On Thu, 14 Feb 2002, Tim Rowledge wrote:

> Bob Arning  <arning at charm.net> is claimed by the authorities to have written:
>
> > You could change your compiler to emit normal message send bytecodes for any of these and then you would get the ability to change the implementation in squeak, at the expense of things being slower and taking more memory

> There is a 'fix' for this to be considered for VI4. It just changes the
> special bytecode to do a normal send of #class. No evaluation of any
> performance change has yet been done.

Actually, during analysis, #species is surprisingly expensive. By
removing the check for 'self species == other species' in
MethodReference>>=, I got several percent speedup for fulltext indexing of
methods.

I thought the change was dodgy, but, it is a noticable improvement.

Scott