Linux distros need configuration and customisation. People want to modify text config files. Some changes need reviews to prevent bad things from happening. Without tool support this is fragile. Problems are then hard to track.

You can version the /etc directory with subversion or the like. But permissions and ownerships is hard to do right there. They can be different from system to system. This is intrusive. You don't want that many versioned directories.

vadm gives you indirect versioning. vadm add /etc/passwd This maps to-be-versioned files into user-specific working copies. It delegates to an underlying versioning system. It gives bidirectional transform of ownership and permission info. History and diffs are available.

vadm: using it

• prerequisites: svn installation, sudo rights for executing user, and a repository
• vadm init file:///sysrepo/mysystem
• vadm add/remove/commit/diff/log path/to/configfile
• Start using it just like subversion

Setup notifications:

• Install a post-commit hook into subversion repo to signal admins of changes.
• daily cronjob may commit any pending changes

We use it to control some 20 systems now. Daily auto-committing has proven to be valuable. It signals changes to others who can then review your changes. You can find out who edited particular lines.

Versioning clusters

• this is future work
• versioning multiple similar machines:
  • distinguish between local and global (cluster-wide) changes
  • allow groups of systems (postfix systems, subversion systems)
  • per system would take preference
• a cluster vadm should only require ssh + local svn config
• vadm is slow now
• don't require a server-side daemon other than sshd?

Suggested semantics

• a URL defines a group of versioned files.
• each system has a stack of multiple such urls
• url marked manual-update or auto-update

Other considerations:

• versioning package installation information. An update of the apache package probably has a new config file. Do this in the same commit.
• speed up versioning of hundreds of system files
• plug in other versioning systems. Subversion is best supported now.

vadm future:

• release planned in 2006, under GPL
• development happens on a demand basis
• training/support possible

Use the source

Joe Developer thinks Twisted is overkill, so he tries to code his own server. But it can only serve one client at a time. But Twisted is overkill, right, so he tries some more clever ideas, with forkingTCPServer. Still doesn't work. So he finally tries out Twisted.

Twisted is:

• high Level: for basic things you need just a few lines. It makes the common case easy.
• with good abstractions. Try defer.Deferred() for processing values that don't exist, which is actually handy. That is a structured way for handling errors.

His Twisted program uses one thread at this point, which is not good when every line received takes long to process. To solve that you can use the process protocol. Or deferred.deferToThread()

You could use the standard python socket library. But Twisted was written because this is hard to use.

Okay, so I gave a talk myself on Europython. My brother wrote about it. You can get the slides at the Zest Software site. Click the red button next to the printer to get a Powerpoint-like presentation.

Holger Krekel thought the examples (adding for example Stories and Tasks to the live site) took too long. He would have preferred to see some more content like that already prepared. I tried that at the Zest office the week before and they told me to do more examples, so the truth is probably somewhere in the middle. :)

Someone else wondered if anyone at Zest Software could talk German. In that case he would probably want to hear this talk in I think the German Zope User Group (dzug). Also vielleicht gehe ich mal nach Deutschland!

I am at the Europython 2006 conference at the CERN in Geneva. I am holding a talk myself Monday. I will be writing about talks I visited here. Hm, I thought about making this an overview page with links to all the talks, but I think my main page on July 2006 should work fine actually. And as usual: see my brother's weblog

Works on CPS, a CMS for Zope. He is in charge of Ajax there.

What is Ajax?

A football club, a cleaner and a Web 2.0 technology: Asysnchronous Javascript And XML. Glue between application.

Examples: del.icio.us, flickr, digg, Google Mail.

Browser sends a Javascript call, server answers with just part of the page. This speeds up interactivity, saves bandwidth, avoids Flash.

Bad things to say about AJAX:

• Search enginges won't index your website right
• The back button is broken.
• Several others

These can be true, but you can work around it, except for the back button.

Choose a good toolkit. Lots of choices.

SSF: Server Side Javascript Generation Frameworks. Javascript dynamically created by the server.

• CrackAjax: change python to Javascript. But untestable and seems dead, but good idea, so keep an eye on it.
• Azax. Young project. Keep an eye on it.

Low level client side frameworks (CSF).:

• Test drive development doable.
• Ready to use
• Community. We choose the low level client side frameworks (CSF) as everyone seems to do that.

Ajax good practices

• Accessibility still matters. Keep it gracefully degrading your UIs. Blind people don't care about Ajax.
• It does bring pleasure.
• Practice Test Driven Development! Javascript deserves it too.
• Practice OO Programming, e.g. with help from Mojikit or other moders toolkits. Reuse.
• Continuous integration. Bundle your javascript tests in your python test suite. The scriptaculous test kit helps here.

CPS Ajax goodies

• nuxeo.javascript: provides all js libs in Zope 3 resources.

Conclusion

Ajax is not a revolution, but it is nice

More reading:

• Nuxeo blogs
• Planet Ajaxian
• Ajaxpatterns with lots of links

You can try it without frameworks, but frameworks help smooth away differences between browsers.