Compiler implementation of speciallized needs annexes.

Compiler implementation of speciallized needs annexes.

[Robert T. Sagris]

I was wondering why only GNAT implements all of the specialized needs 
annex's. All of the other vendors I looked into implement A, B
as they are required. C and D are also generally implemented by
everyone. Rational also implements G.

Is this do to a lack of customer demand for these features or is it 
something else?

I was just wondering

Robbi Sagris

Re: Compiler implementation of speciallized needs annexes.

[Tucker Taft]

Robert T. Sagris wrote:
> 
> I was wondering why only GNAT implements all of the specialized needs
> annex's. All of the other vendors I looked into implement A, B
> as they are required. C and D are also generally implemented by
> everyone. Rational also implements G.
> 
> Is this do to a lack of customer demand for these features or is it
> something else?

Annexes C and D provide functionality in the real-time and
system programming domain, which almost all user's of cross-compilers
need.  The other annexes are more "specialized," and as such,
a critical mass of demand hasn't yet built up for them sufficient
to entice other vendors to implement them.  However, with the
increasing interest in distributed component-based systems, even
among the real-time crowd, I would expect at least the distributed
annex to become more widely implemented.  On the other hand,
Ada/Corba is a viable alternative (e.g. ORBExpress/Ada from Objective
Interface Systems), so that may have reduced the pressure for
the distributed annex.

I would also suspect that the safety-critical annex will be or
is already supported by a number of vendors, even if they
haven't validated against the Annex H tests.
> 
> I was just wondering
> 
> Robbi Sagris

-- 
-Tucker Taft   stt@averstar.com   http://www.averstar.com/~stt/
Technical Director, Distributed IT Solutions  (www.averstar.com/tools)
AverStar (formerly Intermetrics, Inc.)   Burlington, MA  USA

Re: Compiler implementation of speciallized needs annexes.

[Samuel T. Harris]

Tucker Taft wrote:
> 
> Annexes C and D provide functionality in the real-time and
> system programming domain, which almost all user's of cross-compilers
> need.  The other annexes are more "specialized," and as such,
> a critical mass of demand hasn't yet built up for them sufficient
> to entice other vendors to implement them.  However, with the
> increasing interest in distributed component-based systems, even
> among the real-time crowd, I would expect at least the distributed
> annex to become more widely implemented.  On the other hand,
> Ada/Corba is a viable alternative (e.g. ORBExpress/Ada from Objective
> Interface Systems), so that may have reduced the pressure for
> the distributed annex.
> 

I find the CORBA stuff to be daunting indeed. Base on many
conversations with development folks, I find speaking to the
distributed systems annex is much more natural then speaking
about CORBA. CORBA is big and it is not Ada so I have to do
alot of explaining to even begin to have a meaningful
conversation. I don't have this problem when discussing
the distributed systems annex which nicely maps distributed
semantics to the language.

I half expect a compiler vendor to implement the distributed
system annex with CORBA. Given such a beast, I get the best
of both worlds and the distributed systems annex provides
a "natural" intermediary step to getting folks to take
deeper steps into CORBA. From my limited knowledge of the
specifics of CORBA, I don't think this would be a major
effort. If this is true, then any vendor should be able
to support Ada's distributed systems without a huge
investment since CORBA solves many of the nitty-gritty
implementation problems.

-- 
Samuel T. Harris, Principal Engineer
Raytheon, Scientific and Technical Systems
"If you can make it, We can fake it!"

Re: Compiler implementation of speciallized needs annexes.

[robert_dewar]

In article <36D44E24.E8888218@hso.link.com>,
  "Samuel T. Harris" <sam_harris@hso.link.com> wrote:
> I half expect a compiler vendor to implement the
> distributed
> system annex with CORBA. Given such a beast, I get the
> best
> of both worlds and the distributed systems annex provides
> a "natural" intermediary step to getting folks to take
> deeper steps into CORBA. From my limited knowledge of the
> specifics of CORBA, I don't think this would be a major
> effort. If this is true, then any vendor should be able
> to support Ada's distributed systems without a huge
> investment since CORBA solves many of the nitty-gritty
> implementation problems.

It solves a very small part, and rather easy part, of the
whole picture. Note that annex E requires pretty major
compiler work to generate the required stubs. The use
of CORBA may simplify parts of the PCS, but it also
complicates other parts of it. Not at all clear that it
would be a significant help at all in the implementation
effort.

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

Re: Compiler implementation of speciallized needs annexes.

[JP Thornley]

In article: <36D43E60.9231A20@averstar.com>  Tucker Taft 
<stt@averstar.com> writes:
> I would also suspect that the safety-critical annex will be or
> is already supported by a number of vendors, even if they
> haven't validated against the Annex H tests.

Ummm, is there a smiley missing off there? I'm sure that an early 
meeting of the Annex H Rapporteur (sp?) Group (about three years ago?) 
decided that it wasn't possible to define any sensible tests for the 
Annex H facilities.

Phil Thornley.

-- 
------------------------------------------------------------------------
| JP Thornley    EMail jpt@diphi.demon.co.uk                           |
|                      phil.thornley@acm.org                           |
------------------------------------------------------------------------

Re: Compiler implementation of speciallized needs annexes.

[dewar]

In article <727154320wnr@diphi.demon.co.uk>,
  jpt@diphi.demon.co.uk wrote:
> Ummm, is there a smiley missing off there? I'm sure that
> an early  meeting of the Annex H Rapporteur (sp?) Group
> (about three years ago?)  decided that it wasn't possible
> to define any sensible tests for the
> Annex H facilities.
>
> Phil Thornley.

No need to guess! Go look at the ACVC tests. Yes, of course
there are tests for annex H! It is certainly true, as it is
true for all parts of the language, that not all
requirements are tested, and indeed, it is also true that
Annex H has a rather high proportion of untestable stuff,
but that does not mean it is ALL untestable.

Indeed, the tests cannot be entirely trivial, since as
people have noticed that many Ada 95 compilers cannot pass
these tests, and do not claim annex H conformance as a
result!

Robert Dewar
Ada Core Technologies

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

Re: Compiler implementation of speciallized needs annexes.

[dewar]

In article
<36D3A1EF.E7CA2A8C@physics.BLAH.purdue.BLAH.edu>,
  "Robert T. Sagris" <robs@physics.BLAH.purdue.BLAH.edu>
> wrote:
> I was wondering why only GNAT implements all of the
> specialized needs annex's. All of the other vendors I
> looked into implement A, B as they are required. C and D
> are also generally implemented by everyone. Rational also
> implements G.

Actually you should check VSR's carefully, a lot of
compilers do NOT fully implement annex D (for example,
as far as we know, GNAT is the only compiler that fully
implements annex D on VxWorks, including a full
implementation of FIFO_Within_Priorities dispatching
policy.)

> Is this do to a lack of customer demand for these
> features or is it something else?

It is not so much a "lack of customer demand". Rather
it is an issue of how much customer demand there is
compared to the cost of implementation.

Implementing annex E for instance is a very large task,
and one that may well not be justified by customer demand.
GNAT certainly has a number of customers for annex E, but
if the decision had been a purely commercial one, who
knows?

In fact we decided very early on in the GNAT project that
we would implement all the annexes. This decision was made
even before ACT existed, and was part of the commitment of
the GNAT project to providing a *complete* freely available
Ada 95 system.

Since our first validation was for all the annexes, we
decided that we would go for 100% validations on all
targets. That has not always been easy, but it makes for
a consistent guarantee of completeness. Of course not every
annex is required in every situation, and that is why it is
perfectly practical to use compilers that lack annex
features that you do not need. This was after all the whole
point of defining optional special needs annexes in the
first place. For example, the Intermetrics compiler for the
Patriot 2 certainly does NOT need the Information Systems
annex (Patriot missiles are not in the habit of performing
payroll computations as they fly through the air :-)

That being said, it would definitely be nice to see some
other compilers implement some of the "missing" annexes.
For one thing, it would be interesting to see if different
compilers implementing Annex E could really communicate :-)

Most certainly every vendor is driven by customer demand,
so if you see something that is not implemented, then
it is definitely likely to mean that>

> I was just wondering
>
> Robbi Sagris
>

Robert Dewar
Ada Core Technologies

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

Re: Compiler implementation of speciallized needs annexes.

[dennison]

In article <7b2mpq$194$1@nnrp1.dejanews.com>,
  dewar@gnat.com wrote:

> For one thing, it would be interesting to see if different
> compilers implementing Annex E could really communicate :-)

Are they supposed to be able to? That would indeed be very cool, but I didn't
get that impression reading the annex myself.

T.E.D.

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own    

Re: Compiler implementation of speciallized needs annexes.

[Samuel Tardieu]

>>>>> "TED" == dennison  <dennison@telepath.com> writes:

TED> Are they supposed to be able to? That would indeed be very cool,
TED> but I didn't get that impression reading the annex myself.

Well, nothing in the Annex requires (or even encourages!) this, but
other compilers could chose to use GNAT's PCS (Partition Communication 
Subsystem, a.k.a the runtime library) in a compatible way.

  Sam
-- 
Samuel Tardieu -- sam@ada.eu.org

Re: Compiler implementation of speciallized needs annexes.

[Richard D Riehle]

In article <7b2mpq$194$1@nnrp1.dejanews.com>,
	dewar@gnat.com wrote:

>
>In fact we decided very early on in the GNAT project that
>we would implement all the annexes. This decision was made
>even before ACT existed, and was part of the commitment of
>the GNAT project to providing a *complete* freely available
>Ada 95 system.

 To paraphrase another famous line from United States politics,

      "Read my lips, no new syntax"

 Reusable Annexes. Now that would be a nice concept since there is
 no new syntax in the Annexes.  And GNAT versions are largely coded
 in Ada, are they not?

 Perhaps, in the spirit of the FSF, other compiler publishers
 could simply incorporate the GNAT annexes into their product and
 avoid the duplication associated with more implementations.


 Richard Riehle
 www.adaworks.com
 

Re: Compiler implementation of speciallized needs annexes.

[Larry Kilgallen]

In article <7belqo$rl7@dfw-ixnews12.ix.netcom.com>, Richard D Riehle <laoXhai@ix.netcom.com> writes:

>  Perhaps, in the spirit of the FSF, other compiler publishers
>  could simply incorporate the GNAT annexes into their product and
>  avoid the duplication associated with more implementations.

Legal issues aside, I believe multiple implementations provide for a
robust Ada climate.  Even if an annex item _can_ be implemented in
Ada, some implementation might be able to do "better" with their
own version.

Larry Kilgallen

Re: Compiler implementation of speciallized needs annexes.

[dewar]

In article <7belqo$rl7@dfw-ixnews12.ix.netcom.com>,
  Richard D Riehle <laoXhai@ix.netcom.com> wrote:
>  Perhaps, in the spirit of the FSF, other compiler
>  publishers could simply incorporate the GNAT annexes
>  into their product and avoid the duplication associated
>  with more implementations.

To the extent that this is just a matter of acquiring
copies of GNAT run time routines, this is fine, and indeed
for a while at least Aonix was using some of the GNAT
routines in their distribution. Providing the copyright
is properly respected, and the license conditions followed
this usage is welcome.

However, it is unfortunately naive to suppose that this
gets you very far. We have often heard of sales people for
another Ada vendor claiming that it is no problem to have
the System Information Annex, since you can just use the
GNAT routines, but this is quite bogus. The GNAT
implentation depends on some critical intrinsic routines
in the compiler, and also on the full support of 18 digit
decimal fixed-point. These are not run-time library issues.

For the real-time annex, of course the GNAT run-time
routines can be used, but this would require major compiler
changes, to the point where it could hardly make sense.

As for the distribution annex, sure you can use the GNAT
PCS, but there is huge front end work to be done.

In short, Richard's suggestion does not lead to much
simplification in the hard task of implementing all the
Annexes in their entirety in the context of a compiler
other than GNAT.

Robert Dewar
Ada Core Technologies

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own