Show HN: Motif – Scala-like pattern matching with Java 8 | Hacker News Reader

Show HN: Motif – Scala-like pattern matching with Java 8

github.com

51 pointsjohnlcox34 comments

Open articleView on HN

The Cambrian explosion of java 8 functional libraries like this is fun to watch. Others I've come across:

- http://javaslang.com/

- https://github.com/aol/cyclops

- http://totallylazy.com/

- https://github.com/poetix/octarine

- (good ol') http://www.functionaljava.org/

The syntax for these libraries is always a little wonky though. If only these libraries also had some sort of syntax preprocessor to make it more palatable. You could call it, I don't know, Scala, or something.

Exactly. I want to like these things, but they're just way to painful on the eyes.

whoosh?

http://www.scala-lang.org/

He said he wanted something that's not painful on the eyes.

I hardly see how Scala is more painful on the eyes than good old Java.

Random example from a popular project:

     def bimap[C, X[_, _], D](g: Bijection[C, D])(implicit F: Bifunctor[X], evF: F[B] =:= Id[B], evG: G[A] =:= Id[A]): Bijection[X[A, C], X[B, D]] =
        bijection(
          F.bimap(_)(to(_), g.to(_)): Id[X[B, D]]
        , F.bimap(_)(from(_), g.from(_)): Id[X[A, C]]
        )

Hi! I see that you are new to Hacker News. Welcome.

It's unlikely that anyone cares about your opinion re: Scala's syntax and cherry-picked abuses thereof. I certainly do not. My comment was an attempt to point out the sarcasm in the root comment.

Have a nice day.

> I certainly do not.

What a coincidence, I do not care about your opinion either!

And I could cherry pick something just as hideous out of Java. Congrats.

Nobody disputed that but that's off topic. We're discussing the fact that Scala can be pretty ugly too.

Can you show me the equivalent java code? I would guess it's 200 lines of java or more. I would much rather spend the time to learn what that's doing than reading 200 lines of java.

Hence Scala...

Hard to read in comparison to Scala. It's impressive that it was accomplished but some things just aren't meant to happen in Java 8.

If you've ever endured the compilation time of a Scala codebase that has more than, say, 2000+ classes, then you might think that some things just aren't meant to happen in Scala either.

I used Scala for 4+ years at a company that spent a non-trivial amount of time and resources speeding up its Scala builds (way more than 2000+ classes), sometimes doing sadly controversial things[1] to achieve that speedup. It was a real quality-of-life issue, and one of the reasons that compelled me to leave.

I'm using Java 8 now, and yes it has its warts and is nowhere near Scala wrt functional goodness, but I'm glad to be compiling and iterating a lot more quickly with the ability to use even a subset of the Scala stuff I missed.

[1] https://groups.google.com/d/msg/pants-devel/mcdllXQsYNg/96ZG...

If you stick to the same kind of subset in Scala, then compilation isn't that much worse than Java, IME. It's only when you're using a lot of implicits and/or a lot of type inference that compile times go through the roof.

(and FWIW I'd say that 2000+ classes - that's what, 600KLOC? - implies a project that should be split up with clear boundaries (i.e. separate release cycles, documented public APIs between them), if only for developer comprehensibility. How big is a Java project that provides the same amount of functionality, and how long does it take to compile? Crikey, 2000 classes...)

Some companies, like Twitter and Facebook, are doing this development pattern of mono repo... So even if there are architectural lines to be found, they are eschewing that in favor of keeping versions and APIs closer together.

I can't say I understand that strategy completely but I have also never worked on machinery the size and complexity of Facebook or Twitter.

Still at that company, and still enduring long compile times.

Way more than 2000 classes? How many LOC is that?

How much time did it take to compile, and how much time does the same 2000+ classes take to compile in Java?

As a point of reference, compiling ~200 Scala classes at a time would take ~2.5 minutes on average. At my last job, compiling >2000 Java classes took about 10-15 seconds with an ant script doing a top-down build. Unfortunately I don't remember LOC, but I don't remember the average Scala file being overly large at my Scala job, or the average Java file being unusually short at my Java job.

Question - I don't know much about Scala - how often does one need to do a full build? I am talking regular development cycle, ie make a change, save/compile, check if it works etc. I mean typically a compiled language will have some 'incremental compile' to compile just the file you edited. Does that work with Scala?

> Question - I don't know much about Scala - how often does one need to do a full build?

Not often at all.

> I am talking regular development cycle, ie make a change, save/compile, check if it works etc.

sbt[1] supports incremental compilation and projects such as Play[2] leverage this to support "on-the-fly" recompilation while the container remains running.

1 - http://www.scala-sbt.org/0.13/docs/Understanding-Recompilati...

2 - https://www.playframework.com

> As a point of reference, compiling ~200 Scala classes at a time would take ~2.5 minutes on average.

Odd... I just did a full compile of a Scala project I work on which has:

- 259 source files

- Contains dozens of macros[1]

- Results in 2097 classes (includes companion objects and anonymous types)

And it took 75 seconds on my laptop. As I mentioned in another reply, doing a full recompile is rarely needed but was done now to provide quantified metrics.

1 - http://docs.scala-lang.org/overviews/macros/overview.html

In Java, I compile around 250 classes in about 6 seconds on my laptop. It results in around 250 class files being generated because no filea for anonymous inner classes are created from the lambdas in the code. I don't need a full recompile either. An incremental compile in Java is measured in milliseconds. IntelliJ is usable to its fullest instead of being a fancy editor that just autocompletes and jumps to symbol definitions.

You may not be using all of scala's language features. I'm using java's features to the fullest. I compile in 6 seconds. You compile in 75 seconds. You may still prefer that trade off. After working with thousands of scala files (you have 259 source files? That's cute! I wish I could pinch your cheeks!) where a 75 second compile sounds like a fail-fast error, I think I will live happier with what Java 8 has, limited and warted though it may be.

You have no idea my background, how I use Scala, and assume that the small project I reference is the only project I _could_ reference. It was chosen due to being near the size of the situation you describe as "compiling ~200 Scala classes at a time would take ~2.5 minutes."

Being a condescending asshole doesn't help you communicate with others. Frankly, it strengthens my opinion that you don't know how to organize a medium-sized project, since being able to do so obviates requiring to compile 2000+ files for each build _in any language_.

> Unfortunately I don't remember LOC, but I don't remember the average Scala file being overly large at my Scala job, or the average Java file being unusually short at my Java job.

LOC isn't what should count though - there are cases where 1 line of Scala offers functionality that would take 10 or more lines of Java to implement.

Scala's compilation time is not an excuse to write ugly hard to maintain code in Java 8. If you don't like Scala compilation time then write Java 8 but keep it idiomatic and easy to read don't drag parts of Scala into it that don't fit.

Funny, I happened to code something just today very similar, but a lot lighter (just two classes, Match for consuming only, and MatchMap for mapping the optional. No when/get, just a when taking two parameters : the predicate and the mapping function.

  public class MatchMap<T, U> {

    private final Optional<T> opt;
    private final U defaultValue;
    private final AtomicReference<U> result = new AtomicReference<U>();

    private MatchMap(Optional<T> pOpt, U pDefault) {
        super();
        opt = pOpt;
        defaultValue = pDefault;
    }

    public MatchMap<T, U> whenMatch(Predicate<T> pPred, Function<T, U> pMapFn) {
        if (result.get() != null) {
            return this;
        }
        Optional<T> lFiltered = opt.filter(pPred);
        if (lFiltered.isPresent()) {
            result.set(pMapFn.apply(lFiltered.get()));
        }
        return this;
    }

    public MatchMap<T, U> whenEquals(T pExact, Function<T, U> pMapFn) {
        return whenMatch(t -> pExact.equals(t), pMapFn);
    }

    public U resolve() {
        return Optional.ofNullable(result.get())
          .orElse(defaultValue);
    }

    public static <T, U> MatchMap<T, U> of (Optional<T> pOpt, U pDefault) {
        return new MatchMap<T, U>(pOpt, pDefault);
    }

  }

The class Match is very similar, but takes consumer instead of Function and has some logic to allow several matches to be executed (if several predicates are ok), or only allow the first match to be executed.

I'd love to see how code using your version compares to code using the OP's version. Got any similar examples?

I use it especially for enums so it's like:

  MatchMap.of(someOptional, defaultValue)
    .whenEquals(MyEnum.VALUEX, v -> someFn(p1, p2))
    .whenEquals(MyEnum.VALUEY, v -> otherFn(p1, p2))
    .resolve();

I haven't tried it with enums, but it should still be possible with Motif. I'll have to add some tests.

  match(enumValOpt)
      .when(some(eq(MyEnum.VALUEX))).then(() -> someFn(p1, p2))
      .when(some(eq(MyEnum.VALUEY))).then(() -> otherFn(p1, p2))
      .orElse(defaultValue)
      .doMatch();

It's a little longer than your example since there is support for more than just Optional, but it's pretty similar.

You can simplify your code a little bit and add some wildcards :)

https://gist.github.com/anonymous/7d45e0f746b46684ec45

Indeed, thanks!

https://en.wikipedia.org/wiki/Motif_(software)

The javaslang library has a similar construct:

http://javaslang.com/javadoc/1.2.0/javaslang/control/Match.h...

Reply on news.ycombinator.com