capi: Expose frame buffer API

[dwbuiten]

Notes

Currently untested - I will get to testing this tonight after dinner, and will update the PR text after. It's very likely to explode on 4:0:0.

Comments welcome - it's kinda ugly, but such is C interop.

Also of note is that Cargo.toml in the main directory is sitll pointing to av-metrics 0.4.0 for some reason.

[dwbuiten]

Since we hit a cursed Rust -> C ABI issues, I'm attaching a simple test program and files
quick_c_buf_test.zip

that triggers the madness, at @Luni-4's request.

[lu-zero]

What cbindgen produces for those?

[dwbuiten]

It produces exactly the API I wanted:

/**
 * Calculate the `msssim` metric between two frame buffers
 *
 * Returns the correct `msssim` value or `NULL` on errors
 */
const AVMContext *avm_calculate_frame_buf_msssim(const uint8_t *frame1[3],
                                                 ptrdiff_t frame1_strides[3],
                                                 const uint8_t *frame2[3],
                                                 ptrdiff_t frame2_strides[3],
                                                 uint32_t width,
                                                 uint32_t height,
                                                 uint8_t bitdepth,
                                                 AVMChromaSamplePosition chroma_pos,
                                                 AVMChromaSampling subsampling,
                                                 AVMRational pixel_aspect_ratio);

But it doesn't matter, since Rust can't generate a confirming C ABI for it, as discussed on IRC.

[lu-zero]

This is a bug in cbindgen. ptrdiff_t frame2_strides[3] is *mut frame2_strides.

[dwbuiten]

What would be the correct way to write the rust declaration to generate ptrdiff_t frame2_strides[3]? And should we file a cbindgen bug?

[lu-zero]

Let me poke upstream to rustc confirm, but from what you are telling me it seems so and a good idea to tell cbindgen upstream.

[lu-zero]

improper_ctypes is on by default only in extern blocks.

[lu-zero]

improper_ctypes is on by default only in extern blocks.

[dwbuiten]

I mean... not's not really a good answer. Why wouldn't it be on for all extern "C" functions? It makes it almost useless.

[lu-zero]

That's what I had been told today :) I hadn't tested it yet btw.

[lu-zero]

upstream cbindgen would welcome a patch to prevent this and one to add annotations to use the style you'd use here.

[Luni-4]

Now that C-APIs have been removed, are you still interested in re-adding them @dwbuiten? Or can we close this PR?