Emerging Languages

Link: http://www.developerdotstar.com/community/node/144

In the last few days, Martin Fowler posted a few essays on language workbenches and their relation to MDA and MDD. Today John Cowan posted on his Recycled Knowledge blog about traits, as used in Sun's proposed technical computing language, Fortress. I'd been aware of Fortress, but had missed the draft spec coming out.

I don't want to knock Fortress, as it's hard to have a technical computing language better than Fortran and as net-savvy and secure as Java, but I'd agree with Edward G Nilges's blog. Having α := x² isn't enough.

The big thing that Fortress claims as its selling point isn't that it's a secure, robust, transactional, object oriented language. It tries to sell itself on its syntax looking like maths.

Whatever you do with monospaced text, unicode or no, it's not going to look like text-book maths. For that you need a decent document rendering system, and can use block formatting model result from on relation or transforming the abstract syntax of the language, much like the language workbench systems do. Fortress' examples of 'maths like' syntax simply don't do it well enough. There was also a quite sexy demo of Mathematica's means of exploring computational model spaces at Apple's WWDC, which shows what you can do with a compound document style UI in mathematical programming.

But as well as not being quite enough in terms of becoming literate maths, it seems to ignore Fortran and the existing codebase.

The big problems with agile, high productivity technical computing are how to build models quickly, and scripting languages that apply leverage to existing systems (such as SciPy and my own (unfortunately closed source) Jini based system for creating a distributed agile front-end to Fortran code) offer much better reuse of current code by integration rather replacement. You don't want to rewrite your aerodynamic stress library to use it, so these techniques give better return on effort. Which is what high productivity is all about.

That said, there is also some hard to read code out there that results from the lack of data structures in F77 and the aggregation of patches to 20 or 30 year old systems. There is some merit in applying relational KR models, such as conceptual graphs to documented legacy code to extract the intent, and I hope to put aside some time to play with applying something like a pattern match to some of our larger application sets (10⁷ lines Fortran, mainly single purpose stove-piped programs with no shared libraries) for finding redundant code that can be extracted out to a managable codebase.

There's simply no way we can rewrite our code in Fortress, and FFI and code transformation are more important than syntax.


TME