[racket] ~r: formatting 0.0 with 'exponential and #:precision

I came across another thing:

#lang racket

(define x #e100.e-50000)
(rational? x) ; #t
(log x) ; -inf.0
(exact->inexact x); 0.0
(~r x #:notation 'exponential) ; error inexact->exact: no exact
representation
  number: -inf.0

I think this is introduced in line 340 of racket/format.rkt which contains:
(inexact->exact (floor (/ (log N-abs) (log base))))
for the first estimate of the order of magnitude.

As the docs of ~r say "The exactness or inexactness of x does not affect its
formatting.", this behaviour may well be intentional.

It is possible though to avoid the error by subtracting the log of the
denominator from that of the nominator after conversion to exact rational.
Proc order-of-magnitude (in math I think) does this, but is for base 10
only. May be I should adapt order-of-magnitude such as to accept a base
argument?

Best wishes, Jos

> -----Original Message-----
> From: users-bounces at racket-lang.org 
> [mailto:users-bounces at racket-lang.org] On Behalf Of Dmitry Pavlov
> Sent: martes, 07 de mayo de 2013 19:12
> To: users at racket-lang.org
> Subject: [racket] ~r: formatting 0.0 with 'exponential and #:precision
> 
> Hello,
> 
> I just came across the following oddness with formatting
> of real numbers:
> 
>  > (~r 1.0 #:notation 'exponential #:precision '(= 4))
> "1.0000e+00"
> 
> OK
> 
> 
>  > (~r 0.0 #:notation 'exponential #:precision '(= 4))
> "0.e+04"
> 
> ???
> 
> Is there a rational explanation of why did
> the requested precision go into the exponent
> number for 0.0?
> 
> Thanks.
> 
> 
> Best regards,
> 
> Dmitry
> ____________________
>   Racket Users list:
>   http://lists.racket-lang.org/users