New files in git weren't being downloaded during CVS update

If a repository was checked out via git-cvsserver and then later a new file is added to the git repository via some other method; a CVS update wasn't fetching the new file.

It would be reported as a new file as
A some/dir/newfile.c
but would never appear in the directory.

The problem (I think) is that when git-cvsserver detected a new file, it was issuing the new file message then skipping the actual file send part and moving to the next file its list. In fact only an updated file would be transmitted.

The fix is to make the added file section identical to the udpated file section. This additionally makes git-cvsserver behave like a traditional CVS server and will now output
U some/dir/newfile.c
for an added file.

---
This is in response to Simon Schubert's suggestion that T_ADDED is an inappropriate category for a remotely added file. Instead this treats remotely added files the same as remotely changed files.

 git-cvsserver.perl |   10 +++++-----
 1 files changed, 5 insertions(+), 5 deletions(-)

index a33a876..501c182 100755
@@ -879,11 +879,11 @@ sub req_update
 	    elsif ( !defined($wrev) || $wrev == 0 )
 	    {
 	        $log->info("Tell the client the file will be added");
-		print "MT text A \n";
-                print "MT fname $filename\n";
-                print "MT newline\n";
-		next;
-
+			print "MT +updated\n";
+			print "MT text U \n";
+			print "MT fname $filename\n";
+			print "MT newline\n";
+			print "MT -updated\n";
 	    }
 	    else {
                 $log->info("Updating '$filename' $wrev");

1.5.0.rc1.gf4b6c

I'm still concerned that this fix is not right. Having thought more about it, I suspect that this breaks the other "added-file" case, where the file is added locally but not remotely.

There are three key cases:
1. File present locally and remotely. Modified remotely. Response "U"
2. File present locally only. Response should be "A"
3. File present remotely only. Response should be "U"

I think the real problem is that both 2 and 3 are being handled in the same place. Hence, my patch, which has fixed case 3; will have broken case 2.

I need a bit of confirmation from Martin; but I suspect that the correct fix is something like this:

    elsif ( (!defined($wrev) || $wrev == 0) && !defined($meta->{revision})
    {
        $log->info("Tell the client the file will be added");

If I'm correct, this would only run the added section when there is no matching revision in the repository - this would be case 2. Then case 3 would be handled as the else to this if, which handles every other case.

My testing of this works, but I'd like confirmation that I'm right in this thinking? Patch to follow...

Andy

Dr Andy Parkins, M Eng (hons), MIEE
[email protected]

If a repository was checked out via git-cvsserver and then later a new file is added to the git repository via some other method; a CVS update wasn't fetching the new file.

It would be reported as a new file as
A some/dir/newfile.c
but would never appear in the directory.

The problem seems to be that git-cvsserver was treating these two cases identically, as "A" type results.

1. New file in repository
2. New file locally

In fact, traditionally, case 1 is treated as a "U" result, and case 2 only is treated as an "A" result. "A", should just report that the file is added locally and then skip that file during an update as their is (of course) nothing to send.

In both these cases there is no working revision, so the checking for "is there no working revision" will return true. The test for case 2 needs refining to say "if there is no working revision and no upstream revision". This patch does just that, leaving case 1 to be handled by the normal "U" handler.

I've also updated the log message to more accurately describe the operation. i.e. that "A" means that content is scheduled for addition; not that it actually has been added.

---
 git-cvsserver.perl |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

index a33a876..2479f71 100755
@@ -876,9 +876,9 @@ sub req_update
                 print "MT newline\n";
 		next;
 	    }
-	    elsif ( !defined($wrev) || $wrev == 0 )
+	    elsif ( (!defined($wrev) || $wrev == 0) && (!defined($meta->{revision}) || $meta->{revision} == 0) )
 	    {
-	        $log->info("Tell the client the file will be added");
+	        $log->info("Tell the client the file is scheduled for addition");
 		print "MT text A \n";
                 print "MT fname $filename\n";
                 print "MT newline\n";

1.5.0.rc2.g3f8f2

If a repository was checked out via git-cvsserver and then later a new file is added to the git repository via some other method; a CVS update wasn't fetching the new file.

It would be reported as a new file as
A some/dir/newfile.c
but would never appear in the directory.

The problem seems to be that git-cvsserver was treating these two cases identically, as "A" type results.

1. New file in repository
2. New file locally

In fact, traditionally, case 1 is treated as a "U" result, and case 2 only is treated as an "A" result. "A", should just report that the file is added locally and then skip that file during an update as there is (of course) nothing to send.

In both these cases there is no working revision, so the checking for "is there no working revision" will return true. The test for case 2 needs refining to say "if there is no working revision and no upstream revision". This patch does just that, leaving case 1 to be handled by the normal "U" handler.

I've also updated the log message to more accurately describe the operation. i.e. that "A" means that content is scheduled for addition; not that it actually has been added.

---

Apologies; this one is the real replacement. The previous patch didn't include the correction to the update log message which would potentially use an undefined variable when adding new content.

 git-cvsserver.perl |    6 +++---
 1 files changed, 3 insertions(+), 3 deletions(-)

index a33a876..e18e901 100755
@@ -876,9 +876,9 @@ sub req_update
                 print "MT newline\n";
 		next;
 	    }
-	    elsif ( !defined($wrev) || $wrev == 0 )
+	    elsif ( (!defined($wrev) || $wrev == 0) && (!defined($meta->{revision}) || $meta->{revision} == 0) )
 	    {
-	        $log->info("Tell the client the file will be added");
+	        $log->info("Tell the client the file is scheduled for addition");
 		print "MT text A \n";
                 print "MT fname $filename\n";
                 print "MT newline\n";

@@ -886,7 +886,7 @@ sub req_update
 
 	    }
 	    else {
-                $log->info("Updating '$filename' $wrev");
+                $log->info("Updating '$filename' to ".$meta->{revision});
                 print "MT +updated\n";
                 print "MT text U \n";
                 print "MT fname $filename\n";

1.5.0.rc2.g3f8f2

Martin, I think this looks like a sane change.

I do not have anything other than the real CVS running on a Linux box to try this change (most notably I do not do Eclipse nor Tortoise) myself and I am reluctant to touch things I cannot personally test at this stage near the release. I am sure Andy tested this for his own use in his environment, but I would really appreciate a third party Ack from an environment different from the originator of the patch.

It does -- sorry I've been offline for a few days and catching up with work.

Fair enough. I'll give it whirl later tonight. For a quick test, you can always

export CVS_SERVER=/home/martin/src/git/git-cvsserver.perl cvs update

Yup. Will test later tonight.

martin

Yes, that's what I did (what I meant was I can test with cvs as client but not other clients that talk CVS remote protocols).

Thanks.