Re: #578: getting real-ish numbers for option 3

Hi Martin,

On Mon, Oct 27, 2014 at 09:55:42PM -0700, Martin Thomson wrote:
> On 27 October 2014 21:43, Amos Jeffries <squid3@treenet.co.nz> wrote:
> > Statistical significance is (N >= 0.05) .
> 
> I think that this is a good time to remind people that we aren't
> looking for perfection.  The reference set provides a far better
> compression gain than anything I've seen in this thread, in any case.
> We dropped that due to concerns over complexity.
> 
> We've always said that we are merely looking for a robust and
> efficient compression scheme, not a perfect one.

The goal of the proposed change was not to seek perfection at all, but
to fix a defect of the current method which makes it degrade in certain
environments and *will* make it degrade over time as new headers are
introduced. It may for example save us from having to introduce a new
static table in a few years to work around the current defect.

In my opinion it's important to listen to people who invented this
compressor. When one of its authors considers that the current version
is now biased and will not correctly adapt to the future, this has
much more significance to me than the few minutes a handful of early
implementers who do not need to understand how compression works
will have to spend to update their running code (because changing
from draft-09 to the other ones really took minutes here). Also,
Tatsuhiro has already updated his code so for some people it might
simply be a matter of rebuilding the code.

Best regards,
Willy

Received on Tuesday, 28 October 2014 07:06:48 UTC