Git module for capistrano should use checkout instead of merge.

Patch to use git checkout instead of branch & merge

+1 Works for me.

+1 so far so good

+1 git 2 it!

+1

+1 :)

Removing verified tag as that's for rails patches, not capistrano ones.

However, this looks good jamis :)

(In [8901]) use checkout instead of merge for git SCM (closes #10918)

But if the code never changes on the server, like you mentioned, why would there be conflicts?

You removed the "-b" in checkout, which I had originally when I wrote this module, it causes problems in the long run.

IMO, I'd say don't apply this patch, but I guess it's too late.

This patch, accepted into rev 8901, produces failing test cases:

/Volumes/FLASH/dev/capistrano $ svn up -r 8901
U    test/deploy/scm/git_test.rb
U    test/configuration/namespace_dsl_test.rb
U    lib/capistrano/upload.rb
U    lib/capistrano/recipes/deploy/scm/subversion.rb
U    lib/capistrano/recipes/deploy/scm/git.rb
U    lib/capistrano/configuration/namespaces.rb
U    CHANGELOG
Updated to revision 8901.
/Volumes/FLASH/dev/capistrano $ ruby test/deploy/scm/git_test.rb 
Loaded suite test/deploy/scm/git_test
Started
F............FF
Finished in 0.104176 seconds.

  1) Failure:
test_checkout(DeploySCMGitTest)
    [test/deploy/scm/git_test.rb:25:in `test_checkout'
     /Users/garry/sys//lib/ruby/gems/1.8/gems/mocha-0.5.5/lib/mocha/test_case_adapter.rb:19:in `run']:
<"git clone git@somehost.com:project.git /var/www && cd /var/www && git checkout -b deploy HEAD"> expected but was
<"git clone git@somehost.com:project.git /var/www && cd /var/www && git checkout HEAD">.

  2) Failure:
test_shallow_clone(DeploySCMGitTest)
    [test/deploy/scm/git_test.rb:69:in `test_shallow_clone'
     /Users/garry/sys//lib/ruby/gems/1.8/gems/mocha-0.5.5/lib/mocha/test_case_adapter.rb:19:in `run']:
<"git clone --depth 1 git@somehost.com:project.git /var/www && cd /var/www && git checkout -b deploy HEAD"> expected but was
<"git clone --depth 1 git@somehost.com:project.git /var/www && cd /var/www && git checkout HEAD">.

  3) Failure:
test_sync(DeploySCMGitTest)
    [test/deploy/scm/git_test.rb:54:in `test_sync'
     /Users/garry/sys//lib/ruby/gems/1.8/gems/mocha-0.5.5/lib/mocha/test_case_adapter.rb:19:in `run']:
<"cd /var/www && git fetch origin && git merge origin/HEAD"> expected but was
<"cd /var/www && git fetch origin && git checkout origin/HEAD">.

15 tests, 21 assertions, 3 failures, 0 errors

What the heck guys? I wrote those tests so this kind of breakage wouldn't occur.

The Rails patch submission guidelines even say to provide test cases, I figured it was *assumed* that existing tests should pass also.

With these test failures, I must -1 this patch (too late, but oh well).

"But if the code never changes on the server, like you mentioned, why would there be conflicts?"

Here's the scenario where I encountered conflicts:

We have 3 branches: master, release-1.0, bugfix-1.0. bugfix-1.0 has descended from release-1.0, which is descended from master. bugfix-1.0 is the release currently deployed on the server. Now, when we cut release-2.0 from master, which has changed significantly since release-1.0 including the removal of some files that are in bugfix-1.0, if we do a merge onto the deploy branch, it will create conflicts because the new release is not descended from bugfix-1.0.

Using the checkout avoids this entirely. Sorry that I missed fixing those tests - I'll create a patch shortly.

OK, I see where the conflicts can arise, thanks for the explanation.

No need for a patch on those tests, b/c later patches were applied and it rakes clean. The current version doesn't even use your switch from "merge" to "checkout", in fact it uses "reset --hard". I'm not sure why it was changed, I don't like to do reset on anything unless absolutely necessary.

Ah, that makes sense. reset --hard just sets the current HEAD, so it's actually a little safer than checkout because checkout will fail if there are uncommitted changes.