Tuesday, May 01, 2007

Layering of VMs. DLR considered big.

Conceptually, I very much like the DLR - Dynamic Language Runtime.

It creates a sensible layered conceptual model in the heads of software developers that fits dynlangs on top of static langs which, in turn sit on top of bytecode. This is a good way to gently veer developers towards the power of dynlangs.

I remember in the Eighties finding myself gently weened off direct 8086 assembler and into C because I could still code bits in assembly if I wanted to and "glue" them together with an over-arching C-based application. This preserved my ability to use assembler if I needed it but gave me significant extra flexibility and a productivity boost to boot.

Same model today - just one level further up the stack of layers. Nowadays, you can (should![1]) be coding up whatever modules require C#/Java/C++ static typing directly in that language and then glueing the modules together in a dynlang framework. That gives you the ability to directly code in C#/Java/C++ when you need it but gives you significant extra flexibility and a productivity boost to boot.

If I were operating in full dynlang propaganda mode I should probably stop now but I'm not so I won't. There is a further developmental stage to this model. It goes like this...
    At first, I used to code up modules in assembly and link them together with my C modules in an overall, C-based application.

    Over time, the number of times I would resort to diving into assembler got fewer and fewer. Eventually, I was coding end-to-end C applications. The drop in run-time speed was more than compensated for in terms of productivity.

Now. Extrapolate.

[1] In my opinion :-)

No comments: