RSS and namespaces in 2002

Yesterday we tripped over what may be a very serious issue in XML. As with many things related to XML, it's highly politicized, so I'll try to cut through that and stick with what we know and hope that people will add data they have where I got things wrong.

Here's the deal -- there are some XML parsers that don't properly deal with namespace attributes on the top-level element of a source. For these guys, just introducing an xmlns attribute is enough to make them reject the feed. So while they could handle a 0.92 feed, as soon as we introduced the xmlns attribute, they gave up. This was a problem for our users, to put it mildly. We immediately released a new version of the serializer, omitting the xmlns attribute, and (knock on wood) all seems to be going well.

Now I have a straight question. Presumably RSS 1.0 doesn't have the same problem we tripped over yesterday with RSS 2.0. So I looked at a few RSS 1.0 feeds, and guess what, they do the same thing we were doing with the 2.0 feeds. Amazingly, they generally have several xmlns attributes on the top-level element, not just one. You could knock me over with a feather I was so surprised. Therefore I conclude that the same broken parsers that didn't like the 2.0 feeds with the xmlns attributes, must also not like the 1.0 feeds.

If this is true, we can't design using namespaces until:

1. All the parsers are fixed, or

2. Users/content providers expect and accept this kind of breakage (I don't want to be the one delivering that bit of bad news, got burned not only by the users, but by developers too, people generally don't know about this problem, or if they do know are not being responsible with the info).

Anyway it looks to me like there's a big problem in the strategy of formats that intend to organize around namespaces.

Dave