Re: Stuck in a train -- reading HTTP/2 draft.

As Martin said, this isn't in-scope for HTTP/2. If people want to propose separate, backwards-compatible extensions to HTTP in general, they're free to do so, but I'd ask that they use a clear Subject: line.

Regarding Date -- it's end-to-end, not hop-by-hop, and removing it is not backwards-compatible; it's used extensively by the caching model <http://httpwg.github.io/specs/rfc7234.html>, and also for other purposes (e.g., the recent discussion in TLSWG about relying upon it rather than the timestamp in TLS).

Regards,


On 19 Jun 2014, at 8:03 am, Matthew Kerwin <matthew@kerwin.net.au> wrote:

> Date is used to calculate Age. I imagine it's actually in use by caches (?)
> 
> Coding it into the headers frame makes sense, but is there a rule
> about it being set in requests?
> 
> On 19/06/2014, James M Snell <jasnell@gmail.com> wrote:
>> Yep. Using a variable length encoding 5 bytes is the max we would need for
>> all practical purposes.
>> 
>> I'm all for dropping Date entirely, tho. That doesn't help us with the
>> other date headers, but it helps. Last-Modified and If-Modified-Since are
>> both great candidates for five-byte encoding.
>> 
>> It's too bad the WG didn't pick up on such an obvious improvement but, oh
>> well I guess.
>> 
>> - James
>> On Jun 18, 2014 1:36 PM, "Poul-Henning Kamp" <phk@phk.freebsd.dk> wrote:
>> 
>>> In message <
>>> CABkgnnVT8zGSiU8fDqNtiaL+f2ziBytyP_SUGyPSL2anZf546Q@mail.gmail.com>
>>> , Martin Thomson writes:
>>>> On 18 June 2014 13:17, Poul-Henning Kamp <phk@phk.freebsd.dk> wrote:
>>>>> In that case we should transfer the time as a POSIX time_t in the
>>>>> HEADERS frame.  Wasting time huffman encoding dates and still
>>>>> using 24 bytes where 8 would be plenty is just plain stupid.
>>>> 
>>>> Yeah, that was discussed and rejected, though not permanently.  I
>>>> think that James worked out that 5 bytes was enough in the short term
>>>> with a little epoch tweaking.
>>>> 
>>>> And yes, we are plain stupid.  Think of the cost of parsing that stuff
>>>> as opposed to doing ntohl().
>>> 
>>> Indeed.  Ascii Timestamps amount for about 30% if Varnish CPU load :-/
>>> 
>>> --
>>> Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
>>> phk@FreeBSD.ORG         | TCP/IP since RFC 956
>>> FreeBSD committer       | BSD since 4.3-tahoe
>>> Never attribute to malice what can adequately be explained by
>>> incompetence.
>>> 
>>> 
>> 
> 
> 
> -- 
>  Matthew Kerwin, B.Sc (CompSci) (Hons)
>  http://matthew.kerwin.net.au/

--
Mark Nottingham   https://www.mnot.net/

Received on Thursday, 19 June 2014 00:32:34 UTC