With HTTP/1.1 the common assumption is that error responses are allowed before the request is completed.

With HTTP/2, the spec seems to suggest that 2xx responses are also allowed before the request is completed, as long as the server doesn't care about the rest of response. This makes sense, if one considers the status representing the causal result of the server seeing all the received request data (but incomplete).

I believe this issue belongs to http-semantics.

===

I can live with the HTTP/2 interpretation.

It will be useful to spec out how proxies may behave, e.g. what to do with request data that arrives after the status has been received?