Re: Practicalities of Ada for app development

["Randy Brukardt" <randy@rrsoftware.com>]

"Adam Beneschan" <adam@irvine.com> wrote in message 
news:3e8c1c78-31b0-462e-8341-6d661fbc8378@googlegroups.com...
>On Sunday, June 10, 2012 2:36:46 PM UTC-7, Nomen Nescio wrote:
>> "Dmitry A. Kazakov" wrote:
>> >
>> > > Please, why have you defined Code_Point as a modular type?
>> >
>> > Because code point is unsigned. When you declare something signed
>> > you do that because you want negative inverses. Signed integers have
>> > the base type with 0 in the middle of its range.
>>
>> That's no reason to use mod.  You can put boundaries where you want
>> them, integer, or mod.
>
>This isn't quite true because of subtleties of the language.  When you 
>define an  integer "type" with
> range 0..(2**32-1), the *type* you declare is actually a larger type than 
> that.  It has to be at least
> -(2**32-1) .. (2**32-1), which is a type that requires 33 bits (in 
> practice, it will probably be
> -2**64 .. 2**64-1).  You then get a subtype with range 0..(2**32-1).  But 
> some calculations
> have to be done using the base type, which means that using a signed 
> integer type can require
> some extra conversions, and can require additional processing time on a 
> machine that doesn't
> have 64-bit arithmetic instructions.  It probably wouldn't matter much if 
> the type is being used
> to represent a Unicode code point.  But in a more general case, you have 
> to be careful.  It's
> not really accurate to simply say "you can put boundaries where you want".

Right, but the real question here is why is 2**32 being used at all? Unicode 
specifically limits the code points to 2**21 (or something close to that, 
I'm not going to look it up right now); anything beyond that is garbage and 
not allowed in UTF-8 encoding. So the limit of the type ought to be that, 
and (based on the discussion), the numeric operators should be "removed" by 
declaring them abstract. In which case there is no problem using either form 
of declaration.

(If you're keeping the numeric operators, then the form of the declaration 
should be signed, since using mod means you also lose overflow checking --  
and that's almost never what you want.)

                                                  Randy.