When receiving a GOAWAY frame, the client:
* immediately closes the connection if there are no active requests
* refuses to open streams with stream IDs larger than the stream ID in
the GOAWAY frame
* closes the connection once the stream count drops to zero
* http3: rename RoundTripper to Transport
* http3: rename SingleDestinationRoundTripper to ClientConn
* http3: construct the ClientConn via Transport.NewClientConn
* http3: add HTTP Trailer support for clients
This change only adds support to read HTTP trailers sent to clients.
* chore: add protection against some out-of-spec behavior + tests
* chore: re-add test accidentally overwtitten
* chore: empty commit to re-trigger ci
* fix: address some review notes (wip)
* fix: simplify code in stream.Read by using a callback from requestStream.ReadResponse
* restructure where trailers are read and parsed
* WIP simplify trailer parsing design
* chore: refactor to use simpler trailer parsing strategy
* make gofumpt happy
* Update http3/headers.go
Co-authored-by: Marten Seemann <martenseemann@gmail.com>
* remove stray TODO
---------
Co-authored-by: Marten Seemann <martenseemann@gmail.com>
The stream exposes two methods required for doing an HTTP request:
SendRequestHeader and ReadResponse. This can be used by applications
that wish to use the stream for non-HTTP content afterwards. This will
lead to a simplification in the API we need to expose for WebTransport,
and will make it easier to send HTTP Datagrams associated with this
stream.
The stream hijackers only need to be able to associate the stream with
the underlying QUIC connection. They are not supposed to call any
functions on the quic.Connection. As such, the better API is to just
pass them a unique identifier.
For some requests, the client is required to check the server's HTTP/3
SETTINGS. For example, a client is only allowed to send HTTP/3 datagrams
if the server explicitly enabled support.
SETTINGS are sent asynchronously on a control stream (usually the first
unidirectional stream). This means that the SETTINGS might not be
available at the beginning of the connection. This is not expected to be
the common case, since the server can send the SETTINGS in 0.5-RTT data,
but we have to be able to deal with arbitrary delays.
For WebTransport, there are even more SETTINGS values that the client
needs to check. By making CheckSettings a callback on the RoundTripOpt,
this entire validation logic can live at the WebTransport layer.
* http3: use a mock roundTripCloser in tests
* http3: correctly handle failed clients
Specifically,
* immediately remove a client when a request errored
* if that error was an idle error, and the client was a reused client
(from an earlier request that already completed the handshake),
re-dial the connection
