bug#26106: Defining a method named '-' with one parameter

[Andy Wingo]

On Wed 15 Mar 2017 14:35, Alejandro Sanchez <address@hidden> writes:

> If I define a method named ‘-‘ which only takes in one parameter, the 
> expression ‘(- v)’ gets rewritten to ‘(- 0 v)’. Here is a minimal example:
>
>       (use-modules (oop goops))
>       
>       (define-class <vector2> ()
>         (x #:init-value 0 #:getter get-x #:init-keyword #:x)
>         (y #:init-value 0 #:getter get-y #:init-keyword #:y))
>       
>       (define-method (* (n <number>) (v <vector2>))
>         (make <vector2> #:x (* n (get-x v)) #:y (* n (get-y v))))
>       
>       (define-method (- (v <vector2>))
>         (* -1 v))
>       
>       (define v (make <vector2> #:x 1 #:y 2))
>       (* -1 v)  ; Works fine
>       (- v)  ; Throws error
>
> Here is the error message:
>
>       scheme@(guile-user)> (- v)
>       ERROR: In procedure scm-error:
>       ERROR: No applicable method for #<<generic> - (2)> in call (- 0 
> #<<vector2> 10a8e4020>)
>       
>       Entering a new prompt.  Type `,bt' for a backtrace or `,q' to continue.
>       scheme@(guile-user) [1]> ,bt
>       In current input:
>            23:0  2 (_)
>       In oop/goops.scm:
>          1438:4  1 (cache-miss 0 #<<vector2> 10a8e4020>)
>       In unknown file:
>                  0 (scm-error goops-error #f "No applicable method for ~S in 
> call ~S" (#<<generic> - (2)> (- 0 #<<vec…>)) #)

Is (- x) equivalent to (* x -1) ?

Right now there are a few things happening.  The "primitive expansion"
phase in an early part of the compiler turns (- x) to (- 0 x), where
obviously it should not be doing that.  But can it turn it into (* x -1)
?  Note that somewhat confusingly, a later part of the compiler that can
detect when X is a real number will undo that transformation, turning it
to (- 0 x) when X is real.  So that sounds OK from an optimization point
of view but is the (* x -1) tranformation correct from the math point of
view?

Andy