[SciPy-Dev] git on windows (was: scipy.stats)

[Matthew Brett]

Hi,

> My main problem with git was the treatment of the file system, and I
> find it much easier to work with separate branches as in bzr or
> mercurial.

Yes, it is true that the git lightweight branch model takes some time
to get used to.  My experience is that it's quick to get used to the
git way, and once I did, it was a large relief to get rid of all those
branch directories when we switched, but I understand that it is a
leap.

I am sure you know this, but you can replicate the heavyweight
branches of hg and bzr with:

# initial git clone of 'trunk'
git clone git://github.com/nipy/nipy.git
# make a heavyweight branch
git clone nipy my-nipy-branch
# push somewhere
# First add repo for the branch via github interface, then
cd my-nipy-branch
git remote add origin git at github.com:matthew-brett/my-nipy-branch.git
git push origin master

I think you'd agree that it's not a windows / unix difference though.
I'd agree it is a larger conceptual leap from svn to git than it is
from svn to bzr or svn to mercurial.  The git argument is that making
that initial leap gives you a great deal of freedom and flexibility,
but it can be intimidating at first.

> As long as it is possible to stick with the basic workflow of git
> without anything fancy, similar what I have seen while skimming the
> nipy docs, I think it is not a problem on windows.

I think that is true that most of us won't need to go further than the
nipy basic workflow - but we haven't been using git long enough to
know that very well.  I would defer to the git masters out there -
David, Pauli and others - ?

> However, if/when parts of statsmodels go into scipy and I have to do
> maintenance of less isolated code, then I think the Mercurial
> interface might be my preferred choice.
>
> I haven't used Mercurial much yet, but I don't see any problems with it.
>
> So, the bottom line is, that documentation for the hg-git interface
> would be very useful for Windows users (or those that think git is a
> strange/unfamiliar concept.)

So - two issues:

1) The conceptual issues involved in switching mind-set from svn or
bzr to git.  That may require some thought and documentation
2) There might be some technical issues using git on windows - but I
think so far we don't have any reason to think so?
3) Some people may prefer mercurial for other reasons; it would be
good to respect that if possible.

So, it may well be worth making a hg-git doc for numpy when we do the
transition - with the caveats that David raised.

In the meantime, it would be very good to hear of any problems that do
come up specifically using git on windows...

See you,

Matthew