Better Java – Resources for Writing Modern Java

by networked • 11 years ago

202 188 11 years ago

zatkin• 11 years ago

Streams are questionable in performance vs non-stream approaches. I haven't found an article that details performance differences, but I will say that I personally tried doing things using streams and doing the same things using non-streams and have had mixed results, one being faster than the other in some cases.

mabbo6• 11 years ago

The benefits of streams isn't performance improvements for the CPU, but performance improvements for the developer.

chii5• 11 years ago

i have colleges that claim streams don't necesserily make the code more readable. Imagine a triple nested for-loop - making it a triple nested streams (e.g., where your map method is yet another stream), doesn't seem to really make it any more readable.

The only reason to use streams, i think is the possibility to automatically parrelize, and secondarily to make it more readable.

kodablah• 11 years ago

"If you're using Java 8, you can use the excellent new Optional type. If a value may or may not be present, wrap it in an Optional class like this"

While I agree with most things, overuse of Optional like this is an antipattern IMO. Having done a lot of Java 8 development, I find Optional is best for return types, but otherwise forcing callers to wrap values in Optional is unnecessary as opposed to something like Scala. Those that work on the JDK feel the same [1][2] and I think [2] is even overuse with the prevalence of static analysis tools and @Nullable.

Although a tad dated, I find [3] to be a good guide to modern Java all around.

1 - http://stackoverflow.com/questions/26327957/should-java-8-ge... 2 - http://blog.joda.org/2015/08/java-se-8-optional-pragmatic-ap... 3 - http://blog.paralleluniverse.co/2014/05/01/modern-java/

yrro• 11 years ago

I'm curious, which aspects of An Opinionated Guide to Modern Java have dated?

kilink5• 11 years ago

I'm also not a fan of passing Optionals into methods. Most of the time it's easier and cleaner just to write an additional method without that optional parameter.

Peaker4• 11 years ago

It can grow cartesian though.

I think the best compromise is to declare a new data type for the argument, such as:

  data ExecutionDirectory = ExecuteInCWD | ExecuteInDir FilePath

And take an ExecutionDirectory instead of a `Maybe FilePath`.

Of course, you need light-weight data declarations for that.

harryh5• 11 years ago

Keep in mind that Tony Hoare, the inventor of the null pointer, calls this invention his billion dollar mistake. I generally think that he is right and use of null should almost always be avoided.

agumonkey4• 11 years ago

In other paradigms (Lisps) null is a lot less than a billion dollar mistake, it seems it goes better with map/filter/fmap patterns. In imperative languages, data is tied to control, and null means segfault.

userbinator2• 11 years ago

Despite what he says, I don't think Hoare actually invented the "null pointer", as address 0 for "not exists" or "end of list" was probably a common convention ever since linked structures existed and independently discovered by many others. (The other convention would be the all-bits-1 value, but I'd assume that testing for 0 was easier and thus more widely adopted. Also, "0 = nothing" makes great sense.)

IMHO Optional just feels like another typical Java dogmatic overabstractionism. "Null pointer exception? So what? Just catch it." Then again, I mostly use C/Asm.

asveikau2• 11 years ago

By all means, have more clumsy, less straightforward ways to represent the lack of a value in your code, because Tony Hoare says so.

tl• 11 years ago

Watching the evolution of languages / APIs makes me believe that Optional over-use is a necessary step between "our language has no formal Optional therefore crash" and "our language uses null in a sane manner".

C#'s prior behavior gives me hope for languages I'm currently paid to care about like Java and Swift on this issue.

rbolkey• 11 years ago

Have to agree here. The article's use of Optional is well intentioned, but shows use of the construct in a way that it was not intended. It should be used minimally, and only as a return value to signify the absence of a value. When you notice that Optional is not serializable, this starts to sink in.

Here are some more thorough treatments: http://stackoverflow.com/a/23464794 http://blog.jhades.org/java-8-how-to-use-optional/

pron2• 11 years ago

When I get some time I'll update the OGMJ. I think the choice of tools mentioned is still very much up-to-date, it's just that some of the APIs used in the examples have been upgraded.

mahmoudimus2• 11 years ago

I really like OGMJ, it's definitely a great resource and I recommend it to many new Java developers; however, I disagree re: tools mentioned.

After trying dropwizard for a few projects, the superiority of the spring-boot and its various ecosystems just blew me away.

Also, while I appreciated Gradle, Maven is just much more supported through the entire eco-system. As a new comer to Java from C++/Python, it made a big difference.

izzle492• 11 years ago

Please don't use Java 8's default methods in interface's to do multiple inheritance, and code reuse. Prefering this over 'Util' classes is bad as it is inheritance over composition. The default methods were designed to facilitate easier interface migration

gutnor• 11 years ago

That reminds me of "constant interface" anti-pattern - abusing a feature because it brings a small technical benefit.

If you are unhappy with candy-shop utils class and are ready to split them into smaller entities, go the whole way there. Create proper small classes, and inject them using your chose DI technology.

Anyway, kudos to the author, that was a risky article to put together and there was no way to write it without controversial recommendations. That's a good start for what a new java developer need to look for.

lmm10• 11 years ago

Java was designed for TV set-top boxes, but that doesn't mean it's bad to use it on a server. If using a default method in an unintended way leads to clearer, more maintainable code, do it.

izzle492• 11 years ago

how would it lead to more clearer code? You would have many interfaces referencing an instance variable, each with different semantics for that variable. Interfaces cant have variables so they should not be used as traits.

GhotiFish• 11 years ago

Inheritance over composition. I know what these words mean, but I don't see your meaning.

Also, I don't think that's what default methods are for. The new StreamUtil libraries uses default interfaces. It's not migrating anyone, but the default methods describe intrinsic properties about the classes that inherit the interface. A predicate can be in sequence of another predicate, for example.

sigzero• 11 years ago

Guava is on github now. Might want to change the link.

zinxq• 11 years ago

I'm in agreement with the first few comments already - at least some parts of this advice is dubious.

For one - some performance examples (specifically, integer array iteration) is 15 times slower with streams.

Angelika Langer's analysis:

https://jaxenter.com/java-performance-tutorial-how-fast-are-...

coldtea• 11 years ago

For one - some performance examples (specifically, integer array iteration) is 15 times slower with streams.

That's a very misleading example, and even the linked post itself explains why it is so (e.g. not much is done with the integers in the first place).

speakeron• 11 years ago

I would question these results. I ran the same code on a more modern processor (i7-3930K) for 1000000000 iterations and got these averages over 10 loops after a warm-up:

Array: 453 ms Stream: 887 ms Parallel stream: 120 ms

The array loop code is about twice as fast as the sequential stream (parallel stream blows both of them away). The array loop advantage will start to go away once you increase the complexities of what you're doing in the loop and, of course, the stream code is much more readable.

(Edit: I also tested IntStream().of() instead Array().stream() for the sequential stream test, the results were identical in both cases.)

Randgalt• 11 years ago

Nice compendium. I don't agree with everything, but I doubt you can get an Java devs to agree on everything.

fizwhiz6• 11 years ago

> Good alternatives to using Spring is Google and Square's Dagger library or Google's Guice. They don't use Spring's XML configuration file format, and instead they put the injection logic in annotations and in code

Spring has had code-based configurations since 2009[1]; seems a little disingenuous to recommend Guice/Dagger due to this non-existent limitation.

[1] http://spring.io/blog/2009/12/22/configuration-simplificatio...

6502nerdface• 11 years ago

Not sure how you get that reading... the author makes no claim that Spring is limited to XML configuration. In fact, in the paragraph just before the one you quote:

It [the Spring framework] has a either code-based wiring or XML configuration-based wiring.

fizwhiz6• 11 years ago

I read the paragraphs a little like "Spring has code-based config and xml-based config (careful with the xml-based config!) Guice and Dagger don't have xml-based config, so use Guice or Dagger". My gripe is with the hand-wavy treatment of why Guice/Dagger are better alternatives just because they don't have xml-based configs even though the author knows that people can use Spring exclusively without xml-based configs (this is why i said it was disingenuous)

lmm10• 11 years ago

Spring has a million and one ways to do anything; that's the problem with it.

Spring with a very strict, carefully enforced style guide is probably as good as Dagger or Guice. But there is a cost to maintaining and enforcing a style guide. Better to just use something that doesn't have the bad options.

jacques_chester5• 11 years ago

That's what Spring Boot is for.

Everything is autoconfigured and set to sensible defaults. All the dependencies in the starters are levelled and guaranteed to work together.

Spring Boot on Gradle is the closest thing Java has to adding lines to a Gemfile in a Ruby system. Especially when you add other Spring automagical libraries (Spring Data REST, Spring Cloud...).

ryanseys• 11 years ago

The only way it seems you can be effective with Java is with an IDE to do most of the heavy lifting. What would be better is if Java shipped with more focused command-line tools and better defaults that allow me to productive immediately without the need to learn something as hefty as IntelliJ / Eclipse. This could lessen the barrier of entry to Java and widen community adoption.

yogthos2• 11 years ago

Or you could just use modern JVM languages like Clojure.

codarama• 11 years ago

Lisp? Lisp syntax makes less sense than lambda and stream which have ruined Java.

EdwardDiego6• 11 years ago

I like Clojure, but anyway to make it start faster in a scripting context? I wrote the same script in Clojure and Kotlin, and the Kotlin one loads so much faster.

EdwardDiego6• 11 years ago

What would you use instead of an IDE? Learning Intellij is a lot easier than learning Vim or Emacs.

lmm10• 11 years ago

The command-line tools are fine. And Java is the most adopted language in existence.

The problem with writing Java without an IDE isn't one of tooling, it's that the language is so very verbose. They're working on that with lambdas and generics inference, but there's a limit to what you can do and remain backwards compatible.

tootie• 11 years ago

I've been using spring loaded instead of jrebel and it works pretty well and is free.

One thing I'm surprised he didn't mention is exception handling. Namely the growing acceptance that checked exceptions are dumb. I make a point that except for a few rare circumstanes, my code should never have a throws clause. I always created subclasses of RuntimeException (or Spring's NestedRuntimeException) and use them exclusively. Any libs that throw checked exceptions get either handled or rethrown with a custom RuntimeException.

saveliyb• 11 years ago

Nice! Seeing "enterprise" java code anyways makes my blood boil, and I keep reminding myself that it doesn't have to be that ugly.

incepted• 11 years ago

If you really want to describe modern Java, you should probably mention RxJava and TestNG.

fizwhiz6• 11 years ago

TestNG may have been considered a "modern" alternative years ago, but today jUnit4 has caught up with most (if not all) of testNG's features.

Cthulhu_• 11 years ago

Re the first bit about 'struct' style, even better (IMO) is to use Lombok; throw an @Data annotation on your class and all you need to do is put your properties in there (private final if need be). Nobody should be writing those boilerplate constructors, getters and setters. If it's generated by your IDE, use Lombok. The best code is no code.

chii5• 11 years ago

then you suddenly force anyone who uses your code to depend on lombok. I tried to find a compile time only mechanism for lombok - not sure i found it, but it ought to exist!

radq• 11 years ago

Lombok does rewrite bytecode at compile time only. Your dependencies will not have to depend on Lombok.

monksy• 11 years ago

Thats one of the things that'll happen to you if you start writting groovy code. The whole setter/getter boilerplate thing gets really annoying.

Another groovy-ism.. the whole deal with having to copy properties. (In a groovy class they're threated like a map)

vorg4• 11 years ago

properties. (In a groovy class they're threated like a map

Except in hard to remember special cases like .class which gives the class "LinkedHapMap" of the map object instead of the value at key "class". After working with Groovy for a short while you'll realize you're really working around it.

Babylonian2• 11 years ago

Lombok is a crime. Have you seen what happens when someone does a minor reformatting of your code and moves one field declaration above another? You get a nasty little surprise because the two fields of the same type have now reordered in your Lombok generated constructor. When the app runs nobody can figure out why foo is set to bar and bar is set to foo. There's also the issue of Lombok vals which don't conform to the language spec leading to other fun. And it's almost always incompatible with new jdks without code changes to the lombok libs due to the whacky way it relies on implementation specific compiler details.

dguaraglia• 11 years ago

Wow, I didn't know this could happen. I've only recently started using Lombok for some side-project related stuff and this heads up comes at a really good time. Thanks!

hiram112• 11 years ago

I've seen numerous posts lamenting the use of typical Java Beans.

The problem with the struct approach is that numerous tooling / libraries expects beans with typical getter/setters - Jackson Json, Spring binding, etc. Yes it's possible with configurations and/or annotations to get around this, but IMO, saving a few lines of code in exchange for non-default behavior isn't a valid trade-off.

Adding a dependency on Lombock is even worse.

Honestly, I can spot a Bean class within 1 second of opening it up. Unlike Scala or other languages, Java programmers typically use one class per file, and beans rarely get in the way as they're typically contained in some 'model' package anyway.

vbezhenar2• 11 years ago

Another problem with struct approach is inconsistency.

Sometimes you want a computed property and you use non-trivial getter. Sometimes you want to do some work in setter. And then client's code will look terrible: point.x * point.x + point.getY() * point.getY(). There are other problems: binary incompatibility; non-trivial refactoring when I need to introduce a getter or setter.

Java really need lightweight property syntax which would solve all those problems. I don't understand why it isn't introduced yet.

logn• 11 years ago

Sometimes you want a computed property and you use non-trivial getter

Then pre-compute it, make it immutable, and expose the variable. And if that ends up being too hard it's probably because you need a builder and/or to group your properties into smaller classes.

jimueller• 11 years ago

Going from C# to Java, one of the biggest things I miss is the Property syntax.

zmmmmm• 11 years ago

The lack of simple property syntax is probably the main reason I keep using Groovy now that Java 8 has arrived. Yes, state is evil. But there are still plenty of situations where you are just schlepping data around and generating thousands of lines of boilerplate to achieve simple data transfer is such a pointless and ugly hack.

jay-anderson• 11 years ago

Projects exist to to make a light weight property syntax. The most prominent might be @Getter and @Setter from lombok (https://projectlombok.org/), but there are others which have similar goals (e.g. http://immutables.github.io/, https://github.com/google/auto).

That said, I agree that java _really_ needs something like this built in.

PascalW• 11 years ago

I'm surprised I haven't seen anybody mention the "public final fields" approach yet. I use this myself for immutable objects and I quite like it. Granted, you still need to implement equals and hashcode, but these can be generated easily and verified with tools like EqualsVerifier [2].

[1]

  public class MyImmutable {
    public final String field1;
    public final int field2;

    public MyImmutable(String field1, int field2) {
        // assign
    }

     // equals, hashcode, toString
  }

[2] https://github.com/jqno/equalsverifier

jordanpg• 11 years ago

Also, whatever stylistic issues are raised by JavaBeans are largely obviated by modern IDEs.

EdwardDiego6• 11 years ago

Spring binding

Not sure what in particular you're referring to, but Spring supports constructor based dependency injection, and I tend to insist upon it vehemently - getter/setter injection means that you can instantiate your object in an invalid state.

lmm10• 11 years ago

The problem comes when you spot a Bean class and then it isn't. 99.9% of "getX" methods simply read the field. But if you assume that every "getX" method just reads the field, you'll be caught out by the rare one that doesn't - I've seen a production bug happen this way.

mahmoudimus2• 11 years ago

Lombock is one of the most valuable enhancements to Java. Without it, it would make Java development not as fun.

iwwr• 11 years ago

Lombok is also a non-standard dependency and there are costs with having to rely on it.

trbvm23• 11 years ago

Dependency Injection (frameworks) -1

Use constructors. This makes perfectly testable classes and is vastly simpler than encouraging more XML as code. Any proper dependency injection framework should work quite well with regular constructors.

fizwhiz6• 11 years ago

Your comment is unclear. Are you encouraging people to NOT use DI frameworks, or using them only with constructor-based injection?

trbvm23• 11 years ago

The latter. DI frameworks are a convenience item, they should probably not have a big effect on the way the code is written. You get most of the benefits of DI just from not new'ing up things within your classes.

The main advantage of keeping the constructors around is that you retain more flexibility in interacting with that bit of code.

vbezhenar2• 11 years ago

Constructors have their problems. First: constructor calling is not readable because Java misses named parameters. It's fine when there are 2-3 dependencies, but not more. Second: circular dependency is not possible. So I prefer to use setter dependencies, as they don't have those problems.

kilink5• 11 years ago

First: constructor calling is not readable because Java misses named parameters. It's fine when there are 2-3 dependencies, but not more.

If you have more than 2-3 dependencies, it might be time to refactor things, as your class may be doing too many things.

Second: circular dependency is not possible.

Circular dependencies are possible with constructor injection when using a DI framework (e.g., using a Provider in Guice [1], or the equivalent in Dagger).

[1] https://github.com/google/guice/wiki/CyclicDependencies

trbvm23• 11 years ago

I agree on the first. I'm only advocating for keeping the constructor as the DI entry point. A DI framework can then be used for the convenience they provide in wiring things up at main.

Regarding the second: I think circular dependencies should be avoided whenever possible so I view the difficulty constructors create there as advantage rather than a problem.

https://en.wikipedia.org/wiki/Circular_dependency#Problems_o...

hibikir• 11 years ago

Constructor calling is readable either with an IDE, or using distinctive types. If the constructor is unclear, it's a smell.

I'd also say that making circular dependencies not possible is a feature, not a bug. They go against everything we've learned about computing in the last 20 years. Entire languages do not have variables, or at least heavily discourage them, and the gains outweigh the costs.

Setter dependencies have other major disadvantages, like how easy they make creating partially initialized objects: Refactor something in one place, adding a parameter, and good luck making sure that the change is made everywhere else. A constructor guarantees fully initialized objects, and that's a great things.

Java pays for the mistakes of the early 2000s: The Java Bean pattern, and all the infrastructure around it, makes Java code be full of default constructors that force us into mutable objects and nullable fields.

If there is a defense for your argument for setter dependencies, is that we cannot count on any of the nice things that come with, say a Scala case class, because so much Java code out there is built with standards from a less enlightened era that using what would be better standards in any more modern language leads to inconsistencies, due to all the ancient code using Spring, Hibernate, Guava and EJB code that is hanging out there, and that would lead to inconsistencies for those trying to make Java actually use some of its new functional programming inspired features.

usrusr• 11 years ago

I'd rather inject some lazy somethingsomething-provider reference wrapper to break the occasional circular dependency than work with a bunch of nonfinal references that supposedly never change, because DI, but are otherwise indistinguishable from true working state.

lmm10• 11 years ago

I think people use DI just because Java constructor syntax is so incredibly verbose:

    public class MyClass {
      @Inject private String foo;
    }

versus

    public class MyClass {
      private String foo;

      @Inject public MyClass(String foo) {
        this.foo = foo;
      }
    }

cam-4• 11 years ago

Dependency injection is brilliant for testing. Prior to DI you often has to have two constructors, one for prod, one for testing. DI took that away.

kilink5• 11 years ago

I'm surprised he didn't mention AutoValue [1] or Lombok (using @Data and @Builder) in his section about data objects and builders. I prefer AutoValue, but both approaches cut down on a lot of the boilerplate (and they also generate sensible equals(), hashCode(), toString(), etc).

It's also kind of a nitpick, but I disagree with his preference for Guava's Maps.newHashMap() vs new HashMap<>(). I'm pretty sure the only reason those static methods exist is because pre-Java 7 didn't have the diamond operator.

[1] https://github.com/google/auto/tree/master/value

adpreese• 11 years ago

He mentions Lombok in the tools section.

And Guava's collection creation methods are part of an intense desire to not use the new key word anywhere in application code.

kilink5• 11 years ago

I realize he mentions Lombok, but he specifically bemoans the extra boilerplate with writing builders and data classes, but does not mention the @Data or @Builder annotations...

kilink5• 11 years ago

Also, if you look at the Javadocs for some of the creation methods, they explicitly state that they are for Java 6 and earlier, and that modern code should just use the constructor with the diamond syntax.

http://docs.guava-libraries.googlecode.com/git/javadoc/com/g...

Pyxl101• 11 years ago

The other replies kind of hit on this, but I'll expand on it by saying:

The collection creation methods existed because Java could perform type inference on methods but could not on constructors:

  // Java 1.5
  Map<String,Integer> map = new HashMap<String,Integer>();
  // Guava
  Map<String,Integer> map = Maps.newHashMap();
  // Java 1.8 diamond notation
  Map<String,Integer> map = new HashMap<>();

mdasen• 11 years ago

Immutables (http://immutables.org/) is another option which creates nice, immutable value classes. It's a bit more succinct than AutoValue, but it does still require annotation processing in your IDE.

CHY872• 11 years ago

Lombok's a bit sketchy as parts of it rely on some undocumented, non-standard parts of the Sun Java compiler - you have to be really into code generation to want to use it.

Other libraries like Immutables are safer for this purpose.

jonalmeida• 11 years ago

Wouldn't the best way to create an empty map be to use `Collections.emptyMap()`?

carsongross• 11 years ago

There is some good stuff in here, and a lot of bad.

SparkJava (mentioned) is the great hope for sane java web development:

http://sparkjava.com/

Having to convert out to streams for data structure manipulation with lambdas is insane. It would have been very healthy for the core java team to pull in someone with a lot of ruby experience so they could get that API right.

Too bad: the JVM and associated tools are fantastic.

asdf1234• 11 years ago

Having to convert out to streams for data structure manipulation with lambdas is insane

I don't see why. It's explicit and allows you to chain higher order methods without building up a ton of intermediate data structures and convert to any data structure you like very efficiently.

It's still not efficient as imperative approaches but it's close enough and far better than most collection libraries.

potatosareok2• 11 years ago

Have you seen Goldman Sachs collection library? I've been using it for my Java work for last year and found it to work well. Put's a lot of useful methods onto the Collection objects themselves.

joesmo• 11 years ago

'Formatting is so much less important than most programmers make it out to be. Does consistency show that you care about your craft and does it help others read? Absolutely. But let's not waste a day adding spaces to if blocks so that it "matches".'

I do agree that it's less important, but it's so easy to do that there is no excuse for not doing it or giving out such rubbish advice. Anyone that wastes a day on formatting is doing it wrong. The problem with advice like this is that people take it to the extreme and ignore formatting altogether. That leads to someone being frustrated and formatting all the files automatically (as should be done anyway) which then leads to possible conflict and crappy diffs. Instead, you could have had professional programmers work on these. In an industry that's touting how code readability is so important (and it is), this advice is toxic. We're finally getting people to realize that readability is the most important part for most code because it is. Honestly, formatting is important in this inconsequential reply, in essays, in everything we write except short chat messages. Imsurenoonewouldliketoreadsentenceslikethis just like no one wants to read obfuscated, unformatted, crappy code. So yeah, I disagree strongly because formatting is important, get one style, get proper tools to automate it, and then you don't have to spend a day putting in spaces (which is ridiculous).

nobleach10• 11 years ago

After working at two companies filled with Java devs, I don't think any of them will read this article. I can't speak for all Java devs obviously, but it seems unless the dev is a polyglot, they are just fine living in their Spring/SVN/J2EE/Java5 world. They have no idea what's going on in the software development world. I get so excited whenever I hear any of them mention Clojure or even Scala! They just live with this head in the sand mentality. They hate the front end, they think, "these JavaScript developers will go away someday and we can finally go back to writing JSPs". Currently we've updated our JVMs to 8, and that's a big deal, but we've told our devs NOT to use any features from 8 as we're worried about how GC will perform and since we're about to get geared up into shopping season, now is not the time to use shiny new features.

agumonkey4• 11 years ago

Mentioning Scala, or clojure or even Haskell to a HR interview leads to raised eyebrows at best. But that was expected. I'm surprised actual devs are still wearing eye blinders.

nobleach10• 11 years ago

I'm hiring front end JavaScript developers right now. If I see other language experience on their resume, they're getting a call. My last hire had worked through "Learn You A Haskell" and that was enough for my interest to be piqued. Seeing Python, Ruby, C, ObjectiveC, heck even PHP is enough for me to see they're not just a "I did some Angular tutorials" type of candidate. I'm on this site, so I get a nice smattering of different technologies. I don't write Rust, but I know it exists. I've read about it. Same goes for a ton of other technologies/languages/methodologies that I don't deal with. It just floors me that anyone in the development/engineering world doesn't do the same! They must believe their language/tech will exist forever.

fizwhiz6• 11 years ago

What a curious comment. The first portion of your diatribe disparages all java developers, but the latter part of it suggests that you're still working in java shop (perhaps in an ops capacity?) I'm wondering what you meant by JVM 8 because no such thing exists (unless you were talking about the JDK or the JRE.)

You must have worked at some sweatshops to spout such biased vitriol.

nobleach10• 11 years ago

Well, I was an English major, so I'm glad you were engaged by the "diatribe". And if you were unable to piece it together... and I kinda feel bad if you couldn't. I'm referring to Java 8 language features and OpenJDK 1.8. I get the feeling you knew that though and are being pedantic. You've succeeded though, because I responded.

Sadly you missed the part where I said, "Not all Java devs obviously". I have read great blogs by Twitter engineering and many others.

But, since you seem to have some experience in sweatshops, please explain how your current situation is different. Add something to this conversation instead of correcting my stupid generalizations.

thejosh• 11 years ago

This is also how I feel with PHP Developers who work in development agencies or larger companies, either in a small team or by themselves (probably even larger teams, but no first hand experience with that).

Basically they refuse to use version control (FTP their code to the live server), use a terrible dev system, non-existent or terrible staging environments or newer technologies such as composer or other tools which will save them time and effort.

Don't even get me started on the lack of tests, and how buggy and broken their code ends up being - "we don't have time to write tests", yet spend weeks fixing critical issues that would have been found...

/rant

nobleach10• 11 years ago

I see the lack of unit tests constantly. I did a small stint with PHP and one of the first things I did was learn PHPUnit. Having invested heavily in the TDD Koolaid with Rails/Python, I made sure I learned how to apply the same ideas in PHP. And yes, I've seen quite a bit of the "1999 was just fine" culture among PHP developers too. If you can't deploy your code at the touch of a button, you should be working your way to that goal.

Joeri• 11 years ago

On the flip side, the best practice for both Java and PHP is trending towards the same thing. Routes + DI + immutable types + fluent api's. Give it another decade and the only way to tell the difference will be the presence of $ characters.

zeckalpha2• 11 years ago

Even worse, the Java 1.4 world. I've seen people avoid generics and other Java 5+ features for "compatibility reasons".

nobleach10• 11 years ago

Wow, you didn't even get forEach with 1.4, right???

habosa2• 11 years ago

That's a shame. Java is my favorite language because it has the right balance of strictness (although Go is starting to take over for me). When written correctly, Java can be amazingly productive because you learn so much at compile time but the tools (IDEs) make it so that you can autocomplete away most of the verbosity.

nobleach10• 11 years ago

Java as a language is totally fine! It has everything one might need. The culture totally sucks. Oracle leads the march toward what is the "standard". Jersey is a great leap out of this.

metaobject• 11 years ago

What do you mean by: "you learn so much at compile time"?

I'm a C and C++ guy, but I've done a tiny bit of Java a while ago.

brown9-2• 11 years ago

I have a hard time believing that a company stuck on Java 1.5 or SVN is like that because the developers want it that way.

sandGorgon• 11 years ago

What web frameworks are you guys using in the Java 8 world that lets you leverage newer language features?

Is it still Spring ? Are you guys doing things like Socket.io , etc.

And is Hibernate the preferred orm ? Confession: I'm an ex Java programmer, who have been working in rails and python for past several years...so am fairly out of date. I have kept up with things like Dropwizard...but have not seen many other startups do Java development.

jacques_chester5• 11 years ago

Spring is on Java 7 for the 4.x series.

Spring 5 will take Java 8 as the baseline.

namelezz• 11 years ago

I note that it's not safe to pass Java 8 streams around as streams can be closed when invoking terminal operations. If you pass a stream to a method and do not know whether the method is going to close the stream, it's dangerous to reuse it after the method return.

EdwardDiego6• 11 years ago

Well, once a Stream has been consumed, you can't use it again. We have a rule of thumb in my team that our APIs should not consume or return Stream or Optional.

tigeba• 11 years ago

I agreed with most of these suggestions. I'm not sure the suggestion to use a struct pattern is worth the lack of JavaBeans support in argle support libraries. You can always use final properties and only provide getters.

I think asserting that Fluent interfaces are 'more readable' or better is largely a matter of opinion.

eddiezane• 11 years ago

This is a really great resource and I'll certainly be embracing some of it.

I attend quite a few university hackathons as a mentor/sponsor and one of my goals has been to paint Java in a better light as it doesn't have the best reputation in that circuit. I'm always telling people that their goal should be to build something cool and learn something new.

It's much easier to teach someone how to build their first webapp in a weekend when you can do it in a language that they are familiar with and Java seems to be what's taught at most universities.

The Spark Framework mentioned here is very well done and beginner friendly. I recently wrote an intro blog post [0] about it as an attempt to embrace existing knowledge and showcase that it's easy to build something cool with Java.

0: https://www.twilio.com/blog/2015/09/getting-started-with-gra...

potatosareok2• 11 years ago

Looking at Spark framework I'm wondering about your thought of Spark vs Spring Boot. I've not used Spark but I find Spring Boot lets you get an app of the ground quickly, is part of massive Spring ecosystem which gives you a lot of other advantages (for example, Mongo template might be very useful for Hackathon project). Beyond that RestController automatic integration with Jackson can quickly let you bring up a JSON based webservice.

Also I'm not a fan of how in your example the entire app routing/logic is basically crammed into a single main method. I know it might be simple demo app that you're trying to show off Twilo but I don't think you're doing favor by structuring like that.

What do you feel are the benefits of Spark vs Spring Boot (or Dropwizard I guess). In my mind the only reason I wouldn't use Spring-Boot for a new project in Java is if I wanted to use something like Play! since supposedly using Play!/Scala and futures integrate very nicely into that.

bluejekyll• 11 years ago

As a Java Dev of 15 years, through my own experience, almost all of these points I have discovered through hard fought bugs and trying to find a way to reduce the causes of them in the future.

Final by default is the biggest, bc it removes some of the most basic concurrency errors implicitly. Awesome post. It also is pointing out things that in some other languages (e.g. Python) are implicitly impossible to achieve, and others which some new languages (e.g. Rust) have built in from the beginning, like never use null.

Well done.

mabbo6• 11 years ago

95% agree with these opinions, but have a few disagreements, or perhaps the author had missed some great stuff.

Two main ones: Mockito is, IMHO, much nicer than jMock. And Immutability is good, but 'final' is... complicated. I find it's often over used by people who don't know why they're using it.

meddlepal3• 11 years ago

Can you expand on the later point about final please?

mabbo6• 11 years ago

tl;dr: final parameters != immutable and safety, and they look like they ought to.

A lot of people write "final" in front of every. single. declaration. And they do that because someone once told them that was good. But then you remember that Java is a language which all non-primitive types are passed by reference, so your 'final' is actually a final reference and not a const immutable object, like you might get in C.

If someone passes in a final FooBar to a method, and you mutate the FooBar during the method for any reason, then after the method is done the object remains mutated.

Just this week, that exact problem caused my team an entire day of headaches. Two methods were being called concurrently using Futures, passing the same object to each method. One method mutated the object slightly, the other relied on it having not been mutated. Hello race condition. I was so proud to have solved it, until I realized I'm the dolt who wrote the bug in the first place.

Silly, I know. But the point is that final makes you feel safe when you might not be.

lmm10• 11 years ago

jMock isn't great, but Mockito makes it too hard to understand which method you haven't stubbed; I find EasyMock with its explicit replay() step much easier.

twic2• 11 years ago

I rather like Moxie for mocking:

https://github.com/pobrelkey/moxiemocks

cam-4• 11 years ago

JMock at least blows up if you don't mock something. Powermock and Mockito will let it pass. It is a shame JMock's API is so wordy.

mabbo6• 11 years ago

Eesh, let's agree to disagree. EasyMock has some advantages, but I find the readability of Mockito outweighs them.

owlish3• 11 years ago

I used to be a fan of the Builder pattern (or Fluent Interface), but the ease of use for developers comes at the cost of no compile-time checking. Before, you just had to look up the constructor to see which parameters go where, with the benefit of type-checking. Now, you have to look up the Builder's methods to make sure you've filled in all required fields. The only case where I can actually advocate using Builders is when you have a mess of telescoping constructors.

habosa2• 11 years ago

If you have this problem, the library you're using is incorrectly using the Builder pattern. All required parameters should be in the constructor of the builder. Then the other methods on the builder should be optional configuration. For instance on Android you often see a builder that takes a `Context` in the constructor because that's almost always required.

owlish3• 11 years ago

This is contrary to the OP's example, where he passes data and num to the constructor, which is how I've seen it used frequently. The Builder pattern is often recommended to "make code easier to read" and wouldn't be useful if many/most of those parameters were required.

EdwardDiego6• 11 years ago

If you have required parameters in a Builder, you're doing it wrong.

chii5• 11 years ago

Now, you have to look up the Builder's methods to make sure you've filled in all required fields.

the builder should fail with a good error message if there are required fields you didn't supply to the builder.

owlish3• 11 years ago

This doesn't solve the problem where the parameters aren't checked at compile-time. Now, hopefully all your codepaths are tested so your builder doesn't throw exceptions in production.

lliamander• 11 years ago

You can have a type-safe fluent Builder, but it requires having one or more static nested classes with private constructors. The outer class will not have a build() method, but rather that method will live in one of the nested classes. You use the type system to control what parameters have already been set, and what parameters need to be set.

Obviously, this can apply to many sorts of fluent API. This is at least part of what Tony Morris is alluding to in his essay on API design [1].

[1] http://blog.tmorris.net/posts/understanding-practical-api-de...

meddlepal3• 11 years ago

Some good suggestions in here and some very very bad ones. For example, using Optional as a member type, tuples etc.

chii5• 11 years ago

this details a good amount of the problems with optional, but yet why it will still be used despite it: https://developer.atlassian.com/blog/2015/08/optional-broken...

cletus• 11 years ago

I mostly agree with this list but some things I have issues with.

The biggest is DI. Having witnessed what a tangled mess large Guice codebases can become where you really have no idea what is providing what I have to say that Spring here gets a bad rap from the all-XML days and (IMHO) it is actually the cleanest and easiest to work with.

The author mischaracterizes Spring too:

It has a either code-based wiring or XML configuration-based wiring.

Actually Spring can use a mix of XML code configuration and auto-wiring.

A modern Spring application will have precisely an application context XML file with one line per "bean" eg:

    <bean class="com.foo.SomeClass" />

and that's it. Note: there's no classpath scanning. Auto-wiring takes care of the rest. This has two big advantages (over Guice in particular):

1. Everything is in one place. If it's not instantiated here it doesn't exist. Guice can be many levels deep and tends to descend into ModuleBuilders and conditional logic in modules (which the docs recommend against but tends to be used extensively anyway); and

2. Testing. Testing is incredibly easy. I'm talking functional testing here, even integration testing. You simply include another XML file that overrides any definitions from the first. Everything else remains the same.

Testing with Guice can be horrendous and once you start making extensive use of Modules.override() you should abandon all hope.

I'm glad to see this post recommend Maven. It's fun to bash Maven but most detractors ignore the huge benefit that applies to any shop with lots of devs: Maven is opinionated. That's a plus not a minus. Seen one Maven Web application and you've seen them all. There's no really weird directory layout because someone decided they just had to be different.

Lastly Play. I really tried to like Play. I really did. But as of Play 2.0+ it became a Scala not a Java framework. It may still work with Java but some things just require you to know something about Scala. That's a huge negative IMHO. I'm not interested in Scala. Nobody is.

Plus the auto-compile feature of Play tended to randomly break with indecipherable error messages for no readily apparent reasons such that you couldn't tell if it was a bug or not. I get enough of that with C++ template programming thank you.

I also agree that the two books the author recommends are probably the two most important Java books in existence.

jacques_chester5• 11 years ago

A modern Spring application will have precisely an application context XML file with one line per "bean" eg:

I prefer my Spring applications to look like this:

    @SpringBootApplication
    public class Application { ... }

lmm10• 11 years ago

1. Everything is in one place. If it's not instantiated here it doesn't exist.

But that one place is XML. Better to have it be code, where I can use all my normal tooling ("find references" in my IDE (yes some IDEs have spring support), grepping through .java files, automated refactoring...) to work with it.

Guice can be many levels deep and tends to descend into ModuleBuilders and conditional logic in modules (which the docs recommend against but tends to be used extensively anyway)

Have you seen the kind of horrors that are possible in Spring XML with conditionals, proxying, interceptors...? Compare like with like.

And in the worst case, if you must put conditional logic in your config, at least if it's in code then you can unit-test it the normal way.

Testing. Testing is incredibly easy. I'm talking functional testing here, even integration testing. You simply include another XML file that overrides any definitions from the first. Everything else remains the same. Testing with Guice can be horrendous and once you start making extensive use of Modules.override() you should abandon all hope.

Shrug. Not my experience; you get more-or-less equivalent tooling capability with both (that is, you can group your definitions into smaller modules, and you can override specific instances when testing), so if you use them the same way it'll be just as easy or hard.

Babylonian2• 11 years ago

Scala. That's a huge negative IMHO. I'm not interested in Scala. Nobody is.

Their use of Scala was the only reason I gave the play framework a second look. Just because you're not interested in Scala doesn't mean nobody else is. Scala is the most popular plugin for intellij by a very wide margin, that's a lot of 'nobody' downloads. Having used both java and Scala extensively I honestly can't understand why anyone would still pick java by choice other than if it's the only jvm language they know.

jacques_chester5• 11 years ago

I've worked on two Spring Boot projects so far and written somewhere between zero and no XML.

davnicwil2• 11 years ago

On the struct-like data objects part - this is so much nicer than the getter (and setter for mutable objects) pattern, just because readability is so much improved.

Getters/setters is a pattern that it took me a while to reject, having always been told that public fields were really, really bad. But they're simply not. Having logic in getters/setters is the only justification for this added bloat. In my experience, in the overwhelming majority of cases (never say all) having logic in getters and setters is a clear indication that your design is flawed anyway, so most of the time you should never have them.

On this part:

Further, this class is immutable unless you extend it, so we can reason about it easier as we know that it can't be changed.

Well, I'd go further. Make the class final if you really want to guard against people extending it. Of course this is most applicable for writing reusable libraries where communication with consumers may be limited.

If you rely on language features to enforce development practices in a team who can freely communicate, in general you've got a much bigger problems than the few rare bugs that extending an immutable class and making it mutable will cause.

Obviously though, putting guard rails in can only ever be helpful, even when absolutely everyone understands that there's a cliff edge there, and wouldn't conciously try to walk over it anyway.

mark_l_watson• 11 years ago

I don't like to use getters, setters either. I like default package access so instance variables are just visible inside a package.

whitenoice• 11 years ago

For web framework I would also suggest Dropwizard. Its simple easy to use and robust.

mindcrime2• 11 years ago

One thing I'd add, is that Java is very amenable to static code analysis, and there are good tools like PMD and FindBugs for Java. I always encourage teams using Java to use one or both. The biggest downside is that you usually have to tweak the rule-sets to avoid getting lots of false positives. And if you get too many false positives then people tend to stop paying attention to the notifications from the tool.

On a related note, I really encourage Java teams to shoot for a goal of having compiles that finish with no warnings. If there are generic warnings littering your code (like, for example, all the warnings about unqualified references to generic types that you get when using code written before generics existed), bite the bullet and either fix them, or make it explicit that you don't care and add the appropriate annotations to turn them off. At least that way, when a new, interesting warning shows up, maybe people will pay attention to it.

And yes, this is something of a "do as I say, not as I do" thing, as I have definitely written Java code that doesn't follow this rule. Especially on my personal projects or code where I'm not working with a big team. On bigger projects, I do try to push that as much as I can, depending on my role on the project.

lomnakkus3• 11 years ago

[...] Java is very amenable to static code analysis [...]. The biggest downside is that you usually have to tweak the rule-sets to avoid getting lots of false positives.

These two statements seem to be in opposition to each other.

(All your advice is spot on. If you let warnings remain in code, you'll just end up accumulating more and more warnings until it's too much to fix. I guess it's a sort of "broken windows" effect.)

mindcrime2• 11 years ago

I don't know if they're in opposition exactly. The Java language is definitely amenable to static analysis to some extent, probably more so than a more dynamic language. But a lot of the "false positives" are just about what rules the developers of the tool ship by default, and what they define as a "bug" (or "possible bug"). For better or for worse, static analysis tools sometimes conflate simple stylistic issues ("don't use variable names less than two characters") with things that are likely to be actual logic errors. And since the tool will always be limited in how much it can understand about the intent of the code, it's pretty much always just applying heuristics and a probabilistic analysis, even when processing Java.

So yeah, those tools aren't perfect, but my subjective experience has been that they help, as long as you tweak the knobs so that they don't just generate something equivalent to the "broken windows" compiler warnings, etc.

adarshr• 11 years ago

My solution is to use Groovy wherever possible. It has a superb syntax, very low learning curve and compiles into JVM bytecode. There is no issue with getters and setters in Groovy - you just define the fields, which Groovy calls properties and defines implicit getters and setters.

If not for the main application code, I would definitely use Groovy for unit testing in combination with Spock Framework. You just have to try it once to realise what you've been missing all these years with JUnit.

vorg4• 11 years ago

Groovy's always been OK for quick and dirty testing of Java classes, though the statement label hack that Spock uses makes Groovy look trashy. The static compilation that came with Groovy 2.0 is what no-one should touch if they don't want to spend countless hours and days debugging and wondering why they didn't use a language built from the ground up for static compilation, e.g. Java, Scala, Kotlin, Ceylon, and others.

jonhohle• 11 years ago

As a fan of immutable objects, I bemoan the builder pattern. The nice thing about constructors (or factory methods) is that you have a compile time guarantee that you have all required components to construct an object and validate its invariants. With a field oriented builder, only the documentation informs you whether you've met all of the conditions to construct an object. Also, a builder which mutates its own internal state breaks the idea of partial application where several variants can be constructed from a parent.

Unrelated to builders, the stream example could be simplified further with method references

    final List<String> filtered = list.stream()
        .filter(s -> s.startsWith("s"))
        .map(String::toUpperCase)
        .collect(Collectors.toList());

It would be nice if Oracle back filled some predicates to make them more useful in streams (e.g. "s"::isPrefixOf, etc.).

benjiweber2• 11 years ago

You can create immutable "builder" with lambda chains. This also gives you the compiletime guarantee that you have specified the required fields (At the cost of flexibility and ease of adding new fields)

http://benjiweber.co.uk/blog/2014/11/02/builder-pattern-with...

Regarding your isPrefixOf - is there a significant benefit to that being in the standard library? You could define a isPrefixedWith("s") that returns a Predicate<String> yourself.

hepta• 11 years ago

This is great, thanks. I've also used something like this: https://gist.github.com/dwijnand/3741951

jacques_chester5• 11 years ago

The Builder pattern includes setting sensible defaults on the built object before building it. You configure the built object in the constructor of the Builder, before anyone starts updating fields.

You can achieve this through constructors or setters; Builder itself is agnostic.

based2• 11 years ago

Nothing on findbugs, checkstyle, EOL and CVEs?

log4j: now log4j2

jOOQ => $, JPA, MyBatis

tsotha• 11 years ago

Java 8 has a nice stream and lambda syntax. You could write code like this:

  final List<String> filtered = list.stream()
    .filter(s -> s.startsWith("s"))
    .map(s -> s.toUpperCase())
    .collect(Collectors.toList());

>Instead of this:

  final List<String> filtered = new ArrayList<>();
  for (String str : list) {
    if (str.startsWith("s") {
      filtered.add(str.toUpperCase());
    }
  }

>This allows you to write more fluent code, which is more readable.

I don't agree. The second snipped is far more readable. This seems like an effort to use lambdas just to be using lambdas.

Not only that, but the non-lambda version is far easier to maintain. When you get exceptions in code that has the form

  this().that().theOther()

it's generally not obvious where the problem is.

ap22213• 11 years ago

A matter of personal taste, but I greatly prefer the lambda style (though I wish the example had the .stream() on the following line).

I have been coding Java (and C-style) languages for a really long time. And, I still have some cognitive delay when traversing multiple indented scopes. Not a long delay, but long enough to be slightly annoying.

With the lambda style, I can quickly glance down the text column to see what's going on. Stream -> Map -> Filter -> Collect.

After using lambdas enough, one starts to subconsciously cull the stream() and collect() calls. So, when I look at it, I really see Map -> Filter.

But what really, really annoys me is that exception handling can't be applied to the whole stream processing statement. This leads to code that looks like:

    collection
            .stream()
            .map(o -> {
                try {
                    return o.oToC();
                } catch(Exception e) {
                    // handle
                }
                return new C();
             })
            .reduce((c1, c2) -> {
                try {
                    return c1.combine(c2);
                } catch(Exception e) {
                    // handle
                }
                return c1;
             });

For that, I have had to write a bunch of helper functions that wrap and throw the exceptions.

anotherevan• 11 years ago

I find lauding use of Optional over null for return values, but eschewing checked exceptions somewhat inconsistent. The problem with null is it can be returned anywhere, from any function. It is difficult to generate compile time errors for if they go unhandled. But unchecked exceptions are great because you don’t have to then think about them. hmmm.

Addendum: I’m also of the opinion that trying to write java code without an IDE is highly inefficient. With an IDE, you can locate all the throws you’re not taking care of quickly, and usually quickly deal with them to get things to compile. Yeah, it’s a bit of housekeeping overhead, but once they are dealt with, they are dealt with, even if it is just passing them upstream.

Better Java – Resources for Writing Modern Java

Related Stories