We're kinda confusing levels here. There are really three viewpoints of an object:
in reply to Re^5: Language features affect style
in thread Language features affect style
I was speaking from the perspective of #2, not #3. From the #2 perspective, for any normal attribute without an explicit private accessor, $!foo and self!foo are largely interchangeable, give or take a sigil. The MOP may or may not make visible the call that implements $!foo, since $!foo is viewed under #2 as fundamentally a reference to the storage. The MOP is free to not implement method !foo at all if it can determine that it is unneeded,
either by analysis or just by being lazy about autoloading one. From the #2 viewpoint, $!foo is just a variable, and method !foo is a (possibly autogenerated) accessor to the variable.
- The external public view.
- The internal private view.
- The hidden MOP view
From the #3 perspective, sure, the fundamental accessor for the slot could be identical to method !foo, but then you're kinda hosed if the user wants to actually define the !foo method as a wrapper around $!foo, which would be an infinite regress if you confuse the user's slot accessor with the MOP's slot accessor. So unlike the situation with $.foo, the $!foo notation is not just syntactic sugar for a private accessor, and a private method must be called as self!foo.
So why would the user ever want to define their own private accessor? Primarily to mediate how much you want to dynamically "untrust" any other classes you've trusted by declaration.