Preferably, the dispatch should be done as much as possible during compile time such that it takes zero time during runtime.
On Fri, Apr 30, 2021 at 4:50 PM Mikael Djurfeldt <mik...@djurfeldt.com> wrote: > Generic method dispatch is *supposed* to be fast. It was fast once upon a > time. We should fix that. > > On Fri, Apr 30, 2021 at 2:19 PM Stefan Israelsson Tampe < > stefan.ita...@gmail.com> wrote: > >> If performance is important, a goops solution can be slow in vector-ref >> and vector-set! operations due >> to two reasons. (I have pounder an implementation of resizable python >> lists and here is my tips), >> >> 1. slot-ref/slot-set! is slow (I try to fix this using the much more >> difficult struct-ref/struct-set!) >> 2. generic-method-dispatch is slow, I try to make a wrapper function in >> which we short cut for >> python/sheme internal types and if they do not match use the >> generic method. >> >> >> On Fri, Apr 30, 2021 at 1:57 PM Mikael Djurfeldt <mik...@djurfeldt.com> >> wrote: >> >>> On Fri, Apr 30, 2021 at 1:11 AM Damien Mattei <damien.mat...@gmail.com> >>> wrote: >>> >>> > for example in the doc there is: >>> > (define-class <my-complex> (<number>) r i #:name "Complex") >>> > >>> > seems <number> superclass is of no use >>> > >>> >>> Well, it certainly *is* of use in the sense that methods operating on >>> <number> will immediately start to also accept <my-complex> as an >>> argument. >>> That might seem worrisome but is not if there is some agreement on which >>> operations should be implemented for all numbers. So, if you had >>> previously >>> written an algorithm which operates on numbers, there's now a good chance >>> that it would also work for <my-complex> objects. >>> >>
