Re: Converting an Integer into Human Readable String

[Eli Zaretskii <eliz@gnu.org>]

> From: "Pascal J. Bourguignon" <pjb@informatimago.com>
> Date: Fri, 08 Apr 2011 13:41:30 +0200
> 
> Instead, you could have had some fun, and implement yourself the
> required function.   Well, now you've got only what you merit, here I
> had all the fun, you can have the function:
> 
>     (format-human-readable-big-number 123456789012
>                                       *normal-format*
>                                       *exceptional-format*
>                                        "B" t :binary)
>     --> "  114.978 GiB"

That's very impressive, but at closer look, I found the following
problems with this implementation:

 . It cannot be evaluated in Emacs Lisp without commenting out the
   part between "#|" and "|#".

 . It cannot be evaluated on a 32-bit machine without commenting out
   some parts of the integer test values in the test harness, due to
   integer overflows.

 . When the last argument is :binary, it produces wrong results for
   numbers between 1000*2^N and 1023*2^N.  E.g.,

    (format-human-readable-big-number 1023  "%.1f" "%13.3e" "B" t :binary)
      => "   1.023e+003 B"

   whereas I'd expect "1023 B" instead.

 . It always produces results with a fixed number of digits after the
   decimal, determined by the value of *normal-format*.  Thus, with a
   format of "%.1f" it will always produce 1 digit after the decimal,
   even if that digit is zero:

    (format-human-readable-big-number 900  "%.1f" "%13.3e" "B" t :binary)
      => "900.0 B"

   which is IMO ugly; "ls -lh" produces just "900" in this case.  This
   cannot be remedied by using "%.0f" as the normal format, because
   then it will always round to the nearest integral value, and the
   fractions will never be shown; again, this is different from "ls -lh".