intertwingly

Andrew Grumet: Mark, I apologize for calling you a liar. I was feeling defensive and exasperated about a technical issue that I didn't understand. An issue that had very public consequences. You misread the situation and said some hurtful things. Now I understand the technical issue, and am unburdened. It wasn't like you said, but I believe that you believed what you were saying at the time.

Andrew and I finished debugging this problem yesterday.  It took a number of sessions, via AIM, phone, and email, to finally track it down.

Background:  nearly two years ago, ScriptingNews's RSS feed was placed in a new namespace.  This was a serious issue; but not as it was asserted for parsers that did not understand namespaces, but for parsers that do.  Since then, the RSS 2.0 spec clarifies the issue “The elements defined in this document are not themselves members of a namespace”.  Earlier this month, I inquired if the feedvalidator should mark such usages as an error.  Rogers Cadenhead agreed.  I made the change.

Until I made that change, the feedvalidator treated such feeds as valid.  It did so by keeping track of the namespaces to be globally ignored.  Because of the way that CGIs work, this list was reset on every invocation.  By the way that mod_python (the technique the implementation behind rss.scripting.com uses) works, it was not.  The net effect was that a list of namespaces to be ignored accumulated over time.

This bug was fixed in February.  Unfortunately, the validator hosted on rss.scripting.com has not been updated since its inception.  Meanwhile, this bug never affected users of the validator at feedvalidator.org; it only affected rss.scripting.com because that was deployed in an environment that the feedvalidator was never designed for.

Everybody makes assumptions.  Both Andrew and I made, and tested, a number of assumptions in the debugging process.  Some of them assumptions were wild.  The choice of namespaces to be flagged looked very suspicious.  Until this was resolved, I shared the same concern.  But when all was said and done, the answer was rather mundane.

It is my hope and wish that this blog entry does not trigger another round of accusations.  And that the operators of rss.scripting.com consider updating to a more recent version of the validator.  With Andrew's permission, I've committed his front end to the cvs repository.  I'm willing to keep it in synch with the code.  Let me know if there is anything more I can do to help.

Meanwhile, it was fun working together with Andrew to debug this problem.  And kudos to Andrew for apologizing, and finally, thanks for reviewing early drafts of this blog entry.