Re: Ada package registry?

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

"Tero Koskinen" <tero.koskinen@iki.fi> wrote in message 
news:n92uia$do6$1@loke.gir.dk...
> 2.2.2016, 1.22, Randy Brukardt wrote:
...
>> That wouldn't be an *Ada* repository. It would be a *GNAT*
>> repository! There might be use to such a thing, but call a spade a
>> spade here: don't insult those who use Ada, but don't exclusively use
>> GNAT, by calling such a thing an *Ada* tool.
>
> I think you are forgetting who you are replying right now. :)

Yes, surely. :-)

> Most of my Ada software compiles nicely with other Ada compilers
> (Janus/Ada, ICCAda) and I don't think it would be impossible to
> add Janus/Ada or ICCAda support to the package repository tool.
> You just need to make sure you don't try to compile Ada 2005/2012
> code with Janus/Ada, which supports only Ada 95.

But you are the shining exception to the rule. I think you and one other 
author are the only ones that are currently trying for portable Ada code; 
the vast majority don't seem to care or don't know how to do it or 
something. It would help if there were more tools for enforcing portability, 
as well as some way of expressing assumptions. (A package manager would help 
there, of course.)

[What I mean by expressing assumptions is that it's unlikely that any real 
Ada code would really run anywhere. I've been having fun with Gautier over 
his "unconditional portability" claim, as one only needs a single 
counter-example to disprove the claim. After a few obvious problems with the 
choices of Janus/Ada were worked out, I pointed out that his types wouldn't 
work on our old U2200 compiler (as that was a 1's completement, 36-bit 
machine). He decided quite reasonably not to worry about that, but that 
means in his case, that means his code is "unconditionally portable" so long 
as your target supports 32-bit integers and is 2's complement. That might be 
99% of machines, but its not quite unconditional.]

...
>> For more complex systems that need other libraries, I doubt there is
>>  any sensible solution. Unless you're planning to abandon Ada's
>> portability between implementations -- not of much value, IMHO.
>
> Not sure what sort of complexity you are thinking about. Just iterate
> through the dependency chain and build one dependency at time using
> the selected Ada compiler. When building one package, link the
> dependencies, and be happy.

That's trivial: run the compiler's Make. The issues I was thinking about is 
dependencies on things like OpenSSL. One could manage to automate that sort 
of dependency on one system fairly easily, but to do it on all reasonable 
targets (various Windows versions, Linux distributions, OS X, FreeBSD, etc.) 
is a combinational explosion (given that some sort of Ada compiler 
configuration is also involved - which is different for every compiler and 
often for different targets as well).

That's the problem that other people here were talking about. It probably 
isn't the problem YOU were trying to solve - what you described is perfectly 
reasonable. (I'm dubious that it would be that useful for Ada, since 
compilers already include good build tools, and little more is needed to use 
third-party stuff -- but if it helped determine what is and is not portable, 
that certainly would help the non-GNAT users out there.)

                                                    Randy.