Re: ada commercial options

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

<stefan-lucks@see-the.signature> wrote in message 
news:Pine.LNX.4.64.1104210935070.1184@medsec1.medien.uni-weimar.de...
> On Wed, 20 Apr 2011, Randy Brukardt wrote:
>
>> The vast majority of "libraries" written for GNAT aren't portable enough 
>> to
>> use on other compilers, even if they only use Ada 95 features. In my
>> opinion, a library isn't portable unless it is compiled on at least two
>> different Ada compilers.
>
> Randy, could you describe which portability issues with real-world
> libraries you know? Dedicated low-level stuff is hard to write portably,
> but it seems to me that Ada actually makes it easy to write portable
> general-purpose libraries.

Tero got a lot of them.

...
>  1. Avoid gnat-specific pragmas and attributes. E.g., write
>     "Type'Image(Variable)" instead of "Variable'Img".
>
> (The gnat documentation is quite clear about what is gnat-specific, so
> this should be easy for any library author who cares a little bit about
> portability.)

One thing that will help a lot is the new Ada 2012 profile 
"No_Implementation_Extensions", which will prevent using any 
implementation-defined pragmas, attributes, aspects, withing any such 
packages, and using implementation-defined things allowed in the standard 
library. That gets a lot of it right there.

Some of these exist as restrictions pragmas in Ada 2005 
(No_Implementation_Pragmas and No_Implementation_Attributes).

>  2. Avoid gnat-specific libraries.

This also is caught by the above profile.

What is nice about this is it will flag things like implementation-defined 
containers (probably found under Ada.Containers -- the language encourages 
such additions, but of course they hurt portability).

>  3. Avoid constructs whose behavior is defined as implementation-defined
>     by the Ada standard.

This is the hard one, and it is rarely done. Probably because it is 
virtually impossible to avoid such things (specifically, the range of 
Integer, the order that parameters are evaluated), and it is also very hard 
to detect them.

 >  4. Don't assume a certain Size about system constants, such as
>     Integer'Last.

Right.

I think you got most of the high points. The new profile will help, because 
it will at least eliminate all of the obvious non-portabilities. (I've often 
found uses of 'Unrestricted_Access in code I've tried to use, barf.)

                                      Randy.