Re: Ideas for Ada 200X (Ada.Sockets)

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

"chris.danx" <spamoff.danx@ntlworld.com> writes:

> Larry Kilgallen wrote:
> 
> > Standard, as in required of all compilers ?
> > That would be totally bogus in some embedded environments.
> 
> Can't you say "the provision of sockets is compulsory, except for
> environments where such provision makes no sense"?  i.e. it's not
> optional for compilers on networked systems, but not mandatory on
> embedded targets?

It seems better to say "here's the standard way to do it, but it's
optional".  That's what the SN Annexes are for.  Then implementers
can choose whether to implement it based on the needs of their
customers.  After all, the purpose of standards is to encourage
uniformity -- not to encourage implementers to provide functionality
to their customers (the market is supposed to do the latter).

To say "compulsory except where it doesn't make sense" gets standards
bodies embroiled in judging what "makes sense" in various environments,
and that's not what they are good at.

Of course, there's some of that anyway -- *every* feature in the
standard is optional if it's impossible or impractical to implement
in a given environment, and validation authorities sometimes need
to make judgement calls.  For example, one of our (SofCheck's) embedded
Ada targets does not support file I/O, because our customers don't want
it -- their chips are not attached to disk drives.  That did not prevent
validation.

Another alternative is just to have a de-facto standard for sockets.

I use the GNAT sockets packages, even when I'm using non-GNAT compilers,
and that seems to work OK for me.

- Bob