Language Adoption and Lisps

Tagged as LISP, Programming Languages

Written on 2008-04-03 01:12:53

Long-Winded Preface
This is my (marginally informed but mostly inexperienced) personal opinion. I have a habit of thinking about things prematurely and overanalyzing them but I indulge in it. It seems most lispers at some point or another either do a roundup of the available lisps to decide on an implementation, try to figure out why lisp isn't more popular, or try to figure out why other languages seem to be growing towards it. I have an opinion on these issues after obsessing over them for a month or two and in the interest of believing some useful knowledge came from that obsession I'll document my opinions here. I should note that I don't yet consider myself a "lisper" for two reasons. One, I haven't yet encountered a formal definition of that term or a sufficiently common form of usage. Two, I simply haven't written enough lisp yet though I am sold on it to date.

Factors in Language Adoption
There are a few things that seem to drive adoption of programming languages generally but I'm interested in a small subset of programming languages so I'll be covering an appropriate subset of influencing factors. Specifically, I'm interested in languages that are not owned or pushed by a corporation but still have achieved some prominence among language elitists and/or some mainstream success. (4/7/08 Edit: This mostly serves as a disclaimer that I won't cover C, C++, C#, or Java here.)

As far as I can tell, these languages all have some of these critical features:
1. A single or dominant implementation.
2. A module system that works across implementations.
3. A killer app or great libraries.

Case(in_point) ->
The languages I'm thinking of are the following: Python, Perl, PHP, Lua, Ruby, Haskell, Erlang, and OCaml. Smalltalk is a perhaps crucial omission from this list but I count both Smalltalk and Lisp on a different language plane because of the sheer history surrounding them. I simply feel more factors are at play.

That said, all of these languages have achieved some preeminence for one reason or another though Haskell, Erlang, and OCaml are certainly not in the mainstream. Erlang sneaks in because of it's recent hype explosion and the limited use it's seen in industry, OCaml sneaks in on the virtue of it's infamous use at Jane Street alone, and Haskell I'm letting in both for it's proselytizing FP userbase and interesting programs (Darcs, Xmonad).

We can see quite clearly that several of these languages have one standard implementation (excluding ports to the JVM or CLR) including Perl, PHP, Lua, and Erlang, Python, and Ruby. As I already mentioned, Haskell has interesting programs at work and OCaml has some commercial usage. Lua has heavy use in the Video Game industry for scripting among other places. Erlang is used at Amazon, Ericsson, Nortel, and T-Mobile. Python, Perl, and PHP are the languages that built that web and Ruby has taken off since all that Rails business.

Common Objections
Many people complain about the communities being elitist or the implementations being insufficient or the syntax being odd. I'd say those are the three complaints I see the most. Certainly, the syntax does scare a few folks. Certainly, there are cases where a lisper doesn't handle a noob with the proper tact. Certainly, there are cases where it makes more sense to use another language on a project in lieu of Lisp. I do not believe that parentheses will deter those actually interested in stretching their abilities and learning a new language. I do not believe that a few flames are indicative of the general lisp community. I do not believe that lisp not being ideal in a given scenario is necessarily a problem with the available implementations.

The Punchline
What I've failed to talk about is point 2 (a standard module system) which I think is presently the most serious drawback to the Lisps. Lisp will never have a standard de facto implementation. There are plenty of implementations already on the scene and many of good quality, combined with the fact that some implementations are designed around a spec (RnRS). I grant that this entire problem could be solved if the Schemes at least standardized on R6RS, or at least the R6RS module system but apparently that isn't happening.

As I've said, we'll never have a standard implementation and, from what I can tell, the absence of both a standard implementation and a standard module system simply slaughters the code portability that would help a healthy community of code sharing. This precludes the libraries and interesting programs that lead languages out of the elitist ghetto. Most programmers won't touch a language regardless of it's feature set if they can't be productive with it fairly rapidly without writing the sort of libraries they've come to expect. If there is something wrong with lisps, this is it. Very few programmers will seriously try a new language without easily available libraries that make the language have a batteries-included feel to it.

Now, I realize that standardizing all Schemes on R6RS is impractical but we don't need even that. If we could manage to just get the three largest Scheme implementations to unite on one module system we would have made tremendous progress. I'd say the three largest Schemes in userbase with a module system are: PLT Scheme, Gambit-C, and Chicken Scheme or Bigloo Scheme. If we could get even these 3 or 4 implementations to standardize, I think it would be enough to really get things rolling. Even 2 of them standardizing on modules might be enough to pull others in. I can't personally think of a bigger win.

The Inflammatory Bit
I don't mention Common Lisp here because, again I stress personally, I hope we stop encouraging it as a starting point for newcomers. I don't think it's a good way to encourage people to use Lisp. I consider it more fragmented and thorny than Scheme for the beginner though I acknowledge it at least has a dominant implementation, from what I can tell, in SBCL. I realize that it has contributed a great deal and I don't mean to discredit it's achievements. It simply seems something better left to the experienced.

Personally, my disagreement with CL probably stems from two things. One, I dislike the design decision of separate namespaces for functions and arguments. Secondly, I feel that if most of CL's added functionality could be bolted on to a Scheme implementation through the use of libraries then why not have a Scheme implementation with said libraries and allow that to usurp the role played by prior CL implementations. I do grant that would be a lot of work to redo without good cause. At any rate, these are my inexperienced preliminary observations and my opinion may change drastically over the next few years as I have time to give it a fair chance and read through PAIP and On Lisp.

The greatest advantages of Common Lisp over Scheme I would expect to hear as arguments are it's use of CLOS and Namespaces, it's libraries, and it's quality implementations. I believe all of those are solvable issues in Scheme provided a standard module system enters the picture. I also have some opinions about why the Schemes should focus on distribution and concurrency after modules and further opinions on compilation to C versus native code. I leave that for another time.

In short, the central problem lisp must overcome for mainstream success (which may not even be desirable) is a standardized module system. The lack of a module system prevents a culture of code sharing which is preventing the creation of a software ecosystem around lisp. This is, at root, a sociological problem emerging from a technical problem. The lack of a standard module system and shared code produces the social effect of Lisp's continued obscurity. Lisp's obscurity does not stem from some deep difficulty inherent to the language, noob-averse communities, shoddy implementations, or the Lots of Spurious Irritating Parentheses.
comments powered by Disqus

Unless otherwise credited all material Creative Commons License by Brit Butler