> Date: Sat, 16 Jan 2021 10:41:49 -0800 > From: Matt Birkholz <m...@birchwood-abbey.net> > > On Fri, 2021-01-15 at 23:12 +0800, Lingyu Zhu wrote: > > 1 ]=> (paint g1 fovnder) > > ;The object 65536/255, passed as the first argument to integer- > >flonum, is not the correct type. > > This makes me wonder about line 69 of arith.scm: > > (define-standard-unary exact->inexact (lambda (x) x) int:->flonum) > > It seems reasonable to apply exact->inexact to a rational number yet > the procedure uses int:->flonum which, reasonably, objects to > arguments that are not integers so... perhaps screw with (broaden the > domain of) int:->flonum with the following hack?
6001/arith.scm seems to be deliberately defined without rational arithmetic. I don't know what the history behind it is, but the code (exact->inexact (/ n-bins ...)) was probably intended to compute a truncated integer quotient. You could probably get the intended effect by writing (quotient n-bins ...) instead. Perhaps this code should omit (declare (usual-integrations)) so the compiler won't replace (/ ...) by the primitive which does support rational arithmetic (by punting to the runtime via the `generic trampoline').
