Lambda the Ultimate

I think this explains why it's so incredibly hard to popularize new programming paradigms. People have to retrain their whole way of thinking, which is much harder than learning a new syntax.

Whoops! I didn't check it on Google since it seemed to be a little weird and slightly off-topic.

> I think this explains why it's so incredibly hard to
> popularize new programming paradigms. People
> have to retrain their whole way of thinking,
> which is much harder than learning a new syntax.

Incrementally adding features might be less demanding.

...and sometimes it is only *specific features* in an alternative paradigm that are really useful.

One of the Oz papers made the point that all Prolog features are not equally useful. The logical database horn clause feature (a database of ground clause facts, a query, and "back chaining" inference rules that produce a set of predicates that satisfy the clauses) being the useful core of Prolog.

For instance, with this subset of Prolog you can embed business logic in a more declarative fashion in business applications. (One of the columnists at Dr. Dobbs Journal referred to this use of Prolog several timesin his column)

The Sapir-Whorf-like question I'd like a more specific answer to is:

How do specialized DSL syntax and semantics help end-users and programmers conceptualize problems and their solutions in the domain?

As Michael V. posted, the computer languages you know well at any one time do affect the way you think about problems, but that doesn't necessarily say anything about one's ability to learn new languages and approaches. That's really a separate question, and I would answer the above quote with an unequivocal "no". If it were otherwise, that would be an incredibly strong claim, worthy of serious study - it amounts to suggesting that learning one thing prevents you from learning another. There are plenty of counterexamples to falsify this claim.

If anything, the opposite seems to be true: the computer languages one is able to master are determined by qualities of the individual, and the main way in which one's first language(s) matter is that inertia might bind one to them.

How do specialized DSL syntax and semantics help end-users and programmers conceptualize problems and their solutions in the domain?

As Ehud implied, being able to manipulate high-level domain-specific abstractions can provide significant leverage and power. In addition, DSLs allows programmers to focus on the problem domain, not on lower-level details, which can free up cognitive bandwidth. For end-users, the same thing can make it possible to participate more directly in programming, which might otherwise be effectively impossible.

These questions all have direct analogs in math and scientific disciplines. Theoretical advances are often only possible due to layering of abstractions, so science is all about "DSLs". From that perspective, it's rather amazing how much in the computing world (especially "professionally") is done without the benefit of DSLs - it's as though physicists were to shun calculus in favor of manual integration. In fact, in commercial computing it's not unusual to find a strong bias against abstraction, on grounds such as e.g. performance concerns. Of course, this kind of thing has an enormous influence on one's thinking about problems.

On the topic of DSLs: the problem I have with DSLs is that it's a pain to have to learn a whole new syntax for every language domain. That's why I like s-expression syntax: you can mold it into whatever you want and the core syntax doesn't change (you just add new special forms through macros). In fact, Paul Graham advocates lisp as a language for writing DSLs, and he says that in lisp essentially all programming consists of writing DSLs whether they're called that or not. Of course, s-expression syntax is not perfect for all domains, but if I have to learn 100 DSLs I'd be happy if they had some syntactic common ground.

Humans apparently have a very strong ability to learn new human languages between ages 2 and 6. This ability drops off dramatically after about age 6. I don't think we know whether something similar might apply to formal languages such as programming languages, since we don't typically teach them to 2-6 year olds. Perhaps we're all hampered by not having learned ML as toddlers?

Instead, by the time we're learning computer languages, we seem to be applying a generic and rather adaptable ability (at least in some humans) to learn formal languages. Amongst those who are capable of learning multiple formal languages at all, I haven't typically seen much evidence of the first language having a particularly special significance, except for reasons of inertia I mentioned - which admittedly can be strong. I know plenty of people who've managed to move on quite convincingly from having learned BASIC, FORTRAN, Pascal, C, or some assembler as a first language. To use a trivial example, one doesn't often find ex-BASIC or FORTRAN programmers hankering after a GOTO, unless they were simply badly trained...

In fact, Paul Graham advocates lisp as a language for writing DSLs, and he says that in lisp essentially all programming consists of writing DSLs whether they're called that or not.

I agree with you and Paul - however, I think it can make sense to provide a sugared syntax if the language is to be used by end users or programmers familiar with some other syntax, and that's not uncommon with DSLs.