[Python-Dev] Looking for VCS usage scenarios

[Nick Coghlan]

Paul Moore wrote:
> 2008/11/5 Barry Warsaw <barry at python.org>:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On Nov 5, 2008, at 6:09 AM, Paul Moore wrote:
>>
>>> I'll freely admit a (not very) hidden bias here - the slowness of an
>>> initial clone (or going through the "download a shared repo, unpack
>>> it, create a branch and update" rigmarole) makes this a nasty test for
>>> Bazaar. But I do nevertheless think it's an important use case, as
>>> it's all about encouraging casual users to contribute.
>> One thing that we're not taking advantage of on code.python.org for Bazaar
>> is "stacked branches".  This is basically an indirection technique that
>> doesn't require a full revision history download to get started.  Stacked
>> branches greatly improve performance for the initial pull and push, and
>> reduces disk usage on the server.
> 
> While that would be good, my understanding is that stacked branches in
> Bazaar only work (for history operations) while you're online. So they
> make Bazaar work a little like a centralised VCS, I guess. Not sure
> how that's a step forward :-)

You only need full history in order to ask "how did this code get to
this state" type questions. That's something a core dev will
occasionally want to know (which is why we would take to the time to
setup a full checkout no matter how long it takes), but utterly
irrelevant to a casual committer such as the one in your "write and
submit a quick patch" scenario.

Also, the DVCSs shouldn't be compared just against each other - we need
the benchmark numbers against the current SVN repository as well.

Cheers,
Nick.

-- 
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia
---------------------------------------------------------------