cbindgen: Add version defines

[yedayak]

This will allow users to check the version of rustls-ffi without resorting to checking if functions are defined.

In addition to RUSTLS_VERSION_{MAJOR,MINOR,PATCH}, also define RUSTLS_VERSION_NUMBER, which includes each version part in it, bit shifted to the left. This is inspired by openssl[0], c-ares[1]. There are other options for this, for example zstd multiplies each part by a power of 100[2].

We might want to also have the entire version string here, which could help tools that need to parse the header file itself.

[0] https://github.com/openssl/openssl/blob/cdd01b5e0734b0324251b32a8edd97f42ba90429/include/openssl/opensslv.h.in#L92-L102
[1] https://github.com/c-ares/c-ares/blob/42ddbc14ec008e738fa44aa2c16e74cad93742c2/include/ares_version.h#L43-L45
[2] https://github.com/facebook/zstd/blob/3c3b8274c517727952927c705940eb90c10c736f/lib/zstd.h#L115

Fixes #557

[cpu]

Thanks for picking this up! Do you think it's possible to extend the rustls version test to enforce keeping this in sync?

[cpu]

Do you think it's possible to extend the rustls version test to enforce keeping this in sync?

There are some pieces for parsing .h files in docgen/main.rs that might also be helpful for this job.

Writing the test is likely to be more work than it was to implement the change itself so I'm also OK leaving that as follow-up work if you're not keen on tackling it yourself.

[yedayak]

I made the test also verify the defines in the header file

[yedayak]

I'm not sure what to do about the failures in rust 1.73, tree-sitter-language requires rust 1.76.

[cpu]

I'm not sure what to do about the failures in rust 1.73, tree-sitter-language requires rust 1.76.

Ahh crud. I forgot that was one of the motivators for having a separate tools crate for the doc generation tooling.

The simplest thing to do might be to move the whole test to tools. As long as it's run in CI I think that's probably just as good as having it in librustls. WDYT?

[yedayak]

I'm not sure what to do about the failures in rust 1.73, tree-sitter-language requires rust 1.76.

Ahh crud. I forgot that was one of the motivators for having a separate tools crate for the doc generation tooling.

The simplest thing to do might be to move the whole test to tools. As long as it's run in CI I think that's probably just as good as having it in librustls. WDYT?

I moved just the new part to tools, since the current test is dependent on the features of librustls and I couldn't find a good way to use them from tools.
About running it in CI, I'm not sure what exactly is needed to do that, does it need to be added to some cmake target or something?

[cpu]

I moved just the new part to tools, since the current test is dependent on the features of librustls and I couldn't find a good way to use them from tools.

SGTM. Thanks!

About running it in CI, I'm not sure what exactly is needed to do that, does it need to be added to some cmake target or something?

The cmake stuff is rooted inlibrustls so it won't be a good fit for this.

I think the best route here is to add a new job block (maybe after the minver job block?) that runs the tests from the non-default rustls-ffi-tools workspace member. Something like:

tools:
  name: Test rustls-ffi-tools
  runs-on: ubuntu-latest
  steps:
    - name: Checkout sources
      uses: actions/checkout@v4
      with:
        persist-credentials: false

    - name: Install rust toolchain
      uses: dtolnay/rust-toolchain@nightly
        
    - name: Run tools unit tests
      run: cargo test -p rustls-ffi-tools

We don't want to add the cargo test step to the main build job because it would have to be skipped for the MSRV. We could do it as part of ensure-header-updated but it seems clearer to just have a new job. WDYT?

[cpu]

Looks great! I had a last round of nits but assuming those are addressed and nobody else has feedback I think this is ready for merge.

I appreciate you putting in the extra work to land the test. I know it was a lot more effort than the actual .h update 😅