svn or arch

[Colin Walters]

On Sat, 2004-12-18 at 00:29 +0100, Enrico Scholz wrote:
> walters at redhat.com (Colin Walters) writes:
> 
> > One thing that should be clear is that by using a revision control
> > system for RPM packaging, we've already conceptually broken
> > compatibility because the SRPM is no longer the preferred form of
> > modification, to use the GPL terminology.
> 
> CVS can not replace SRPM:

Note that I was talking about prerequisites for changing to a better
revision control system, not how our current system is flawed.  But I'll
answer anyways:

> - SRPM can be signed, CVS not

Right; this is solved directly in pretty much all the distributed RCSes.

> - SRPM are (usually) working, while the CVS checkout might be a completely
>   broken development snapshot

You wouldn't check out CVS HEAD; you'd check out a branch tag
corresponding to a particluar NVR.

> - SRPM give you reproducibility, CVS not

Not true if you can map NVR->CVS tag.

> - SRPM can be better accessed (e.g. in a browsable http/ftp listing);

ViewCVS solves this, I think.  And e.g. Arch includes tools to browse
the repository in the client.

>   for CVS you need tags which are more difficultly to handle

Sure.

> - SRPM are buildable with system-tools (rpmbuild); for CVS you need lots
>   of prerequisites.

Not necessarily.  We could just stick the necessary scripts in the
common/ dir or whatever.  Or just include the necessary tools in an
updated rpmbuild.

> (- a known CVS drawback: cvs checkin/checkout is not atomic)

checkin is unrelated to this discussion, and checkout is solved because
you check out a tag.