From: Wilson Bilkovich <wilsonb@...>
Date: 2006-12-09T01:23:30+09:00
Subject: Re: Rubinius interview with Evan Phoenix

On 12/8/06, Sam Smoot <ssmoot@gmail.com> wrote:
> Since we're kinda on the topic of implementation I've got a question:
>
> Avi Bryant's article in response to JoelOnSoftware said you could get
> method calls down to a jump and compare.
>
> My impression is that MRI basically uses hashtables for method lookup.
>
> Is it in the cards for Rubinius to receive the sort of optimizations
> Avi was talking about? Is it reasonable to expect that an optimized
> version of Rubinius could get method calls into the same performance
> realm as c# or Java?
>
> This way I suppose C-Extensions would be used more as bridges to
> leverage existing libraries, and not for performance reasons.
>

Yes. One way to do this is to make the common case as fast as
possible, even if that assumption 'breaks' some of the dynamic
features of Ruby. At runtime, you then detect when you've violated
that assumption, trap the error, and re-evaluate the code in a slower
mode.

If you pick targets for that that don't trap very often, you can
supposedly gain back quite a lot of speed.  I believe the JVM has
something like this for dynamic invocations, but I haven't looked at
that code, so I might be making a fool out of myself by saying it.