Invalid certificates are not signaled as such when used through rustls

[yorickpeterse]

In the thread of ring being unmaintained I learned about Graviola and gave it a try for Inko. The switch itself is straightforward, but I noticed a difference: I have a test that takes a valid X509 private key and an invalid X509 certificate:

https://github.com/inko-lang/inko/blob/b8ce18ff917379bd469b16d5348ea428ba64ca09/std/test/std/net/test_tls.inko#L310-L317

If I remember correctly, the invalid certificate is generated by generating an initially valid certificate using the OpenSSL CLI and then just chopping off a bunch of data in between the start/end markers. The test private key is as follows:

-----BEGIN PRIVATE KEY-----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-----END PRIVATE KEY-----

And the invalid test certificate:

-----BEGIN CERTIFICATE-----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-----END CERTIFICATE-----

This test then tries to create a rustls ServerConfig object using this Rust code:

https://github.com/inko-lang/inko/blob/b8ce18ff917379bd469b16d5348ea428ba64ca09/rt/src/runtime/tls.rs#L179-L197

When using ring, a rustls.Error is produced as expected. However, when using rustls-graviola no error at all is produced.

Now my question is this: is it expected that graviola won't produce errors in this case, or is that itself a bug of sorts?

[ctz]

Hey, thanks for the try-out. This error behaviour comes from rustls, in https://github.com/rustls/rustls/blob/main/rustls/src/crypto/signer.rs#L185-L196 -- and what you see here results from rustls-graviola not implementing the SigningKey::public_key() method. That should be an easy addition.