[LLVMdev] Version Control Upgrade?

Reid Spencer reid at x10sys.com
Mon Jan 10 12:27:58 PST 2005


On Mon, 2005-01-10 at 10:24, Tanya Lattner wrote:
> >> Considering that everyone is managing just fine with cvs, anything
> >> discussed about the differences between cvs and *X* is not a show stopper.
> >
> > I would have used the term "coping just fine" with cvs. Yeah, it works.
> > But, is it productive? efficient? easy? fast? .. none of those. There
> > are better alternatives and that's why we're discussing it. Don't forget
> > that the experience with CVS in Russia and even Seattle is QUITE
> > different than in the same building as the CVS server.
> 
> CVS is not so horrible slow that its not possible to get anything done.
> And i think coping is a very harsh word.. most people are using it just 
> fine. And you really should not generalize and assume that I work in the 
> same building as the CVS server since you really don't know.

You're right, I don't know where you work. I apologize. But, that
doesn't change the argument much. Regardless of where at UIUC you work,
even NCSA, your network connection is likely to be orders of magnitude
better than my DSL connection 16 hops away. Others have it even worse. 

As far as "coping" goes, I used that word purposefully. When it takes 3
hours to do a commit (this happened recently), I consider it "coping".
Contrast that with updating the ENTIRE content of all my web sites via
Subversion to a server in NJ in about 8 minutes over my same DSL
connection. My websites are 288MBytes. LLVM source is 78MBytes.
 
> 
> The main benefit to cvs is that everyone knows how to use it.

Those same people would find subversion completely comfortable. The
usage model is nearly identical.

> Most people 
> even have it installed by default.

Subversion ships with Fedora Core 2 and 3 and is easily installable with
binary packages available for most Linux FreeBSD and Windows variants.

>  If you go and change from cvs to some 
> other software, you are going to put up another barrier for people to 
> overcome before joining llvm. 

I don't see that happening. Subversion, at least, and several of the
others have huge followings. Look at the responses we got already.
Developers have already used Perforce and other tools. The question is
really what is *effective* for LLVM developers. For remote developers
such as myself, CVS doesn't seem effective when I contrast it with other
tools. And this from someone that has advocated CVS as a great tool for
nearly a decade!

> Learning llvm is a big enough task in 
> itself and contrary to what you may believe.. its not globally known yet. 
> Switching to something at this stage in the game is premature and an 
> uncesssary headache.

That's not to say that it won't become popular. Awareness of LLVM is
increasing daily and the right time to switch, in my opinion, is before
the masses get here. Since we're only investigating this, its probably
six months or more before we actually make the switch.

> In my personal opinion, subversion is better with its features. But I'm 
> skeptical to its stability since its so young. Waiting awhile to switch 
> would give it a chance to mature and for the community to start switching 
> over.

I was skeptical too until the 1.1 release. I think its ready now. I've
been using it fora few weeks, and I'm pretty happy with it. Completely
glitch free, fast, and my repository is running on an old 350MHz P5

Subversion is not the only tool we'll consider for LLVM. Others have
suggested Perforce and I think arch. 

At this stage of the investigation we're looking for feedback on tools
developers have actively used. What more can you tell us about your use
of Subversion, arch or Perforce?

Reid.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20050110/c26a4486/attachment.sig>


More information about the llvm-dev mailing list