irker-svnpoller: Subversion poller and mail filter application for irker

Just as irker’s adoption rate is increasing, I have just completed work on a very simple application for Subversion repositories — two applications, in fact.

irker-svnpoller is a very simple script that polls a single commit log (not data) from a Subversion repository and delivers notifications to any number of channels using an irkerd running on the same host. It mimics the CIA bots’ formatting, much like nenolod’s irker CIA proxy, from which I borrowed a small amount of code.

irker-svnpoller → irkerd

But exactly how is this supposed to be useful to anyone, you may be wondering right now? Well, irker-svnpoller is not really intended to be used stand-alone. A timed poller script that tracks the last notified revision could come in handy, but people could get impatient waiting for their commits to appear in their IRC channels minutes later. I am well familiarized with the defects, quirks, and virtues of my primary audience—the Battle for Wesnoth and Wesnoth-UMC-Dev projects—and this approach would simply not scale well over time.

Enter the first companion script, svnmail-filter. It reads email message headers from stdin to determine a commit’s revision number and the pertinent repository to probe using irker-svnpoller. Configuration is mostly done through a ruleset file using the JSON format.

Of course, svnmail-filter is not that useful on its own either. The idea is that procmail or some other MDA should pipe incoming email headers through svnmail-filter — and preferably, only those from legitimate sources, such as subscribed commit mailing lists. This is actually simpler than it sounds, and it is more or less inspired by CIA.vc’s perpetually broken mail-based SVN poller.

MDA → svnmail-filter → irker-svnpoller → irkerd

Since no service in the pipeline other than irkerd runs persistently in the background, this should be significantly more fault-resilient than CIA.vc’s approach, which apparently required a poller service to listen and act upon local requests. The downside is that the host running irker-svnpoller may need to create many short-lived SVN repository connections for individual commits in a chain. In Wesnoth’s case, SVN commit chains are rare enough, but their size often goes around a dozen individual commits or so. Regardless, this shouldn’t be terribly concerning for a production server with a decent low-latency uplink, and the overhead on the repository provider should be rather small compared to pushing massive commit diffs across the network.

Right now, the Wesnoth and Wesnoth-UMC-Dev projects are using this service as a stopgap measure until their respective providers—Gna.org and SourceForge.net—allow installing a hook that can either talk directly to an irkerd server, or to an instance of the aforementioned CIA proxy using the CIA XML-RPC method.

I am not all that keen on other people using a piece of software I developed and tested within less than three days without any prior experience working with Python. There are also various problems inherent to any application depending upon Subversion and its incompetent network transport layer.

Nonetheless, I published a Git repository on GitHub including a small amount of documentation to get started:

• irydacea/irker-svnpoller on GitHub

I am open to possible improvements coming from people intending to use this on production servers. In particular, if someone out there works with a commit mailing list where revision numbers can’t be found in mail subjects it would be necessary to adapt svnmail-filter a little to handle that case. Perhaps it might even be possible to skip the irker-svnpoller step for mailing lists where the notification message structure is constant and well documented.