server_name: offer ServerName::to_str w/ alloc/std features

[cpu]

Downstream crates may want to represent a ServerName as a String or &str, e.g. for passing through a ffi boundary to external code (for example, in rustls-ffi, or in rustls-platform-verifier).

This commit implements support for converting a pki_types::IpAddr to a std::net::IpAddr (to use the to_string impl), and adds a ServerName::to_str that can return a Cow<'_, str> for a DnsName or an IpAddr.

[djc]

It seems to me that ServerName is an input-only type from the perspective of webpki/rustls so it's not that obvious why it needs non-trivial implementations to change it back to string form. What's the ffi/platform verifier use case, exactly?

[cpu]

It seems to me that ServerName is an input-only type from the perspective of webpki/rustls so it's not that obvious why it needs non-trivial implementations to change it back to string form. What's the ffi/platform verifier use case, exactly?

For rustls-ffi the ServerName is translated to a C string to pass through the FFI boundary to a verification callback. Previously that was using the AsRef<&str> impl that existed on rustls::server::DnsName and to_string from the Std IP address. Those went away with the pki-types migration.

For rustls-platform-verifier the ServerName is translated to a Rust String that can be based to Java code over a FFI boundary to do additional revocation checking. It too was using the DnsName::AsRef and Std IP to_string approach (and also needed an ugly unreachable!).

Since there are two separate uses and the functionality existed before moving to the pki-types arrangement it seemed sensible to implement. Does the above help convince you?

[cpu]

Published rustls-pki-types v0.2.3 at registry crates-io

[cpu]

Argh - I broke semver calling this 0.2.3 😢 af0db93 was in main and added a Debug bound on SignatureVerificationAlgorithm.

That makes it incompatible with rustls-webpki 0.102.0-alpha.7 because RingAlgorithm isn't Debug.

I could cut a new 0.2.4 that includes this branch's changes but not af0db93, or we could roll forward and fix webpki. I'm not sure what the best course of action is given this crate is pretty internal to the Rustls ecosystem.

[cpu]

Here's a webpki fix PR: rustls/webpki#212

[djc]

I guess we should add a cargo-semver-checks job? :D I suppose the webpki fix will be good enough for now while this only causes issues for alphas.

[cpu]

I guess we should add a cargo-semver-checks job?

It was there, but CI wasn't running 😭 I should have fixed that before I merged the other work. Silly mistakes.