[rfc] bzr-colo into core

[John Szakmeister]

On Wed, Mar 23, 2011 at 1:23 AM, Eli Zaretskii <eliz at gnu.org> wrote:
>> From: "Stephen J. Turnbull" <stephen at xemacs.org>
>> Cc: Nicholas Allen <nick.allen at onlinehome.de>,
>>     mbp at canonical.com,
>>     bazaar at lists.canonical.com
>> Date: Wed, 23 Mar 2011 11:38:33 +0900
>>
>> Erm, getting additional workspaces is trivial, as even git[1] users know.
>> Just clone the repo again.  Every git user does this, even for small
>> projects.  What's the big deal?
>
> Not a big deal, but clearly inferior to what we already have in bzr:
> separate branches in a shared repo.  For starters, disk space savings
> (that are claimed to be one reason why colocated branches are an
> improvement) will go down the drain.  More generally, emulating
> something is always going to be inferior to the thing itself.

I disagree.  The shared repository abstraction leaks sometimes.  Such
as when hosting a branch and trying to put access controls on it.
Instead of proxying requests through the branch url, it finds the
shared repo and communicates at that level.  Additionally, I find that
set up harder to use because of the setup involved.  I taught half a
dozen people how to use Bazaar at my workplace, and their first
inclination was almost always wrong (as in way too much disk space--we
use virtual machines a lot, and slow).  We used the shared repository
set up and switch to bzr-colo later because it was so much easier to
get up and running.  I personally don't think the shared repos setup
is *that* hard, but I can tell you that having bzr-colo made a huge
difference in usability for the team.

> Anyway, I think the consensus here is that colocated branches is a
> Good Thing to have, even as the default model, but the ability to have
> separate branches is better in some important use cases, and should
> stay.

Agreed.  But I think at some point some of the models need to get
trimmed out, as it adds a lot more edge cases for testing, and many
more ways for the user to get confused.  Such as the whole '--local'
commit thing... it should really just go away.

-John