(Re)learning Java - My takeaways

Stuck at Java 1.4

(Re)learning & upskilling to Java 18

Effective Java

Java Pub House

Java Concurrency course

Read a lot of Java code

Observed Java code reviews

Took ownership of a roadmap-Java project

The good

• Static factory methods and public constructors both have their uses, and it pays to understand their relative merits. Often static factories are preferable (cleaner and easier to read), so avoid the reflex to provide public constructors without first considering static factories.
• Java streams were utterly new to me. I love how they can help to produce more compact code. But it's challenging to set the line of when enough is enough. Overusing streams can have the opposite effect: Code becomes more complex and challenging to understand. And it is so easy to parallelize the computation of streams by "just" marking the stream as .parallel() (more on that later in this post).
• Overall, object-oriented languages tend to include more and more functional paradigms. The functional interfaces, which Java provides now, are fantastic. Their full powers shine in combination with the use of streams. An entire book can be written about Java functional interfaces, so I leave it to you to do any further digging.
• Local type inference help to reduce even more boilerplate code. E.g. instead of Hash<String,Hash<String,String>> foo = new Hash<String,Hash<String,String>>(); it's possible to just write var foo = new Hash<String,Hash<String,String>>();
• Class inheritance isn't the preferred way anymore to structure reusable code. Now, it's composition over inheritance. E.g. use dependency injection (inject one object to another object through its constructor) or prefer interfaces (which now also support default implementations of methods) over class inheritance. This makes sense to me as I do that already when I program in Ruby.
• I learned the try-with-resources pattern. Very useful in ensuring closing resources again correctly. No need anymore for complicated and nested finally-blocks, which used to be almost impossible to get right previously in case of an error condition (e.g. I/O error somewhere deeply nested in an input or output stream).
• Optimize only when required. It's considered to be cleaner to prefer immutable variables (declaring them as final). I knew that already, but for Java, it always seemed to be a waste of resources (creating entirely new objects whenever states change), but apparently, it's okay. Java also does many internal tricks for performance optimization here, e.g. interning strings.
• I learned about the concept of static member classes and the difference between non-static member classes (also sometimes known as inner classes). Non-static member classes have full access to all members of their outer class (think of closure). In contrast, static member classes act like completely separate classes without such access but provide the benefit of a nested name that can help group functionality in the code.
• I learned about the existence of thread-local variables. These are only available to the current thread and aren't shared with other threads.
• I learned about the concept of Java modules, which help to structure larger code bases better. The traditional Java packages are different.
• I learned to love the new Optional type. I already knew the concept from Haskell, where Maybe would be the corresponding type. Optional helps to avoid null-pointers but comes with some (minimal) performance penalty. So, in the end, you end up with both Optional types and null-pointers in your code (depending on the requirements). But I like to prefer Optional over null-pointer when "no result" is a valid return value from a method.
• The enum type is way more powerful than I thought. Initially, I felt an enum could only be used to define a list of constants and then to compare an instance to another instance of the same. An enum is still there to define a list of constants, but it's also almost like a class (you can implement constructors, and methods, inherit from other enums). There are quite a lot of possible use cases.
• A small but almost the most helpful thing I learned is always to use the @Override annotation when overriding a method from a parent class. If done, Java helps to detect any typos or type errors when overriding methods. That's useful and spares a lot of time debugging where a method was mistakenly overloaded but not overridden.
• Lambdas are much cleaner, shorter and easier to read than anonymous classes. Many Java libraries require passing instances of (anonymous) classes (e.g. in Swing) to other objects. Lambdas are so lovely because they are primarily compatible with the passing of anonymous classes, so they are a 1:1 replacement in many instances. Lambdas also play very nicely together with the Java functional interfaces, as each Lambda got a type, and the type can be an already existing functional interface (or, if you got a particular case, you could define your custom functional interface for your own set of Lambdas, of course).
• I love the concept of Java records. You can think of a record as an immutable object holding some data (as members). They are ideal for pipe and stream processing. They are much easier to define (with much less boilerplate) and come with write protection out of the box.

The bad and the ugly

• Finalizers and cleaners seem obsolete, fragile and still, you can use them.
• In many cases, extreme caution needs to be taken to minimize the accessibility of class members. You might think that Java provides the best "out-of-the-box" solution for proper encapsulation, but the language has many loopholes.
• In the early days, Java didn't support generics yet. So what you would use is to cast everything to Object. Java now fully supports generics (for a while already), but you can still cast everything to Object and back to whatever type you want. That can lead to nasty runtime errors. Also, there's a particular case to convert between an Array of Object to an Array of String or from an Array of String to a List of String. Java can't convert between these types automatically, and extreme caution needs to be taken when enforcing so (e.g. through explicit type casts). In many of these cases, Java would print out warnings that need to be manually suppressed via annotations. Programming that way, converting data between old and new best practices, is clunky.
• If you don't know what you do, Java streams can be all wrong. Side effects in functions used in streams can be nasty to debug. Also, don't just blindly add a .parallel() to your stream. You need to understand what the stream does and how it exactly works; otherwise, parallelizing a stream can impact the performance drastically (in a negative way). There need to be language constructs preventing you from doing the wrong things. That's so much easier to do it right in a purely functional programming language like Haskell.
• Java is a pretty old language (already), so there are many obstacles to consider. There are too many exceptions and different outcomes of how Java code can behave. In most cases, when you write an API, every method you program needs to be documented so the user won't encounter any surprises using your code. Writing and reading a lot of documentation seems to be quite the overhead when the method name is already descriptive.
• Java serialization is broken. It works, and the language still supports it, but you better not use Java's native way of object serialization and deserialization. Unbelievable how much can get wrong here, especially regarding security (injecting arbitrary code).
• Being a bit spoiled by Golang's Goroutines, I was shocked about the limitations of the Java threads. They are resource hungry, and you can't just spin up millions of them as you would with Goroutines. I knew this limitation of threads already (as it's not a problem of the language but of how threads work in the OS), but still, I was pretty shocked when I got reminded of them again. Of course, there's a workaround: Use asynchronous sockets so that you don't waste a whole thread on a single I/O operation (in my case, waiting for a network response). Golang's runtime does that automatically for you: An OS thread will be re-used for other tasks until the network socket unblocks. Every modern programming language should support lightweight threads or Coroutines like Go's Goroutines.

Conclusion

Generated by Gemtexter 2.1.0-release | served by OpenBSD/httpd(8) | Site Mirrors