Language-oriented or Metadata-driven?
Sergey Dmitriev has a great
article on what he, and others, have called Language Oriented Programming. Go read it, it’s a great piece. As we're heavily engaged in building
Domain Specific Language tools at present we're obviously swimming in the same sea. However, I'm keen to look at LOP and DSLs as only one aspect of the bigger picture - I'm especially reticent to attach too strongly to any meme that appears to put the focus solely on programming.
Enterprise system development these days has almost as much emphasis on configuration of middleware and packages, specification of deployment options and physical rig topologies. It is also hugely influenced by the need to meet non-functional requirements, a subject on which programming languages have typically had remarkably little to say. This is a major reason why
UML isn't the whole answer.
What seems to be needed is a consistent metadata substrate encompassing all of these things. Then tooling and transformations can readily be used to link together different islands within the substrate, validate consistency, manage dependencies etc. Clearly all of the components of a heterogeneous system aren't going to use the same technology to manage their metadata anytime soon, so good adapters are necessary to make various stores available.
However, this huge sea of metadata is intimidating and full of detail. I see Languages as a vital enabling technology here to help abstract this problem into something that humans can manage more reliably. Languages will drive the generation of custom application code. Languages will configure middleware. Languages will allow specification of performance requirements. Crucially, statements in the same language will bridge all, allowing the human who knows they want a new Database to specify that intent. A different human will have codified what detailed information is needed to populate the islands of metadata and what supporting frameworks are required to make the right things happen at deployment time, at application run time, at monitoring time or any other part of the lifecycle.
Gradually we'll grok this approach for abstracting and coordinating the complexities of our horizontal technologies and move on into the realms of our vertical domains - then we'll really be getting towards our Software Factories vision.
