Re: PROPOSAL: Move to git, now that bzr is no longer a req.

["Eric S. Raymond" <esr@thyrsus.com>]

Andreas Schwab <schwab@linux-m68k.org>:
> "Eric S. Raymond" <esr@thyrsus.com> writes:
> 
> > Are you aware of any CVS conversion artifacts that should be cleaned
> > up?
> 
> I did a lot of cleanup when I converted the old CVS repository to bzr.

Great.  Less work for me. Maybe none; that would be nice.  

What did you use for the cleanup? reposurgeon?  Something else? 

> > I'm thinking of, in particular:
> >
> > 1. Unmerged commit cliques.
> 
> I don't know what that means.

Many converters don't do a perfect job of detecting CVS file commit cliques
that ought to be merged to changesets.  The symptom of this is runs of 
single-file commits with identical comments.  One way it can happen is if
the CVS exporter's commit-time-difference window was set too small.

(It has many other uses now, but I originally wrote reposurgeon to make
cleaning up this kind of artifact easy to do.)

> > 2. Out-of-place release tags.
> 
> I think I fixed them all up.

Oh, good.  One less tedious task.
 
> > 3. Fossil CVS commit references in commit comments.
> 
> I didn't change any of the commit comments, on purpose.

That is one philosophy. 

I, on the other hand, have way too much experience with repositories that
have geological strata of crap from multiple previous conversions in
them. My goal that for someone browsing a history I have converted,
the VCS transitions should be *invisible*.

This means:

1. Changing source-system ignore files to target-system conventions, not
just in the head revision but in the entire history.

3. Fixing up commit references in change comments so they are either in
the target system's native format or in a VCS-independent form.

We may decide these things should not be done in this case.  But I have
the tools and experience to do them, and that option should be on the
table.
-- 
		<a href="http://www.catb.org/~esr/">Eric S. Raymond</a>