`windows` now returns `HRESULT` directly. Is it possible to return `Result` again?

[Berrysoft]

Suggestion

It makes it more wordy to use question mark ? on the result. Previously I can use ? directly, but now I have to use .ok()?.

[Fulgen301]

It seems the Result transformation has disappeared for functions that aren't part of a COM interface. BOOL still has an ok method, but the rest needs manual error handling...at least according to the examples.

windows-rs/crates/samples/windows/samples/examples/kernel_event.rs

Lines 8 to 17 in acaaa37

	unsafe {
	let event = CreateEventW(None, true, false, None);
	if event.0.is_null() {
	return Err(Error::from_thread());
	}
	SetEvent(event).ok()?;
	WaitForSingleObject(event, 0);
	CloseHandle(event).ok()?;
	}

[Berrysoft]

That's not good for Win32 APIs. I understand that we are now switching to our own metadata generation, but it lacks many convenient features, e.g., returning Result, and passing slices directly.

[RossSmyth]

This change is pretty bad. If error handling is just C-style error handling, then there's not a ton of reason to use the wrapped API. Just use windows-sys and your compile times will be faster. I mainly use -sys since it is lighter, but I used the wrapped API in one project. With this change I'll move that project to -sys too since the entire reason I use the wrapped API was for Rust-style error handling.

[kennykerr]

The new bindings preserve the Windows SDK headers rather than supplementing them with separately maintained policy. Raw HRESULT keeps meaningful success codes such as S_FALSE - callers that only need error propagation can use .ok()?. Result<T> and slices are generated only when header annotations describe them faithfully, since output buffers, shared counts, and uninitialized storage often cannot be represented safely as Rust slices. We plan to improve projections by adding semantic information to the headers themselves rather than maintaining an additional metadata layer.

[kennykerr]

You can find a summary of the plan to shift the semantic information back into the headers here: microsoft/win32metadata#2295