Re: Straw Poll: Restore Header Table and Static Table Indices

[Mark Nottingham <[e-mail]>]

I think we need to see a proposal here if we’re going to take this seriously — i.e., either fairly complete text or a pull request.

Are you interested, and can you do so in a small number of days?

Ill say outright that it feels to me that we’re in the weeds hyper-optimising the format, whereas we’ve agreed many times that doing so isn’t a high priority. Thus, I suspect that it’s going to be difficult to get consensus on such a proposal — but (as always) I’m happy to be proven wrong by the WG.

Cheers,


On 16 Oct 2014, at 8:16 pm, Willy Tarreau <[e-mail]> wrote:

> On Thu, Oct 16, 2014 at 09:09:55AM +0000, RUELLAN Herve wrote:
>>> -----Original Message-----
>>> From: Willy Tarreau [mailto:[e-mail]]
>>> Sent: mercredi 15 octobre 2014 22:16
>>> To: Amos Jeffries
>>> Cc: [e-mail]
>>> Subject: Re: Straw Poll: Restore Header Table and Static Table Indices
>>> 
>>> On Wed, Oct 15, 2014 at 06:20:00PM +1300, Amos Jeffries wrote:
>>>> -----BEGIN PGP SIGNED MESSAGE-----
>>>> Hash: SHA1
>>>> 
>>>> On 15/10/2014 6:11 p.m., Adrian Cole wrote:
>>>>>> If an argument can be made that 2 byte encodings are still too
>>>>>> large for dynamic headers, then instead of flipping back let's
>>>>>> investigate how the 1 byte slots can be shared between static and
>>>>>> dynamic.
>>>>> FWIW, I'm happy to implement an alternate approach, if one comes
>>>>> out. Thanks, Greg.
>>>>> 
>>>> 
>>>> Alternative approach has already been proposed. That the first bit of
>>>> the index is used as a flag to indicate static or dynamic table for
>>>> the remaning 7+ bits.
>>>> 
>>>> That not only puts both on an even bias, but expands the range of
>>>> values getting 1-byte encodings in either table and removes the need
>>>> for the math complexity people are disliking.
>>>> 
>>>> 1 stone, 3 birds.
>>> 
>>> I totally agree with this except that I don't see where you take that
>>> spare bit from, that's what initially led me to rethink the encoding.
>>> So if you konw where to find one bit, +1 for me obviously!
>> 
>> Everywhere where there's an index into the static/dynamic table, reduce the
>> index size by 1 bit and use this bit as a flag to find which table is used.
> 
> OK that way it's clear. Initially I didn't want to do it that way (when
> I tried to adjust the encoding to save more bits) because I believed we
> needed to have all the static headers in 1 byte. But assuming the most
> common ones have the lowest indexes, it makes sense to do so, as we can
> encode the first 32 of them with 1 byte, same for the dynamic ones.
> 
> Thus +1 for me.
> 
> Willy
> 
> 

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