it should return false but not raise a fatal error at all, but there is no valid reason to return true because you can't access it outside the class

The reason (as I said in the bug report) is a consistency with property_exists and method_exists, see https://3v4l.org/7PYOQ

you still compare different things here and for consistency they all should return false outside of the context $this because it's simply wrong behavior leading to errors like this one which couldn't work beause for the current scope it don't exist

if(property_exists($foo, 'bar')))
{
 $foo->bar = 10;
}
_________________________________

$foo = new Foo;
var_dump(property_exists($foo, 'bar'));
var_dump(method_exists($foo, 'bar'));

defined('Foo::BAR');
_________________________________

I agree with you it would be better if a current scope were taken into account for all three methods.

But you can even see in a changelog of property_exists 
( http://cz2.php.net/manual/en/function.property-exists.php ) that "This function checks the existence of a property independent of accessibility." So I suppose there was some good reason for this behavior and therefore I think "defined" function should behave in the same way.

I believe the closer analogon to defined() on constants is isset() on properties and is_callable() on methods, both of which will take visibility into account. property_exists() and method_exists() are more like reflection methods, from a time where reflection didn't exist yet.

@nikic@php.net: but in no case it has to raise a exception / fatal error because such things are typically used to test if something exists instead blindly call it and then fail - so the current behavior defeats the whole purpose

Yes, obviously. My comment was referring to the discussion whether the return value should be true or false for an inaccessible class constant.

This has been fixed at some point: https://3v4l.org/qUMeI