[MERGE] 'bzr pack' tells the index builder to optimize

[Andrew Bennetts]

John Arbash Meinel wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Andrew Bennetts wrote:
> > bb:tweak
> > 
> > The tests feel slightly weak, because they are mainly about testing that
> > a flag is or isn't set on an object, rather than testing that
> > for_size=True and friends actually cause a change in behaviour.  I'm
> > don't think it's easy fix that, so I'm fine with this, but I felt I
> > ought to mention it.
> 
> Well, I considered that we could create a btree format repository, and
> then check the size of the index. If we force known data into it, we
> should be able to craft something that will pack better with the
> advanced settings. That said, it is going to be a bit voodoo, and I
> wouldn't want to expect that zlib will produce exactly the same bytes
> across all platforms.
> 
> So if we made the data nearly trivially redundant (keys of all the same
> character, etc), then I know the old form would only minimally compress
> them, and the new form would pack more per page. It still feels like a
> "soft" test to me, do you want me to put in the effort?

That's what I thought.  I don't think it's necessary.

-Andrew.