Re: Anouncing the (somewhat broken) E v0.7.1 release Bill Frantz (frantz@communities.com)
Wed, 11 Nov 1998 18:24:19 -0800

At 11:28 AM 11/11/98 -0800, Mark S. Miller wrote:
>
> static private final String FOO = "foo".intern();
>
> ...
>
> E.send(target, FOO, args...)
>
>for when it's just gotta be fast even under broken jvms, even at some cost
>in maintainability.
>
>What do people think of using ELib in this style? The reason I'm asking?...
>
>
>Should measurements prove it to be a win (as should be expected from other
>dynamic oo language implementations) I'm thinking of introducing a "call
>site caching" mechanism into the E language's implementation. In order for
>the ELib programmer to use it as well, they'd have to use code much like
>the above. Just replace the first line with
>
> static private final Name FOO = new Name("foo");

I would like to see the benchmark results. Three versions need to be measured: 

(1) Broken intern sends
(2) Interned string sends
(3) new Name("foo") sends

I am certainly willing to go to some inconvenience if it results in significantly better performance. (As an assembly programmer, I better be willing. :-)