[bt-devel] Misgivings about git (and Gitorious)

[Greg Hellings]

On Sat, Aug 21, 2010 at 10:10 AM, Gary Holmlund <gary.holmlund at gmail.com> wrote:
> On 08/21/2010 07:55 AM, Raoul Snyman wrote:
>>
>> On Saturday 21 August 2010 15:07:18 Martin Gruner wrote:
>>
>>>
>>> Thanks for giving us your opinion! If it really is the case that after a
>>> testing period several people say that they cannot work with Git, then
>>> we will definitely reconsider it.
>>>
>>
>> I certainly can't use it - I have to use the bzr-git bridge in order to be
>> able to work the way I want to.
>>
>> I know that there are others who would prefer working in something other
>> than
>> git, but they haven't voiced it on the mailing list, only in the IRC
>> channel.
>>
>
> As I said in IRC, I can work with either git or bzr. I know that there are
> some good and bad points about both. I am certainly willing to switch to bzr
> if others want it.

While I can work in either, bzr is certainly the conceptually easier.
I know most people are comfortable with the multiple-directory
approach from having used SVN (and doing several checkouts).  It fits
well for many projects, like while I was getting the Python and Perl
bindings working for sword, where the git branch model does not work.
Getting comfortable with git's workflow is much more laborious and
takes far more time.  The same can be achieved with git, it's just
more cumbersome.

Also, to claim that git is so much bigger and more popular than bzr is
sort of moot.  There are some major projects that use bzr as well -
the whole Ubuntu distribution as well as MySQL are both hosted there.
True, most of the other big projects have gone over to git - Gnome,
KDE, kernel, Qt, etc.  But bzr still has sizable players and
commercial backing.  It's very unlikely to go anywhere so long as
Ubuntu and Canonical are backing it.

The issues of size and performance between git and Bazaar have been
rapidly narrowing.  With the move to the 2.0 and 2.1 series, pushing
and pulling are still a bit slower, but the other processes are now on
a par with git's performance.  And the relative difference in speed
for the size of our pushes and pulls comes to git's ~1s compared to
bzr's ~4s.  For much larger projects, the relative performance
gap/difference is much narrower.

For my own work with SWORD and other subversion repositories, I use
bzr.  Both because I do like the discrete directories it is designed
around and because its integration with svn feels more robust, after
that fiasco of converting the BibleTime and all its tags and branches
over.  I don't know about the differences for just developing a new
feature - I'm sure that probably can get away with the single branch
at a time in the repo that git advocates, but for doing build system
additions and changes, having the two versions in parallel to work on
that bzr does is a major help.

I suppose my point is that there are very solid reasons to go with
both.  I lean slightly towards bzr, since my hands are almost
exclusively in the build system and being able to build two branches
in parallel is a big help with that.  But I can also finagle git to my
whim without too much trouble using its clone command for local
cloning.

--Greg