Re: [quicwg/base-drafts] Make using static table and Huffman encoding in QPACK opt-in (#3622)

[Victor Vasiliev <[e-mail]>]

> How should a server that has no intention of enabling this setting respond to HEADERS that include other instructions? If the answer is to blow up the connection then the (unintended?) side effect will encourage clients to start connections by sending all requests as "Literal Header Field Without Name Reference" until they receive the setting. But that stage of a HTTP/3 connection is the exact place where static and huffman have the most value; we groomed the static table to suit best the first flights of client requests.

Well, I suggest this mechanism as an opt-in, rather than an opt-out, so it's less of "encourage" and more of "require".

This will in fact affect the first flight of requests if the client does not receive server's SETTINGS frame fast enough.  That does sound like a problem, but it will be eventually fixed when we fix #3086, so I'm not particularly concerned.

> My concern is that by trying to accommodate some smaller segment of HTTP clients, we cause a serious regression compared to HTTP/2.

That's a fair point.  On the other hand, HTTP/2 did come with a lot of features that we decided to make optional in HTTP/3, so that would be another step in this direction.

> This sort of thing seems better suited to a version that is "profile of HTTP/3" that disables some things and mandates others.

I am not sure how that kind of profile would look in practice.  The reason I am filing this for v1 is that in general, you can add complexity as an opt-in, but opting out of it does not really work.

-- 
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/3622#issuecomment-623568022