Bug #12344: check_funcall_respond_to behavior seems incorrect

• Status: Rejected
• Assignee: -
• Target version: -
• ruby -v: ruby 2.3.1p112 (2016-04-26 revision 54768) [x86_64-linux]
• Backport: 2.1: UNKNOWN, 2.2: UNKNOWN, 2.3: UNKNOWN

[mbedalov (Matt Bedalov)]

Prior to ruby 2.3, and specifically I believe prior to the implementation of this change:

Thu Aug 20 14:13:27 2015 Nobuyoshi Nakada [e-mail]
vm_eval.c (check_funcall_respond_to)
share the behavior with rb_obj_respond_to. [ruby-core:70460] [Bug #11465]

vm_method.c (vm_respond_to)
extract from rb_obj_respond_to and merge r39881.

When a global method_missing was defined, I believe that "check_funcall_respond_to" would not yield that the to_hash method would be defined for objects of type String. In Ruby 2.3, it behaves as if to_hash is defined on a String just due to the fact that we've added a global method_missing. This is hugely problematic when calling a gsub as the second parameter is checked (in str_gsub - string.c:4593) to see if it will act like a hash - i.e. is it a Hash or does it have a to_hash method. It appears that eventually vm_respond_to() calls METHOD_ENTRY_BASIC() which then makes the false statement that the object responds to the to_hash method.

I believe that a possible fix is to have vm_eval.c - check_funcall_respond_to be changed from:

static int
check_funcall_respond_to(rb_thread_t *th, VALUE klass, VALUE recv, ID mid)
{
    return vm_respond_to(th, klass, recv, mid, TRUE);
}

to:

static int
check_funcall_respond_to(rb_thread_t *th, VALUE klass, VALUE recv, ID mid)
{
    return rb_obj_respond_to(recv, mid, TRUE);
}

With the current behavior, the following in my opinion is incorrectly throwing the NameError in the method_missing.

def method_missing(name, *args, &block)
     raise NameError, "#{name} not defined on #{self.inspect}"
end

puts "foo".gsub(/o/, "bar")

[mbedalov (Matt Bedalov)]

• Description updated (diff)

[nobu (Nobuyoshi Nakada)]

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

It's a common mistake, to define method_missing but not respond_to_missing?.