A good Git technique for referring back to original files

Hi,

I have a client with an SDK product. Normally the SDK is used in its unpackaged form by the end-user, and that is the directory structure and set of files in which development work on the SDK functionality is performed.

However the SDK directory and content is generated from a packager which first runs on numerous other version controlled projects (currently CVS projects - this is unlikely to change).

This means that once changes to the unpackaged SDK have been tested, they have to be cross-referred back to the original projects and the changes ported back.

I have found it most convenient to control my in-SDK changes with git.

However it's still a royal pain to cross-refer and diff the changes back to the originals, especially since duplicate file names exist across the original projects which get filtered down to one relevant instance by the packager.

Since git is so good at tracking file content, I wondered whether there was any technique using git that would simplify the back-referencing task.

Failing a method using git 'normally', perhaps building a script on top of the git file system might be a possibility - if that is feasible ...

Thanks, MikeW

MikeW <mw_phil@yahoo.co.uk> writes:

I'm not sure I understand the question, but if you want to add meta-data to Git commits (e.g. "this Git commit is revision 42 in CVS repository foo"), then have a look at git-notes. It won't give you directly "reference to other VCS", but at least can be used as a storage mechanism to store these references.

Matthieu Moy <Matthieu.Moy <at> grenoble-inp.fr> writes:

there was any

Thanks for the reply.

In my work environment both the SDK and the original files are available (in an enclosing directory).

--SDK_content
  |
  SDK_subproj1-- ...
  |            |
  |            content
  |
  SDK_subproj2- ...
  |            |
  |            content
  |
  SDK_subprojN- ...
  |            |
  |            content
  |
  Working_SDK ... (under git, baseline generated from subproj1..N)
               |
               content derived from subproj1..N

What I had in mind was something I could run over, say, SDK_content (alternatively, from within Working_SDK, referring back to SDK_content) which would note the changed files in Working_SDK and locate the original files in SDK_subproj1..N letting me merge the changes back.

Hi Mike,

I think git-cvsimport and git-subtree could help you here.

Roughly:

# Create a git version of each SDK_subproj git cvsimport -r upstream -d $CVSREPO1 $CVSMODULE1 -C SDK_subproj1 git cvsimport -r upstream -d $CVSREPO2 $CVSMODULE2 -C SDK_subproj2

# Create your Working_SDK git init Working_SDK cd Working_SDK

# Import the SDK_subprojN repos git subtree add --prefix=subproj1 ../SDK_subproj1 upstream/master git subtree add --prefix=subproj2 ../SDK_subproj2 upstream/master

# Edit and commit your files # N.B. when committing don't commit to more than one subproj in a single commit

# Update from the upstream CVS as needed git cvsimport -r upstream -d $CVSREPO1 $CVSMODULE1 -C ../SDK_subproj1 git subtree pull --prefix=subproj1 ../SDK_subproj1 upstream/master git cvsimport -r upstream -d $CVSREPO2 $CVSMODULE2 -C ../SDK_subproj2 git subtree pull --prefix=subproj2 ../SDK_subproj2 upstream/master

# Push your changes back to SDK_subproj repos into a branch other than master git subtree push --prefix=subproj1 ../SDK_subproj1 new-branch git subtree push --prefix=subproj1 ../SDK_subproj2 new-branch

# Prepare patches to apply to a real CVS copy or submit upstream (cd ../SDK_subproj1 && git format-patch upstream/master..new-branch) (cd ../SDK_subproj2 && git format-patch upstream/master..new-branch)

Hope that helps.

-- Paul

On Tue, Feb 12, 2013 at 10:19 AM, MikeW <mw_phil@yahoo.co.uk> wrote:

-- Paul [W] Campbell

Paul Campbell <pcampbell <at> kemitix.net> writes:

That looks very interesting, had not considered git subtree and it looks like the right kind of method.

Thanks. Mike

... Super-Snip ...

MikeW <mw_phil <at> yahoo.co.uk> writes:

The only alternative I can think of is to scan through Working_SDK and replace the files there with symlinks back to the matching files within the original subprojects - such scripts exist !

Then any changes in Working_SDK will update the (baselined) originals in-place.

But then no explicit use of git except for tracking work prior to pushing back to CVS.

Oh well, thanks for ideas, will see which work best.

Mike