Treat return in block in class/module as LocalJumpError

[jeremyevans]

return directly in class/module is an error, so return in
proc in class/module should also be an error. I believe the
previous behavior was an unintentional oversight during the
addition of top-level return in 2.4.

[enebo]

@jeremyevans I was responsible for that current spec but I really think we should not leave in that specified behavior (for 2.4-2.6). Your new spec for 2.7 on seems like semantic behavior and good to be in ruby/spec. My original test just happens to verify something which the implementation happens to do.

My recommendation is to remove the pre 2.7 spec because it specs what I consider to be an unintended behavior.

[eregon]

I'm neutral on removing the spec for Ruby < 2.7, both are fine for me.

[enebo]

I am fine either way as well other than having an opinion. We will just tag it out if it stays (until JRuby 9.3.x which will be 2.7.x). I just don't like seeing non-semantic behavior in ruby/spec (and I am to blame for adding it originally :) ).

[eregon]

Let's remove the <2.7 spec then since it seems clearly unintended behavior.

[ko1]

• it was merged, but semantic change. don't you need Matz's check?
• it seems we can check it at compile time.

[ko1]

... i agree it is unexpected behavior, too.

[jeremyevans]

I'm sorry. I thought it was a bug, since we do not allow return directly in class/module. I don't think we can check it at compile time:

class Foo
  class << self
    alias proc lambda if ARGV.first == 'lambda'
  end
  proc { return }.call
end

I can definitely revert and wait for matz's decision if you think it is best.

[ko1]

I agree, too. Wow!

[enebo]

@ko1 do you still think Matz needs to check this? It feels like this is just accidental behavior and not intended.

[ko1]

Now I don't think we need to ask Matz. I'll ask Matz when I meet him if I remember :)