[PATCH] Git SCM for Capistrano 2.0

The existing patch doesn't support the remote_cache deployment strategy. I've enhanced it to do that and will upload an enhanced patch with tests soon.

http://pastie.textmate.org/100640

I've modified topfunkys patch with the ability to specify a depth for a shallow clone for a much faster clone if not using the remote_cache strategy.

http://pastie.textmate.org/101098

+1 - I've tested both garry's and topfunky's versions and they both work

Thanks for the updates guys. I'll take both versions and attach an updated patch.

topfunky,

There's an issue with the sync() method if someone specifies "origin/cool_branch" as the branch to deploy. All branches will be remote branches cloned from the push-only repo specified by :repository. Nothing new there. But, the sync() would generate:

git pull origin origin/cool_branch

which would be incorrect, "origin/" needs to be stripped off the branch name since origin is already specified as the remote (1st arg to pull).

Alternatively, since string manipulation does not sound like a good solution, we can modify the sync() like so:

git fetch origin && git merge origin/cool_branch

This does about the same thing, only difference is that the remote branch updates are stored, instead of thrown away with 'pull'. Requires more disk space but not much of an issue otherwise, IMO.

If no branch is specified at all, then the 2nd half of that command would reduce to "git merge HEAD", which does nothing useful. So I have to work that out.

Your thoughts?

Gentlemen,

I've updated the attached .diff with the contributions from topfunky and Scott. I had to modify sync() a bit, but the general idea is still there. I added tests for sync() and :git_shallow_clone.

Thanks guys for your contributions.

Oh, and I answered my own question by playing around with a test repo. "git merge HEAD" doesn't do anything, but "git merge origin/HEAD" does indeed make the local branch up-to-date w/ the remote origin. Assuming we never checkout'd a specific branch (just left it at the default), origin's HEAD is the remote branch we'd be tracking.

+1

This works for me. Originally I was confused by the -b deploy on initial clone, but it makes sense now. I tried it with master and with other branches and it works correctly.

The only thing is that one may need to delete the repository_cache if deploying a different branch. I use this:

desc "Delete repository cache on server"
task :delete_repository_cache do
  run "rm -rf #{shared_path}/#{repository_cache}"
end

Yep, without "-b deploy" you'll get a warning each time you checkout.

Regarding the repistory_cache, they'll indeed need to delete the cache if a different branch is deployed. I wonder if we could detect what branch they "came from" (sibling of "deploy"), and if a different one is deployed, it'll do a checkout instead of merge. I'll play around with this some.

(In [7695]) Git SCM module (closes #9635)

Thanks for the excellent patch! I decided to apply it to the 2.1 release after all, given that it doesn't touch any existing functionality and looks well tested and reviewed.