From: Gregory Brown <gregory.t.brown@...>
Date: 2009-05-29T02:50:24+09:00
Subject: [ruby-core:23593] Defining #name= at the class level

Hi folks,

I'd like to just confirm from Matz or one of the core Ruby folks that
defining #name at the class level should be an undefined behavior.

Almost every implementation I've tried allows the following code to
work (MRI, YARV, JRuby, MacRuby):

>> @class = Class.new
=> #<Class:0x267690>
>> @class.instance_eval { class << self; attr_accessor :name; end }
=> nil
>> @class.name = "My Special Class"
=> "My Special Class"
>> @class.name
=> "My Special Class

However, Rubinius fails, like this:

>> @class = Class.new
=> #<Class:0x13c>
>> @class.instance_eval { class << self; attr_accessor :name; end }
=> nil
>> @class.name = "Foo"
TypeError:   Tried to use object of type String (48) as type Symbol (49)
   from Object# (__eval__) at (irb):3

Although this may seem esoteric, I ran into the issue because the
test/spec library uses this to assign human names to anonymous classes
created in unit tests.    I am curious whether this behavior should be
considered:

1) Defined to work as MRI and YARV, in which case Rubinius would not
be compatible for this feature
2) Undefined, in which case, test/spec should be patched to not use
#name=, as it cannot reliably work across implementations
3) Forbidden.  In this case, it'd be nice to see the other
implementations fail as Rubinius does.

I am guessing this is undefined behavior, no different from changing
the way any other existing Ruby code works via monkeypatching.   If
that's the case, I'll just patch test/spec and go away happy.
However, if it's one of the other options, I'd like to know.

-greg

-- 
BOOK:  http://rubybestpractices.com
TECH: http://blog.majesticseacreature.com
NON-TECH: http://metametta.blogspot.com