Re: [Webtransport] A comment on draft-kinnear-httpbis-http2-transport-02

[Martin J. Dürst <[e-mail]>]

I have a question somewhat similar to Bernard's question.

On 2019/11/19 11:41, Bernard Aboba wrote:
> This is a straightforward draft, so my questions don't relate to the
> content such much as the context:
> 
> The draft extends the CONNECT handshake to allow "bytestream" to be
> provided for the :protocol pseudo-header field, instead of "websocket".
> 
> Is there corresponding JS API work contemplated (or ongoing) to support
> this?
> 
> The "WebSocketStream API design"
> <https://docs.google.com/document/d/1La1ehXw76HP6n1uUeks-WJGFgAnpX2tCjKts7QFJ57Y/>
> indicates "a 1:1 correspondence between chunks and WebSocket messages" so
> that it doesn't appear to take "bytestream" into account.

The WebSocket protocol and API use messages. The draft provides the 
"datagram" :protocol pseudo-header for message-based communication. The 
implementation is very simple: each HTTP/2 DATA frame is a message aka 
datagram. So far so good.

But the WebSocket protocol distinguishes two types of messages: Text 
messages and Binary messages (in terms of implementation, the 
distinction is carried on frames, which make up messages, but the type 
has to be the same for all frames of a message). Also, text messages are 
strictly restricted to always use UTF-8 (see 
https://tools.ietf.org/html/rfc6455#section-1.2).

The distinction between binary and text datagrams, and the limitation of 
text datagrams to UTF-8, considerably simplifies the creation of 
datagram-based protocols and increases interoperability.

I wonder why this wasn't included in the Webtransport draft, and hope 
that it can be added.

[cc.ed to the W3C i18n WG]

Regards,   Martin.