Coding Convention

In this document I have gathered a set of very simple coding conventions. These apply to all org.simantics projects and are recommended for anyone developing with Java. I try to address some of the common mistakes that recur in a multi-developer project. A vast number of bugs can be back-tracked to failure of common understanding, a situation of mis-matching assumptions.

Sometimes a developer makes code where the rules or assumptions have changed along the way. It is natural that our handiwork changes as we grow in experience. For co-developers and future selves it is a confusing trail. There is room for speculation, and the only approach left is reaction style coding: guessing and fixing-on-error. The worst are products where there are no rules or consistency at all. Bad quality is propagated.

By following a common set of rules, a load of hours are saved in avoided debugging. Our developer minds can focus on relevant issues instead of second-guessing.

Argument Assumption

• All method arguments are non-null unless explicitely stated otherwise in documentation.

The default assumption is that an argument is non-null. This applies to undocumented methods too.

<syntaxhighlight lang="java">

   /**
     * Read the object from a file.
     *
     * @param file 
     */
   void read(File file);

   // and
   void read(File file);

</syntaxhighlight>

A null possibility must be explicitely stated.

<syntaxhighlight lang="java">

   /**
     * Write or remove existing value.
     *
     * @param newValue new value or null</t> to remove the existing value
     */
   void setValue(Object newValue);

</syntaxhighlight>

Return value assumption

• All return values are non-null unless explicitely stated otherwise in documentation.

The thumb rule is that the return value is non-null. It applies to undocumented methods aswell.

<syntaxhighlight lang="java">

   /**
     * Get the value
     *
     * @return the value
     */
   Object get();

   // and
   Object get();

</syntaxhighlight>

Null option as return value is always explicitely documented.

<syntaxhighlight lang="java">

   /**
     * Get possibly existing value
     *
     * @return the value is exists, otherwise null
     */
   Object get();

</syntaxhighlight>

Trust your assumptions

• You have a code of conduct - give it a chance.
  

The callee can trust the caller. Remember VM will do null checkings anyway.

<syntaxhighlight lang="java">

   BigInteger multiply(BigInteger a, BigInteger b) throw IOException {
       return a.multiply(b);
   }

</syntaxhighlight>

And the caller the callee.

<syntaxhighlight lang="java">

   System.out.println( multiply(a, b) );

</syntaxhighlight>

In most cases there is no good reason to do redundant checking, especially at run-time.

<syntaxhighlight lang="java">

   BigInteger multiply(BigInteger a, BigInteger b) throw IOException {
       if ( a == null || b == null ) throw IllegalArgumentException("Non-null argument is expected");
       return a.multiply(b);
   } 

</syntaxhighlight>

Nor caller.

<syntaxhighlight lang="java">

   Object x = multiply( a, b );
   if ( x != null ) System.out.println( x );

</syntaxhighlight>

Use assertions if you must. Checking does improve quality a bit and helps in early detection of problems, but is seldom needed if convention is followed. Assertion is not considered as run-time checking as they can be disabled from the VM.

<syntaxhighlight lang="java" style="background: #dfd;">

   BigInteger multiply(BigInteger a, BigInteger b) throw IOException {
       assert( a != null && b != null );
       return a.multiply(b);
   }

</syntaxhighlight>

Maintenance and migration

These rules apply to code that is published and in wide use.

• API doesn't change between minor releases.

In case of faulty design, old methods are preserved and are marked @Deprecated. They can be removed in the next major version release.

<syntaxhighlight lang="java">

   @Deprecated
   Object getValue(Object newValue);

</syntaxhighlight>

• Documentation is correct, the implementation is faulty.

In situation where there is a mismatch between the documentation and the implementation, the documentation prevails and the fault is in the implementation.

In this example there is a problematic method that may return an unexpected null.

<syntaxhighlight lang="java">

   /**
    * Deserialize an object from an input stream.
    *
    * @param is source stream
    * @return the object
    **/
   Object deserialize(InputStream is) {
       try {
           int x = is.read();
           ...
           return result;
       } catch (IOException e) {
          return null;
       }
   }

</syntaxhighlight>

The assumptions that can be derived from the documentation are left unchanged and the implementation is corrected.

<syntaxhighlight lang="java">

   /**
    * Deserialize an object from an input stream.
    *
    * @deprecated use deserialize2, it has better error control
    * @param is source stream
    * @return the object
    * @throws RuntimeIOException in case of IO problems
    **/
   Object deserialize(InputStream is) throws RuntimeIOException
   {
       try {
           int x = is.read();
           ...
           return result;
       } catch (IOException e) {            
           throw new RuntimeIOException( e );
       }
   }

</syntaxhighlight>

The faulty method can be replaced with correct one in the next major version release.

<syntaxhighlight lang="java">

   /**
    * Deserialize an object from an input stream.
    *
    * @param is source stream
    * @return an object
    * @throws IOException in case of problems
    **/
   Object deserialize(InputStream is) throws IOException
   {
       int x = is.read();
       ...
       return result;
   }

</syntaxhighlight>

--
Toni Kalajainen