Update objc2 to v0.6, and use new objc2-core-graphics crate

[madsmtm]

objc2 now provides an interface to CoreGraphics, which has automatic memory management, and interfaces better with the other objc2-* crates. Note that this doesn't have the CGDataProvider helper we were using before, but I'd argue that's actually a benefit, since it forces us to implement the data provider ourselves (and makes it very visible that that's a place where there's room for improvement).

Another notable change is the introduction of Retained::downcast, which we use to ensure that our types are correct in the key-value observer (instead of blindly casting, and hoping that they are).

This requires an MSRV bump to Rust 1.71 to get access to extern "C-unwind" functions.

[ids1024]

Looks like the released winit only requires 1.70.0 (master requires 1.73), but the bump here is still well within the rust-windowing MSRV policy, so it seems perfectly reasonable.

Note that this doesn't have, but I'd argue that's actually a benefit, since it forces us to implement the data provider ourselves

I think there's a word or two missing in this sentence after "doesn't have"? CGImage maybe?

[madsmtm]

Note that this doesn't have, but I'd argue that's actually a benefit, since it forces us to implement the data provider ourselves

I think there's a word or two missing in this sentence after "doesn't have"? CGImage maybe?

Added "the CGDataProvider helper we were using before", thanks.

(I'd have linked to the helper, but it's marked #[doc(hidden)], which makes it a bit difficult).

[ids1024]

It was a while ago, but I was wondering if we could do something custom with the CGDataProvider to swap between buffers instead of allocating each present (#83 (comment)). So yeah, not having that helper in objc2 isn't necessarily a bad thing.

Testing on an M1 Mac Mini, I see an index out of bounds error in the winit example trying to resize the window, but that seems to happen on master too... not something I see on Linux/Wayland, anyway. Perhaps the same issue as #253.

[ids1024]

I can't claim expertise with the CoreGraphics APIs, but most of this seems like a straightforward change from the old API, and the into_raw/from_raw part seems correct.

[madsmtm]

Thanks for the reviews!

Testing on an M1 Mac Mini, I see an index out of bounds error in the winit example trying to resize the window, but that seems to happen on master too... not something I see on Linux/Wayland, anyway. Perhaps the same issue as #253.

I think that has been fixed with newer Winit, try cargo update in this repo?

[ids1024]

Ah right. Forgot to try that, and of course I have an old Cargo.lock. Seems to work fine after a cargo update.