Interesting and possible buggy behavior in GNAT regarding Annex E pragmas.

Interesting and possible buggy behavior in GNAT regarding Annex E pragmas.

[Shark8]

Hi everyone.
I was experimenting with partitioning-pragmas trying to decide on a
good way to set up my PostScript interpreter so that the parser could
be a remote object possibly servicing multiple clients and I came
across a possible bug that is bothering me. In certain instances using
one of the pragmas will cause the compiler to error-out, as is
expected, but in other cases the compiler does not error out.

The only way I can see this happening is if, by WITH-ing these
children units, the parent units are LIMITED WITH-ed; but that is
counter-intuitive and does not allow for USES (which still compile) of
the parent unit.

I have reduced things to the minimal compilable example and the source
is as follows:

---------------------- Example_1
With Ada.Tags;
Package Example_1 is
   -- Simple package spec which has a Stub function using the
Ada.Tags.Tag-type
   -- in a parameter; compiles.

   Procedure Stub( This_Tag : In Ada.Tags.Tag );

End Example_1;

---------------------- Example_2
With Ada.Tags;
Package Example_2 is
   -- Simple package spec which has a Stub function using the
Ada.Tags.Tag-type
   -- in a parameter and the Pragma Remote_Types has been added;
   --	THIS DOES NOT COMPILE
   --	example_2.ads:1:06: cannot depend on "Tags" (wrong
categorization)


   Pragma Remote_Types;

   Procedure Stub( This_Tag : In Ada.Tags.Tag );

End Example_2;

---------------------- Example_3
With Ada.Tags.Generic_Dispatching_Constructor;
--Use Ada.Tags; -- Uncomment this line and it STILL compiles...
Package Example_3 is
   -- Simple package spec which has a Stub function using the
Ada.Tags.Tag-type
   -- in a parameter and the Pragma Remote_Types has been added; but
the WITH
   -- for that tipe is implicit in the WITH-ing of the Dispatching
Constructor.
   --	THIS DOES NOT COMPILE
   --	example_2.ads:1:06: cannot depend on "Tags" (wrong
categorization)


   Pragma Remote_Types;

   Procedure Stub( This_Tag : In Ada.Tags.Tag );

End Example_3;

---------------------- Example Bodies
With Ada.Text_IO;

Package Body Example_# is
   -- Simple package body which has a Stub function using the
Ada.Tags.Tag-type
   -- in a parameter; this is the same (excepting the name) for all
examples.

   Procedure Stub( This_Tag : In Ada.Tags.Tag ) is
      S : String:= Ada.Tags.External_Tag( This_Tag );
   begin
      Ada.Text_IO.Put_Line( S );
   end Stub;

End Example_#;

---------------------- Experiment.adb

With
Ada.Text_IO,
     Example_1,      -- This compiles, there is nothing special about
it
--     Example_2,    -- This does not compile, as expected, the
Categorization is wrong
     Example_3       -- This *DOES* compile though...
;

with ada.Integer_Text_IO;
Procedure Experiment is

   Use Ada.Text_IO;
Begin

   Put_Line( "Terminating." );

End Experiment;

Re: Interesting and possible buggy behavior in GNAT regarding Annex E pragmas.

[Ludovic Brenta]

Shark8 writes on comp.lang.ada:
> Hi everyone.

> I was experimenting with partitioning-pragmas trying to decide on a
> good way to set up my PostScript interpreter so that the parser could
> be a remote object possibly servicing multiple clients and I came
> across a possible bug that is bothering me. In certain instances using
> one of the pragmas will cause the compiler to error-out, as is
> expected, but in other cases the compiler does not error out.
>
> The only way I can see this happening is if, by WITH-ing these
> children units, the parent units are LIMITED WITH-ed; but that is
> counter-intuitive and does not allow for USES (which still compile) of
> the parent unit.
>
> I have reduced things to the minimal compilable example and the source
> is as follows:

I've been able to reproduce this with gnat-4.4 (=4.4.6-1) on Debian
unstable.  This looks like a genuine bug to me; please report it on
http://gcc.gnu.org/bugzilla/.  What version of GNAT are you using?

-- 
Ludovic Brenta.

Re: Interesting and possible buggy behavior in GNAT regarding Annex E pragmas.

[Michael R]

Hi,

I've noticed the same behaviour on Ubuntu 11.04 (recently upgraded from 10.10 where I was using the AdaCore distribution).  For AUnit, I worked around it by replacing "gprbuild" with "gnat make" and "gprclean" with "gnat clean" in it's Makefile's.

Take care,
Michael.

Re: Interesting and possible buggy behavior in GNAT regarding Annex E pragmas.

[Stephen Leake]

Michael R <michael@zanyblue.com> writes:

> I've noticed the same behaviour on Ubuntu 11.04 (recently upgraded
> from 10.10 where I was using the AdaCore distribution). For AUnit, I
> worked around it by replacing "gprbuild" with "gnat make" and
> "gprclean" with "gnat clean" in it's Makefile's.

As Ludovic pointed out, both AUnit and gprbuild are now Debian packages,
making installation even easier.

-- 
-- Stephe

Re: Interesting and possible buggy behavior in GNAT regarding Annex E pragmas.

[Shark8]

On May 25, 3:21 pm, Ludovic Brenta <ludo...@ludovic-brenta.org> wrote:
> Shark8 writes on comp.lang.ada:
>
> > Hi everyone.
> > I was experimenting with partitioning-pragmas trying to decide on a
> > good way to set up my PostScript interpreter so that the parser could
> > be a remote object possibly servicing multiple clients and I came
> > across a possible bug that is bothering me. In certain instances using
> > one of the pragmas will cause the compiler to error-out, as is
> > expected, but in other cases the compiler does not error out.
>
> > The only way I can see this happening is if, by WITH-ing these
> > children units, the parent units are LIMITED WITH-ed; but that is
> > counter-intuitive and does not allow for USES (which still compile) of
> > the parent unit.
>
> > I have reduced things to the minimal compilable example and the source
> > is as follows:
>
> I've been able to reproduce this with gnat-4.4 (=4.4.6-1) on Debian
> unstable.  This looks like a genuine bug to me; please report it onhttp://gcc.gnu.org/bugzilla/.  What version of GNAT are you using?
>
> --
> Ludovic Brenta.

I'm using 4.4