You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Jani Averbach <ja...@jaa.iki.fi> on 2005/01/14 18:39:46 UTC
Release, bindings (Was: svn commit: r12731 - trunk/notes)
On 2005-01-14 12:12-0600, sussman@tigris.org wrote:
>
> Document that our release process should at least *compile* bindings.
>
> No, it's not the same as real testing, but at least it avoids any
> brown-paper-bag releases like svn 1.1.2. It's one thing to ship a
> tarball with bugs; it's a whole separate level of embarrassment to
> ship a tarball that doesn't fully compile.
I would like to see that our release process mandate that we won't cut
tarball from branch until 24-48h hours after the last merge.
This way someone else has a fighting chance to test branch, and our
smoke testers has time to kick in.
At the moment, svntester runs unit tests for swig-pl and javahl, and
verify that swig-py compiles and installs.
BR, Jani
--
Jani Averbach
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
Greg Hudson <gh...@MIT.EDU> writes:
> On Fri, 2005-01-14 at 13:56, kfogel@collab.net wrote:
> > Let's just do a "-beta" release for every tarball. That way testing
> > is not limited to those who are comfortable building from a
> > checked-out tree. Anyone could download the beta tarball and test it.
>
> I think that would be overkill. Let's stick to small changes to the
> release process until we've demonstrated a need for large changes.
IMHO, the current Perl bindings situation was a small demonstration of
exactly that. I'm happy to wait for another demonstration or two --
as long as, when they occur, we take the lesson to heart.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Greg Hudson <gh...@MIT.EDU>.
On Fri, 2005-01-14 at 13:56, kfogel@collab.net wrote:
> Let's just do a "-beta" release for every tarball. That way testing
> is not limited to those who are comfortable building from a
> checked-out tree. Anyone could download the beta tarball and test it.
I think that would be overkill. Let's stick to small changes to the
release process until we've demonstrated a need for large changes.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
Jani Averbach <ja...@jaa.iki.fi> writes:
> I would like to see that our release process mandate that we won't cut
> tarball from branch until 24-48h hours after the last merge.
>
> This way someone else has a fighting chance to test branch, and our
> smoke testers has time to kick in.
>
> At the moment, svntester runs unit tests for swig-pl and javahl, and
> verify that swig-py compiles and installs.
I think this is a *great* idea -- but if we're going to do it, why not
go all the way?
Let's just do a "-beta" release for every tarball. That way testing
is not limited to those who are comfortable building from a
checked-out tree. Anyone could download the beta tarball and test it.
Thoughts? (From breser especially?)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Jani Averbach <ja...@jaa.iki.fi>.
On 2005-01-14 11:27-0800, Ben Reser wrote:
>
> a) How many people would really test the branch? I didn't think anyone
> was running automated tests on the branches either.
1.1.x is tested at least by these (I might have missed some of Brane's
machines):
i686-pc-linux-gnu
i386-unknown-freebsdelf5.0
x86_64-unknown-linux-gnu
> b) Testing beta tarballs seems not very useful.
In the ideal world, we should pretest our tarballs, but this brings a
lot more bureaucracy to the process. Moreover, I could not promise
that I have time to test a tarball at the some particular time, but I
could almost sure to promise that my svntest system is up and running
to test the branch. With little pause between last merge and tarball
cutting we will gain a lot for free.
BR, Jani
--
Jani Averbach
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Friday, January 14, 2005 1:36 PM -0600 kfogel@collab.net wrote:
> Are you saying that it would be difficult to build a source -beta
> tarball followed by a true release tarball, such that the only
> difference between them is the version label? (That's surprising, if
> so.)
I believe there should be no differences whatsoever between what is
approved for release and what is actually released. If you make any
changes to the release file, something could go wrong. I think the release
in its final state should be approved as-is by the committers.
> That's an independent proposal, I think. If we want to have 3 or 4
> developers test it, we can do that as easily with beta tarballs as
> with true-but-slightly-secret tarballs. The difference is that we
> can't let the latter kind of tarball out where other people could test
> it too. That seems a pretty large disadvantage...
No, I think the issue is to introduce it as a forcing factor: if three
committers approve the release, we can say that we put forth due diligence.
At this point, we don't have such safeguards. -- justin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Reser <be...@reser.org>.
On Fri, Jan 14, 2005 at 02:05:13PM -0600, kfogel@collab.net wrote:
> Ben Reser <be...@reser.org> writes:
> > So I'm skeptical of the usefulness of releasing beta tarballs.
>
> Note that doing "-beta" tarballs for non-patch releases still serves a
> purpose. But after that, we should *still* do the 3 +1's routine on
> the final non-beta tarball.
Yeah I was meaning in this limited context. When we're doing a new
minor/major the changes are significant enough it warrents more
eyeballs.
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
Ben Reser <be...@reser.org> writes:
> So I'm skeptical of the usefulness of releasing beta tarballs.
I wrote:
> +1 on this plan, scratch the "-beta" stuff then.
>
> (I see your point about the potential dangers of recutting the
> tarball, Ben.)
Note that doing "-beta" tarballs for non-patch releases still serves a
purpose. But after that, we should *still* do the 3 +1's routine on
the final non-beta tarball.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Reser <be...@reser.org>.
On Fri, Jan 14, 2005 at 01:36:13PM -0600, kfogel@collab.net wrote:
> Are you saying that it would be difficult to build a source -beta
> tarball followed by a true release tarball, such that the only
> difference between them is the version label? (That's surprising, if
> so.)
I'm of the opinion that if re-cut the tarball you have to start all over
with the tests. You're not controlling for potential mistakes that I
might make cutting the tarball.
If someone else is cutting it without a chroot environment, they might
have changed something in their environment that breaks things. We have
some very specific requirements for software that we need to generate
the tarballs. If they aren't followed it breaks things in obscure ways.
Unfortunately, such is the nature of libtool and autoconf.
> That's an independent proposal, I think. If we want to have 3 or 4
> developers test it, we can do that as easily with beta tarballs as
> with true-but-slightly-secret tarballs. The difference is that we
> can't let the latter kind of tarball out where other people could test
> it too. That seems a pretty large disadvantage...
You're assuming that other people would test it. And that their testing
results would be useful. Having had random people hanging around in the
channels try to test my released tarballs only to run into really odd
problems that are more of an issue of environment than anything wrong
with the tarball.
Heck even sussman is having difficulty trying to test stuff. The more
people testing the better. But we want people that know what they're
doing. We don't want to spend a lot of time figuring out testers
mistakes.
So I'm skeptical of the usefulness of releasing beta tarballs.
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
Ben Reser <be...@reser.org> writes:
> b) Testing beta tarballs seems not very useful. This ignores that the
> tarball production setup is fairly brittle. You guys are used to one
> person doing it and doing it in a very controlled environment (I build
> tarballs out of chroot that never does anything but build tarballs so
> there aren't any unexpected changes to the release environment).
Are you saying that it would be difficult to build a source -beta
tarball followed by a true release tarball, such that the only
difference between them is the version label? (That's surprising, if
so.)
> What I'd much rather see is, that we cut tarballs as though they were
> releases and distribute them to developers for testing. When several
> developers sign off on testing then we post them for everyone else.
> I've always been thinking 3 or 4 developers.
>
> While this cuts out anyone that just randomly wants to test, it isn't
> based upon the presumption that if it's there people will test it. We
> know who tested it. And presumeably we know what procedures they
> followed to test it.
That's an independent proposal, I think. If we want to have 3 or 4
developers test it, we can do that as easily with beta tarballs as
with true-but-slightly-secret tarballs. The difference is that we
can't let the latter kind of tarball out where other people could test
it too. That seems a pretty large disadvantage...
> Additionally, I've wanted us to always produce multiple GPG signatures
> for every release. One person's GPG key being compromised shouldn't be
> sufficient to sign a hacked release. As things stand now if my key was
> ever compromised so are most of the releases we've made (or at least we
> can't be assured that they aren't).
+1, I'll be happy to sign.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
Ben Reser <be...@reser.org> writes:
> Tossing the version number is fine with me... It eliminates the need for
> renaming the files and eliminates kfogel's issue with not being able to
> allow anyone to test them.
+1 on this plan, scratch the "-beta" stuff then.
(I see your point about the potential dangers of recutting the
tarball, Ben.)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Reser <be...@reser.org>.
On Fri, Jan 14, 2005 at 12:12:12PM -0800, Justin Erenkrantz wrote:
> I'm not exactly sure how multiple signatures would work in practice, but
> that'd be a fine way to record who has signed off on it. -- justin
IMHO this is how I'd handle this.
Create a new file somwhere (probably on the approriate branch) for
tracking the tarball. It'd probably similar to STATUS. People would
vote, leave comments, etc...
At the bottom of the file we'd have a section for each tarball/zip file
where people would append their signature which they made by doing:
gpg -b --armor subversion-1.1.x.tar.bz2
(and so on for the other files).
The .asc file that we ship can have multiple sigs and GPG will check
each one. So I would simply take and append all the sigs together into
the .asc file before final posting.
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Friday, January 14, 2005 12:01 PM -0800 Ben Reser <be...@reser.org>
wrote:
> On Fri, Jan 14, 2005 at 11:54:39AM -0800, Justin Erenkrantz wrote:
>> Exactly. This follows the httpd convention: 3 developers must give a +1
>> to the official tarball before it can be publicly posted. If it fails
>> (or has a brown-paper bag bug), the version number is tossed and never
>> released to the general public.
>
> Tossing the version number is fine with me... It eliminates the need for
> renaming the files and eliminates kfogel's issue with not being able to
> allow anyone to test them.
Yes.
>> I really think that'd be an excellent idea and simpler than issue a
>> 'beta' tarball that reports itself as such. Post a 1.1.3 on dev@ and
>> wait until it receives 3 +1s before posting it and sending email
>> announcements. --
>
> Yup, I'd just like to expand the +1 to being a GPG signature. Helps
> fulfill my wish for multiple sigs.
I'm not exactly sure how multiple signatures would work in practice, but
that'd be a fine way to record who has signed off on it. -- justin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Reser <be...@reser.org>.
On Fri, Jan 14, 2005 at 11:54:39AM -0800, Justin Erenkrantz wrote:
> Exactly. This follows the httpd convention: 3 developers must give a +1 to
> the official tarball before it can be publicly posted. If it fails (or has
> a brown-paper bag bug), the version number is tossed and never released to
> the general public.
Tossing the version number is fine with me... It eliminates the need for
renaming the files and eliminates kfogel's issue with not being able to
allow anyone to test them.
> I really think that'd be an excellent idea and simpler than issue a 'beta'
> tarball that reports itself as such. Post a 1.1.3 on dev@ and wait until
> it receives 3 +1s before posting it and sending email announcements. --
Yup, I'd just like to expand the +1 to being a GPG signature. Helps
fulfill my wish for multiple sigs.
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Justin Erenkrantz <ju...@erenkrantz.com>.
--On Friday, January 14, 2005 11:27 AM -0800 Ben Reser <be...@reser.org>
wrote:
> What I'd much rather see is, that we cut tarballs as though they were
> releases and distribute them to developers for testing. When several
> developers sign off on testing then we post them for everyone else.
> I've always been thinking 3 or 4 developers.
Exactly. This follows the httpd convention: 3 developers must give a +1 to
the official tarball before it can be publicly posted. If it fails (or has
a brown-paper bag bug), the version number is tossed and never released to
the general public.
I really think that'd be an excellent idea and simpler than issue a 'beta'
tarball that reports itself as such. Post a 1.1.3 on dev@ and wait until
it receives 3 +1s before posting it and sending email announcements. --
justin
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
Ben Reser <be...@reser.org> writes:
> Ohh you mean not actually building it so that svn --version reports that
> it's a beta. Just renaming the tarball file and then renaming it back
> when the devs are okay with it?
Well, I meant that 'svn --version' would report beta, yes. (Is that
hard?)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Reser <be...@reser.org>.
On Fri, Jan 14, 2005 at 01:35:03PM -0600, Ben Collins-Sussman wrote:
> Right, the idea is that a 'beta' tarball is posted, and when N
> developers have tested it and officially +1'd it, then we remove the
> beta-label and release to the world. We just need to choose a value of
> N.
Ohh you mean not actually building it so that svn --version reports that
it's a beta. Just renaming the tarball file and then renaming it back
when the devs are okay with it?
If that's the case I'd rather we not call it "beta" as that's mixing the
use of term. We do release beta releases that are for public
consumption. How about calling them "subversion-1.1.3-devtest.tar.gz"
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Collins-Sussman <su...@collab.net>.
On Jan 14, 2005, at 1:27 PM, Ben Reser wrote:
>
> What I'd much rather see is, that we cut tarballs as though they were
> releases and distribute them to developers for testing. When several
> developers sign off on testing then we post them for everyone else.
> I've always been thinking 3 or 4 developers.
>
Right, the idea is that a 'beta' tarball is posted, and when N
developers have tested it and officially +1'd it, then we remove the
beta-label and release to the world. We just need to choose a value of
N.
Isn't this what the httpd project does?
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Reser <be...@reser.org>.
On Fri, Jan 14, 2005 at 11:39:46AM -0700, Jani Averbach wrote:
> I would like to see that our release process mandate that we won't cut
> tarball from branch until 24-48h hours after the last merge.
>
> This way someone else has a fighting chance to test branch, and our
> smoke testers has time to kick in.
>
> At the moment, svntester runs unit tests for swig-pl and javahl, and
> verify that swig-py compiles and installs.
I'm not sure how I feel about this or the beta tarball thing.
a) How many people would really test the branch? I didn't think anyone
was running automated tests on the branches either.
b) Testing beta tarballs seems not very useful. This ignores that the
tarball production setup is fairly brittle. You guys are used to one
person doing it and doing it in a very controlled environment (I build
tarballs out of chroot that never does anything but build tarballs so
there aren't any unexpected changes to the release environment).
What I'd much rather see is, that we cut tarballs as though they were
releases and distribute them to developers for testing. When several
developers sign off on testing then we post them for everyone else.
I've always been thinking 3 or 4 developers.
While this cuts out anyone that just randomly wants to test, it isn't
based upon the presumption that if it's there people will test it. We
know who tested it. And presumeably we know what procedures they
followed to test it.
Additionally, I've wanted us to always produce multiple GPG signatures
for every release. One person's GPG key being compromised shouldn't be
sufficient to sign a hacked release. As things stand now if my key was
ever compromised so are most of the releases we've made (or at least we
can't be assured that they aren't).
--
Ben Reser <be...@reser.org>
http://ben.reser.org
"Conscience is the inner voice which warns us somebody may be looking."
- H.L. Mencken
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by Ben Collins-Sussman <su...@collab.net>.
On Jan 14, 2005, at 12:39 PM, Jani Averbach wrote:
> On 2005-01-14 12:12-0600, sussman@tigris.org wrote:
>>
>> Document that our release process should at least *compile* bindings.
>>
>> No, it's not the same as real testing, but at least it avoids any
>> brown-paper-bag releases like svn 1.1.2. It's one thing to ship a
>> tarball with bugs; it's a whole separate level of embarrassment to
>> ship a tarball that doesn't fully compile.
>
> I would like to see that our release process mandate that we won't cut
> tarball from branch until 24-48h hours after the last merge.
>
> This way someone else has a fighting chance to test branch, and our
> smoke testers has time to kick in.
>
> At the moment, svntester runs unit tests for swig-pl and javahl, and
> verify that swig-py compiles and installs.
>
kfogel was also saying today that he thinks we should release 'beta'
tarballs of bugfix releases, so that they can be tested for a few days.
I guess it's a bit silly that we finish merging patches, roll the
tarball, test for an hour, then ship it. We should allow more time for
testing tarballs -- even the bugfix releases.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by kf...@collab.net.
David Summers <da...@summersoft.fay.ar.us> writes:
> Yes, please, give a few days before "official" releases.
>
> [...]
Yup. I think everyone now agrees that more careful release procedures
are needed. We'll probably adopt something like the system Apache
httpd uses, where people test the actual release tarball, and some
minimum number of developers GPG-sign it, before the release is
announced. Ben Reser and Justin have mentioned some of these things
on this list already; we'll codify the procedures soon (meaning: I'll
write it up in HACKING unless Ben or someone beats me to it :-) ).
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Re: Release, bindings (Was: svn commit: r12731 - trunk/notes)
Posted by David Summers <da...@summersoft.fay.ar.us>.
On Fri, 14 Jan 2005, Jani Averbach wrote:
> On 2005-01-14 12:12-0600, sussman@tigris.org wrote:
> >
> > Document that our release process should at least *compile* bindings.
> >
> > No, it's not the same as real testing, but at least it avoids any
> > brown-paper-bag releases like svn 1.1.2. It's one thing to ship a
> > tarball with bugs; it's a whole separate level of embarrassment to
> > ship a tarball that doesn't fully compile.
>
> I would like to see that our release process mandate that we won't cut
> tarball from branch until 24-48h hours after the last merge.
>
> This way someone else has a fighting chance to test branch, and our
> smoke testers has time to kick in.
>
> At the moment, svntester runs unit tests for swig-pl and javahl, and
> verify that swig-py compiles and installs.
>
Yes, please, give a few days before "official" releases.
Several times, releases have been put out and I've been busy enough that
I've not had time to try to build the RPMs, only to find out later that
they fail compiling. (1.1.2 was the latest example).
Also I would like to point out that my RPM build for RedHat 7.3, 8.0, 9.0,
FC1, and WBEL3/RHEL3 exercise the compile, the perl and python bindings
compiles and any tests they have, and all the unit tests for Subversion
for ra_local, ra_dav, and ra_svn. While not "perfect", just building the
RPMs exercises a lot of subversion and dependent package compiles, tests,
and functionaltiy. Of course that is only for these platforms and not
others, but for devs with those platforms it is a pretty good test.
When these problems happen then I have to release a "special" build that
includes the path and %changelog entries to detail exactly how the RPM was
built.
Normally, when everything works right, I just check out the tag and go to
the appropriate packages/rpm/<distro> directory and do "make RELEASE=1"
and wait a couple of hours (the tests take a long time but are worth it.)
--
David Wayne Summers "Linux: Because reboots are for hardware upgrades!"
david@summersoft.fay.ar.us PGP Key: http://summersoft.fay.ar.us/~david/pgp.txt
PGP Key fingerprint = 0B44 B118 85CC F4EC 7021 1ED4 1516 5B78 E320 2001
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org