RE: I ran across this while working on the spec.

If you maintain the reference set into the static table as a bitmask, we're talking about eight bytes per connection.

From: Roberto Peon [mailto:grmocg@gmail.com]
Sent: Thursday, October 17, 2013 11:47 AM
To: Martin Thomson
Cc: Jeff Pinner; RUELLAN Herve; Fred Akalin; Roberto Peon; ietf-http-wg@w3.org Group
Subject: Re: I ran across this while working on the spec.

I have (that was what Herve proposed :) ), but I do worry: A set typically
requires two pointers per element, plus the size of the element itself plus
pointer to the index itself.
If we call that 8+8+4 bytes of overhead per entry, we're talking about a kilobyte of potential overhead. That is noticable.

-=R

On Thu, Oct 17, 2013 at 11:43 AM, Martin Thomson <martin.thomson@gmail.com<mailto:martin.thomson@gmail.com>> wrote:
On 17 October 2013 11:17, Roberto Peon <grmocg@gmail.com<mailto:grmocg@gmail.com>> wrote:
> Lets call this #6.

For #6, I think that you want to allow indexed representations to use
the static set, otherwise you don't get to take advantage of the
values there, just the keys.  You might say, however, that indexing
these doesn't add them to the reference set.

I really don't have a particular preference here, but I do note that
there is a strict upper bound on the size of the reference set after
evicting all dynamic entries.  That is, it's the size of your
reference, times the size of the static table.  Have you considered
the "don't worry"?