[ruby-core:98307] [Ruby master Bug#16850] Object#hash doesn't behave as documented

Issue #16850 has been updated by jeremyevans0 (Jeremy Evans).


Dan0042 (Daniel DeLorme) wrote in #note-4:
> Would it be feasible to raise some sort of error when trying to redefine Integer#hash ?
> Not just for this case, but I feel there were other cases where ruby bypasses the method call to use the built-in implementation directly. Can't recall exactly atm.
> Making certain methods un-overridable could be a useful signal that "this is not going to work as you expect".

It's probably feasible, but I don't think it is a good idea.  You can do it in pure Ruby if you want:

```ruby
module HashOverrideRaise
  def method_added(method)
    raise "don't override hash" if method == :hash
    super
  end

  def include(*)
    super
    raise "don't override hash" unless instance_method(:hash).owner == Kernel
  end
end

Integer.extend HashOverrideRaise
```

However, this doesn't handle the case in Ana's example, because you are defining Object#hash, and not Integer#hash.

The cases where Ruby bypasses method calls to use built-in code are different than this case.  Ruby will default to use the faster built-in code, but if you define a method, Ruby will recognize it and start calling the method.  This approach is usually referred to as deoptimization.

In this case, I don't think it makes sense to change Ruby's behavior.  There is no reason to override #hash for the cases where Ruby uses a built-in implementation.  So the fact that Ruby does not call #hash in all cases should not be a problem.  If you override #eql?, Ruby will still respect that and treat objects differently:

```ruby
h = {}
s = ''
def s.eql?(_) false end
h[s] = 1
h[s] = 2
h
# => {""=>1, ""=>2}
```

----------------------------------------
Bug #16850: Object#hash doesn't behave as documented
https://bugs.ruby-lang.org/issues/16850#change-85546

* Author: ana06 (Ana Maria Martinez Gomez)
* Status: Closed
* Priority: Normal
* ruby -v: master
* Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN
----------------------------------------
From [Ruby 2.7 Object class documentation](https://ruby-doc.org/core-2.7.0/Object.html#method-i-hash):
>The hash value is used along with eql? by the Hash class to determine if two objects reference the same hash key.

From this I expect that overwritting the `Object#hash` method changes the Hashing algorithm. But this is only the case for some objets. Consider the following code:

```Ruby
class Object
  def hash
    i_dont_exist
  end 
end

class Ana 
end

hash = {}
hash[3] = true
hash[3] = false
puts 'Integers are not using the new hash method'
hash[Ana.new] = true
puts 'this will not be executed because the class Ana uses the new hash method'
```

The output of executing this code is:
```
Integers are not using the new hash method
Traceback (most recent call last):
        1: from a.rb:13:in `<main>'
a.rb:3:in `hash': undefined local variable or method `i_dont_exist' for #<Ana:0x000055626f0b7a30> (NameError)
```

This proves that Integer hash method is not used to determine if two objects reference the same hash key, as said in the documentation. Check the [`any_hash` method](https://github.com/ruby/ruby/blob/master/hash.c#L188) for the other classes affected.

I think that either the behavior should be modified (aiming for a consistency along different classes and following the documentation) or the documentation updated.



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:[e-mail]?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>