Looking at this change and prompted by #1395 (comment), what happens if you still pass a body in, i.e. Response(status_code=204, content="hi") in the test?

My skimming of the relevant ASGI 3 spec suggests the ASGI server may add Transfer-Encoding: chunked, which then becomes RFC-invalid again, as the presence of either of Content-Length or Transfer-Encoding signals a body, which is disallowed for 204 and related responses.

Perhaps the implicit expectation is that the ASGI server will use "may" here and recognise response codes for which Transfer-Encoding should not be added (and also implicitly drop the body data, perhaps)? Although that seems like a gap in the spec that it's not explicit, and a gap in behaviour that the same thing is not done for Content-Length, as the same HTTP spec restrictions apply.

I apologise if I'm asking a question with an obvious/known answer already. I'm new to the ASGI ecosystem, and may be misunderstanding the division of responsibility or otherwise.

what happens if you still pass a body in, i.e. Response(status_code=204, content="hi") in the test?

Probably you've just got yourself an RFC-invalid response.

We could potentially guard against those kinds of cases, and I'd be happy for us to discuss that as a follow-up issue if anyone's sufficiently motivated to do so.

The most important thing for this pull request tho, is to make sure we're doing the right thing for the sensible cases.