Java memes which refuse to die | Hacker News Reader

Java memes which refuse to die

javacodegeeks.com

35 pointsiliastsagklis27 comments

Nearly all of the times I've seen StringBuilder used incode reviews, using the + operator or a better design would have been more appropriate. There are a few cases (like looping over a collection), where StringBuilders make sense, but usually they are needlessly verbose.

StringBuilders are verbose, but they do make a measurable difference in your code.

The problem with no using them is that then you need to be sure your code is never called by something else that runs it in a loop.

A fair while ago (JDK 1.5 had just been released, so I was still looking at StringBuffer) I did some analysis on the bytecode generated by concatenation.

There is a pretty significant difference:

http://nicklothian.com/blog/2005/06/09/on-java-string-concat...

The example in that article is fairly dumb (why would you use concatenation as an input to a StringBuffer). The Java compiler converts concatenation to StringBuilder in Java 1.6. In most cases (eight the exception of loops and federated, organized string creation) the + operator is identical in performance to StringBuilder (because that's what it compiles to).

There are a lot more then 4. I didn't even see anything about Factory Factories.

I haven't actually seen a Factory Factory in a while. I think Dependency Injection has pretty much killed that anti-pattern.

Factory factories USED to be a problem in my Java programming, but they aren't anymore because I created a configurable builder class for generating my factory factories. I call it FactoryFactoryBuilder, and it completely eliminated the factory factory problem.

That sounds like a great idea. I bet you could even use some XML configuration files to make it even more extensible.

I would add boilerplate getters/setters to the list. Almost always, a simple public field works just fine.

Using a public field removes one of the major benefits of object-oriented design. Don't misunderstand me: in many cases, really all you need is a collection of related fields and making them public is indeed OK - like a struct in C.

But I tend to err on the side of safety with my fields - private member, public accessors - while I work out my modeling and refactoring. When the design settles, then I can relax access if it's OK. And when something changes in the design (atomicity requirements, immutability, etc) I already have most of the boilerplate in place. Mostly though, I keep my accessors.

On a related note, do Java IDEs not provide boilerplate code? I'm more an Objective-C guy and the language now supports "here's a property" and "create the accessors" alongside my needs for custom ivar names, custom setters, etc. Even if Java the language doesn't have similar features, I'd expect IDEs to give you this kind of boilerplate stuff.

Eclipse has a "Generate Getters/Setters" option. You choose the variables, and voilà!

They do allow code generation, and that's precisely the problem. You end up with these ridiculously bloated files, most of the code of which hasn't ever even been looked at by a human.

But generating getters and setters based on fields is putting the cart before the horse. You're right, it is very important for your classes to have a stable, clean public API that's distinct from the private implementation. But if that public API is intended to expose data at the field level, then just expose the fields without confusing the issue. If you don't want to expose the fields, then don't.

Having private-but-not-really fields with a lot of auto-generated getters and setters doesn't help you provide a clean, small API: instead, it bloats the public API. Also, it obfuscates what's actually happening... is a getter just a simple data value, or is it doing a bunch of complicated shit behind the scenes? And if it is doing a bunch of complicated shit, just make it a regular method on the public API, not a getter that's logically tied to a field that the client knows is there.

TL;DR: Fields are fields, methods are methods, auto-generating getters/setters blurs the line in unhelpful ways.

Except when you want to change behavior in the future, e.g. adding a check to the getter or doing more than simply setting a value in the setter.

If you started out with getters/setters in the first place, you can do this without changing the API and ABI.

Well if you are changing the behaviour of the getter or the setter then you are already changing the public API contract.

I would rather want a compile time error instead of a silently not changing value in a setter for example after a library upgrade

Changing the underlying implementation is not changing the API contract. The API makes no guarantees about where your arguments are shipped off to nor where the return value came from. It simply says "you give me X, Y, Z and I'll return you M" and that's it. This is the very reason for specified APIs: the underlying implementation changes (for caching, efficiency, new storage models, etc) and external code goes on its merry way doing what it did before-- running.

if that public API is intended to expose data at the field level

> That doesn't make a lot of sense. An API is an interface, which in this case exposes data. The fact that this data is held by objects at the field level as nothing to do with the purpose of the API: it's an implementation concern. The consumers of this API do not, and should not, need to know whether a field or anything else is used to provide the requested data. If the public API gets bloated when using getters/setters then blame Java, not this basic encapsulation principle.

is a getter just a simple data value, or is it doing a bunch of complicated shit behind the scenes?

> As you said, and as a rule of thumb, never do a bunch of complicated shit inside a getter: always expose a method for this operation. Just because getters exist does not mean methods should disappear.

An API is an interface, which in this case exposes data.

Yes. The confusion comes from complecting[1] an object api and data. In general, having a class that is both a data structure (a bean or a struct) and a service class is a bad idea.

My contention is that if a class is Just Data, exposing public fields is fine. And if it is a service class that needs a consistent, stable API, it shouldn't be exposing fields via getters and setters either.

[1]http://www.infoq.com/presentations/Simple-Made-Easy

TL;DR: Fields are fields, methods are methods, auto-generating getters/setters blurs the line in unhelpful ways.

That's not really true in Java. The getter/setter pattern is baked into a lot of the standard Java libraries.

For example, Javabeans are:

a Java Object that is serializable, has a 0-argument constructor, and allows access to properties using getter and setter methods.[1]

Things like JPA, Jersey, Guice etc all rely on the Javabeans convention.

Yes, Java should have real properties like Object Pascal & C# does, but it doesn't, and getters & setters are harmless to anything except aesthetic sensibilities.

TL;DR: Getters & setters are a useful convention that allow many things that are not possible with public fields.

[1] http://en.wikipedia.org/wiki/JavaBeans

Java IDEs are leagues ahead of Xcode when it comes to refactoring and code generation.

Except that a lot of reflection-based Java code relies on the Javabean property pattern (getters & setters) by default.

You can often customize this (it is Java after all!) but sometimes it's better to just go with the expected standard in the language.

Good luck adding Validation at a later date... And the fact many libraries rely on the pattern.

Java rely needs properties...

The original without all of the noise:

http://vanillajava.blogspot.gr/2012/08/java-memes-which-refu...

Can someone with a better understanding of the Java memory model comment on this, referring to StringBuffer:

"Say you had the code

  // run in two threads
  sb.append(key).append("=").append(value).append(", ");

Each append is thread safe, but the lock could be release at any point meaning you could get [1]

  key1=value1, key2=value2,
  key1=key2value1=, value2,
  key1key2==value1value2, ,

What makes it worse is that the JIT and JVM will attempt to hold onto the lock between calls in the interests of efficiency.[2]"

[1] is weirdly worded, or even wrong (I don't think "the lock could be release at any point" is correct) but the sample result does make sense to me if there are multiple threads banging on it at once. StringBuffer's append methods are synchronized, but without external synchronization, multiple calls could be interleaved.

I don't recall ever having heard of [2] before, and my copy of Java Concurrency in Practice isn't immediately at hand.

I don't know if that's the answer you're looking for, but a call to append is synchronized:

    public synchronized StringBuffer append(/* Something */);

... but chaining them doesn't mean that the chain is synchronized.

   sb.append(key).append("=").append(value).append(", ");

is executed the same way as

   sb.append(key);
   sb.append("=");
   sb.append(value);
   sb.append(", ");

Two concurrent threads running this part of code might do this:

   sb.append(key);   // # 1
   sb.append("=");   // # 2
   sb.append(value); // # 3
   sb.append(", ");  // # 4

   // Thread 1 executes #1
   // Thread 2 executes #1
   // Thread 1 executes #2
   // Thread 1 executes #3
   // Thread 2 executes #2
   // Thread 2 executes #3
   // Thread 2 executes #4
   // Thread 1 executes #4

This can be executed in many more different ways.

To ensure that only one thread executes those `append` at once, you would have to synchronize on the `StringBuffer` instance:

    synchronize(sb){
        sb.append(key).append("=").append(value).append(", ");
    }

Then only one thread will do the appending at a time. But this introduce yet another lock in your code, and render the `synchronized` methods useless and redondant.

Thanks, I looked up the source and saw that append was synchronized too. I was mostly asking about this claim from the OP: "the JIT and JVM will attempt to hold onto the lock between calls in the interests of efficiency". I don't think that's true.

It seems bizarre that there are a lot of equivalent classes that are thread safe or not with names that are simply synonymous. Is it just me?

It seems bizarre, but it's just because they were part of different versions. StringBuffer was in Java 1.0, and StringBuilder was added in Java 1.5. If they were both part of the original Java libraries they would hopefully have been given better names. I suppose they could have called StringBuilder "UnSynchronizedStringBuffer" or "NotThreadSafeStringBuffer", but those are problematic on their own.

You have the same problems with Vector and ArrayList I guess, although since ArrayList was added as part of the collections libraries it makes a bit more sense.

s/which/that/g

Reply on news.ycombinator.com