InternalCompilerError: undefined method `compatible?' for nil:NilClass

[abscondment]

Updating my project to 0.0.10 from 0.0.9 showed a new compilation snag:

Mirah::InternalCompilerError: undefined method `compatible?' for nil:NilClass
         infer at /Users/brendan/code/mirah/lib/mirah/typer/simple.rb:280
         infer at /Users/brendan/code/mirah/lib/mirah/ast/flow.rb:376
    resolve_if at /Users/brendan/code/mirah/lib/mirah/ast.rb:248
         infer at /Users/brendan/code/mirah/lib/mirah/ast/flow.rb:374
         infer at /Users/brendan/code/mirah/lib/mirah/typer/simple.rb:275
         infer at /Users/brendan/code/mirah/lib/mirah/ast/structure.rb:34
          each at org/jruby/RubyArray.java:1614

Git bisect between 0.0.9 and 0.0.10 pins it on 677e4aa

I don't quite grok what's going on here, but I have a test case to help in fixing things: https://gist.github.com/1478479

[abscondment]

Interestingly, if I tweak the return value/type of that while block, this error goes away.

[abscondment]

A more succinct test case:

class ContentType
  def getContentLength:long
    begin
      return getLengthOfParts
    rescue Exception => e
      return long(0)
    end
  end

  def getLengthOfParts:long
    long(1)
  end
end

[baroquebobcat]

Funnily, these are actually two separate bugs. The first one is because we do inference on ensure as though it is an expression, even though ensure's never have a return type.

https://github.com/mirah/mirah/blob/master/lib/mirah/ast/flow.rb#L373-377

    class Ensure < Node
      child :body
      child :clause
      attr_accessor :state  # Used by some compilers.

      def initialize(parent, position, &block)
        super(parent, position, &block)
      end

      def infer(typer, expression)
        resolve_if(typer) do
          typer.infer(clause, false)
          typer.infer(body, true) # << true should be false here
        end
      end
    end

The second will take a bit more to tease out, but I'm guessing it has something to do with return not having a type & it needing to be special cased.

[abscondment]

What do you think of abscondment@2fddffe for the 2nd issue?

[baroquebobcat]

looks ok to me, add some test cases in test/jvm/test_rescue.rb that cover the new cases and open a pull request.

[baroquebobcat]

Thanks.