geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Greg Stein <>
Subject Subversion (was: geronimo-dev Digest 12 Aug 2003 18:20:52 -0000 Issue 37)
Date Tue, 12 Aug 2003 20:42:54 GMT
On Tue, Aug 12, 2003 at 06:20:52PM -0000, wrote:
> From: "Jeremy Boynes" <>
> Subject: SVN?
> To: <>
> Date: Tue, 12 Aug 2003 10:39:51 -0700
> Noel J. Bergman wrote:
> > Want to try using SVN?  Is it ready?
> Would love to if it isn't too hard to get people up and running. My biggest
> concern would be client support (availability of IDE integration etc), but
> maybe the server-side benefits (like move :-) outweigh that.

There are SVN integrations with Eclipse and with DevStudio.NET. There are
some early GUIs, but the most robust client is the command-line client.

> Do we have any SVN experts around?

Heh :-)

> Is there a sandbox?

Oh, such as :-)

> From: Brian McCallister <>
> Subject: Re: SVN?
> To:
> Date: Tue, 12 Aug 2003 14:02:44 -0400
> We experimented briefly with SVN a month or so ago and decided it 
> wasn't ready for production use yet:
> 1) None of the servers we were willing to certify for production met 
> its requirements for up-to-dateness of packages (it takes a lot of 
> brand new, with particular patches, packages).

We already have it running at, so this isn't a problem. The
client is *very* minimal in its demands.

> 2) Database corrupts far too often. It is an "easy to fix" corruption, 
> and they include the tool to do it, but... no thank you.

Woah. Easy here. There is a HUGE difference between Berkeley DB getting a
bit confused, and data corruption (i.e. your choice of words definitely
leads to the wrong impression). In the past two years, since SVN started
self-hosting and other people really started to use it, there has only been
*one* recorded case of data loss, and that was due to a problematic BDB on
the MacOS X platform.

Subversion itself has been self-hosting for nearly two years, and we've
never even had a close call. It happily cranks away, and we haven't even had
to "dig in" to get data out. The code has Just Worked(tm).

That said, if you exit the process abnormally (e.g. httpd crashes or you
'kill' a local client), then BDB doesn't get its data flushed to disk. You
then need to run a "recovery" so that BDB figures out where it left off. But
note that BDB is a *journaling* system. It never gets left in a hosed state.
The worst that'll happen is that a commit will not have "taken". But also
note that commits are atomic; unlike CVS, it is impossible to end up with a
partial commit.

> Yes, Subversion is around for testing.  I don't know if it is considered
> suitable enough, but the impression I have is that *IF* SVN were to be used,
> it would be better to start with it, and not migrate.  I get the feeling

Too late now :-)

> that the Subversion core is more ready than the CVS migration tools.
> However, I am not an SVN expert, and there are others here who are, which is
> why I phrased my message as a question.

The Subversion core has been ready for day-to-day use for over a year now.
The migration tools are just fine, especially if you have no branches or
tags (such as the current geronio CVS module). The branch/tag conversion
code is "relatively new" although it has already received quite a lot of
testing. To confuse it, you have to have a *really* screwy CVS repository; I
mean, like nightmare-screwy. Where you do partial branches, shifting tags
around, and marking multiple revisions as "dead".

> Client support includes,
>, and of course, the svn command line
> interface.

And and


Greg Stein,

View raw message