Forum: Ruby-core Question on scoped constant resolution Class vs Module

I asked this on ruby-talk, but didn't get anywhere.  Someone suggested
I post over here.

Resolution of fully scoped constants behaves differently if there is a
class versus a module in the name:

   XYZ = 10

   module M
   end

   class C
   end

   p defined? M::XYZ  # => nil
   p defined? C::XYZ  # => "constant"

A few questions:

1: Is this difference intended?
2: If so, what is the rationale?
3: Should defined?(C::XYZ) behave differently than C.const_defined?
(:XYZ) ?
4: Is the draft ruby spec correct, or MRI behavior correct, or did I
miss something in the following:

I think the behavior with M::XYZ does not conform to section 11.4.3.2
of the draft ruby spec:

11.4.3.2
  (a) the primary-expression is M,
  (b) M is a module, do not raise TypeError
  (c 1) XYZ is the constant-identifier
  (c 2) XYZ is not one of the constants defined in M
  (c 3 i) There are no included modules in M (skip)
  (c 3 ii) N/A
  (c 3 iii) Goto step e of section 11.4.3.1

11.4.3.1
  (e) M is not a class
  (e 1) Search Object for a binding for XYZ

So, according to the spec, it looks like M:XYZ should resolve to the
XYZ in Object and MRI is incorrect to return nil.

Did I miss something?


The real world occurrence of this was:

    return unless defined? RDoc::VERSION # RDoc 1 does not have VERSION
    # code that blows up if RDoc 1 being used...

Which does not currently work in MagLev, since MagLev returns
"constant" for defined?RDoc::VERSION.  One could argue that the proper
test, which works for both classes and modules in all ruby
implementations, is:

    return unless RDoc.const_defined? :VERSION

--
Peter McLain
peter.mclain@gemstone.com

On 3/4/10, Peter McLain <peter.mclain@gemstone.com> wrote:
> I asked this on ruby-talk, but didn't get anywhere.  Someone suggested
> I post over here.

Here's the original thread on ruby-talk:
http://www.ruby-forum.com/topic/205339

Hi Peter,

2010/3/5 Peter McLain <peter.mclain@gemstone.com>:
> 1: Is this difference intended?

I think so.


> 2: If so, what is the rationale?

- constant defined in toplevel belongs to Object class
- scoped constant reference tries to find binding towards ancestors
- ancestors of M does not include Object (M does not inherit from
  Object)
- ancestors of C includes Object (C inherits from Object)


> 3: Should defined?(C::XYZ) behave differently than C.const_defined?(:XYZ) ?

Module#const_defined? in 1.8 checks whether there is binding in the
module itself.  Indeed it is not good semantics.  1.9 is improved
so that the method finds binding in all ancestors by default.


> 4: Is the draft ruby spec correct, or MRI behavior correct, or did I miss
> something in the following:

Good point.  I bet the draft is wrong.
But note that the above answers are just my opinions.

Matz or Shugo Maeda, could you please answer this question?

On Mar 6, 2010, at 8:06 AM, Yusuke ENDOH wrote:
>> 4: Is the draft ruby spec correct, or MRI behavior correct, or did  
>> I miss
>> something in the following:
>
> Good point.  I bet the draft is wrong.
> But note that the above answers are just my opinions.
>
> Matz or Shugo Maeda, could you please answer this question?

   If the draft is wrong, then I will file a comment on it, but I will
wait until we get a definitive answer.

   Thanks

--
Peter McLain
peter.mclain@gemstone.com

Hi,

Sorry for the delay.
I missed this thread.

2010/3/7 Yusuke ENDOH <mame@tsg.ne.jp>:
>> 4: Is the draft ruby spec correct, or MRI behavior correct, or did I miss
>> something in the following:
>
> Good point.  I bet the draft is wrong.

I agree with you, and will fix the draft.
Is it enough to fix 11.4.3.2 c)-3) as follows?

i) Let L be the included module list of C. Search each element of L in
the reverse order for a binding of a constant with name N.
ii) If the binding is found, the value of the
scoped-constant-reference is the value of the binding.
iii) Otherwise, if C is an instance of the class Class, search for a
binding of a constant with name N from Step e of §11.4.3.1.
iv) Otherwise, create a direct instance of the class Symbol with name
N, and let R be that instance. Invoke the method const_missing on C
with R as the only argument.

On Mar 10, 2010, at 6:32 PM, Shugo Maeda wrote:
>
> 2010/3/7 Yusuke ENDOH <mame@tsg.ne.jp>:
>>> 4: Is the draft ruby spec correct, or MRI behavior correct, or did  
>>> I miss
>>> something in the following:
>>
>> Good point.  I bet the draft is wrong.
>
> I agree with you, and will fix the draft.
> Is it enough to fix 11.4.3.2 c)-3) as follows?

   Yes, your wording will fix the spec.  Thanks.

--
Peter McLain
peter.mclain@gemstone.com

Hello,

2010/3/12 Peter McLain <peter.mclain@gemstone.com>:
>  Yes, your wording will fix the spec.  Thanks.

I have fixed it.  Thank you.