Useful side effects of the perl testing culture
I am currently in the middle of trying a new mechanism to deploy perl applications (several which are catalyst based) on servers at work while attempting to keep the level of supporting system (ie perl and the required modules) maintenance work to a minimum whilst being able to use modern perl - and we are running on Centos systems by default, so currently the system perl is 5.8.8. I intend to write about my proposed solution to this in a few weeks when I have had more of a chance to see how it works...
However as part of this I have been building perl and a whole raft of supporting modules under a set of build system scripts - encouragingly the whole build appears to be around one hour in length.
What I wasn't expecting to see was test failures when building the modules.
I really wasn't expecting to see a set of failures down to the tests failing in XML::Parser - thats a widely used and slow changing module.
So I went digging... and finally got to the Red Hat Bugzilla entry #556415.
A undertested security fix to expat had caused a regression - which went unnoticed other than by the XML::Parser test suite.
Thats a big win for the perl testing culture.
I've found it's good just to include a Perl distribution in the deployment process - it's easy with local::lib. Since you're having to compile modules to architecture anyway, compiling Perl allows you to not be dependent on RedHat's impossibly slow upgrade drifts.
I think I've been hit by the same bug. I was wondering what was going on…