Added a (failing) test-case for two sequential git-cvs pushes without…

… a git-cvs pull in between

This was prompted by it being pointed out that two pushes fail, since
git-cvs pushes everything since the last cvsimport.

t/von-1.t

@@ -0,0 +1,94 @@
+#!/usr/bin/perl
+use strict;
+use warnings;
+use FindBin qw($Bin);
+use lib "$Bin/lib";
+use MyTest::Dirs;
+use MyTest::Replay;
+
+use Test::More tests => 2;
+
+# This tests pushing twice without pulling in between
+# Currently it fails!  
+
+# Define some directories
+my %D = MyTest::Dirs->hash(
+    data => [],
+    temp => [cvs_repo => 'cvs',
+             cvs_work => 'cvs_work',
+             git_repo => 'git'],
+);
+
+my $cvs_module = 'module1';
+
+# Create a cvs repo and working dir
+my $cvs = MyTest::Replay::CVS->new(path => $D{cvs_work},
+                                   module => $cvs_module,
+                                   cvsroot => $D{cvs_repo});
+
+$cvs->playback(<<ACTIONS);
+## Check in a couple of files
++one
++two 
+*cvs add one two
+*cvs ci -m "added one and two"
+ACTIONS
+
+# Create a git repo, which explicitly uses our dist's git cvs
+my $git = MyTest::Replay::Git->new(path => $D{git_repo},
+                                   exe_map => {
+                                       'git-cvs' => "$Bin/../bin/git-cvs",
+                                   });
+
+
+$git->playback(<<ACTIONS);
+## Init .git-cvs and make the first import from CVS
++.git-cvs cvsroot=$D{cvs_repo}
++.git-cvs cvsmodule=$cvs_module
+
+*git-cvs pull
+?one
+?two
+ACTIONS
+
+$git->playback(<<ACTIONS);
+## Now add a file in git
++alpha
+*git add alpha
+*git commit -m "added alpha" 
+
+*git-cvs push
+ACTIONS
+
+$git->playback(<<ACTIONS);
++beta
+*git add beta
+*git commit -m "added beta" 
+
+*git-cvs push
+ACTIONS
+
+__END__
+Firstly, thank you and congratulations on being the first self-confessed user of git-cvs!  I wasn't aware of anyone else using it so far.
+
+Now, to try and help you - something more specific would help me replicate your problem.  I'm guessing you're doing something like this?
+
+$ mkdir repo_name
+$ cd repo_name
+$ echo cvsroot=$CVS_ROOT      >.git-cvs
+$ echo cvsmodule=$CVS_MODULE >>.git-cvs
+$ git-cvs pull
+$ echo hello > foo
+$ git commit -a -m "comitted foo"
+$ git-cvs push
+$ echo hello again >>foo
+$ git commit -a -m "comitted foo again"
+$ git-cvs push # some error here?
+
+I've created a unit test which tries to replicate this, called t/von-1.t.  You might want to check this does indeed replicate your problem (I've checked it in).
+
+What I think you have found here is a "misfeature", which I personally avoid by git-pulling after each push (followed by a git reset --hard master to remove the merge point).  This keeps the git repo's master branch  in sync with the CVS repo HEAD branch (assumng you're on master).
+
+I did experiment with recording the last push with a tag, but shelved this idea, I think because correctly deducing the right commits to export in all circumstances seemed hard to get right.
+
+I'll think a bit more about tomorrow, when it might all seem much simpler.  I need sleep right now...