Switching to Plan J (From Scala back to Java) | Hacker News Reader

Switching to Plan J (From Scala back to Java)

alarmingdevelopment.org

95 pointscfontes65 comments

What the the article seems to miss is that Scala can be used strictly as a "better Java" without using any of the advanced type features or idioms if that is what you want.

(Anecdotal)War Story: Once upon a time I worked with a team of scientists (not professional developers) who had some legacy research code in Java, had tried moving to Scala and got badly bitten by the advanced type theory bits and so were considering moving back to Java. When I joined the team I taught them how to use the "Scala as a better Java" approach, (which would correspond to approximately the L1 level in Martin Odersky's blog post at http://www.scala-lang.org/node/8610) and to this day they are happily building new stuff in Scala (and Java).

Just as you can use Common Lisp without completely mastering macros, the condition system etc, to get stuff done at the cost of some redundancy when compared to what the code would look like if an expert were to write it, so can you use subsets of Scala to get serious work done.

I suspect problems arise when you have the attitude that you must completely master the language before you start using it to build stuff. This excludes all but the ultra simple languages from consideration. Reductio Ad Absurdum, Imagine a world where the only programming language in use is VB (or even C) !

I offer the hypothesis that when using any new and "large" language on a real life project you have to work with a subset. This is not a bad thing and "subset" is not a dirty word (as long as the subset is clearly identified and at least as good as the alternative languages). As you gain experience with the subset and actually build useful systems with it you can then go deeper into the language and learn more advanced features, which you can then use to refactor your code to use those advanced features where appropriate, in a virtuous cycle.

Imagine you are learning a non programming language (say French) or even musical notation. If you wait till you have wrapped your head completely around whatever you are learning before you start using it, you are (a) going to wait a long time (b) very unusual in your approach. Is having an ultra simple syntax or a small surface area a virtue? Yes it is. It is one aspect of a design but not the only one.

The lack of an IDE is a more valid complaint, (especially if you plan to have big teams of mediocre programmers) but then you restrict yourself to using only core mainstream languages with very strong vendor ecosystems (say Java, VC++, C# and VB). It would be nice if every new language sprang from their creators brows fully equipped with ultra functional IDEs but that maybe a tad unrealistic a wish ;-).

I also suspect that most of the carping comes from people who haven't actually used Scala in any serious fashion. Complaints about "too complex a type system" from people who have never used Scala to actually build something have about as much legitimacy as people who refuse to use Lisp because it has "too many parantheses" and just "know" that parantheses are confusing.

I think the "subsetting the language" approach is a red herring. Yes, you can decide you'll only use feature X, Y, and Z. Many people do that in similarly complex languages like C++.

But the programmers will, one way or the other, get exposed to the features you considered esoteric. Be it through libraries, or sample code, or whatever. And then they have a piece of code in the language they use for daily work that they don't understand, can't debug, and are alienated by strange compiler errors - not good.

It's IMHO also quite hard to come up with the subset you like. I mean, programmers like to have endless discussions about completely irrelevant topics, like brace placement; wait and see the discussions about actual language features to (not) use! Like I said in the other comment, most features are in themselves completely reasonable, so it's hard to make a cut here.

I also think the 'mediocre programmer' is a red herring. Not that there aren't mediocre programmers that will have trouble to understand certain language features at all - there certainly are.

But I'd wager that even for very good programmers, having many complex language features (that interact!) in a code base will be problematic. Of course you can understand what's going on with some effort on your side, but the problem is with the effort. Programming languages are IMHO a bit like user interfaces in this regard - "don't make me think". Having code in a code base that's complex and/or hard to understand is a really, really bad thing. People will shy away from reading and modifying it, and over time it'll grow into an unmaintained monstrosity.

Again, this is about the complexity budget. Original Java code was dead easy to understand, now with generics it's a bit harder (but overall I think generics are less bad than you hear all the time). That was and still is one of the really killer Java features. Not only for mediocre programmers, but for general readability, and thus usability of the language.

"Of course you can understand what's going on with some effort on your side, but the problem is with the effort. Programming languages are IMHO a bit like user interfaces in this regard - "don't make me think"."

By this logic no one would use the command line and Windows click thru wizards blow Unix away. I agree that guis are good for non programmers. Java is better than Scala for some subset of programmers. So is VB. The problem arises when that statement is modified to use the Universal Quantifier.

The thing about a "powerful" programming language though (or for that matter a command line) is that it is a power tool. No one forces you to use it and you shouldn't if you feel it is too complex, but using some mythical "complexity budget" as if it were some kind of inviolable scientific fact is (imo) bogus. The "complexity budget" of anything is a function of the context of its use. As cemerick points out in another post here, Scala or Clojure (or Haskell or Erlang) will always have a "complexity" higher than VB or Java and that is fine.

"Again, this is about the complexity budget."

There is no such thing independent of the intent of the language designer and the context of the use of the tool. Scala and Erlang and Haskell are not intended to occupy the same "complexity point" as Java 1.4. So what?

Pascal has a lesser "complexity budget" than C. That doesn't make the former uncontestably "better" independent of context.

And anyway my original point was only that once you choose a language which you can't grasp all at once, for whatever reason, you should start programming with the subset you do understand rather than wait till you completely grok it. You can always choose to work with "low complexity" languages if that is what you want.

"As cemerick points out in another post here, Scala or Clojure (or Haskell or Erlang) will always have a "complexity" higher than VB or Java and that is fine."

I find the complexity budget of Clojure much lower than Java, and maybe VB, too (haven't programmed VB in a long time).

In many ways, immutability by default, first class functions, and a library of higher order functions in the standard API make code much less complex with fewer concepts to understand than the corresponding Java code would be. The problem for many coming to Clojure is having to "unlearn" the "object oriented" Java concepts they already know (true for other functional languages, too, of course).

Clojure leaves type declarations completely optional. It does not try to assemble disparate programming paradigms and integrate them into a coherent whole. The built in data structures all implement the seq interface, so you can use the same functions on all of them.

Rich Hickey spent his complexity budget very well. I would rank Clojure with Scheme and Smalltalk in that regard.

"I find the complexity budget of Clojure much lower than Java, and maybe VB, too (haven't programmed VB in a long time)."

I am not disputing that you find Clojure easier than Java but if you include complete mastery of macros and laziness and functional programming idioms into the "complexity budget" it is very arguable that some Joe Blow developer would find Java easier to learn and read and consider Clojure "too complex". I am just pointing out that they are bogus reasons to reject a language. The OP is claiming that the deep type theory constructs of Scala are too difficult for him. Of course they are, just as deep macro fu would be hard for a Clojure beginner.

Please don't mistake me. I prefer Clojure to Java any day. I am reacting to the idea that "too many parantheses" or "but macros are too complex" or "laziness is too hard to wrap my head around" or "Scala has a more powerful type system than I am used to" or "Erlang has too weird a syntax" etc are sufficient reasons to reject them.

Back to Clojure. My (original) point is that you don't need to understand macros (say) in all their glory before you build something in Clojure. Use a macro-less or macro-lite subset and grow into macros as you gain experience. Yes you will write less than the best Clojure for a while. (Imho) that is fine (and unavoidable). Rejecting Clojure because you couldn't completely understand macros (or read "complex" code with lots of macros) is (imo) not an optimal strategy.

> Pascal has a lesser "complexity budget" than C.

Well, the point of the "complexity budget" idea is that you just have a fixed budget, and can spend it on things. So everybody gets the same budget - maybe Pascal just didn't spend parts of what'd been available.

Regardless, as I said, this is not about good or bad programmers. The shell (i.e., not the horrible abomination called "command line" on Windows) is not a complex tool. The number of different concepts in a shell like bash is actually quite low - commands, functions, environment variables, substitution, globbing, if, for, pipes & streams. Escaping is a bit idiosyncratic, but hey. I'd even argue that using a well designed shell command is actually simpler than using a Windows wizard - the number of concepts and hoops you have to jump through are less, you just have to learn a limited number of concepts once.

Same goes for Erlang. That's actually a remarkably simple language. Once you wrap your head around functional programming, to the point where it could use some more complexity IMHO (like better typing/structs).

Understood like that, the "complexity budget" is actually a valid concept. Languages have features (concepts), and they tend to interact with each other (e.g. static, initializers, final and constructors in Java). If you have to many of them, the interactions get unpredictable and messy, and the language has too many concepts to keep in mind easily at the same time.

I'm not arguing for dumbing things down to the minimum here. Complex features are fine if they give power, but having too many of them, and possibly ill designed ones, will yield an unnecessarily difficult language. The difficulty of language design is IMHO what to take, what to leave out, and how to pick features to give a maximum of power/expressivity with the minimum complexity.

I think you're missing the point. Of course ultimately a programmer has to master (to a point) the language he uses, even if he doesn't use all the features.

I think GP's point was more "don't wait until you understood everything to start coding", and i must say i agree wholeheartedly

Java was not considered easy to understand back in 1997. There were cries about it being "too object-oriented". Developers were mostly used to procedural coding at the time. Thanks to hundreds of OO and Java books, along with the internet, developers came to see OO as a standard way of developing - Java benefited from that paradigm shift. (Good marketing helped as well.)

We are entering another paradigm shift into functional programming. People are complaining about Scala's complexity but they are really more mystified by FP. Given time and lots of code sharing, FP and its representative languages will seem absolutely normal.

> But the programmers will, one way or the other, get exposed to the features you considered esoteric. Be it through libraries, or sample code, or whatever. And then they have a piece of code in the language they use for daily work that they don't understand, can't debug, and are alienated by strange compiler errors - not good.

Alternatively, this is a learning opportunity. Not automatically bad, either, especially if the resources are in place (especially mentoring) to really exploit it.

And, (opinion warning!) unlike C++, in Scala the techniques & principles of the more pointy-headed stuff can often be reused in other settings.

First on Java generics: I don't understand all the complaints about generics anyway. They're very useful and not really hard to understand, basically. Besides the readability gain, you get compile time errors, or even hints in an ide for types that don't fit compared to casting from object to the type you expect to get. Also APIs are a lot more readable and expressive when you don't have a bunch of methods taking objects as parameters.

I think the "subsetting the language" approach is a red herring.

I don't think that the compatibility in syntax is meant for subsetting, but merely for approaching a language one step at a time. When you learn programming in a new language that has this advantage, you can start right away, using familiar concepts and learn more features as you need them. You can grow into the language.

At least that was my approach in learning languages like VB.Net(oh yeah..), C# and Java for example. The specific advantages of each language come in when i try to simplify things or do something in a better way.

Making the wall merely a ramp you can go up slowly instead of climbing right uphill.

Also I'm interested in the way some programmer could/should be able to learn a whole language in depth before getting his hands dirty on a real project, working in a company.

I think that you lose readability with Java generics, rather than gain it, because of the ugly angle brackets which stand out far too much visually with respect to the other glyphs.

By the way, I think the same about the "enhancements" proposed for Java 7 and 8, regardless of their technical merit.

I had to laugh at the "ultra simple languages" bit in reference to VB or C.

C, at least, is not an ultra simple language. Logo, maybe. But C, no. There's all kinds of nooks and crannies to explore (or get tripped up by).

"The lack of an IDE is a more valid complaint, (especially if you plan to have big teams of mediocre programmers)"

Firstly, I take issue with your assertion that IDEs only help mediocre programmers - a good IDE also makes good programmers even more productive. In fact, I'd go so far as to say that the difference might be greater for more sophisticated developers than more mediocre ones. They tend to take advantage of more features of the IDE than lesser mortals, who often use IDEs as "TextPad with a project view".

Also, the language subsetting concept that you're promoting helps the end user, but doesn't help the IDE developer develop the world-class tools that new languages really need to get popular. An IDE for Scala can't support just a subset of the language, it really has to support the whole thing to be useful. This language complexity massively raises the bar for a good IDE - I totally agree with the OP point that this really has to be a consideration in modern language design. Please, make it easy to parse with a top-down parser, and if at all possible design your toolchain to allow language tools to take advantage of it. The benefits you will reap are huge.

it seems to boil down to the "ide divide" [http://osteele.com/archives/2004/11/ides] - people who feel that the language is what gives them power are usually surfing the leading edge of languages that have matured enough to be used in production, but not enough to have proper ide support. by the time IDEs catch up to scala, they'll likely have moved on.

Maybe so, but I think that the fact that the people designing languages don't think of tool support as a first class language feature makes that gap much larger than it needs to be. There's no reason a powerful language has to be difficult to parse or reason about, it's just that the people designing the language don't place it as highly on their list of priorities as (IMO) they should.

You don't know what Scala code is doing unless you're familiar with the specific libraries it invokes; syntax isn't much of a guide.

You don't know what Ruby code is doing unless you're familiar with the specific libraries it invokes; syntax isn't much of a guide.

You don't know what C++ code is doing unless you're familiar with the specific libraries it invokes; syntax isn't much of a guide.

Heck, you don't know what annotated Java code is doing unless you're familiar with the build tooling that processes the annotations, and the libraries it invokes; syntax isn't much of a guide.

There's a pattern here. I'm not sure it has much to do with the language.

> You don't know what C++ code is doing unless you're familiar with the specific libraries it invokes; syntax isn't much of a guide.

i'd say this about Perl or Ruby, but C++?

Yeah. With operator overloading, automatic type conversions, template expansions and so on it's a fair criticism.

C = A + B;

In C++ what does this line do? The list of things that can effect what this line of code does are: Types of A,B, and C, single arg constructors, explicit conversion operators, operator +, operator =, and copy constructors might come into play. I more than likely have forgotten something from that list.

Macros.

That's essentially the argumentation of Torvalds, why i likes C. In C you know, what the code does. Higher macro magic is rare.

And his point (among other things) is that the lack of namespacing is a win in this respect because it fosters the practice of explicit naming like: `smtp_connect()` rather than `connect()`.

This is a silly excluded middle argument, no?

Left out Lisp macros.

Isn't this to be expected when you add abstraction? I'll argue that not "knowing what [the] code is doing" isn't such a bad thing. In fact, I don't think we can have advances in language design without relaxing the constraint you're alluding to.

Consider an extreme scenario where I've designed a language where the following line of code does what you'd expect:

    ask for an input, assume it is a number, calculate its factorial, and display it

Notice that you have never seen this language before and you have no idea what's going on under the hood to make this happen. However, you know what the code is expected to accomplish.

I think that's a worthwhile tradeoff. It's what all of these languages are trying to do - to allow the programmer to write down their thoughts as close to human language as possible. You only need to care about how things work at the lower level of abstraction when the abstraction has a bug or needs changing. If I can work 80% of my time at the level of abstraction of the "code sample" above (diving into the nasty internals 20% of the time) I'd start using that language immediately.

This is what the Java designers always talk about: languages have a "complexity budget".

You can choose to spend your complexity budget on different features (static typing, generics, closures, manual memory management, auto casts, overloaded operators, ...). Some features can take away complexity (GC). But once you've overdrawn your budget, the language gets creaky to use and everybody starts talking about "subsetting" the language. C.f. C++ - all those features do serve a purpose, and looking at the feature alone, it might be reasonable.

I think Scala has way blown its complexity budget. They've innovated on too many different fronts. I'd been happy to have Java with closures, first class functions, and maybe better generics.

There was a lot of buzz around DSLs at some point of time, and they decided to go all in with implicits, which I think was a bad idea. I guess they were trying to market the language...

I don't understand what you mean when you say they went all-in on DSLs with implicits. I far more often use implicits for integrating Java libraries or papering over clunky apis than for putting together DSLs.

Just take a look at how specs [1] works:

  "Some element" should {
      "behave in some way" in {
          // ...
      }
  }

According to this, the String type has a "should" method and an "in" method, but it really doesn't. Instead, there is an implicit conversion from a string to another type that actually has those methods. They idea here is to make the DSLs more fluent by removing the need to worry about complex type interactions while designing the interface. It lets the users use the pre-existing types they're already familiar with.

[1] http://code.google.com/p/specs/

These DSLs in Scala can be awful. I mean:

    val ivySvn = "ivysvn" % "ivysvn" % "2.1.0" from "http://maven.twttr.com/ivysvn/ivysvn/2.1.0/ivysvn-2.1.0.jar"

Oh, but if you use a different maven convention, just double up the first percent sign:

    val ivySvn = "ivysvn" %% "ivysvn" % "2.1.0" from "http://maven.twttr.com/ivysvn/ivysvn/2.1.0/ivysvn-2.1.0.jar"

Readable, yes; intuitive, no.

Implicits are obviously useful for DSLs, I'm not disputing that. My point is that they're obviously useful for other things as well, and so it doesn't make sense to say their inclusion in the language is a sign that the designers are going "all in" for DSLs - that designation would seem to be reserved for features like macros or parser combinators.

I've been digging into scala recently, and although i'd have to agree, I couldn't say necessarily where this budget was blown. I love that the generics are improved over java (I usually don't need the IDE to check complex generics for me in scala as I do in java), but the overall type system can be daunting. First class function and closures are a easy win for me as well. I also love some of the more "advanced" functional constructs, and lazy evaluation (Stream! I love Stream!).

What gets me is the flexibility of the language. I don't want multiple ways to declare and/or bind the same function, and I don't want multiple ways to declare a lambda (why do I need a language shortcut, and then a shorter shortcut to choose from?). I guess where I'm going with this, is that I think the "complexity budget" would have been manageable, still with most of this feature set, if the language wasn't a DSL factory, and there was a single obvious way to write your program (obvious python bias on my part here).

Oh, and while I'm ranting; XML Literals? ಠ_ಠ

I think basically everyone agrees that XML literals are a bad idea. There is a little more background described here: http://www.quora.com/Will-Scala-ever-remove-XML-literals

"I'd been happy to have Java with closures, first class functions, and maybe better generics."

But that would have made Scala just about equivalent to modern C#. That is, a language balanced for the current masses but still with some modern features. This is not Scala.

Scala is amongst those languages that are blazing the paths for the future masses. There is a friendly subset but it is meant for those that are not afraid to play with the experimental at the expense of the farmiliar. Basically Language hobbyist and startups with no legacy code base. It is not hard, I know this because I taught someone with no experience whatsoever in programming and they found it easier than Java. Someday it will be considered clunky and mundane but that day is not today.

But my complaints on Scala could not have been put better than ominus_prime, too many way to do simple things. Another complaint is how well it interacts with the Java ecosystem - Ive noticed C#/F# is a loot smoother than Java/Scala. Simple Enums in Scala are also something that is too hard (for say if you need to interact with java or just dont need full blown algebraic types). Things are a lot better in this regard since 2.8 but still not there IMO.

Enums are awful in Scala. I usually just use this pattern:

    abstract case class MyEnum(id: Int)
    object MyEnum {
      object Foo extends MyEnum(0)
      object Bar extends MyEnum(1)
      object Baz extends MyEnum(2)
    }

> I'd been happy to have Java with closures, first class functions, and maybe better generics.

What you want is Fantom. It's an incremental improvement over Java with lots of features that doesn't blow the complexity budget IMHO.

http://fantom.org/doc/docIntro/Tour.html

I'm pretty far removed from the Java ecosystem, but that's more or less what Groovy shoots for too, right? Or is Groovy more complex?

The disaffection of ex-Java-devs towards these newer JVM languages is only really surprising if you were expecting any other language to overtake Java for Java developers. I used to think that would come to pass, but I've been convinced otherwise.

By the time Java 8 comes around (don't expect Oracle to drag their heels on releases as prior regimes have), there will be little inherent reason for the typical Java developer to use Scala or Clojure. Only the biggest vendors (Oracle, Microsoft, IBM, et al.) have the firepower and perspective to produce/support whole-solution languages and ecosystems that go mainstream with tooling, training, and general-purpose facilities approachable for the vast middle.

That is, the next Java is Java′.

Of course, none of those typical devs are reading this or TFA. The real question is, which language is most applicable to particular problem domains and usable within particular organizations by atypical devs solving atypical problems? (Added bonus goes to the language that makes bidirectional interop with that vast middle of Java devs a piece of cake, for those atypical devs looking to provide libraries, tools, and products to be used by those Java devs.)

Few points I can recollect (without answering "which language is most applicable...."):

1. Future of Java-the-language is limited by backward compatibility. This imposes a barrier on the scope of innovation in the language.

2. #1 can be helped by the newer JVM languages.

3. Current trend to increasingly adopt Ruby, Python or PHP for turnaround time is the result of a momentum and has to mean something. There is clearly demand for faster turnaround time in some segments.

4. The JVM languages that can reach feature parity with Ruby/Python/PHP ecosystem and tooling will have a better chance of adoption IMO.

5. The next Java is certainly not Java alone. Probably a mix of languages, depending upon which ones have a better VM, tooling, libraries, turnaround time etc.

I program in Scala because it's my pseudocode.

Filter this. Transform that. Collect these. I need an object that does such and such. I'd like some new functions on an object from another library.

For the most part, Scala has an immediate, direct way to do whatever task I need in a type-safe manner. The type safety is important - it really does catch bugs and makes refactoring code a lot easier than in a dynamic language. This matters when the project is larger than 2 developers.

Some others liken Scala to C++. I don't think Scala is as dangerous as C++ because of immutability and garbage collection, but I'll readily admit it's a complex language.

You can write Scala as Java with some nice stuff (done at http://sna-projects.com/kafka/, a distributed pub/sub messaging queue) or you can go full on with the Scala like Norbert (http://sna-projects.com/norbert/), an RPC/cluster management system used for search and the social graph at LinkedIn. I don't think there's a problem with either.

If you need Eclipse-level tools to be productive, you probably shouldn't be trying new programming languages. It takes hundreds or thousands of man-years of effort to replicate those features for a new language. Python and Ruby only really got there recently.

I've really had to play with enough languages in the last few years that IDEs are just more trouble than they are worth to me. The editing isn't great, the tools are sketch, and the design of visual space tends to be poor. And always being on the tool learning-curve gets old...

So I stick with emacs and my primary IDE-esque tools are syntax highlighting and hippie-expand. The only time I break out of my emacs world is to run debuggers.

you should give GUD a try (M-x gdb).

Thats the reason Clojure is so great. Its an simple language. The only complex things (if you get FP) are the macros and in the clojure world macros aren't used often. The Clojure Community learnd from the old lisp makro overuse. Everybody talks about not using makros if you don't really have to. Look at this talk by C.Grand: (not= DSL macros)http://clojure.blip.tv/file/4522250/

If you really want to see what happens with your code just expand it. Its extremly transparent and easy (most IDEs should be able to to this).

In a talk Simon Peyton-Jones once said something like: "Macros are like missile. You don't need them often but if you want to destroy a small village nothing can be better."

> "Macros are like missile. You don't need them often but if you want to destroy a small village nothing can be better."

I'm partial to "Macros are like second amendment rights. You hope you never have to use them, but if you've got no other choice then you will be very glad you have them."

I successfully used Scala as prototyping language for the first time recently. It was nice. I kinda liked it. But, I will have rewrite what I did in it soon because the VM startup cost for scala is too high for what I want to do. The lack of IDE support that the OP complained about didn't really bother me as I am more a vi/kate{+vi mode} kinda guy. That said, I could definitely see how really good IDE support would be nice to have when integrating with large Java libraries.

EDIT: I forgot to mention the one thing that is nice about Java from a research perspective is it is a really easy language to work with. It's syntax is relatively easy to deal with, it doesn't have a preprocessor. Its semantics are straight forward. It has type annotations. There is a lot of code in it so you can test your research on real code bases. This makes Java the best language to conduct research on languages in. Even if it is not the best language to write it.

Nailgun might help you with the startup time: http://martiansoftware.com/nailgun/

(Last changelog entry: 2005.)

JRebel is a must when doing Scala web development. It's not free but it definitely speeds up the development process as it eliminates almost all instances where one would have to restart the app server for code changes.

Actually, you can get a free JRebel license for Scala development: http://sales.zeroturnaround.com/

Thanks, Heiflo, I was going to say the same thing!

One of the key design principles behind Java was to be a cleaner C++: remove all the hippie stuff which make programs too hard to read back, which makes one thing look like it's something else entirely, etc. The result is a boring but reliable language, which gave birth to a wide ecosystem of good quality libraries.

Scala puts back all the hippie stuff into Java. Together with them, it brings back the associated problems which plagued C++. OK, some stuff are done in a less broken way in Scala, largely because they don't try to be a superset of C. But still, we end up with a language hard to read, hard to trust, hard to support for IDEs... not something you want to use for big teams or long-term programming.

And my natural question is: have you actually used Scala for real work, or are you just theorizing?

I had to make some evolutions in an Eclipse plug-in which was written in Scala (the module wasn't intended to manipulate Scala code, it just happened to be the implementation language). I initially had a positive prejudice, as a OCaml enthusiast who used to get paid for developping in OCaml; moreover, as someone who's been exposed to languages with serious abtraction power, I easily get irritated by Java.

The code was a true pain in the ass to work with: integration with OSGi was cumbersome, lots of black magic features were used gratuitously, the possibility to define stuff-which-look-like-methods-yet-aren't-in-the-class made quite a mess of code exploration, and of course tooling support was abysmally bad.

I know the usual rebuttal: it's the bad developer's fault, not the language's. I don't buy this, a "good" language for which you can't find decent coders isn't any good.

- Scala-like languages tend to attract guys who will mess with all the quirky features of the language;

- In real life, developers are a given and the language is the adjustment variable, not the other way around. Most people don't have the luxury of only working with rockstars, and most of the cornerstones of the Internet haven't been developed by rockstars anyway.

- The average developer will do an average job of writing maintainable Java, and an awful job of writing acceptable Scala, if only because there's no widespread consensus ov what it means for Scala code to be acceptable. Java is dull and uneventful, and I'd rather take over a syntactically dull and uneventful code base than an "interesting" one.

- I've heared several times that "you can use Scala as a Java with closures". It's a variant of the argument that there's a 5 times smaller language hidden in C++ that's beautiful: nobody ever agreed over which 20% should be selected.

The two languages which gave me the impression of trying to support every feature ever invented, and of making it very difficult for people to acquire and share a notion of standard, maintainable code are C++ and Scala. An iconic counter-example would be Python.

More anecdotically, I spent a couple of days studying how Lift works, for personal curiosity. I found the code hard to navigate into. due to too many non-orthogonal ways of composing code together. A good tooling support would have helped, but if it doesn't exist after all those years, I might mean that there's something wrong with the language: it's not as if the need wasn't clearly perceived.

In this same thread Martin Odersky announces he is working in a new version of the Scala plugin for Eclipse, to be released with Scala 2.9. A statically typed language really shines within an IDE.

I get this sentiment a lot ...

     Secondly, it is the ultimate language to date for writing libraries 
     and  embedded DSL’s with static types. ... These abilities go 
     way beyond any prior statically typed language.

The author conveniently left out dynamic languages; but regardless of that, I don't like Scala because:

1) you can't override the method dispatching mechanism to make use of those types in interesting ways (e.g. multimethods are possible with DLR in .NET)

2) You can't get the syntax-tree of a block of code, then recompile it to something else (e.g. LINQ - and F# supports this as a general construct). There was a compiler-plugin at some point, but it's deprecated and incomplete and full of bugs

3) All OOP languages should have optional dynamic typing; OOP just doesn't blend with static types, being one reason why Haskell is not OOP (while providing other similar tools for polymorphism)

What you do get in Scala is a flexible syntax, that can lead to awkward behavior (I often find myself asking whether some line needs a semicolon). And the type-system, while better than the one in Java, is also too complex, too verbose and definitely not elegant.

You know what would be elegant? Being able to test if some object is an instance of List[String] (and yes, I know about manifests and implicit parameters: it's still an ugly hack that doesn't solve all problems).

One could say that Scala is a pragmatic language, but so is C++; and C++ has no replacement, while Scala does have many.

I've been doing my scala stuff using sbt and vim. So far, it's working out quite nicely. I haven't played with sbt plugins for various IDEs (I'm too used to vim), I would assume it should still work.

I did some research (but not actual programming) to compare Java to Scala, specifically in the area of generics. If you are hoping for some language to make static typing palatable again, Scala is not it: it's so good at static typing that it can statically type anything, but it takes intricate details and more knowledge of its type system than most would want to care about to get it right. In general, I felt there were too many niggly little details in the syntax to be a nice language to use.

Of course, I'm used to Python, so I may well be spoiled.

I agree with the characterization of Scala's generics as complicated, but I disagree that that ends up making its static typing unpleasant or day-to-day programming difficult. Type inference is a big win, and other language features add enough that I find myself not reaching for generics for my own code anywhere near as often as I did with Java. Aside from the occasional container type missing from the standard API, I just pretty much don't use generics in Scala except as a consumer, and I don't miss them.

I actually do programming in Scala and I'd have to disagree. While there's the freedom to build a "type castle", you don't have to. For the most part Scala gets things right.

When you do need/want to get fancy with types, you get some of the best tools out there to do it.

Scala seems to be the academic kitchen sink of programming languages.

This exactly summarizes why I stopped using Scala. All I wanted was a cleaned up Java, but it's much more and riskier than that.

Maybe you'd like Mirah?

http://www.mirah.org/

I never understood why anyone would want a hybrid of OO and functional programming.

I see OO as basically: Encapsulation, Inheritance and Polymorphism.

Functional programming provides better facilities for encapsulation and polymorphism. Interface inheritance is pretty much a subset of implicit traits or type-classes. I don't think implementation inheritance is ever the right thing.

So if you have the full power of FP (including existential types and type-classes), what is it in OO that you would yearn for?

Groovy is good alternative JVM language if you just want some improvements on Java without a completely new language to learn. It addresses many of the small frustrations I have with Java (lack of list and map literals, lack of anonymous functions) without introducing many new paradigms or complexities. I'd recommend it to anyone wanting a quick prototyping language on the JVM.

Scala subsetting (as mentioned in the comments) sounds like a promising avenue. Especially if it were to result in better performing tools. Unfortunately for all the great features scala does have, Java has a few killer features that it lacks i.e. fast (instantaneous) and reliable compilation and debugging.

Reply on news.ycombinator.com