Closed cpu closed 7 months ago
Nice work! Maybe throw in a proactive version bump?
Good call. Added one and proactively updated the changelog.
I also double checked this was a regression (though I didn't bother digging up which commit was the breaker. (Edit: I think it was me in 30489d74833e8230bb1bc1d9412d17b776ce05bd)). Backporting the unit test from this branch to 0.12.1 shows it passing, just like described in 261.
crates-io
Thanks for publishing the fix :+1:
Previously when issuing a certificate with an authority key identifier (AKI) extension that's signed by an issuer certificate we had a small bug where we used the to-be-issued certificate's param's
key_identifier_method
to derive the key identifier of the issuing certificate to use for the issued certificate's AKI. Instead we should be using the issuer certificate's param'skey_identifier_method
, taking care to mind the pre-specified variant.We missed this with our unit testing of the pre-specified key identifier method because we only issued a self-signed test certificate, never issuing a certificate signed by the CA that has the customization. We principally exercised that the subject key identifier (SKI) of the self-signed cert matched the pre-specified value, but never tested that an issued cert's AKI matches the pre-specified SKI of the issuer.
This branch fixes the bug and extends test coverage to prevent further regression.
Resolves https://github.com/rustls/rcgen/issues/261