JMRI Code: Recommended Practices

This page contains miscellaneous info and pointers for JMRI developers.

Class Library Preferences

We use Java Swing for our GUI development. It's a lot more powerful than the original AWT, and the price is right. We have a recommended organization and usage pattern that's documented on another page.
JMRI uses the jSerialCommn to support serial communications on Windows, Linux and macOS. Before JMRI 5.7.1 we used the PureJavaComm libraries for this. PureJavaComm references are slowly being removed. Before JMRI 4.7.5, we used the RXTX and SerialIO libraries for this, but these were then removed.
Take a few moments to learn about the different types of Java collections that are available ( List, Deque, HashMap, etc) in the java.util package. Not everything needs to be an array! Only use a Vector when there's a specific reason for it such as compatibility with an existing API; they're slow and memory-intensive compared to e.g. an ArrayList. A Deque can be a good default solution to holding data that won't ever contain 'null'. Consider using a TreeSet for data that's best kept in some sorted order; there's are very few good reasons for sorting data during normal running in JMRI. If none of these are adequate, consider incorporating a specialized collection type from Apache Commons Collections instead of rolling your own.
JMRI uses Java generics extensively to reduce errors and improve understanding. With lots of people writing and sharing code, using generics instead of casts makes it much easier to understand the code, and allows the compiler to catch many misunderstandings about how things should be used. Most of the important information on them can be found on this page from the Java Tutorial.
Enums, added in Java 5, are a better way of representing N selections than using integer constants. The entire idea is to not have them be just another form of Integer or String, so don't provide methods to convert them from or to Integer (int) or String. If you need to provide an ordering or comparison, add a public method to the enum to do that; it can work with the internal structure of the enum without exposing it.
If you need to use comma-separated variable (CSV) files, please use the Apache Commons CSV API if possible. We are already using that in a number of places, and will probably use it in more. If that does not provide enough functionality, we might eventually move to the opencsv API, but since we only want to use one, the conversion will be a lot of work.
We currently don't use the Java assert keyword in our production code. (We use the JUnit Assert in testing, but that's different). We may use it in the future, but for now please log a message at the Error level if you encounter an internal consistency problem during normal operation; if there's no way to continue, throw new AssertionError("description"); as that's a step toward our eventually figuring out how to use Java assertions in such cases.

Collections

Take a few moments to learn about the different types of Java collections that are available ( List, Deque, HashMap, etc) in the java.util package.

Please don't use fixed-size arrays for holding variable sized data in objects. That generally ends up wasting huge amounts of space at runtime, and we try to keep the JMRI memory footprint small when we can.
Only use a Vector when there's a specific reason for it such as compatibility with an existing API; they're slow and memory-intensive compared to e.g. an ArrayList.
A Deque can be a good default solution to holding data that won't ever contain 'null'.
Consider using a TreeSet for data that's best kept in some sorted order; there's are very few good reasons for sorting data during normal running in JMRI.

Code Format

The Java Code Conventions (if that link is broken, try this one from the Internet Archive) for names, formatting, etc are really useful. If you find that you can't read a piece of code, these will help make it better.

Note that we have a few local conventions beyond those in the Java recommendations. You'll find them on other pages in this section, but for example, we recommend that you define the logger reference at the bottom of each file.

Deprecating Code

As development proceeds, sometimes old ways of doing things have to be replaced by new ways. In many cases, you can just change all the using code in our repository, and move forward. For general interfaces that might be used externally to JMRI, such as in scripts and CATS, we prefer to leave the old interface in place for a while, marking it as "deprecated" so that people can discover that it will eventually go away. (The current list of those is available as part of the Javadocs) The sequence is then:

Write the new methods.
Mark the old methods with both @deprecated and @Deprecated (see below). The associated Javadoc comment should tell people where to find the replacement.
Note that if you mark an interface or super-class (i.e. abstract) class or method as deprecated, you should also mark all implementations of it as deprecated. That way, they won't themselves show as deprecated usages to fix, but code that uses them will.
Start generating warnings for users (especially scripting users) by adding:
```
            jmri.util.Log4JUtil.deprecationWarning(log, "myMethod()");
```
to the start of each deprecated method. This will put a warning in the log if that method is used during normal operation.
Remove all uses in JMRI code, except tests, of these deprecated classes and/or methods until the deprecations report during compilation is clean. Tests are kept to make sure the code keeps working until the code is removed.
In rare cases, this might not be practical. In that case you can add a line like
```
            @SuppressWarnings("deprecation")  // Thread.stop
```
just before the usage. The comment is required. It should indicate the name of the deprecated member being used.
We also use lines like this when using deprecated Java methods and deprecated methods from external libraries to indicate we know that we're doing it, and to add them to the technical backlog tracked by Jenkins.
If you have tests for that method (you should!) you may need to modify the direct tests of the deprecated method; see the instructions on the JUnit page.
Don't forget to migrate the Jython sample scripts!
These changes can then be included in the next test release. As soon as possible, you should PR them so that people get notice of the deprecations. Make sure that the PR includes something for the release note about the deprecations to draw people's attention.
Traditionally, we wait for four production releases (two years) before removing major/public deprecated methods and their tests. Implementation methods can be removed earlier, perhaps after two production releases (one year). (Truely internal methods don't need to go through this process at all; just remove them) When you do remove the deprecated code, make sure that the PR to do this includes an update to the release note about the removal; if it's a significant change, consider adding something to the warnings section.

Note that a deprecated item is meant to still work. Deprecated should only mean that you can't count on the deprecated item working in the future, so that it would be good to code away from it while it's still working.

There are two forms of marking something as deprecated in the code: Javadoc @deprecated tag and the @Deprecated annotation. Both allow you to add additional information. A nice discussion of the technicalities is here with additional discussion here. We strongly recommend using both of them like this:

/**
 * (Other Javadoc comments)
 * @deprecated As of 5.3.1, use {@ link #foo()} instead
 */
@Deprecated(since="5.3.1")

where the comment and since parameter contain the version in which the deprecation is applied. That lets the user easily know how long ago it was deprecated.

You can add a forRemoval=true clause to @Deprecated. We encourage you to do that once there are no references in the JMRI code. It will make the deprecation much more visible to people who attempt to use it because it defeats @SuppressWarnings.

Exceptions

Throwing Exceptions

When checking inputs (i.e. for valid parameter values) and you find a problem, what should you do?

You can throw an unchecked exception. IllegalArgumentException is a great example: Most of the time it doesn't really have to be thought about. It's rare to explicitly try/catch for it. When an IllegalArgumentException (or whatever) is thrown, it works up to some high-level handler, where there will be some default handling (usually just logging) it, and the various libraries will get a chance to clean things up.
You can thrown a checked exception, i.e. a JmriException subclass or some on-target Java checked exception class. This forces all writers of method-using code to think about how to handle problems. They might wrap with a try-catch or just declare it to go upward, but they have to think about it.
Don't use error codes or other "check the return value" approaches. They just make it likely there will be invisible problems that bite people in complicated ways (because you can ignore them, leaving junk behind)

Generally, JMRI developers tend to throw an unchecked exception, i.e. IllegalArgumentException or similar.

Catching Exceptions

SpotBugs will object to code like this:

  try {
     // do something here
  } catch (Exception e) {
  }

with a REC_CATCH_EXCEPTION and/or a DE_MIGHT_IGNORE (less often DE_MIGHT_DROP). This is an example of two problems:

Catching Exception instead of more specific types
Having nothing in the catch block

Let's discuss those separately:

Catching the Exception class

There are two subcases here:

You're catching Exception because a lot of different specific things can be thrown, and it's a pain to write multiple catch blocks for each of those. That's actually obsolete thinking, though, because now there's syntax for combining those:
```
  try {
    // do something here
  } catch (IOException | JDOMException e) {
    // and do something, see below
  }
```
That's a much better way to convey what this code is intended to do.
You want to catch any unchecked exception, such as NullPointerException, that is thrown. Those are all subclasses of RunTimeException, so the clean way to do this is to catch that:
```
  try {
    // do something here
  } catch (RunTimeException e) {
    // and do something, see below
  }
```

Empty catch block

What's an empty catch block trying to say?

Ideally, the code should handle the exception, in the sense of reacting to it so that the right stuff still happens for the user. So do that as a first strategy.
But sometimes, that's complicated, or the error is really something that isn't understood, or perhaps is routine and doesn't matter.
If you can't do anything else, at least consider logging that the code has caught something. Typically, this would be a warning:
```
log.warn("can't do anything useful with this:", e);
```
Adding the Exception object e to the log logs all its content, including the traceback. That way, if something weird happens later on, at least there's a clue in the log.
If it's really routine, log at debug or even trace level. That way if it turns out later to not actually be routine, turning on additional logging will help track it down.
Consider not catching the error, so that it propagates up. (The methods for targeting the catch block more narrowly in the prior section can help here) It will eventually be caught, and will contain a more useful stack trace. This might require adding the exception to the signature of the current method so you can throw it upwards.
Or, if absolutely can't change the signature, consider re-characterizing it, e.g.:
```
  try {
     // do something here
  } catch (IOException e) {
     // do something to clean upp
     ...
     // But still need to signal that this failed.
     // Let's blame the arguments we are processing:
     throw new IllegalArgumentException("couldn't process because ...", e);
  }
```
If it really is "ignore this, it's going to be OK", you should add an annotation so that others know why it's OK and don't have to worry about it in the future:
```
@edu.umd.cs.findbugs.annotations.SuppressFBWarnings(value={"DE_MIGHT_DROP", "DE_MIGHT_IGNORE"},
    justification = "(Something about why this is OK)")
```

Use of Optional

The Optional class is the right way to provide "a method return type where there is a clear need to represent 'no result,' and where using null is likely to cause errors". There's some good background on use of Optional in an introductory Java Magazine article. There's more info about how null makes errors likely is dicussed in a related article. As Tony Hoare - one of the giants of computer science - wrote, "I call it my billion-dollar mistake. It was the invention of the null reference in 1965. I couldn't resist the temptation to put in a null reference, simply because it was so easy to implement."