Re: Ada generics, numerics, and conversions

Re: Ada generics, numerics, and conversions

[Robert I. Eachus]

In article <1993Jul15.223407.3677@software.org> smithd@software.org (Doug Smith
) writes:

  > I could not find a language note, or LRM reference for the
  > following unusual behavior.  Would anyone like to comment
  > on whether this is a compiler problem, or a language
  > feature :-)

    (Code deleted...)

   The results of executing the program are:
    2147483647
   ** MAIN PROGRAM ABANDONED -- EXCEPTION "numeric_error" RAISED

  > Interestingly enough, I could not find a justification for
  > the exception being raised.  Paragraph 4.6(7) says "...The
  > conversion of a real value to an integer type rounds to the
  > nearest integer..."

  > Well, for my machine, 2147483647 is the nearest integer to
  > 1.0/0.0000000000000000001.  However, it seems inconsistent
  > for the generic form to behave differently than the
  > non-generic form.  This can be a sign of an erroneous
  > program, but I don't see it yet.

     First of all, the nearest integer (small i) to
1.0/0.0000000000000000001 is 10000000000000000000, assuming
Float'MANTISSA >= 45.  The nearest value of the type Standard.Integer
is irrelevant.

     Second, 4.6(2) and 4.6(12) are explict about when
CONSTRAINT_ERROR must be raised on a type conversion which is out of
range.  The permissions in 11.6(6) and 11.6(7) do not really apply,
but, especially given the recent ARG discussions on this issue, I
would hesitate to call any action absent a CONSTRAINT_ERROR or others
handler a bug.

     Integer'IMAGE on the other hand is not required to raise an
exception for a value out of range.  In particular 'IMAGE operates on
the base type, so Positive'IMAGE(-3) should print "-3" and
Integer'IMAGE(10000000000000000000) is allowed by 11.6 to print 
" 10000000000000000000".

     NUMERIC_ERROR can be raised almost anyplace you are dealing with
a value outside the range of the base type, so the second result is
acceptable.  It is the first result which should be questioned, since
printing Integer'LAST is never right.

--

					Robert I. Eachus

with Standard_Disclaimer;
use  Standard_Disclaimer;
function Message (Text: in Clever_Ideas) return Better_Ideas is...

Re: Ada generics, numerics, and conversions

[Doug Smith]

> In article <1993Jul15.224245.3907@software.org> smithd@software.org (Doug Smi
th) writes:
> > [example deleted]
> > The results of executing the program are:
> >  2147483647
> > ** MAIN PROGRAM ABANDONED -- EXCEPTION "numeric_error" RAISED
> > 
> > [reference to  4.6(7)]
> > 
> > Well, for my machine, 2147483647 is the nearest integer to
> > 1.0/0.0000000000000000001.  However, it seems inconsistent
> > [...]

> From: pelakh@convex.com (Boris Pelakh)
> 2147483647  is nowhere near 1.0/0.0000000000000000001, its just integer'last
> for a 32 bit representation. I think your second assignment is doing the righ
t
> thing, ie raising numeric_error (to be constraint in 9x). Whereas your generi
c
> is losing it. Look at the assembly, see what it does ...

> From: eachus@spectre.mitre.org (Robert I. Eachus)
>      First of all, the nearest integer (small i) to
> 1.0/0.0000000000000000001 is 10000000000000000000, assuming
> Float'MANTISSA >= 45.  The nearest value of the type Standard.Integer
> is irrelevant.

Thank you.  Yes I had misinterpreted the LRM to mean the nearest
Standard.Integer.

But Robert I. Eachus continues, and suggests that Constraint_Error
is the correct behavior according to the LRM:

>      Second, 4.6(2) and 4.6(12) are explict about when
                   ^3 in my annotated LRM
> CONSTRAINT_ERROR must be raised on a type conversion which is out of
> range.  The permissions in 11.6(6) and 11.6(7) do not really apply,
> but, especially given the recent ARG discussions on this issue, I
> would hesitate to call any action absent a CONSTRAINT_ERROR or others
> handler a bug.
> 
> [Interesting points about 'Image]

Unless there is some more response, the conclusion seems to be that
the compiler is incorrect if it does not raise Constraint_Error
in both cases.  I suspected the first case was incorrect and
appreciate the clear explanations (which will also help when
submitting the bug report.)

Thank you.

Doug Smith