Though the href attribute permits relative URIs and is clear on how to resolve them to full URIs, client software sometimes does not look for a base element or treat a document's URI as the base. Additionally, some feeds won't have a URI when delivered by means such as e-mail. If a base URI has not been specified in such feed data, relative URIs will not work. The best way a publisher can be assured of autodiscovery working is to use a full URI as the href value.
Proposed addition to Section 4.3:
To account for client software that overlooks the base URI in relation to autodiscovered links, publishers SHOULD identify feeds with full URIs.
If this is adopted, the examples in sections 7.4 should be revised to employ the recommended use of full URIs.
This approach is consistent with the RSS Autodiscovery specification recently published by the RSS Advisory Board:
Ideally, the two groups will ultimately offer the same guidance on autodiscovery. Whether that means Atom follows the RSS lead or vice versa is an open subject.
(Posted by James Snell on Rogers Cadenhead's behalf)