Subject: Re: keeping pkgsrc up-to-date
To: None <tech-pkg@NetBSD.ORG>
From: Marton Fabo <>
List: tech-pkg
Date: 07/02/2002 21:10:38
> > Well, that's a big problem. I had a machine for example, with (quite
> > unstable) cable net access, and I tried supping pkgsrc for weeks without a
> > single successful session. That suxx.
>So, your complaint against cvs, sup, and perhaps rsync (though you
>didn't explicitly mention the latter), is really only that they hold
>connections open for quite a long time without any activity and that
>this utilization pattern doesn't work well with your available
>connectivity.  Thus your complaint is really only about your own network
>connectivity, not any given method of updating a local copy of the
>NetBSD sources (or repository).  :-)

Actually, my problem is with the current system's usability in certain 
circumstances, for example with slow/unreliable connetcivity, which I 
happen to have. Anyway, the _real_ problem is that any of the three 
mentioned methods (tarball, cvs, sup) are hardly optimized... Why don't 
optimize, if optimization _can_ be made?

Anyway, I sent another mail (originally it was sent a few hours ago, but 
the address was misspelled, so now I sent it again), in which I opt for a 
solution, instead of various hacks... (:

>I really can't imagine what your problem is with using the tar archive,
>unless perhaps you do keep local changes in your pkgsrc tree and you
>don't want to wipe them out (in which case you should instead be using a
>local CVS repository and importing the new (unpacked) tar archive and
>then merging into your local working tree).

As someone mentioned, it is only weekly updated. Not the best option in 
case of a OpenSSH remote-root SA... (: And anyway, with clever solution, 
the problem could be solved with much less load. Why to settle with the 
imperfect hacks currently available?

>I think CVSup could be the best all-round solution for everyone (because
>with one tool it can do all the different types of updates, and it's
>very efficient at what it does), except unfortunately it's not (yet)
>usable by everyone because it won't work on many NetBSD platforms.  You
>could try it if it's supported on your platform.

Well, as you already may have guessed, I just started to play around with 
the NetBSD package system, that's why I'm posting so many dumb (or not so 
dumb at times) messages. And because in the "About the package collection" 
on the web site only the above three methods were mentioned, I didn't even 
know about the fourth. To be honest, I wasn't really aware that CVS and 
CVSup are two distinct things...

>BTW, your suggested use of e-mail based updating assumes the diffs for
>changes are actually sent in e-mail.  Traditionally in FreeBSD this kind
>of mechanism has been called "CTM", or "CVS Through Email" (or as the
>manual page says "Current Through Email").  The scripts proposed by Sean
>and Thomas are a grossly inefficient way of doing CTM in a "pull"
>fashion.  I don't believe they will in any way solve your real problems.

That's OK. Although this fact doesn't say much about the case when the 
diffs themselves are sent. And, this has nothing to do with providing a 
more optimal synching method. I'll try this CVSup thing (if someone, 
off-list, explains how to use it (a pointer is enough)). Maybe that's what 
I'm looking for. Maybe a real solution should be made...