Why Clojure will win
michaelochurch.wordpress.com
michaelochurch.wordpress.com
This seems to be a popular meme, right until someone asks for details – crickets chirp.
That is good and bad. Good: you can probably find good use for every feature Scala has. Bad: On a sufficiently big project with one or more extra clever developers most probably every feature will be used and that will make it extra hard to understand for normal software developers that have to maintain it for long years after the original extra smart developer already left the company. See it a bit like that: On a hard to understand scale with 1 being easy to understand und 10 being extremely hard to understand Java projects can go from 1 to 6 and Scala projects can go from 1 to 10.
Personally I like simple languages more. And I don't need a compiler to catch errors that a unit test would also catch.
But if one compares Scala to languages which are actually used in the wild like Java, C#, PHP or C++; Scala's footprint is just incredibly small.
All of the features you mentioned are available in languages like Java or C#, too. The difference is that in most cases the implementation is so poor and incoherent, that the feature is next to useless.
I guess one just "sees" more language features being used in Scala, because almost everything which ships in the language makes some sense, while in Java/C#/PHP/C++ you have tons of utterly useless "features" which don't add any benefit, but still force people to learn them in case somebody else used them.
I have never seen "On a sufficiently big project with one or more extra clever developers most probably every feature will be used" happen in real life. In my experience one of the strengths of Scala is the way one can use some advanced feature where it makes sense, without having it bleed through the whole code. Compare that with Java's, C#'s "Oh, you used some complicated Generic type over there? Guess what, we will make every user suffer from it throughout your whole code base!"
In a sense, I prefer a language (Scala) which ships with 100 features where 90 of them make sense to a language which ships with 400 features where only 50 make sense, like C# (multiply "feature" count by 10 for C++).
> And I don't need a compiler to catch errors that a unit test would also catch.
And of course: Using a statically typed language is no excuse to not write unit tests. Unit tests serve as a safety net checking if the behavior of the implemented unit is (and stays) correct. As a side effect this will also find all type errors a compiler will find. I have programmed a lot in Java and also in dynamic programming languages and in my opinion type errors just don't happen often enough to justify the additional amount of work and complexity that comes with static type systems. However I'd agree that if you want to get the best possible performance you probably need a statically typed language as the compiler can then optimize the generated code better.
> Java has no type inference, higher order functions,
> coroutines, implicits (it has some standard implicit
> conversions but not implicit conversions that can be
> added by the user) and no Turing complete type system.
int bar = new Object(){ int foo = 42; }.foo;
// Magic? No. Type inference.
http://download.java.net/jdk8/docs/api/java/util/stream/Stream.html
Looks pretty higher-order to me.
Also possible in older version in Java with stuff like Google Guava or FJ.
http://stackoverflow.com/questions/2846428/available-coroutine-libraries-in-java
Having them built into the language is even worse.
In Scala you could just de-import them and have them gone.
Anyway, pick C# as an example instead if you want to have
user-defined implicit conversions.
interface Interface<T> {}
class Bang<T> implements Interface<Interface<? super Bang<Bang<T>>>> {
static void bang() {
Interface<? super Bang<Byte>> bang = new Bang<Byte>();
}
}
Looks pretty undecidable to me. In a way, it is even
worse than other languages: You are unable to tell
whether the type checker will terminate in finite time
(like some other languages), but don't get any additional
expressivness in return (unlike those other languages).
> As a side effect this will also find all type errors
> a compiler will find.
> I have programmed a lot in Java and also in dynamic
> programming languages and in my opinion type errors
> just don't happen often enough to justify the
> additional amount of work and complexity that comes
> with static type systems.
If I have a function: def upcase(s: String): String... The compiler with its static type system can tell me if e. g. I am trying to call the upcase function with an Int argument. But in my experience something like that just does not happen very often. Most programmers are not so stupid that they try to fit a square peg in a round hole. ;-)
A unit test can check if the returned string is really the upcase version of the input parameter. Plus if the programmer would really assume that he could upcase an Int it would just as well give the feedback that his assumption was wrong-like the compiler.
> It seems we have different definitions of what type
> inference, higher order functions and Turing
> completeness means. :-)
I think we can agree that the way Java supports those features ranges between unusable and pretty terrible in practice, but that's not what your claims were about.
Your notion of Scala having more features than a "standard" mainstream language might even be correct, it is just that the points you have given don't support your point here.
> Regarding coroutines in Java: I don't know if libraries
> using native continuation mechanisms or bytecode
> manipulations prove that Java (the language) has
> continuation support. :-)
> I don't have anything against Scala.
It's just that you have claimed a few things which I think are incorrect, either factually or in the way they were worded, and I'm offering some data points against it. This all happens in good faith and in the hope that we both will have gained some knowledge at the end of this discussion.
> As you noticed with Java 8 Java is already borrowing a lot
> of useful things from Scala. I am sure conservative
> corporate decision makers will see less and less reasons
> to switch to Scala as time passes by.
> If I have a function: def upcase(s: String): String [...]
> A unit test can check if the returned string is really
> the upcase version of the input parameter.
def upcase(s: String): UppercasedString
Ok, then let's just ignore continuations as they are somehow available in both languages although the Scala one feels a bit more officially endorsed to me.
> def upcase(s: String): UppercasedString
Even with that type signature of course the compiler can not check that for input "a" the output will be "A". Your code could have a bug and upcase "a" to "B" (Off-by-one errors are quite common) and your compiler would be perfectly happy. That is the whole point why unit test are more valuable than compiler type checks: checking specific code behavior and not only static types.
That's the thing about the Clojure community. It doesn't have wackjob fantatics like you.
> Before I learned about Clojure, I certainly wouldn't have
> believed it if someone told me that there was a Lisp-like
> language without the pretentious assholes.
Here's a speculative reason why it may not win:
Java, Ruby, PHP, Python all won because intermediate level programmers could, all of a sudden, successfully build 2000-era sized apps.
These languages are not doing so hot for 2013-era sized apps. It now takes extreme skills to successfully ship 2013-sized apps which is why a small portion of the programmers are making most of the money.
The next language to win will be the one that lets an intermediate team ship apps again. Java let your average C++ programmer build much bigger apps without crashing. In 2000, Java forced you to program right by 2000 standards; the programmer needn't have figured it out himself.
A language that lets a strong architect say, "NO! you CANNOT do I/O in the view, that goes in the controller!", that lets an architect shape what good code should look like and have the language enforce it, can lead a team of intermediate programmers to success.
Clojure is like a musical instrument - a brilliant artist can do amazing things, but a mediocre artist will produce mediocre output. Clojure works great if you have 5 brilliant architects collaborating - Datomic, core.async, ClojureScript, all these are outputs of brilliant minds collaborating; the company behind these products has a reputation for only hiring the best of the best.
But I think a business can make more money by putting 5 brilliant architects on 5 separate projects and giving them a bunch of intermediate programmers. The intermediate programmers learn; the architect multiplies; the pointy haired boss gets rich, everybody wins.
That said, I can never see Clojure "winning" as in how Java "won". For all of Clojure's strengths, it's at best an intermediate step up from Ruby, Python, Perl... It's not that next order of magnitude leap of productivity that is necessary to "win".
The research just isn't going in that direction. The research is going into type systems and static guarantees. And I think the real productivity boosters of the future will be in very smart tooling - a conversation with your IDE I would say. Maybe Clojure could have different views (different surface syntaxes). Maybe it could be the basis of something like intentional programming. But straight Clojure with Emacs isn't going to be "winning" in the big way.
> but it will win in the next 15 years in a major way,
> because it is already one of the most interesting
Anyway, I think they have already won in a major way: The managed to build a community which is supportive and friendly. This is a huge deal when you have a look at the poisonous behaviours of individuals and the wide-spread arrogance typical in most Lisp communities.
Before I learned about Clojure, I certainly wouldn't have believed it if someone told me that there was a Lisp-like language without the pretentious assholes.
> An enterprise Scala project, for example, is typically
> going to present the frustrations that come with
> components that live outside the language. Learning Scala
> isn’t that bad, but Maven and Spring and Hibernate? Ouch.
> In Clojure, most of the typically “external” needs are
> provided with idiomatic libraries that can be reasoned
> about within the language.
> while every static language worth its salt has an
> interactive mode (“REPL”) it doesn’t have the same
> first-class-citizen feel
I think the opinion voiced in the article is interesting, but past experience tells us that the chances of a Lisp to get even slightly popular is low to non-existing.
Is that really true of most Lisp communities, or just Common Lisp's?