Re: GNAT GPL is not shareware

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

"Shark8" <OneWingedShark@gmail.com> wrote in message 
news:eFZrw.1020853$No4.1016889@fx19.iad...
> On 09-Jan-15 15:03, Randy Brukardt wrote:
>> And it probably would have problems
>> with macro-expanded generics vs. code-shared generics (again, which ASIS 
>> has
>> in spades) - either you duplicate the definitions, so you no longer have 
>> a
>> 1-to-1 relation of identifiers to source, or you don't, and you have
>> multiple versions of the same thing.
>
> I haven't read anything about the DIANA based systems having trouble with 
> generics/instantiations in my research. -- Do you have some experience or 
> stories about them? Could you elaborate on the problems of ASIS w/ 
> generics?

DIANA certainly wouldn't have any issues, because any individual 
implementation simply does whatever it does, which is fine. The problem with 
ASIS is that is a disaster for portability. ASIS 99 essentially allows the 
implementation to do whatever it does for generics -- so how do you write 
portable code for that? (Ans: you have to write it both ways and use query 
functions. Gag.) The alternative is to require it to emulate a 
macro-expansion. But of course then an ASIS implementation is near 
impossible on a sharing implementation. (After all, if the implementer 
wanted to implement the full nastiness of a macro-expansion implementation, 
they probably would have done that in the first place. And they'll have to 
implement 95% of it in order to properly implement ASIS.)

The other problem with ASIS is that the name "Ada Semantic Interface" is 
nearly a lie; ASIS has very little semantic information in it. It's almost 
exclusively about matching syntactic constructs with their syntactic 
definitions. It does badly with anonymous things for this reason (the worst 
being T'Class, which is 100% implicit), and it makes it very hard to deal 
with visibility and the like. It also means its a bad mapping for any 
implementation that does not keep the syntax around. For instance, in 
Janus/Ada, we eliminate most static matching issues by collapsing subtypes 
whenever possible. So there is never more than one subtype object with each 
static semantics, even if those come from different declarations. (If you 
can't tell the difference from the Ada perspective, why waste space on 
storage?) One can't implement ASIS that way; we'd have to change the 
underlying way subtypes are handled in order to separate the different 
declarations.

A true semantic interface would talk about semantic entities only, the 
associated syntax would not enter into it.

                             Randy.