Is JRuby's implementation of Kernel#__method__ broken?

问题

This is the description of Kernel#__method__ according to Ruby-Doc.org (emphasis added):

Returns the name at the definition of the current method as a Symbol. If called outside of a method, it returns nil.

Now consider the following code snippet:

DEFINITION = proc { __method__ }

class C
  define_method :one, DEFINITION
  define_method :two, DEFINITION
end

o = C.new

When I run the following using MRI v1.8.7+ I'm getting the expected results:

o.one  #=> :one
o.two  #=> :two

However when I run the same code using JRuby 1.7+ (I haven't tested the previous versions):

o.one  #=> :two
o.two  #=> :two

Could this be considered a defect in JRuby's implementation or is it simply a different interpretation of Kernel#__method__?

回答1:

It may be a defect in JRuby's implementation of __method__, or it may be a bug in the implementation of define_method, or it may be strictly limited to the use of the two together. Look what happens if you cast the Proc object into a block by using the & operator:

DEFINITION = proc { __method__ }

class C
  define_method :one, &DEFINITION
  define_method :two, &DEFINITION
end

o = C.new

Now in MRI, as before:

o.one  #=> :one
o.two  #=> :two

However, in JRuby, it's fixed:

o.one  #=> :one
o.two  #=> :two

Given the internal implementation of MRI's define_method, which includes the handling of Proc arguments vs block arguments, if JRuby's is at all similar, it's also possible that that is where the problem could lie.

Either way, there are no parallels to be found by replacing __method__ with self, binding, object_id, or any combination or permutation of them, so the issue is certainly localized to uses of __method__.

UPDATE: Twist Ending

This was a known bug in MRI 1.9.2, and JRuby's implementation mirrors that behavior.

来源：https://stackoverflow.com/questions/26966547/is-jrubys-implementation-of-kernel-method-broken

标签

ruby

jruby

defects