Feature #8546: super errors in UnboundMethods

• Status: Closed
• Assignee: -
• Target version: -

[saturnflyer (Jim Gay)]

UnboundMethods are unable to call super

module M
  def hello
    puts "hello from M"
  end
end

class O
  def hello
    puts "hello"
  end
end

o = O.new
o.hello #=> "hello"
M.instance_method(:hello).bind(o).call #=> "hello from M"

module M
  def hello
    super
  end
end

M.instance_method(:hello).bind(o).call #=> TypeError: self has wrong type to call super in this context: O (expected M)}))

Given that the non-super method works, I would expect super to work.

[nobu (Nobuyoshi Nakada)]

• Tracker changed from Bug to Feature
• Description updated (diff)

Seems interesting.

[headius (Charles Nutter)]

I don't see how it would know the order in which to do the super logic. What do you expect to happen?

First off, if the "supering" M was actually included into O, there would still be nothing to super to since its method would be above O's. So I assume you don't want the Method-based "super" to error the same way.

If you would expect the "supering" hello to call O's hello, I'd like to see some justification. That would emulate behavior as if M had been prepended on to O. As it stands, M and O have no hierarchical relationship at all.

I don't think there's other possibilities. What do you want it to do?

Holy canolli, since which version is it possible to bind methods to incompatible object types in the first place?

Fuck me running, from 2.0.0. And you keep such an important change for yourself? Do you realize what did I go through trying to cheat 1.9.3 into binding methods to incompatible objects? Lame attempts such as redefining #kind_of?. And in 2.0.0 you say, uh oh, it wasn't a good idea to keep it confined to the source object progeny, we're allowing it now for everyone, sorry for the inconvenience? And nobody mentions this in the lectures about 2.0.0?

[saturnflyer (Jim Gay)]

headius (Charles Nutter) wrote:

I don't see how it would know the order in which to do the super logic. What do you expect to happen?

First off, if the "supering" M was actually included into O, there would still be nothing to super to since its method would be above O's. So I assume you don't want the Method-based "super" to error the same way.

If you would expect the "supering" hello to call O's hello, I'd like to see some justification. That would emulate behavior as if M had been prepended on to O. As it stands, M and O have no hierarchical relationship at all.

I don't think there's other possibilities. What do you want it to do?

My expectation is that it would call whatever hello is first available on the singleton_class of o which in this case is that defined on O.
In other words, my current understanding of how this works is that binding the method is at the most immediate level (the equivalent of prepending a module).

module M
  def hello
    "hello from M"
  end
end

module P
  def hello
    "prepended #{super}"
  end
end

module B
  def hello
    "binded #{super}"
  end
end

class O
  include M
  prepend P
  def hello
    super
  end
end

o = O.new
puts o.hello #=> "prepended hello from M"
puts B.instance_method(:hello).bind(o).call #=> self has wrong type to call super in this context: O (expected B) (TypeError)

With the new ability to bind methods to arbitrary objects, I was expecting the last line to return the string "binded prepended hello from M"

[nobu (Nobuyoshi Nakada)]

• Description updated (diff)
• Status changed from Open to Closed

Fixed in 2.1.

[saturnflyer (Jim Gay)]

Nobuyoshi Nakada wrote:

Fixed in 2.1.

I tried the above sample code in 2.1.2 and the error has changed from a TypeError to a NoMethodError: "NoMethodError: super: no superclass method `hello' for #<O:0x007fb799046c50>"
Is that the fix?
From my perspective it would be not a fix but a refusal to implement using super when binding an UnboundMethod to an object. Have I misunderstood something?

[nobu (Nobuyoshi Nakada)]

Which revision did you try?