Skip to Content.
Sympa Menu

sm-discuss - Re: [SM-Discuss] Package management with distributed SCM

sm-discuss AT lists.ibiblio.org

Subject: Public SourceMage Discussion List

List archive

Chronological Thread  
  • From: Jeremy Blosser <jblosser-smgl AT firinn.org>
  • To: sm-discuss AT lists.ibiblio.org
  • Subject: Re: [SM-Discuss] Package management with distributed SCM
  • Date: Sat, 16 Jun 2007 23:09:42 -0500

On Jun 16, Daniel Drake [dsd AT gentoo.org] wrote:
> Gentoo uses a centralized SCM (CVS) for the package tree.

Just so you're aware you're at least the 3rd or 4th Gentoo developer who
has approached us about our use of git. You guys might want to compare
notes at some point, left hand talk to the right hand and all that...

> So, I am wondering if SourceMage have benefited similarly through
> maintaining grimoire in git.

> I hunted through your website but didn't even find a mention of git. I
> moved to the wiki and found a git guide, but this only detailed
> developer access. However, I found a webserver running on the same
> system that hosts the developer git access, which appears to provide the
> git repositories over http. However, as git-update-server-info had not
> been run there, it was not possible to clone this way.
>
> Next, I tried Google. This pointed me towards a gitweb interface to your
> trees, but not to any clone access. I'm also surprised that I have yet
> to find a reference to this gitweb on either your site or wiki.
>
> I next asked on IRC, and Eric Sandall provided me with a public access
> git:// URL. Unfortunately this was not working either, but Eric (or
> someone else) fixed the http access so I could clone that way.

I swear git-update-server-info was enabled on those at one point, but I see
now it isn't... it doesn't matter, http:// is more trouble than it's worth
for git, and accessing our repository that way is not supported. I should
just turn it off but I keep thinking I had left it on for some other
reason...

git:// is supported and should always work. It was down for a few hours
one night last week, which is the only significant downtime it's had since
it was put online (to replace the http:// access).

> To help anyone else on a similar path, I added details about the above
> resources to the GitGuide page on the wiki, but my additions were
> removed a few hours later. I apologise if I was out of line here and
> there is some reason you want to keep the public access information
> relatively quiet.

"So you're the one", heh. I reverted it because a) git:// should work, and
if I'd known it was down I would have fixed it, and I did fix it as soon as
I woke up the next morning and saw the nagios alerts it was down, b) we do
not currently advertise our anonymous access because we have a bandwidth
cap on that server and I don't want to deal with it getting crawled or
otherwise overused. Work is in progress to get it mirrored on a non-capped
server, at which point the URL can be shown in a Super Bowl(tm) commercial
for all I care. Meanwhile I don't care if people ask in IRC and find out
about it and use it but that at least keeps the excessive use down a little
bit.

> Anyway, with a clone of grimoire.git in front of me, I started looking
> through to get a better idea of if you have been able to leverage
> contributors outside of the developer community with the ease that git
> provides (although I wasn't too confident I'd find promising results due
> to my own troubles with finding and cloning the tree):
>
> - I found a total of 42 authors of the whole repository (after manually
> filtering those with slight name/email variations).
> - After removing sourcemage.org email addresses, this list decreased to
> 22.
> - After removing names found on the developers list on your website, the
> list decreased to 4.
> - After removing names with only single commits, this list decreased to
> 2.
>
> The remaining 2 people have made a total of 67 commits, or 0.3% of the
> total.
>
> So, my initial thoughts are that you haven't really had people outside
> of the developer community asking you to pull from their trees. Is this
> a mostly accurate assumption? I do realise that you could have had
> external git contributors but you later promoted them to developers.

To your general question: we haven't been using git very long relatively
speaking, and while we have talked about the benefits we could get from
having non-developers make changes in their own tree and e.g. publish those
to use in bug reports instead of attaching patches, we haven't made a push
to work that way and none of our users have done it on their own. It's
something some of us are quite interested in seeing happen but it's just
too early to really see it happening yet.

To the specific points you're raising, it's just the case that pretty much
anyone capable of writing a working patch or two and submitting them is
offered (and usually accepts) direct commit access. We are quite different
from gentoo here, and it well limit how much value we *ever* realize from
a distributed SCM's ability to have non-developers run their own trees.
It's hard to fathom anyone capable of running their own git tree *not*
becoming a regular developer. Removing people with single commits probably
excluded anyone who did publish a fix or two either via patches or their
own tree before they then became regular developers based on that work.

However, our SCM instance is divided into multiple component repositories
and ACLs, and what you *do* see happening from time to time is people who
work in one component but don't have commit access to another publishing
their fixes to that other component in their own tree. For what it's
worth.

> The 2 names remaining on the list are Armando Vega and Matthew Clark.
> I'd be interested to know if these people are actually unlisted
> developers (i.e. people with commit access to the central grimoire.git),
> or if they really are external contributors.

They are both former developers.

> I'm also interested in any other general comments regarding the
> pros/cons of using a distributed system for package management.
> Congratulations on reaching a first in the distro world :)

The biggest advantage we probably currently realize is the ability to work
offline, including commits, diffs, and any other operation. Several
developers have taken advantage of this capability, which was sorely
lacking in our previous, non-distributed SCM.

Attachment: pgpK2PfMF8IVJ.pgp
Description: PGP signature




Archive powered by MHonArc 2.6.24.

Top of Page