Interfaces.Ada

Interfaces.Ada

[Kyongsuk Pace]

Let me try the same question again.  The organization I work has tons of
Fortran programs and we really need to have the interface capability
between Fortran and Ada.

I tried to use pragma import in GNAT to use some Fortran programs.
However, pragma for Fortran is not implemented yet according to Robert
Dewar.  I got this response from Robert Dewar after I sent the question to
gnat-report!  I read about interfaces.Fortran in one of Mike Feldman's
postings in info-Ada.  Can you (Mike Feldman) tell me where I can get it
and how to use it with GNAT (i.e., does it require any changes before it
will work)?  I am using Sun Sparc.  Thank you.

Susie Pace
=====================================================================
    Kyongsuk Pace
    Fleet Numerical Meteorology and
    Oceanography Center                         pacek@fnoc.navy.mil
    7 Grace Hopper Ave, stop 1                  408-656-4367
    Monterey, CA 93943-5501                     fax:408-656-4489
=====================================================================

Re: Interfaces.Ada

[Michael Feldman]

In article <9409317836.AA783621232@smtpgw.fnoc.navy.mil>,
Kyongsuk Pace  <pacek@FNOC.NAVY.MIL> wrote:
>Let me try the same question again.  The organization I work has tons of
>Fortran programs and we really need to have the interface capability
>between Fortran and Ada.
>
>I tried to use pragma import in GNAT to use some Fortran programs.
>However, pragma for Fortran is not implemented yet according to Robert
>Dewar.  I got this response from Robert Dewar after I sent the question to
>gnat-report!  I read about interfaces.Fortran in one of Mike Feldman's
>postings in info-Ada.  Can you (Mike Feldman) tell me where I can get it
>and how to use it with GNAT (i.e., does it require any changes before it
>will work)?  I am using Sun Sparc.  Thank you.

I think Robert Dewar has commented on the best way (currently) to
interface Fortran programs to GNAT. Interfaces.Fortran, and especially
the data structures interface - which will provide column-mapped
multi-dimensional arrays, etc. simply has not been implemented yet.

I suggested to CLA readers that they look in RM9X 5.0 to see what
these language interfaces will look like in an implementation that
supports them fully. I did not (mean to) imply that GNAT was currently
supporting anything but C.

I am not at all familiar with GCC F77, so I'm not a good one to ask for 
details on how to do the interfacing to GNAT (especially while we wait
for the full Interfaces.Fortran to be done). Robert Dewar has suggested
that the scalar data types are the same in all the GNU languages;
Fortran generally requires parameters to be passed by reference; typically 
this has been done in Ada 83 by using the 'Address attribute to pass an
address from Ada to Fortran. (NOTE: an Ada access type is not
necessaily the same thing, as has been discussed at length here recently).

Handling arrays is something you'll have to look at closely, until there's
a full Interfaces.Fortran to make it easier.

If you are really hurting to get started, maybe the NYU folks can help, but
I'm hesitant to suggest this because they are up to their ears
finishing the compiler. Maybe Robert will comment again on CLA.

BTW - I've always advocated that each organization set up a small group
or even a single individual who becomes the local expert on language
interfacing, and that this group or person be tasked with writing
whatever interfacing "glue" is necessary to make the Ada side of the
interface as clean and clear as possible to the rest of the project.
Try to avoid distributing 'Address calls and suchlike throughout a
big program, for example - go through a project-specific interface 
layer if need be. Do not assume that the performance hit will be
unacceptable.

Is there a CLA reader out there who has experience in interfacing
to Fortran?

Mike Feldman
------------------------------------------------------------------------
Michael B. Feldman -  chair, SIGAda Education Working Group
Professor, Dept. of Electrical Engineering and Computer Science
The George Washington University -  Washington, DC 20052 USA
202-994-5919 (voice) - 202-994-0227 (fax) - mfeldman@seas.gwu.edu (Internet)
------------------------------------------------------------------------
         Ada on the World-Wide Web: http://lglwww.epfl.ch/Ada/
------------------------------------------------------------------------
"Non illegitimi carborundum." (Don't let the bastards grind you down.)
------------------------------------------------------------------------

Re: Interfaces.Ada

[Matt Kennel]

Michael Feldman (mfeldman@seas.gwu.edu) wrote:
: >will work)?  I am using Sun Sparc.  Thank you.

: I think Robert Dewar has commented on the best way (currently) to
: interface Fortran programs to GNAT. Interfaces.Fortran, and especially
: the data structures interface - which will provide column-mapped
: multi-dimensional arrays, etc. simply has not been implemented yet.

I've always wondered about one thing:

  When people make new langauges why do they implement the C storage
  layout for multi-dimensional arrays instead of that of Fortran?  E.g.
  as it seems, Ada9x, Modula-3.

  Given that most matrix algorithms are in Fortran why be backwards?  (and
  ones in C usually use some sort of non-compatible dope-vector thing anyway
  because C doesn't have variable matrix sizes.)

  Why not make the default just like Fortran?  I don't think one way is
  particularly more "natural" than the other so now that Fortran chose one
  convention why not stay with it?

  The only thing I can figure out is that it is a childish attitude
  like "bleah fortran sucks so we'll be different".  I don't believe this
  is true so what is the real reason?  Is it that compelling?

: Mike Feldman

--
-Matt Kennel  		mbk@inls1.ucsd.edu
-Institute for Nonlinear Science, University of California, San Diego
-*** AD: Archive for nonlinear dynamics papers & programs: FTP to
-***     lyapunov.ucsd.edu, username "anonymous".

Re: Interfaces.Ada

[Tucker Taft]

In article <396agd$9bt@network.ucsd.edu>,
Matt Kennel <mbk@inls1.ucsd.edu> wrote:

>Michael Feldman (mfeldman@seas.gwu.edu) wrote:
>: >will work)?  I am using Sun Sparc.  Thank you.
>
>: I think Robert Dewar has commented on the best way (currently) to
>: interface Fortran programs to GNAT. Interfaces.Fortran, and especially
>: the data structures interface - which will provide column-mapped
>: multi-dimensional arrays, etc. simply has not been implemented yet.
>
>I've always wondered about one thing:
>
>  When people make new langauges why do they implement the C storage
>  layout for multi-dimensional arrays instead of that of Fortran?  E.g.
>  as it seems, Ada9x, Modula-3.

Note that Ada doesn't "implement the C storage layout" for
arrays.  Ada 83 doesn't specify the order.  However, implementors
naturally have chosen the layout implied by the way that
array aggregates are written, where the second index varies
fastest as you move through memory.  

For example, in the following array aggregate:

    ((1,2,3), (4,5,6), (7,8,9))

One would naturally expect the second dimension to vary
fastest, by just looking at the syntax for the aggregate.
But to reiterate, there is nothing in Ada 83 that requires
a particular layout.

In Ada 9X, the Convention pragma may be used to specify 
Fortran-style array layout.

Also, it is a bit funny to call the non-Fortran order the
"C storage layout" for multi-dimensional arrays.  C was
certainly not the first language to vary the last dimension
fastest.  Algol 60 used the non-Fortran order (or at least suggested
it in the same sense that Ada's array aggregate notation
suggests it).  So it has a long history, though clearly not
as relevant a history as Fortran's choice to modern numerics.

It is also of course possible to simply reverse the indices
when defining and using a multidimensional array in an implementation
that has the non-Fortran order, so it doesn't seem like
an insurmountable barrier in any case.

>  Given that most matrix algorithms are in Fortran why be backwards?  (and
>  ones in C usually use some sort of non-compatible dope-vector thing anyway
>  because C doesn't have variable matrix sizes.)

Calling the non-Fortran order "backwards" reminds me of the
typical non-Britisher comment about the "backwards" nature
of the British drive-on-the-left rule ;-).

>  Why not make the default just like Fortran?  I don't think one way is
>  particularly more "natural" than the other so now that Fortran chose one
>  convention why not stay with it?

The array aggregate notation suggests that the non-Fortran order
is somewhat more natural in Ada.

>  The only thing I can figure out is that it is a childish attitude
>  like "bleah fortran sucks so we'll be different".  I don't believe this
>  is true so what is the real reason?  Is it that compelling?

It didn't come out of any desire to be different.  I believe it
was simply that Ada comes more from the Algol school of languages
than the Fortran school of languages.

>: Mike Feldman
>
>--
>-Matt Kennel  		mbk@inls1.ucsd.edu
>-Institute for Nonlinear Science, University of California, San Diego
>-*** AD: Archive for nonlinear dynamics papers & programs: FTP to
>-***     lyapunov.ucsd.edu, username "anonymous".

-Tucker Taft  stt@inmet.com
Intermetrics, Inc.
Cambridge, MA  02138

Re: Interfaces.Ada

[Mats Weber]

In article <396agd$9bt@network.ucsd.edu>, mbk@inls1.ucsd.edu (Matt Kennel)
wrote:
>   When people make new langauges why do they implement the C storage
>   layout for multi-dimensional arrays instead of that of Fortran?  E.g.
>   as it seems, Ada9x, Modula-3.

There is a technical reason for this, which appears in Pascal, where

   type Matrix is array [1 .. n, 1 .. m] of Real;

is defined as equivalent to

   type Matrix is array [1 .. n] of array [1 .. m] of Real;

so it is quite natural to chose an ordering different from FORTRAN.

BTW the above is not true in Ada, and the layout of arrays in memory is
not sepcified by the language.

Mats

Re: Interfaces.Ada

[Stephen J Bevan]

[ Since the following isn't Ada specific, I've redirected followups to
  comp.lang.misc - bevan ]

In article <396agd$9bt@network.ucsd.edu> mbk@inls1.ucsd.edu (Matt Kennel) writes:
   ... I've always wondered about one thing:

     When people make new langauges why do they implement the C storage
     layout for multi-dimensional arrays instead of that of Fortran?  E.g.
     as it seems, Ada9x, Modula-3. ...

Perhaps because the people who make new languages with Fortran
incompatible arrays don't necessarily care about Fortran
compatibility?


     Why not make the default just like Fortran?  I don't think one way is
     particularly more "natural" than the other so now that Fortran chose one
     convention why not stay with it?

Perhaps because some do find the row major order more natural.  The
following is taken from the section 2.2.6 "Arrays and Orthogonal Lists"
in Knuth's Fundamental Algorithms pp 296 :-

  The most natural (and most commonly used) way to allocate storage is
  to let the array appear in memory in the ``lexiographic order'' of
  its indices, sometimes called ``row major order''.

Re: Interfaces.Ada

[Robert Dewar]

It is actually the columnwise choice in Fortran that is peculiar. In most
languages it doesn't matter how arrays are stored (you can't normally
tell), and rowwise is somewhat the more natural choice for ENglish speakers
(note how you are reading rows of this message and not columns).

Why Fortran chose column-wise I don't know, but the choice of using
row-wise is well established (all Algol compilers I know of used row
wise), and basically we just have another one of those incompatiilities
which is a nuisance.

At this stage, obviously you can't be compatible with everyone automatically,
but the pragma Convention (Fortran seems a nice approach in Ada 9X (it was
actually my idea to put it in in this general form, so I'm of course biased
in liking the way this is done :-)

Re: Interfaces.Ada

[Michael Feldman]

In article <39agv3$870@schonberg.cs.nyu.edu>,
Robert Dewar <dewar@cs.nyu.edu> wrote:

>At this stage, obviously you can't be compatible with everyone automatically,
>but the pragma Convention (Fortran seems a nice approach in Ada 9X (it was
>actually my idea to put it in in this general form, so I'm of course biased
>in liking the way this is done :-)

I like it too. As you know, I've griped for years that Ada compiler 
implementers missed the boat in not providing something like this in 
their Ada 83 compilers (via an implementation-dependent pragma, say), 
as a way of gaining entree into the Fortran engineering world, which
rejected PL/I, as I recall, in part because of the lack of array 
compatibility. I'm very glad that Ada 9X has focused on the issue.

('Course now with Fortran 90 a real thing, perhaps it's too late.
I hope not. It would be really nice to have - and for engineers to use -
a really smooth Fortran-to-Ada interface. Hope GNAT fills in the 
Convention gap soon.)

Mike Feldman

Re: Interfaces.Ada

[Robert Dewar]

"Fortran generally requires parameters to be passed by reference"

a common myth, but one that has never been true, not even array parameters
are required to be passed by reference in Fortran.

In fact the Ada permission to pass composites either by copy or by
reference is conciously copied from the Fortran standard!

Back to GNAT and Fortran. The only piece of functionality that I see
missing is the columnwise arrays (which has nothing to do with either
pragma Import, or with Interfaces.Fortran, but rather with pragma
Convention applied to an array.