git-svn: Branching clarifications

git-svn: Branching clarifications

[Russ Brown]

I have a few questions about how/when to use git branches when using
git-svn (I'm a tad confused...)

Say I've initialised and fetched a git repo involving trunk and one
branch (say branch1) from an svn repository.

If I do git branch -a, I see similar to the following:

* master
  branch1
  trunk

(branch1 and trunk are in red for me, which I figure means they're
remotely tracked or something like that?)

OK, so that's telling me that I currently have master checked out into
my working copy. My question is: where did master come from? Is it a
local branch of trunk?

Moving on, say I want to work on branch1. Can I simply issue git
checkout branch1? If I do so I get this:

$ git branch -a
* (no branch)
  master
  branch1
  trunk

Which is a bit scary. It seems my working copy is orphaned...

OK, so let's assume I'm supposed to create a local branch of each remote
branch I want to work on. So:

$ git branch local/branch1 branch1
$ git checkout local/branch1

$ git-branch -a
* local/branch1
  master
  branch1
  trunk

Am I supposed to have used --track when creating  this branch? What are
the implications for specifying or not specifying that flag when using
git-svn?

So I do some editing on this branch, commit and dcommit. The changes
appear as expected in the repo.

At this point if I checkout master, the contents look like
local/branch1, which isn't what I'd suspected (that it would be a branch
of trunk). What does master represent?

So I checkout local/trunk, and create a new file, commit and dcommit.
Umm, it's been committed to branch1 on the repo: not trunk,

So I figure I'm quite obviously doing something wrong here. Could
someone give me a hand and tell me what it is I'm getting wrong?

Thanks!

-- 

Russ

Re: git-svn: Branching clarifications

[Eric Wong]

Russ Brown <pickscrape@gmail.com> wrote:
> I have a few questions about how/when to use git branches when using
> git-svn (I'm a tad confused...)
> 
> Say I've initialised and fetched a git repo involving trunk and one
> branch (say branch1) from an svn repository.
> 
> If I do git branch -a, I see similar to the following:
> 
> * master
>   branch1
>   trunk
> 
> (branch1 and trunk are in red for me, which I figure means they're
> remotely tracked or something like that?)

Yes, that seems to be the case (I just enabled color.branch=auto in
.git/config for the first time).

> OK, so that's telling me that I currently have master checked out into
> my working copy. My question is: where did master come from? Is it a
> local branch of trunk?

git-svn sets "master" to the most recently committed-to branch
in SVN the first time it fetches.  "git-log master" will tell
you (look at the git-svn-id: lines).

After you do your initial fetch/clone, it should say something like:

  ----------------------------------------------------------------------
  Checked out HEAD:
    svn://my-repository-here/branches/foo r12345
  ----------------------------------------------------------------------

> Moving on, say I want to work on branch1. Can I simply issue git
> checkout branch1? If I do so I get this:
> 
> $ git branch -a
> * (no branch)
>   master
>   branch1
>   trunk
> 
> Which is a bit scary. It seems my working copy is orphaned...

Yes it is.  Branches under the refs/remotes/ hierarchy were created
back in the day to tell the local user they should not commit to
them directly.

> OK, so let's assume I'm supposed to create a local branch of each remote
> branch I want to work on. So:
> 
> $ git branch local/branch1 branch1
> $ git checkout local/branch1
> 
> $ git-branch -a
> * local/branch1
>   master
>   branch1
>   trunk

That's correct.  You can also use "git checkout -b local/branch1 branch1"
instead of those two commands.

> Am I supposed to have used --track when creating  this branch? What are
> the implications for specifying or not specifying that flag when using
> git-svn?

--track has no effect with git-svn.  dcommit will automatically figure
out which branch it should commit to[1].  Running "git-svn dcommit -n"
with 1.5.3 will tell you which URL you'll commit to.

> So I do some editing on this branch, commit and dcommit. The changes
> appear as expected in the repo.
> 
> At this point if I checkout master, the contents look like
> local/branch1, which isn't what I'd suspected (that it would be a branch
> of trunk). What does master represent?

(see above)

> So I checkout local/trunk, and create a new file, commit and dcommit.
> Umm, it's been committed to branch1 on the repo: not trunk,
> 
> So I figure I'm quite obviously doing something wrong here. Could
> someone give me a hand and tell me what it is I'm getting wrong?

If you run "git-log local/trunk", does the first commit to show
a "git-svn-id: " line have the URL pointing to trunk or branch1?

Again, if you're unsure about where you're committing to,
"git-svn dcommit -n" in 1.5.3 is your friend.

[1] - as long as you don't use git-merge or git-pull.  If you decide to
      do those things, make sure you have Lars's latest patches
      that enables --first-parent.
      Otherwise stick with format-patch/am/cherry-pick/fetch/rebase

-- 
Eric Wong

Re: git-svn: Branching clarifications

[David Kastrup]

Eric Wong <normalperson@yhbt.net> writes:

> git-svn sets "master" to the most recently committed-to branch
> in SVN the first time it fetches.  "git-log master" will tell
> you (look at the git-svn-id: lines).

Sigh.  Another "surprise the user by an arbitrary looking choice that
might possibly correspond to what he wants done because it something
obscure in the commit history suggests so" design decision.

I don't want my master set according to something that a coworker (or
even myself) happened to commit last to.

Please.  git-svn is told how to find the trunk on its command line.
Nothing makes sense (short of an _explicit_ wish otherwise for which
it might make sense to create a command line option) than to map
master to the trunk.

As a design rule: don't second-guess the user, _ever_, and
particularly not on decisions with large consequences.  A tool should
not have a mind of its own but do what it is told.  And if it can't
figure out what it is told, by simple, user-understandable criteria,
barf.  And of course have a way to _direct_ it when it can't figure it
out on its own, or if the simple and obvious default would not do the
right thing.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum

Re: git-svn: Branching clarifications

[Eric Wong]

David Kastrup <dak@gnu.org> wrote:
> Eric Wong <normalperson@yhbt.net> writes:
> 
> > git-svn sets "master" to the most recently committed-to branch
> > in SVN the first time it fetches.  "git-log master" will tell
> > you (look at the git-svn-id: lines).
> 
> Sigh.  Another "surprise the user by an arbitrary looking choice that
> might possibly correspond to what he wants done because it something
> obscure in the commit history suggests so" design decision.
> 
> I don't want my master set according to something that a coworker (or
> even myself) happened to commit last to.
> 
> Please.  git-svn is told how to find the trunk on its command line.
> Nothing makes sense (short of an _explicit_ wish otherwise for which
> it might make sense to create a command line option) than to map
> master to the trunk.

Keep in mind that command-line arguments for trunk, branches and tags
are _all_ optional to git-svn.

If only trunk or nothing is specified, the current behavior will always
be correct.

There's also a case if only branches and/or tags are specified, with no
trunk given.  That would need to be handled, somehow...

I've also tracked several (both OSS and closed) projects that have a
policy of doing all work on branches with a trunk that's almost never
up-to-date.

Tracking the last-committed branch was the easiest to code, and we even
tell the user which branch they're on.  I guess I could add a message
telling them all the other refs they can "git reset --hard" to if they
don't like their current one.

> As a design rule: don't second-guess the user, _ever_, and
> particularly not on decisions with large consequences.  A tool should
> not have a mind of its own but do what it is told.  And if it can't
> figure out what it is told, by simple, user-understandable criteria,
> barf.  And of course have a way to _direct_ it when it can't figure it
> out on its own, or if the simple and obvious default would not do the
> right thing.

git-svn used to never check anything out and leave HEAD dangling.  I was
happy with that, but I got a lot of user complaints from that, too.

-- 
Eric Wong

Re: git-svn: Branching clarifications

[David Kastrup]

Eric Wong <normalperson@yhbt.net> writes:

> David Kastrup <dak@gnu.org> wrote:
>
>> Please.  git-svn is told how to find the trunk on its command line.
>> Nothing makes sense (short of an _explicit_ wish otherwise for
>> which it might make sense to create a command line option) than to
>> map master to the trunk.
>
> Keep in mind that command-line arguments for trunk, branches and tags
> are _all_ optional to git-svn.
>
> If only trunk or nothing is specified, the current behavior will
> always be correct.

Sure, since Subversion does not distinguish trunk, branches, tags, or
even projects from each other: they are just naming conventions in the
repository and nothing enforces them.  So if you check out a single,
named directory, it is natural that this will be master, and tracked.

> There's also a case if only branches and/or tags are specified, with
> no trunk given.  That would need to be handled, somehow...

Just barf unless a --master=thisbranch option is given.  If you want
to, you can allow multiple --track=remotebranch options as well,
giving you a tracking branch for the specified remote branches under
their original name.

Something like that.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum