Switch to windows-bindgen

[djc]

No description provided.

[thomaseizinger]

Out of curiosity, is the benefit of windows-link that we are decoupling from the particular version of the windows crate? I came here because hostname is the last crate in our dependency tree that still pulls in [email protected] and it would be great to eliminate that.

[djc]

Yup, that's the idea. It also prevents downloading big wads of code that end up largely unused via feature flags...

I've been struggling with getting the codegen lined up correctly, something about the formatting is not working right. Want to take a look?

[thomaseizinger]

I've been struggling with getting the codegen lined up correctly, something about the formatting is not working right. Want to take a look?

Instead of the current approach, would it make more sense to run the binding generation in CI and then assert that git diff doesn't produce any changes?

[djc]

I've been struggling with getting the codegen lined up correctly, something about the formatting is not working right. Want to take a look?

Instead of the current approach, would it make more sense to run the binding generation in CI and then assert that git diff doesn't produce any changes?

That is kind of the current approach? I've used similar code in a bunch of projects and it mostly works well. But for some reason the formatting isn't sticky here...

[thomaseizinger]

I've been struggling with getting the codegen lined up correctly, something about the formatting is not working right. Want to take a look?

Instead of the current approach, would it make more sense to run the binding generation in CI and then assert that git diff doesn't produce any changes?

That is kind of the current approach? I've used similar code in a bunch of projects and it mostly works well. But for some reason the formatting isn't sticky here...

Kind of yeah. I was thinking something like:

• Run binding generation (perhaps still as a test)?
• Run cargo fmt on the project
• Run git diff --exit-code

If whitespace is still an issue, we could also consider using --ignore-all-space on git diff to just ignore whitespace altogether.

[thomaseizinger]

Unfortunately, CI doesn't run on PRs in this repo so I can't test whether my changes in #30 work as intended.

Edit: I pushed a follow-up commit that enables CI for PRs, needs maintainer approval though.

[djc]

That feels more like a workaround, though! I think I figured it out, the rustfmt component wasn't installed.