cvsps_import can think that trunk checkins happened on a new branch

[Max Kanat-Alexander]

I have been regularly rsync'ing the Bugzilla CVS repository to my local machine and running bzr cvsps-import on it, using the following commands:

 TZ=UTC CVSROOT=`pwd`/mozilla-cvs/ cvsps --root `pwd`/mozilla-cvs/ mozilla/webtools/bugzilla -u -A > bugzilla.cvsps.log 2> /dev/null
 bzr cvsps-import --quiet --cvsps-dump=bugzilla.cvsps.log mozilla-cvs mozilla/webtools/bugzilla bugzilla-import

However, the new BUGZILLA-3_2-BRANCH seems to have checkins that only happened in HEAD, in it. Here's probably why:

1. We made BUGZILLA-3_2-BRANCH in CVS.
2. After branching for 3.2, we did some checkins on HEAD that had been waiting for us to branch.
3. We didn't check anything into the new BUGZILLA-3_2-BRANCH for several days.

So now my bzr import of the 3.2 branch has all of the HEAD checkins we did for those several days, when it really shouldn't.

[Max Kanat-Alexander (mkanat)]

Here's the cvsps output for the bugzilla import. The HEAD parts that shouldn't be in the branch start at PatchSet 8195.

[John A Meinel (jameinel)]

This seems to be a basic problem with cvsps's ability to divine what actually happened. It seems like it is only focusing on timestamps, and not recognizing that when you create a branch, the tag of each file is important.

There isn't much that cvsps-import can do, other than push it to 'cvsps' to do a better job.

If we did get cvsps fixed, then we would have to look at changing cvsps-import to not think it had already converted those patches. Which should be possible, just remove stuff from the map files, and it won't know about the work it has already done, and then it should re-do it with different revision ids.

[Alexandre Garnier (zigarn)]

The problem is that branch creation is detected by cvsps only at first commit on the branch.
In cvsps 2.2b1, it introduce new notion to detect branch creation at last commit before (http://www.cobite.com/cvsps/CHANGELOG):
"add a new line item in the PatchSet output which indicates which
  branches fork off from after this PatchSet. This is actually
  the only way to correctly indicate the branching. The 'Ancestor
  Branch' thing is actually broken by design, and is now deprecated."

It would be nice to use this new "Branches:" line to create branches at right time : after last commit before their creation

[John A Meinel (jameinel)]

It would certainly be good to support this, I don't have the specific time right now to work on it, but I'd be happy to look at patches.