I pulled HOW much?

[Matt Nordhoff]

Matthew D. Fuller wrote:
> So I happened to look at the pack files after I pulled bzr today, and
> something seems awful messed up.
> 
> I was at r4439 (last rev before ~today; 4 behind).  The commit mails
> with diffs add up to ~150k for the group.  Obviously, the total of the
> revisions and all will probably be bigger even after compression, but
> still, it's a ballpark.
> 
> After the pull from http://bazaar-vcs.org/bzr/bzr.dev/, I've got a new
> packfile:
> 
> -rw-r--r--  1 fullermd  user    7304959 Jun 15 14:26 fb2c2b3623949c212c43653c3957ae27.pack
> 
> WHAT?  7 megs for a few hundred k of changes?  What the frell?
> 
> 
> To experiment, I created a clean standalone bzr.dev:4439 and tried
> again.  Same size to the byte.
> 
> 
> Then I did the same thing, and pulled from lp:bzr (bzr+ssh).  That
> created a much more sensible:
> 
> -rw-r--r--  1 fullermd  wheel     411542 Jun 15 14:37 7a475eea60d2eb3f26f139cd2fac98f0.pack
> 
> 
> Well, let's try it again from launchpad, but over http:
> 
> -rw-r--r--  1 fullermd  wheel    6473856 Jun 15 14:41 5f38d8f463b267a342e251d5814ef95c.pack
> 
> 
> Now, after doing a pack, it compresses down to a combined size that
> isn't nearly as insane (143 + 4.2 + 7.0 = 144).  But why is the pack
> grabbed over HTTP an order of magnitude bigger than the data out of it
> that we really need?  I presume it's fallout from some tradeoff made
> for number of round trips to a dumb server, but still, 10x the data
> transfer?

I've been noticing this for a while now, too. On my American ADSL
connection, downloading ~5 MB of data to pull a handful of revisions is
*really* annoying.

I just ran some test pulls on a copy of r4442 of bzr.dev with various -D
options. There was only 1 mainline revision to pull, but it downloaded
about 8.2-8.5 MB of data. FWIW, I put up a copy of .bzr.log:

<http://cheezum.mattnordhoff.com/tmp/.bzr.log>
--