Excise the HasRaw*Handle traits

[notgull]

After reading #135, I've realized that the HasRawDisplayHandle and HasRawWindowHandle traits have little value in a post-borrowed-handle world. Borrowed handles do everything better, and the raw handle can be extracted from the borrowed handle. Therefore it makes sense to remove these traits.

Closes #135.

[madsmtm]

This looks perfect honestly.

I could potentially see how an implementer might not want to provide the strong guarantees about validity that HasWindowHandle, but HasRawWindowHandle already had those guarantees, they were just more implicit, so I think this will be fine!

[notgull]

Removing it outright might be a bit too harsh. I've opted to deprecate it for now.

[madsmtm]

Maybe you could add a blanket unsafe impl<T: ?Sized + HasWindowHandle> HasRawWindowHandle for T { ... }? That would mean less work for crates wanting to implement these traits.

[notgull]

Maybe you could add a blanket unsafe impl<T: ?Sized + HasWindowHandle> HasRawWindowHandle for T { ... }? That would mean less work for crates wanting to implement these traits.

Banger idea, I've implemented this

[Lokathor]

What would you say people who don't have borrowing lifetimes on their handles do

[madsmtm]

What would you say people who don't have borrowing lifetimes on their handles do

I'm not sure I understand the question, but what would an example of this be? Like, what would be a situation where you, as a "provider" type, would like to say "hey, I have this window handle, its lifetime is not tied to me (&self), but I don't know the lifetime of it either"?

[Lokathor]

So SDL gives you *mut SDL_Window when you make a window.

and because pointers are Copy the "lifetime" of that pointer, as the compiler sees it, is not anything sensible because there's all sorts of spots where you'll accidentally get a copy of the value instead of extending the life of a borrow on the pointer, or some other fiddly thing like that.

So there's certainly some span of time that's correct (until you call SDL_DestroyWindow), but Rust's borrow checker won't know what the correct time is.

[madsmtm]

Makes sense. In this case, don't implement Has[Raw]WindowHandle for the pointer, because it doesn't have a window handle for its full lifetime.

Instead, expose an unsafe method like this:

/// # Safety
/// The window must not be destroyed for the lifetime of the handle
pub unsafe fn get_handle<'a>(window: *mut SDL_Window) -> WindowHandle<'a> { ... }

Note that WindowHandle<'_> implements HasWindowHandle, so graphics crates that take T: HasWindowHandle will still work. Note also that the lifetime here is unbounded, so the user can choose 'static if that's needed by their graphics crate.

[Lokathor]

Alright, as long as we've thought about it and we have some answer, then we can proceed

[notgull]

It seems to me that everyone is fine with this, so I'll merge it.