Sunday 23 August 2009

Minting MIME types

MIME types are central to the REST architectural style because they help define the interface between client and server. However they have to be registered with a central body, which is an incumberence when creating new MIME types.

Benjamin Carlyle has a suggestion for how we might bypass the bottleneck of registering MIME types with the IANA. Stefan Tilkov has been following similar proposals for some time.

There are two separate problems in this debate that I think are being confused.

One is the identification of new resource types. The other is providing a definition of the format of resources e.g. an XML schema.

Benjamin's proposal involves using a URI to identify resource types. If the resource definition URIs of two resources differ, then the client must interpret them as different resource types. Dereferencing a resource definition URI will yield a definition of that format.

The difficulty as I see it is that multiple URIs could point to the same resource definition. Furthermore, the data format might be defined in multiple places and in multiple ways e.g. a DTD and an XML schema. Ideally conneg would be used to put equivalent definitions behind a single URI, but in practice conneg is oft-neglected.

The process of agreeing on a canonical URI to use for a given format is no simpler than agreeing on an "x-" prefixed custom MIME type. So while Benjamin's proposal helps solve the resource type definition problem, I don't think it makes much progress on the resource type identification problem.

Benjamin quite rightly points out that a key problem in resource identification is when resource types 'grow up' and move beyond the boundries of the organisation where they were created. However I fear that using URIs to identify formats makes this transition more difficult, because if the URI of the format description is changed then all clients using the format will have to be updated.

If the original URI of the format was internal to the organisation's LAN, then the URI will be forced to change. At some stage during the format's development the URI will probably have to be changed anyway.

In short, URIs are good for pointing to definitions of resource formats but are problematic for establishing the identity of resource types.