GnuCash+git development process with multiple branches?

[Derek Atkins]

John Ralls <jralls at ceridwen.us> writes:

[snip]
>> So I guess this boils does to more of a general git question than
>> specifically gnucash, but ---- how do you work with multiple branches
>> simultaneously?  This is particularly important because of having to
>> dcommit back upstream.
>> 
>> My first attempt was to do a git clone --bare and then figuring I could
>> git-clone from there for my branches.  It converts my 'svk sync' to a
>> 'git fetch' in the bare repo and 'svk update' to a 'git-update' (or git
>> pull) from my WCs.  Alas, this doesn' *quite* work because the git fetch
>> fails with:
>> 
>> Fetching origin
>> fatal: Couldn't find remote ref HEAD
>> error: Could not fetch origin
>> 
>> This is probably because refs/heads/ is empty in the bare repo?
>> 
>> So what do you all do?  I'm thinking that once I figure it out I'll
>> update the GnuCash Git Wiki Page, although honestly I'm probably the
>> last dev to migrate to git :-P
>> 
>
> Whew! I was afraid you were going to say that upgrading to perl 5.16 blew up
> git-update!

Nope, that is probably okay, otherwise I think Geert would have noticed
that already.  :)

> I just keep one git repo and flip around the branches as needed.

Yeah, alas I really need to be able to have multiple branches "checked
out" simultaneously.  I don't want a single tree that I'm constantly
swapping between branches.

> Left to itself, git pull will update all of the checked out branches
> in the local repo, but we can't let it do that because each branch
> needs to have its refs updated (particularly for new tags), so
> git-update only rebases the currently checked out branch, so you have
> to git-update each branch in turn. More important, you need to fetch
> and update between dcommits when you're working in more than one
> branch (like backporting a change) or git-svn gets all confused and
> you spend a bunch of time resetting it back and rebasing before it
> will let you dcommit again.

So what does "git-update" do that "git fetch" (or git pull) does not?
Could I theoretically run a 'git fetch' into the bare repo, clone that
into each checkout I need, git svn init in each WC, and then dcommit
from there back to SVN?  So I'd run 'git-update' from the WCs (instead
of 'git pull'), but still run 'git fetch' from the bare repo?  It does
mean that I'm holding around multiple copies of the git repo, but I
could use --local to try to conserve space.

> None of which explains why having a local bare clone wouldn't work. In
> fact, I just tried it and it seems to work -- though there aren't yet
> any changes to update, it didn't whine about anything:
>   $ athena:/Users/john> git clone --bare git://github.com/Gnucash/gnucash.gitCloning into bare repository 'gnucash.git'...

What if you do it against code.gnucash.org instead of github?

>   $ athena:/Users/john/gnucash.git> git fetch
>   From git://github.com/Gnucash/gnucash
>    * branch            HEAD       -> FETCH_HEAD
>   $ athena:/Users/john/gnucash.git>

I notice in my checkout that:

[warlord at mocana gnucash]$ more gnucash.git/HEAD 
ref: refs/heads/master
[warlord at mocana gnucash]$ more gnucash.git/FETCH_HEAD 
[warlord at mocana gnucash]$ cat gnucash.git/refs/heads/master
cat: gnucash.git/refs/heads/master: No such file or directory
[warlord at mocana gnucash]$

> I'll leave it and try again tomorrow after somebody commits something.
>
> Regards,
> John Ralls

-derek

-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord at MIT.EDU                        PGP key available