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

I did some quick testing and the second proposal from Willy has almost no impact on the compaction size (on the HR test set from the compression testing suite).

I implemented it in a slightly different way: I use a 1-bit flag to determine if an index is into the static table or the dynamic table. It seemed much simpler to implement and less prone to errors.

Hervé.

> -----Original Message-----
> From: Nicholas Hurley [mailto:hurley@todesschaf.org]
> Sent: mardi 7 octobre 2014 18:36
> To: Willy Tarreau
> Cc: Mark Nottingham; HTTP Working Group
> Subject: Re: Straw Poll: Restore Header Table and Static Table Indices
> 
> On Tue, Oct 7, 2014, at 09:26, Willy Tarreau wrote:
> > For the first proposal I can understand, but for the second one, where's
> > the big change in just applying an offset to the index number (since it's
> > basically what it boils down to, by having only the tail of the static
> > header table not fit in the single-byte encoding) ?
> 
> My issue is that it's totally untested, and the fact that we have
> offsets at all in HPACK (the 1-based indexing versus most languages
> being 0-based) has already caused enough interop problems in my
> experience. It's silly, in my opinion, to introduce the possibility for
> even more interop problems of the same ilk as those we have already
> fixed. That's what makes it significant.
> --
> Peace,
>   -Nick

Received on Tuesday, 7 October 2014 16:49:59 UTC