i had had issues with it so i could have worked out the garbage i was doing earlier

AAqua how i have had 0 issues with what i've been doing i've got no idea, i wish i *ha...

TechPizza•1/6/23, 10:07 PM

if you created large arrays they probably got sent to LOH and rarely moved

Aqua•1/6/23, 10:07 PM

ah
yeah
since it is texture data

Aqua•1/6/23, 10:08 PM

wait

Aqua•1/6/23, 10:08 PM

this is a problem

Aqua•1/6/23, 10:09 PM

okay gc handles it is i guess
hopefully

PerkseyOP•1/6/23, 10:10 PM

what's wrong with just storing a byte[]byte[] again?

Aqua•1/6/23, 10:10 PM

it needs to be an unmanagedunmanaged type

Aqua•1/6/23, 10:10 PM

hm though

PerkseyOP•1/6/23, 10:10 PM

TechPizza•1/6/23, 10:10 PM

why use an array when you can just make one thicc TextureData<T>

Aqua•1/6/23, 10:10 PM

i might be able to remove that constraint now
i'll test

Aqua•1/6/23, 10:11 PM

i considered that but i'd like it to be friendly for the user
i went the directx route of taking an array of subresourcedata

Aqua•1/6/23, 10:12 PM

yeah i can remove that constraint lol
because my internal createtexture function just takes a pointer to the texturedata struct anyway
it used to be that it took an array which is what necessitated that type constraint

Aqua•1/6/23, 10:16 PM

i think i'll still go the gchandle route though since i'd like textures to be able to use raw pointers if desired

TechPizza•1/6/23, 10:16 PM

a GCHandle can't pin a ref though :)

TechPizza•1/6/23, 10:16 PM

so you have to use a pointer + a construct that can free a GCHandle if array was provided

PerkseyOP•1/6/23, 10:17 PM

i need to figure out the implications of making the vast majority of structs in Silk.NET a ref struct tbf

Aqua•1/6/23, 10:17 PM

ok i officially hate garbage collected languages
why is this so much pain
i just want to store a pointer to some data lmao

TechPizza•1/6/23, 10:17 PM

then just use a pointer when

Aqua•1/6/23, 10:19 PM

b-but that's what i was doing

basically

struct TextureData {
  void* DataPointer:
}

public Texture CreateTexture(blah stuff, TextureData* datas);

struct TextureData {
  void* DataPointer:
}

public Texture CreateTexture(blah stuff, TextureData* datas);

is the gist of what i'm doing
while i'd like the user to be able to use byte[]byte[] and whatnot, i'd like to internally store it all as raw pointers
which i seem to be making an absolute mess of

Aqua•1/6/23, 10:20 PM

but i'm guessing that's not something you can easily do without resorting to spans and ref structs
... that i take an array of so i can't do that

Aqua•1/6/23, 10:21 PM

i'm guessing the simple answer is that i can't use pointers for this
which is really heckndarn annoying

TechPizza•1/6/23, 10:22 PM

meh, i wouldn't bother making immutable texture creation methods nice-to-use

store a ref

Similar Threads