Finding methods that return null instead of a collection (in Java)

This involves a bit of work with pmd, but I believe that this is a valid case for writing a custom rule.

The way that custom rules in pmd work are to "compile" the Java code into the xml version of the abstract symbol tree and then run xpath queries against it.

To get an idea of the structure, the non-tag xml looks something like:

class Example {
 void bar() {
  while (baz)
   buz.doSomething();
 }
}

gets 'compiled' to:

CompilationUnit
 TypeDeclaration
  ClassDeclaration:(package private)
   UnmodifiedClassDeclaration(Example)
    ClassBody
     ClassBodyDeclaration
      MethodDeclaration:(package private)
       ResultType
       MethodDeclarator(bar)
        FormalParameters
       Block
        BlockStatement
         Statement
          WhileStatement
           Expression
            PrimaryExpression
             PrimaryPrefix
              Name:baz
           Statement
            StatementExpression:null
             PrimaryExpression
              PrimaryPrefix
               Name:buz.doSomething
              PrimarySuffix
               Arguments

For example: //WhileStatement[not(Statement/Block)] will identify while statements that are not followed by a statement or a block with statements - an empty block.

While it would take a bit more digging into the specifics, it should be possible to write the appropriate xpath query to identify a method deceleration where the type is something and that it has a return statement nested under it that returns a null. Writing one that flags any return null; should be trivial (a starting point to refine for the rule). I have written a pmd rule in the past (don't have it with me anymore) that identified static SimpleDateFormat ... decelerations which could cause problems as SimpleDateFormat isn't thread safe).

If the xpath isn't powerful enough to identify the specific rule, its possible to write a class that gets plugged into pmd that will get the AST and any transformations can be done on that to identify problems. This is a bit more difficult to plug into pmd though (compared to a xpath rule).