eval 'return' not working and not failing exactly either...

Punting to post 1.0. Nontrivial, and no reported app bugs because of it.

Fix for 1.1 along with JRUBY-966.

This is going to require the long-awaited Jump/LocalJumpError unification. I think this should be in 1.1, but maybe not in the beta.

Punting issues from 1.1 RC2 to 1.1 final.

Punting...this is another case never reported in the wild, and pretty weird to boot. Nontrivial fix, post 1.1.

It appears that jruby 1.7.0(dev) and 1.6.4 do the right thing. eval('return') inside a lambda just plain returns, eval('return') inside a bare Proc raises LocalJumpError, as it should. See:
http://ruby-doc.org/docs/keywords/1.9/Object.html#method-i-return

This is indeed pretty weird, but it looks suspiciously as if it is no longer a bug, per se.

I will be submitting a pull request to rubyspec with a test for this (a test which both MRI and JRuby pass).

Martin,

I am not convinced. How did you come to that conclusion?

On my machine, the snippet shown in the description of this ticket still does not work the way MRI does.

The code snippet from the code I submitted to rubyspec looks like this:

293 foobar = false
294 begin
295 Proc.new

.call
298 rescue LocalJumpError
299 foobar = true
300 end
301 foobar.should == true

I just added this to the test:

foobar = false
begin
eval("return 3")
rescue LocalJumpError
foobar = true
end
foobar.should == true

and it looks as if it passes because it does indeed go flying off into space: the foobar.should == is never executed. Interesting. I will have to put on my thinking cap and try to figure out how to make this test fail for jruby.

So, to summarize: bug still valid (and disturbing), but only as a naked eval("return") in the begin...end block. Put it in a Proc, and it raises LocalJumpError as expected.

MRI raises LocalJumpError (at least for 1.8.7 and 1.9.2).

This is not limited to eval.

This snippet:

begin
return
end

will fail with a LocalJumpError in MRI, but complete silently and happily in jruby 1.7.0(dev).

So should this be a "return" bug rather than an "eval" bug?

For that matter, the one-line ruby script:

return

behaves the same way. MRI raises the exception and (since there's no handler in that single line) issues the error message, JRuby does not.

See JRUBY-2760. eval("return") is behaving exactly like a bare return here. JRuby "returns" from the top-level script rather than raising LocalJumpError when it sees a return in the top-level (outside a method). MRI raises LocalJumpError.

rubyspec now includes a test for this behavior. The test fails as expected on JRuby, and passes on MRI Ruby.

There'a another bug for this.