Re: [Monotone-devel] partial pull #2 - gaps instead of a single horizon

[Nathaniel Smith]

On Wed, May 30, 2007 at 12:41:36PM -0700, Rob Schoening wrote:
>    Also speaking as a user, it seems to me that the real use-case that drives
>    this feature is simply not transmitting "ancient" history in cases where
>    the user is unlikely to ever want it.  The common use case consists of
>    the the guy who is not a monotone expert and wants to check out the head
>    of a given branch but doesn't understand why he needs to download 100s of
>    megabytes of data to do so.

The thing is, our experience with monotone and reports from other
systems (including ones that have been scaled farther, e.g. BK) all
say that with appropriate delta compression, your average repository
is within a small constant factor of the size of a checkout.  (Delta
compression _really_ works.)

So in practice, unless you do have decades of history, partial pull
will only save you, say, a factor of 3 -- very different from the
"100s of mysterious megabytes" idea.

>    Rather than make this feature pull-er specified, would it be possible
>    (offline) to employ a checkpoint/archive approach on the part of the
>    developers committing work?  At the far extreme, one can do this today, by
>    simply abandoning the history and creating an entirely new repository
>    consisting of the current heads.  Before you laugh too hard, remember that
>    Linus did exactly this when he moved to Git.

"Nightly snapshot tarballs"? :-)

-- Nathaniel

-- 
Eternity is very long, especially towards the end.
  -- Woody Allen