From: murphy <murphy@...>
Date: 2007-07-07T20:05:16+09:00
Subject: Re: Optimizing Symbol#to_proc

> I don't what the "old Ruby version" was, but I tried to implement mine and 
> to my surprise my custom version is *much* faster. Does anybody know why ?
Yes. Your version doesn't handle multiple arguments like the Ruby 1.9
version does. enjoy Haskell-style folding:

  [1, 2, 3].inject(&:+)  #=> 6

This ongoing discussion about a Rails ticket was the source of my
question: http://dev.rubyonrails.org/ticket/8818
lifofifo provides benchmarks for Ruby 1.8/1.9 and some versions of
Symbol#to_proc there. To sum it up, it seems that the arguments in the
call are the key, not whether it is implemented in C or Ruby.

- calling a proc with one parameter is fast
- calling a proc with a catch-all parameter is slow

So Symbol#to_proc could be much faster for the common case (one
argument). But cutting the power of a feature only for performance
reasons is sad.

My question to those who understand the core and the new VM better than
me is:

* Is it possible to handle :&x faster without restricting it to one
  argument?

It wouldn't be very nice to present a great new addition to the Ruby
slang and warn about it being slow at the same time ;)

[murphy]