[PATCH] Don't propagate SKIPPED warning to future commits.

[PATCH] Don't propagate SKIPPED warning to future commits.

[Carl Worth]

Here's a simple bug fix for the (currently undocumented) -S flag to
git-cvsimport.

The -S flag allows an import to proceed in the face of a missing ,v
file in the CVS repository. It marks this by adding a SKIPPED warning
in the relevant commit.

Without this patch, the SKIPPED warning propagates to all future
commits, (and accumulates with future SKIPPED warnings).

Martin did the original work on -S to fix an import of the cairo
repository.

However, I just discovered that a correct fix for the cairo import
problem was for me to just "rm -rf ~/.cvsps". (I suppose calling with
a -x option would have worked too.)

I don't know how common it is for CVS repositories to change in ways
such that the cvsps cache gets totally confused as in my case, but it
might be worth adding something to cvsimport to help this. Some
options could be:

 * Use cvsps -x always, to ignore the cache.

 * Warn the user, (perhaps emit some description of stale cache
   possibility if the import fails?).

 * Add a note to the documentation.

Signed-off-by: Carl Worth <cworth@cworth.org>

---

 git-cvsimport.perl |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

c816d743a1faf65fb209273a471b15c7b51b55df
diff --git a/git-cvsimport.perl b/git-cvsimport.perl
index 6b63aa2..25db0ed 100755
--- a/git-cvsimport.perl
+++ b/git-cvsimport.perl
@@ -749,6 +749,7 @@ my $commit = sub {
 	if (@skipped) {
 	    $logmsg .= "\n\n\nSKIPPED:\n\t";
 	    $logmsg .= join("\n\t", @skipped) . "\n";
+	    @skipped = ();
 	}
 
 	print $pw "$logmsg\n"
-- 
1.1.6.g9da5

Re: [PATCH] Don't propagate SKIPPED warning to future commits.

[Junio C Hamano]

Carl Worth <cworth@cworth.org> writes:

> Here's a simple bug fix for the (currently undocumented) -S flag to
> git-cvsimport.
>
> The -S flag allows an import to proceed in the face of a missing ,v
> file in the CVS repository. It marks this by adding a SKIPPED warning
> in the relevant commit.
>
> Without this patch, the SKIPPED warning propagates to all future
> commits, (and accumulates with future SKIPPED warnings).

I might be missing something fundamental, but isn't prapagating
the warning a good thing?  If an earlier commit is suspected to
lack some path that needs to be there (I think that is what
"missing ,v file" means), wouldn't later commits built on top of
that one be by definition missing that path?

Re: [PATCH] Don't propagate SKIPPED warning to future commits.

[Carl Worth]

[-- Attachment #1: Type: text/plain, Size: 1006 bytes --]

On Fri, 10 Feb 2006 18:21:30 -0800, Junio C Hamano wrote:
> 
> I might be missing something fundamental, but isn't propagating
> the warning a good thing?

Yes, that would be fine.

The problem with the current behavior is that the warnings accumulate
for every commit that affects the particular file. So, after import,
the head commit will have a commit message ending in:

	SKIPPED: broken-file
	SKIPPED: broken-file
	SKIPPED: broken-file
	...

With one line per affected commit. So an alternate patch that simply
kept the warning unique per file would be fine too.

The discussion may be entirely moot since I don't know when it is
reasonable for a file to be missing in this way. In my specific case I
think the missing ,v file happened due to a combination of factors,
including old data in ~/.cvsps, and that data being stale due to some
manual surgery on the ,v files, (copying or renaming or so).

It would probably be fine to leave the current behavior until someone
runs into this again.

-Carl

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]