Fix missing re-exports

[gretchenfrage]

Closes #2012

In that issue, we noted that there doesn't seem to exists any automated tooling to automatically detect types from external crates which are visible in a crate's public API but which the crate does not re-export. So I went ahead and attempted to create such a tool: [cargo should-be-public-checker]. It's still pretty crude right now but it was enough to detect a handful of items that I confirmed were missing reexports in Quinn and proto. That said, it's possible that there are some things it missed.

should-be-public-checker output for quinn-proto (on main):

visible but not importable:
- quinn_proto::Chunk::bytes::Bytes
- quinn_proto::ClientConfig::with_root_certificates::RootCertStore
- quinn_proto::ClientConfig::with_root_certificates::VerifierBuilderError
- quinn_proto::ConnectionClose::frame_type::Type
- quinn_proto::ConnectionHandle::`<_ as Index<Some(AngleBracketed { args: [Type(ResolvedPath(Path { name: "ConnectionHandle", id: Id(4627), args: Some(AngleBracketed { args: [], constraints: [] }) }))], constraints: [] })>>`::Output::ConnectionMeta
- quinn_proto::ConnectionId::from_buf::Buf
- quinn_proto::PartialDecode::new::BytesMut
- quinn_proto::ServerConfig::with_single_cert::CertificateDer
- quinn_proto::ServerConfig::with_single_cert::PrivateKeyDer
- quinn_proto::crypto::rustls::HandshakeData::negotiated_cipher_suite::CipherSuite
- quinn_proto::crypto::rustls::QuicClientConfig::with_initial::ClientConfig
- quinn_proto::crypto::rustls::QuicServerConfig::with_initial::ServerConfig
- quinn_proto::crypto::rustls::QuicServerConfig::with_initial::Suite
- quinn_proto::crypto::rustls::TlsSession::`<_ as VZip<Some(AngleBracketed { args: [Type(Generic("V"))], constraints: [] })>>`::MultiLane
- quinn_proto::transport_parameters::TransportParameters::write::BufMut

should-be-public-checker output for quinn (on main):

visible but not importable:
- quinn::ApplicationClose::reason::Bytes
- quinn::ClientConfig::initial_dst_cid_provider::ConnectionId
- quinn::ClientConfig::initial_dst_cid_provider::ConnectionId::from_buf::Buf
- quinn::ConnectionClose::error_code::Code
- quinn::ConnectionClose::frame_type::Type
- quinn::ConnectionStats::frame_tx::FrameStats
- quinn::ConnectionStats::path::PathStats
- quinn::ConnectionStats::udp_tx::UdpStats
- quinn::EndpointConfig::cid_generator::ConnectionIdGenerator
- quinn::SendStream::write_chunks::Written
- quinn::StreamId::new::Dir
- quinn::StreamId::new::Side
- quinn::Transmit::ecn::EcnCodepoint
- quinn::VarInt::from_u64::VarIntBoundsExceeded
- quinn::crypto::CryptoError::`<_ as VZip<Some(AngleBracketed { args: [Type(Generic("V"))], constraints: [] })>>`::MultiLane
- quinn::rustls::ConfigSide::Sealed
- quinn::rustls::RootCertStore::roots::TrustAnchor
- quinn::rustls::client::EchConfig::new::EchConfigListBytes
- quinn::rustls::client::ServerCertVerifierBuilder::with_crls::CertificateRevocationListDer
- quinn::rustls::client::Tls13ClientSessionValue::`<_ as Deref<Some(AngleBracketed { args: [], constraints: [] })>>`::Target::ClientSessionCommon
- quinn::rustls::client::verify_server_cert_signed_by_trust_anchor::CertificateDer
- quinn::rustls::client::verify_server_cert_signed_by_trust_anchor::SignatureVerificationAlgorithm
- quinn::rustls::client::verify_server_cert_signed_by_trust_anchor::UnixTime
- quinn::rustls::client::verify_server_name::ServerName
- quinn::rustls::crypto::cipher::OutboundOpaqueMessage::read::MessageError
- quinn::rustls::crypto::cipher::OutboundOpaqueMessage::read::Reader
- quinn::rustls::crypto::cipher::PlainMessage::payload::Payload
- quinn::rustls::crypto::hpke::HpkeSuite::kem::HpkeKem
- quinn::rustls::crypto::hpke::HpkeSuite::sym::HpkeSymmetricCipherSuite
- quinn::rustls::crypto::hpke::HpkeSuite::sym::HpkeSymmetricCipherSuite::aead_id::HpkeAead
- quinn::rustls::crypto::hpke::HpkeSuite::sym::HpkeSymmetricCipherSuite::kdf_id::HpkeKdf
- quinn::rustls::crypto::ring::sign::any_ecdsa_type::PrivateKeyDer
- quinn::rustls::crypto::ring::sign::any_eddsa_type::PrivatePkcs8KeyDer
- quinn::rustls::crypto::verify_tls13_signature_with_raw_key::SubjectPublicKeyInfoDer
- quinn::rustls::server::ClientHello::server_cert_types::CertificateType
- quinn::rustls::unbuffered::TransmitTlsData::may_encrypt_early_data::MayEncryptEarlyData

Summary of changes:

• quinn_proto::frame::Type becomes public. So, in addition to simply re-exporting it:
  • Renamed it to FrameType
  • Added doc comment
  • Made its associated constants pub(crate) rather than pub
• In terms of crate re-exports: Quinn already re-exports rustls and udp. Following this trend, I also added re-export of rustls to proto, and re-export of bytes to both.
• I changed crate re-exports from pub use to pub extern crate. I'm fine changing this back if you like, but I think this is better because it's more clear that it's more clear that it's a crate being re-exported.
• Other than that, simply pub used all the missing items.

Of course, fixing the missing re-exports in rustls is out of scope, and the thing about MultiLane is some irrelevant blanket-impl.

[djc]

Very cool that you created a tool for this, definitely good to improve on this in a more structural way!

• I changed crate re-exports from pub use to pub extern crate. I'm fine changing this back if you like, but I think this is better because it's more clear that it's more clear that it's a crate being re-exported.

This definitely does not seem like an improvement, IMO extern crate should basically never be used in modern Rust, and I think it's plenty clear enough that these are crate re-exports. If not, just add some comments.

I don't think we should re-export bytes. It is a 1.0 dependency that IMO is pretty unlikely to ever get a semver-incompatible bump, so there's not the same benefit like with rustls that matching the used version is hard.

My remaining question here is, are these changes semver-compatible? (In particular, the renaming of frame::Type.)

[gretchenfrage]

If not, just add some comments.

Unfortunately, doc comments on re-export items doesn't seem to be working with rustdoc? It's fine, though.

[gretchenfrage]

:)

• Removed reexport of bytes.
• Changed pub extern crate back to pub use.

My remaining question here is, are these changes semver-compatible? (In particular, the renaming of frame::Type.)

quinn_proto::frame::Type is not importable on main. the struct Type is pub, but the mod frame is default visibility, and Type is never re-exported. You can confirm that in the current docs.rs for eg. ConnectionClose, the Type in frame_type: Option<Type> is not a link, and Type cannot be found by searching. Therefore, renaming it should not cause semver problems.

The only changes other than that rename are adding new re-exports. I don't imagine this is something that's considered to cause semver problems.

It may be hard to tell at a glance because cargo fmt alphabetizes it, but the new re-exports added to Quinn in the last commit are:

• quinn_proto::Dir
• quinn_proto::Side
• quinn_proto::ConnectionId
• quinn_proto::Written
• quinn_proto::TransportErrorCode
• quinn_proto::FrameStats
• quinn_proto::VarIntBoundsExceeded
• quinn_proto::PathStats
• quinn_proto::UdpStats
• quinn_proto::ConnectionIdGenerator
• quinn_proto::EcnCodepoint
• quinn_proto::FrameType

[djc]

Thanks, this looks good to me!

(If you want to do a similar pass for rustls, that would be much appreciated!)

[Ralith]

Thanks! Would be cool to have this running in CI.