Sun May 10 14:26:08 CEST 2009
eta-reduction on rpn-lambda
Making abstractions with rpn-lambda with a single term at this moment
introduces abstraction overhead. I wonder if this can lead to
memoization problems due to the fact that (macro: m) != (macro m).
To do this purely generative instead of by postprocessing requires a
different approach. Currently the information is lost: what you want
is to fish out the function that is applied, not the application.
Maybe it isn't really a problem: performance wise i don't think it
would matter since the compiler is probably smart enough to eliminate