Re: C'est pourtant évident.

Ça me rappelle un entretien accordé par Brian Kernighan :

M: I have a question about research in language design. It's interesting for instance that Java is very much hyped and the community is split among the merits and flaws of the language. The language has indeed acquired some nice features proposed by researchers in the area (like garbage collection), but also the researchers point some of its weaknesses (like the arrays which are covariant and they shouldn't be). There's a whole body of research done in programming languages nowadays, and a very interesting body of research in functional programming languages, but you don't see this research to really influence the real world, i.e. what people are really using on an everyday basis. Instead all sorts of ad-hoc languages pop up like Perl or Python or stuff like that. Where do you see the fault; what's not right?

K: That is unfortunately a very good question, and there's a certain amount of discussion here at Bell Labs between a very strong group in functional programming languages and a group using very much ad-hoc, pragmatic languages. I honestly don't know why the functional languages don't succeed. For instance ML, which is arguably the best combination, perhaps the one that ought to succeed: in spite of being a very well designed language, thought hard about by a lot of good people over a very long time, embodying an enormous amount of effort of compiler technology, still does not seem to be broadly used. I will oversimplify a lot, and probably offend my friends, by saying that the only thing people do with ML is to make ML compilers. [laughing] I'm overstating intentionally, but it has some of that flavor, and I don't really understand why. I think, speaking only for myself, part of the reason that ML in particular, and functional programming languages in general have not caught on more broadly, is that they're aimed at people who have mathematical sophistication, who are able to think in more abstract ways, that lots of other folks, and I include myself, have trouble with. Whereas languages like C are very operational, you can see how every single piece of them maps into what's going on in the machine in a very very direct sense. If I had been brought up at a different time and in a different environment perhaps I'd be totally comfortable in ML and would find C unsafe, a little dangerous, not very expressive. But my sense is that the functional languages come out of a fairly mathematical community and require a fairly mathematical line of reasoning and therefore are difficult for the people on the street.

M: So I guess, the suggestion is for the researchers to somehow lower the language level, to promote the good qualities?

K: I didn't really answer the other part of your question, why research in languages has not had as much effect as it should have. I think actually it has had an effect, in places like parser technology, code generation, no matter what the language is: research had a big effect on building language tools but less on the design of languages per se.

The languages that succeed are very pragmatic, and are very often fairly dirty because they try to solve real problems. C++ is a great example of a language that in many ways has serious flaws. One of the flaws is that it tried very hard to be compatible with C: compatible at the object level, compatible very closely at the source level. Because of this there are places where there's something ugly in the language, weird syntactic problems, strange semantic behaviors. In one sense this is bad, and nobody should ever do that, but one of the reasons that C++ succeeded was precisely that it was compatible with C, it was able to use the C libraries, it was usable by the base of existing C programmers, and therefore people could back into it and use it fairly effectively without having to buy into a whole new way of doing business. And this is not the case for ML, which was being done at about the same time and, at least partly, in almost the same place, but which took a very different view of the world. As a pragmatic thing, C++ is extremely successful but it paid a certain price by being compatible with the previous language.

Personnellement, j'ai naguère essayé de me mettre à Ocaml, malheureusement tous les tutos que j'ai alors trouvés étaient effectivement très orientés "matheux" ; le genre plein de formules incompréhensibles pour moi et se paluchant longuement sur les suites de Fibonacci, sans jamais expliquer comment simplement récupérer/reverser les données dans un fichier (le B-A-BA du truc utile, quoi).