shrink-revlog bakeoff: preliminary results

[Greg Ward]

On Thu, Feb 4, 2010 at 9:20 PM, I wrote:
> My opinion:
>  * we should add the ability for pluggable shrink algorithms so that
> it's easy to repeat these kind of experiments in future (I've already
> posted the patches, and will repost soon once I've tweaked them a bit)
>  * we should pick one of parentdelta and reversepostorder as the
> default algorithm and ditch the other.  No need to have equivalent
> algorithms kicking around.
>  * we should remove branchsort, since it is clearly inferior to
> parentdelta/reversepostorder
>  * we might want to keep postorderreverse in our back pockets: in its
> current form it is unsuitable because of its instability, but perhaps
> it's possible to improve and make it more consistent.  (It
> demonstrates that it is possible to do better than parentdelta on at
> least some data.)

Sorry about not following up on this.  I've had multiple hardware
failures and still haven't gotten these patches off the disk where
they are lying dormant.  Soon, I hope!  Just need a working
motherboard that I can hook that disk up to.  ;-)

Greg