The most insightful text on the linked page is the first (currently only) comment: this is unnecessary and confusing.

In fact, this is one of the "features" of Ruby that I hate -- it's frequently difficult to know what a particular piece of code does, because the Ruby community's golly-gee fascination with interfaces and dynamic mixins tends to lead to inconsistent behavior in different contexts.

The idea of separating interface and implementation is not new. Even us crusty, abused C++ programmers can override the bracket operator; we just happen to believe that overriding that operator in a way that leads to unexpected behaviour for particular subscript values is a Bad Idea.

The principle of least surprise is an important one in programming; I wish that more people would obey it as a law.

You've missed my point.

I understand that [] is a method in Ruby. In fact, operator[] is also a method name in C++, and can be overloaded, overridden, etc. Ruby did not invent these features. Likewise, in C++, the type of an object and the method name invoked determine unambiguously the method that is called.

The difference -- and this is critical -- is that in the C++ community, there is a strong bias against syntactical games that make code behave inconsistently. Having operator[] behave differently for different subscripts would be a mortal sin in C++ design (not to mention that the type system would probably punish you for it).

We can debate the relative merits of strong/weak typing, but the principle of least surprise is a good one, and should be respected, regardless of what a language may allow you to do (cf. Perl).

First, of course the principle of least surprise is good, but it doesn't apply here. No experienced Ruby programmer would look at 'a[0]' and think - oh, an index into an array. They know that it could mean many things, depending on the receiver, not the subscript value.

Also in Ruby there is no overloading. Overloading means that a different method is chosen depending on the types of the arguments. That means that in C++, it is not true that "the type of an object and the method name invoked determine unambiguously the method that is called".

In practise there is also little overriding - there seems to be no need. In C++ there is a lot more, and because of multiple inheritance, things can get confusing. (Overriding and overloading are two ways in which OO-languages violate the principle of least surprise).

This means that the reasons why using operator overloading is bad in C++ don't (automatically) apply to Ruby.

I'm going to ignore your more pedantic points (e.g. method "signature" vs. method "name", and the silly notion of "overloading" in a dynamically typed language), and focus right on the meat of the argument:

Every "experienced Ruby programmer" may well come to understand that standard expressions fail to mean standard things when used in Ruby; that doesn't make it a good idiom. The problem here isn't that the bracket operator fails to reference into an array -- it's that the bracket operator might behave inconsistently for certain subscript values.

The STL map class is a perfect analogue: rather than implement an operator[] than could fail in certain contexts, the C++ standard demands that map::operator[] must always return an object. This is consistent behavior. Allowing operator[] to fail for subscript N-1 when it works for N? That's inconsistent (and confusing), but apparently, it's the Ruby Way.

> To know which method is called, you need to know the class of the receiver and the method name

Unless someone's been using metaclass trickery, eg.

    array = []
    def array.length; "banana"; end
    array.length # => "banana"
    [].length # => 0

(you can do the same thing with []=, the example is just longer)

I'd argue that this involves changing the class of the receiver (I think that is what actually happens under the covers).

Also, I think this would count as a mortal sin in production code (but a lot of fun nevertheless).

DataMapper, an ActiveRecord replacement where I am one of the coders,uses the same Person[1] syntax. Check it out sometime: http://datamapper.org/

It's been part of Og for a while, perhaps from the beginning.

There seems to be some Ruby communal understanding that the use of [x] should follow Array-ish behavior, but when used with a class constant it's apparent that this is not simply an array, and that some other sort of indexing is going on.