Re: Unclear error message - please help

[Robert A Duff <bobduff@shell01.TheWorld.com>]

Dennis Lee Bieber <wlfraed@ix.netcom.com> writes:

> On Mon, 10 Oct 2005 08:02:07 +1000, Brian May <bam@snoopy.apana.org.au>
> declaimed the following in comp.lang.ada:
> 
> > 
> > I read the above links but didn't see why a range 0..65535 requires 17
> > bits.
> 
> 	I haven't dug up the reference books, but I seem to recall that, for
> a non-MOD type, the /base/ storage is assumed to be signed. So...
> 0..65535 implies a base of -65535..65535; with a run-time check that no
> negatives are allowed.
> 
> 	Ah... Page 39 of my old reference manual
> 
> """
> A signed_integer_type_definition defines an integer type whose base
> range includes at least the values of the simple_expressions and is
> symmetric about zero, excepting possibly an extra negative value. A
> signed_integer_type_definition also defines a constrained first subtype
> of the type, with a range whose bounds are given by the values of the
> simple_expressions, converted to the type being defined.
> """
> 
> Keywords: base range; symmetric about zero.

Right.  But don't confuse the base range with the memory size of
values/objects of that subtype.

The base range determines how arithmetic works: overflow will not
happen, so long as you stay within that base range.  So, for example,
(X+Y)/2 will not overflow, so long as the value of X+Y is within the
base range.

But the object X might be stored in a smaller number of bits than the
base range requires.  Normally, the compiler will store X in a register
or memory location that matches the base range, for efficiency, but you
can request smaller memory size with all kinds of rep clauses/pragmas.

Anyway, I think the original post showed a [sub]type derived from
Integer, so it will have the same base range as Integer (perhaps 32
bits).

- Bob