Forums : Technical Issue Help

Hi,

We have got the old problem again: Project Samba 4 (3599) counts lines together with Project Samba (4558).

This has started happening, since we changed our repository to git. Samba and Samba 4 share a common repository, each Project has its own branch (v4-0-test for Samba 4 vs v3-2-test for Samba). The Samba3 branch (v3-2-test) is the default branch of the repository. Might this be a cause of the problem?

I also have the impression that this oscillates, i.e. happens - doesen't happen - happens - doesn't happen... This might be an illusion, though.

Cheers - Michael

Michael Adam about 17 years ago

To put it more precisely: The commits of v3-2-test are added to the commits counted for v4-0-test. Not vice versa.

Michael Adam about 17 years ago

Hi Michael,

I can smell that this is going to be complicated.

Can you point me at a specific commit on Ohloh in project 3599 which should not be there? That would be very helpful.

I tried walking through the commits manually, but did not spot anything out of place -- but I'm not familiar with the project. I did see some common commits between the two branches, but upon inspection they appeared to be commits that were intentionally cherry-picked from one branch into the other.

If this is an oscillating on-again off-again problem, it might just happen to not be present at this instant.

You can't see it from the web front end, but we do have multiple versions of the reports for Samba/Samba 4 stored here. Given a specific commit I can go through the history and figure out where things are confused. It may be that some versions of the report show the bug but others do not.

Thanks,
Robin

Robin Luckey about 17 years ago

Robin,

by filtering out my own commits, I found the first wrongly assigned commit about one month ago on 2008-01-25:

http://www.ohloh.net/projects/3599/commits/19573289

That commit was not done in the v4-0-test branch but in the v3-2-test branch.
From that on, there on, there are lots of wrongliy assigned commits, apparently all commits to v3-2-test.

At first I suspected our switch from SVN to
git in samba 4, but branch v4-0-test was created on 2008-01-10 - two weeks earlier.

Our repo can be inspected at http://gitweb.samba.org/

Thanks, Michael

Michael Adam about 17 years ago

Here is the corresponding commit in our gitweb:

http://gitweb.samba.org/?p=samba.git;a=commit;h=dcd42a1e0642c69348adfaeecef7f7f2f074ac30

Michael Adam about 17 years ago

Hi Michael,

After some digging, it turns out that while our database thinks we are following v4-0-test branch, when I examine our copy of the git repository on our server hard disk, it turns out it believes it is tracking the v3-2-test branch.

I'm not sure how we got into this state; I'm really hoping this is a one-time error, but if it can happen here, it might be happening on other projects as well.

I'm still digging for the root cause and will keep you posted.

Thanks,
Robin

Robin Luckey about 17 years ago

OK, I found the bug. There was a point in the incremental update (under some very particular circumstances) where we left the branch unspecified.

The good news is that the the repository we cloned to our server is still sound; it's just the database report that becomes confused.

I'll schedule new clean reports for Samba 4, and look around for other projects with this problem.

Thanks for your help Michael. This was a tricky but important bug and I appreciate the detailed info.

Robin

Robin Luckey about 17 years ago

Hi Robin,

thanks for your investigations. The explanation sounds very plausible for the the phenomena observed: our v3-2-test branch -- being the default branch of the repository -- was used as a fallback when incremental update lost track of the branch it was actually counting.

Let's hope that this subtle error is fixed for good now!

Cheers - Michael

Michael Adam about 17 years ago