From: "Matthias Wächter" <matthias@...>
Date: 2007-10-03T15:46:26+09:00
Subject: Re: hash key performance

On 02.10.2007 16:55, Marcin Mielżyński wrote:
> Matthias Wächter wrote:
> 
>> But here it comes: Using false and, even worse, nil (in either
>> literal or variable form) is way slower (a good 20%) than using
>> integers and symbols, even if they follow the same "immutable
>> object" philosophy of fixed object_ids.
>>
>> Why is that? What's so special about false and nil to have such a
>> degraded performance as a hash key?
>>
> 
> hash.c:rb_any_hash is the answer to your question (funcall('hash') is
> discarded only for Fixnums, Symbols and Strings.

So the first two cases of "switch (TYPE(a))" should be enhanced with
checks for nil, true and false like this:

    switch (TYPE(a)) {
      case T_FIXNUM:
      case T_SYMBOL:
+     case T_NIL:
+     case T_FALSE:
+     case T_TRUE:
	return (int)a;
	break;


Objections?

- Matthias