This lovely hack for CecConnectionCfg::open appears to work:

    pub fn open(mut self) -> CecConnectionResult<CecConnection> {
        ...
        let mut devices: [libcec_sys::cec_adapter_descriptor; 10] = unsafe { std::mem::zeroed() };
        let num_devices = unsafe {
            libcec_sys::libcec_detect_adapters(
                connection.1,
                &mut devices as _,
                10,
                std::ptr::null(),
                true as i32,
            )
        };

        if num_devices < 0 {
            panic!("no devices found")
        }

        if unsafe { libcec_open(connection.1, devices[0].strComName.as_ptr(), open_timeout) } == 0 {
            return Err(CecConnectionResultError::AdapterOpenFailed);
        }
        ...
    }

Rusty wrapper for detection has not been implemented, no. You can find it in the list of unimplemented functions here:

Currently port name needs to be supplied by the user.

Pull requests and API design are most welcome :)

I made it work, the timeline went:

• Failed to build cec-rs due to libcec failing to build
• Failed to build libcec manually
• Reversed the libcec build process
• Fixed the libcec-sys build script
• Realized libcec-sys rebuilds on every change which causes a rebuild of libcec taking about 10 minutes
• Rewrote the build script to use precompiled binaries
• Realized cec-rs's use of #[repr(u32)] on enums breaks the windows build due to differences in C enum representations on MSVC
• Fixed the enums
• Cried

Anyhow, my fork diverged substantially and I'm unsure how to upstream it.

Links:

• cec-rs
• libcec-sys

It is sad to read how painful the windows developer experience is here. This reveals the current setup is not yet very polished/production grade even though the build and ci passes in github with windows.

Similar CI/CD optimizations have been discussed/proposed in ssalonen/libcec-sys#45 although not sure if the person will follow it through

This is first time I hear incompatibility with msvc, can you file ticket about it, please? How does it surface? I can see how this lead to deeper rabbit whole and introducing the bindgen as part of the build loop (as opposed to static runs of bindgen + python like now)...

re upstreaming: indeed, sounds like this needs quite a bit of work to split into manageable and logical chunks

Similar CI/CD optimizations have been discussed/proposed in ssalonen/libcec-sys#45 although not sure if the person will follow it through

I ended up using the precompiled binaries mentioned in that issue.

This is first time I hear incompatibility with msvc, can you file ticket about it, please? How does it surface?

As far as I can tell, I:

• Installed VS2019 as per the instructions
• Ran bindgen on the vendored/precompiled cecc.h with the same options as bindgen.sh
• Compared the output with the committed bindings and observed the constified enums were i32 instead of u32

I can see how this lead to deeper rabbit whole and introducing the bindgen as part of the build loop (as opposed to static runs of bindgen + python like now)...

In my experience, running bindgen in the build script is far less annoying and surfaces issues faster compared to committing the bindings.

Compared the output with the committed bindings and observed the constified enums were i32 instead of u32

Did this somehow generate build errors or runtime errors? What makes it really "incompatible"?

As far as I can understand, both types are compatible with the enum values. Some enums are signed when the c type contains negative values.

Re. Precompiled binaries

Instead of 3rd party precompiled binaries, I would instead prefer to use the dll's distributed via the official installer.

The approach details were discussed in the same issue

Did this somehow generate build errors or runtime errors? What makes it really "incompatible"?

It was build errors. From memory it was incompatible because u32 can't be losslessly cast to i32.

Coming back to this...

As far as I can tell, I:

Installed VS2019 as per the instructions
Ran bindgen on the vendored/precompiled cecc.h with the same options as bindgen.sh
Compared the output with the committed bindings and observed the constified enums were i32 instead of u32

I understand you might get build failures if you change the signatures...

But what the very first build error is MSVC, before you changed any of the code?

@ssalonen Hey, sorry for the delay. The first set of build error was with your code, untouched by me. If you make a dummy app that depends on cec-rs on Windows it should fail to build.