Feature/ws over h2

[drymus]

Description

• Add h2 CONNECT specs
• Update h2 binding to support CONNECT

Fixes # (issue)

[jfallows]

Please add a newline at end of file.

[jfallows]

These scripts should move to rfc8441 directory.

[jfallows]

Here we are just tracking the presence and count of the pseudo-headers, so they can be checked in decodeHeaders. So we likely need to add a count for protocol and include in the check.

Note we are validating the pseudo-headers themselves, not the values.

[drymus]

I think I misread this part of the spec in rfc8441:

The :protocol pseudo-header field MUST be included in the CONNECT request, and it MUST have a value of websocket to initiate a WebSocket connection on an HTTP/2 stream

I now think it means that if we have a CONNECT request, before we create a websocket connection we must ensure that the request also contains the :protocol pseudo-header and it must have the value websocket.

Maybe we don't need to do any validation in the http2 binding in relation to a connect request.

[drymus]

Until we find a better place for it, I have added a check to allow a :protocol pseudo-header in a CONNECT request which is otherwise not allowed by rfc7540. This seems to make sense to be in the http2 binding because the binding is always advertising the ENABLE_CONNECT_PROTOCOL setting.

[jfallows]

Agreed, just make sure you are doing the simple count here, and the enforcement in decodeHeaders, aligned with approach for other checks.

[jfallows]

Use http:matchBeginEx when reading, and only include the headers that matter per spec, such as :status above.

[jfallows]

If we are going with websocket handshake, then payload should either be omitted or follow websocket framing instead of Hello, world.

[drymus]

I expect the first octet to be 0x81 for text but the server is returning 0x82 (binary). I couldn't find any tests that demonstrate receiving plain text from the engine.

[jfallows]

Given that text is a subset of binary, we default to binary to ensure correctness.

However, if WsDataEx on DATA frame has flags 0x81 then we send a text frame instead.

You can see this in action via ws.echo example. Sending a text frame causes a text frame to be echoed back, because the WsDataEx for the decoded text frame, is sent back from echo binding, and then ws binding encodes DATA as a text frame.

[jfallows]

Btw, missing newline at EOF.

[jfallows]

Btw, when you are done making the changes, please also test end-to-end using a local build with ws.echo example and changing values file to set the image tag to develop-SNAPSHOT, similar to mqtt.kafka.reflect example.

Looks like Chrome first shipped support for WebSocket over HTTP/2 in v91.
https://chromestatus.com/feature/6251293127475200

[jfallows]

Missing newline at end of file.

[jfallows]

Given that text is a subset of binary, we default to binary to ensure correctness.

However, if WsDataEx on DATA frame has flags 0x81 then we send a text frame instead.

You can see this in action via ws.echo example. Sending a text frame causes a text frame to be echoed back, because the WsDataEx for the decoded text frame, is sent back from echo binding, and then ws binding encodes DATA as a text frame.

[jfallows]

Instead of performing a UTF-8 decode to String then comparing to String constant, please create a DirectBuffer constant for HEADER_NAME_PROTOCOL and compare without the need for a decode.

Note: one easy way to create the right DirectBuffer constant is new String8FW(":protocol").value().

[jfallows]

Agreed, just make sure you are doing the simple count here, and the enforcement in decodeHeaders, aligned with approach for other checks.

[jfallows]

Given the vastly different approaches to the handshake, suggest you avoid all the if/else logic and instead have separate Ws11Server and Ws2Server, both extending generic abstract WsServer if there is enough overlap in implementation beyond the handshake.

[jfallows]

Missing newline at EOF.

[jfallows]

Missing newline at EOF.

[jfallows]

Missing newline at EOF.

[jfallows]

Missing newline at EOF.

[jfallows]

Missing newline at EOF.