Representational Computing


Completing a task has a representation, such as 'render image' and if requested will select a program to do so depending on criteria. If the input data is in PNG format then it must pick a program representation that can render PNG. The inputs and outputs of programs are representations themselves so the type matching works upon programs themselves.

If an operation on a particular format has not been implemented but can be readily converted into a format that supports the operation, the system can follow a chain of conversions that allow an operation to be satisfied transparently.
RC is akin to a program requesting a library that provides 'X' or 'can do X to Y' and having that library injected in. (see representations are tests)
There are many different ways of completing a task and they can be approached from different perspectives. Given a set of inputs and the intended goal, mappings or wrappers can provide for treating one representation as another but in a more formal way than industry practice. What something wraps and what programs are wrappers is in the heads of people and not known by the machine.
There needs to be some form of proof that two libraries will be interoperable. Representations as tests could provide this promise as a developer will import the representation of a task and any library promising to satisfy this representation would have to pass the tests dictated by the representation.
Existing:

Negotiation protocols
Aspect oriented programming
Type systems, Type systems in the operating system and desktop environment
Dependency Injection
Testing in the field (production code that runs tests to make decisions, such as Modernizr)
Weaving code into binaries by purpose