Some unscientific timing results (on the Python source tree)

[Matt Nordhoff]

Barry Warsaw wrote:
> It's very possible that we (Thomas and I) didn't set up code.python.org
> most optimally.  A brief description of what we did follows; suggestions
> for improvement are welcome!
> 
> First, we wanted to use as much of the existing svn infrastructure as
> possible.  Write access to svn is controlled by a common user, with a
> single authorized_keys file containing a command to run svnserve with
> the proper arguments (including -t and --tunnel-user).  We simply
> extended all that infrastructure, creating a new shared user with a
> different authorized_keys file.  This latter used a command of "bzr
> serve" with arguments translated to make sense for Bazaar.
> 
> http access was even simpler; we just set up an alias pointing into the
> Bazaar shared repository directory.
> 
> If anybody is interested and has ideas on how to tune the server side,
> ping me on irc.  I'm 'barry' on irc.freenode.net, and usually hang out
> on #mailman and #launchpad (and now, #bzr).
> 
> -Barry

You could set up an (anonymous, read-only) HTTP smart server (bzr+http).
It requires CGI or FastCGI or mod_python or somesuch.

See doc/en/user-guide/http_smart_server.txt in the source tree, or
<http://doc.bazaar-vcs.org/latest/en/user-guide/index.html#serving-bazaar-with-fastcgi>.
There's a scary warning about security problems, though. AFAIK, there
are no known problems, but it hasn't been thoroughly audited to be sure.

Obviously, it requires more resources on the server, and may or may not
be much of a performance improvement. Packs are pretty fast.

(BTW, until a couple months ago, the HTTP smart server had a serious
path handling bug. It wasn't a security issue, but it meant it didn't
work in many setups. Perhaps that's part of why it's not very popular...)
--