[Kde-scm-interest] Braindump: fixups

[Oswald Buddenhagen]

On Sat, Jul 19, 2008 at 10:43:31AM -0300, Thiago Macieira wrote:
> Oswald Buddenhagen wrote:
> >On Fri, Jul 18, 2008 at 11:07:15PM -0300, Thiago Macieira wrote:
> >>  -> must not let this tool execute on the CVS branches imported since
> >>  cvs2svn did a lousy job.
> >
> >i'm aware of one actual breakage (which i would try to fix retroactively
> >if i get some time allocated for that). or do you see some more
> >fundamental problem?
> 
> The symptom is that the old branches have a lot fewer files than the 
> corresponding trunk at the time. I don't know if it's cvs2svn's fault or 
> if we failed at tagging/branching correctly with CVS back when.
> 
yes, that's the problem i'm referring to. that's purely coolo's and my
fault.

> There's a smaller one which is that the dates and SVN revisions are
> completely out of whack.
>
that might be related to the missing files, too. would have to
investigate that.

> the tool sees those branchings as a lot of 
> individual "copy with history" in the same commit, which is completely 
> unrepresentable in Git (which is irrelevant, since it is supposed to 
> represent one state in history, despite what cvs2svn did).
> 
yep. cvs2svn trunk can do direct conversions to git now, so it had to
struggle with that problem, too. the rather insane picking of sources
was a problem in itself, which is fixed, too.
the problem is that cvs2svn trunk basically does not have a single line
of code in common with 1.3 (which i based my modifications on), so it
would have to be re-patched from scratch basically, and in the hope that
it will produce at most as many revisions as the old one so it can be
retro-fitted into the repository (this seems likely, though).

> It might be possible to fix history for the old CVS tags later on by 
> re-doing the job.
>
wouldn't this invalidate all sha1s?

-- 
Hi! I'm a .signature virus! Copy me into your ~/.signature, please!
--
Confusion, chaos, panic - my work here is done.