Re: [quicwg/base-drafts] PUSH_ID as a frame (#2526)

[Luca Niccolini <[e-mail]>]

> The varint type and length of a frame is also unframed. 

@martinthomson those are in the frame header -- so they are part of the frame. I am not sure what your point is here then.

Here I am assuming that an implementation has generic code to parse frames, and that parsing anything that is not part of a frame requires special handling. 
While this is implementation specific and there can definitely be counter examples, from an abstraction and consistency point of view I see no reason for doing it the way we are doing it now. 

I believe there should be two type of streams, framed (control, request, and push streams) and unframed (QPACK, and potentially extension streams). The push stream right now is a hybrid, that only becomes a framed stream after receiving that first PUSH_ID. If I am recalling correctly when we discussed potentially having DATA frames with zero-length to transition to unframed data, it was also considered a bit unusual. Here I am arguing that we are sort of doing the reverse here where we have an unframed stream that then transitions to framed. I think this generates complexity. And really the benefit we get is saving two bytes and one frame type.

@lpardue I do vaguely remember that discussion and not 'wasting' frame types may have been one of the arguments when we made that decision, but that is not a problem now that we have a 62bit space for frame types. 


-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/2526#issuecomment-473722917