Re: Odd error with access to derived types

Re: Odd error with access to derived types

[Niklas Holsti]

A. Imran wrote:
> Hello,
> 
> I'm getting a very odd error from my compiler (GNAT 3.15p, also tried
> the latest GCC based GNAT).
> 
> Would anyone point out why the following is / is not valid ada? TIA.

The following works for me with GNAT 3.15p, Debian:

package testp is

    type a_type is tagged record
       foo : natural;
    end record;

end testp;

package testp.b is

     type b_type is new a_type with private;

     type b_access is access all b_type;

private

     type b_type is new a_type with record
         dummy : boolean;
     end record;

end testp.b;


Did I misunderstand which variant of your code fails?

Niklas Holsti
Tidorum Ltd

niklas holsti tidorum fi
       .      @       .

> 
> -- begin testp-b.adb --
> package testp.b is
> 
>     -- TO FAIL: Comment the next line and leave 
>     -- the line after that uncommented.
> 
>     -- XXX 0    
> --  type b_type is new a_type with null record;
>     type b_type is new a_type with private;
> 
>     type b_access is access all b_type;
> 
> private
> 
>     -- TO FAIL: Keep the following uncommented
>     -- XXX 1
>     type b_type is new a_type with record
>         dummy : boolean;
>     end record;
> 
> end testp.b;
> -- end testp-b.adb --
> 
> I'm getting an error that says that the definition of the private type
> (XXX 1)  conflicts with the access type (XXX 0).  Any ideas?
> 
> 
> 
> Here is the rest of the code:
> 
> -- begin testp.adb --
> package testp is
> 
>     task type a_task;
>     
>     type a_type is abstract tagged limited record
>         dummy_task : a_task;
>     end record; 
> 
> end testp;
> -- end testp.adb --
> 
> -- begin test.adb --
> -- Just used for compiling library units
> with testp;
> with testp.b;
> 
> procedure test is
> begin
>     null;
> end test;
> -- end test.adb --
> 
> 

Re: Odd error with access to derived types

[Stephen Leake]

A.Imran <me5@privacy.net> writes:

> Hello,
> 
> I'm getting a very odd error from my compiler (GNAT 3.15p, also tried
> the latest GCC based GNAT).
> 
> Would anyone point out why the following is / is not valid ada? TIA.

I get this:
gnatmake -k -g -O0 -gnatf -gnato -gnatwa -gnatwe -gnatwL -gnatVa -I..  test -largs   -bargs -E -cargs 
gcc -c -I./ -g -O0 -gnatf -gnato -gnatwa -gnatwe -gnatwL -gnatVa -I.. -I- ..\test.adb
test.adb:1:11: warning: unit "b" is not referenced
testp-b.ads:7:10: "b_type" conflicts with declaration at line 6
testp-b.ads:15:10: completion of nonlimited type cannot be limited

you are declaring 'b_type' twice. But perhaps that's not what you
meant. Perhaps you meant "I get an error with the second definition of
b_type". But I'm not going to guess. Please post _exactly_ the code
that generates the error, so we can help.

> 
> -- begin testp-b.adb --
> package testp.b is
> 
>     -- TO FAIL: Comment the next line and leave 
>     -- the line after that uncommented.
> 
>     -- XXX 0    
Is this the line you want uncommented? I don't think so.

> --  type b_type is new a_type with null record;
>     type b_type is new a_type with private;

Did you want both definitions of 'b_type'? I don't think so.

-- 
-- Stephe

Re: Odd error with access to derived types

[A. Imran]

Thank you for looking at the code.  Here is the error message:

~> gnatmake test
gcc -c test.adb
parent-child.ads:10:09: "derived_access" conflicts with declaration at
line 6
gnatmake: "test.adb" compilation error


Relevant lines from parent-child.ads:

     6		type derived_access is access all derived_type;

    10		type derived_type is new base_type with record
    11			dummy : boolean;
    12		end record;

Complete code:

package parent is

	task type a_task;

	type base_type is abstract tagged limited record
		dummy_task : a_task;
	end record;

end parent;



package parent.child is

	type derived_type is new base_type with private;

	type derived_access is access all derived_type;

private

	type derived_type is new base_type with record
		dummy : boolean;
	end record;

end parent.child;


The driver I use to compile the code:

with parent;
with parent.child;

procedure test is
begin
	null;
end test;

Re: Odd error with access to derived types

[Stephen Leake]

A.Imran <me5@privacy.net> writes:

> Thank you for looking at the code.  Here is the error message:
> 
> ~> gnatmake test
> gcc -c test.adb
> parent-child.ads:10:09: "derived_access" conflicts with declaration at
> line 6
> gnatmake: "test.adb" compilation error

That is certainly a confusing error message. I'd call it a compiler
bug; submit this to report@gnat.com.

As for working around it, this compiles:
package parent is

        task type a_task;

        type base_type is abstract tagged limited record
                dummy_task : a_task;
        end record;
        type Task_Access is access Base_Type'Class;

end parent;
package parent.child is

        type derived_type is new base_type with private;

--        type Derived_Access is access derived_type;

private

        type derived_type is new base_type with record
                dummy : boolean;
        end record;

end parent.child;

with parent.child;
procedure test
is
   My_Task : Parent.Task_Access := new Parent.Child.Derived_Type;
begin
        null;
end test;



-- 
-- Stephe

Re: Odd error with access to derived types

[Brian May]

>>>>> "Stephen" == Stephen Leake <stephen_leake@acm.org> writes:

    Stephen> A.Imran <me5@privacy.net> writes:
    >> Thank you for looking at the code.  Here is the error message:
    >> 
    >> ~> gnatmake test gcc -c test.adb parent-child.ads:10:09:
    >> "derived_access" conflicts with declaration at line 6 gnatmake:
    >> "test.adb" compilation error

    Stephen> That is certainly a confusing error message. I'd call it
    Stephen> a compiler bug; submit this to report@gnat.com.

Was the previous code valid? If not, why not?

    Stephen> As for working around it, this compiles: package parent
    Stephen> is

Another three work arounds I found:

a) move definition of derived_access after derived_type (i.e. make it private).

b) make definition of derived_type public.

It would appear that derived_access needs the full definition of
derived_type in order to work because it contains a task. I don't
understand why.

c) make dummy_task an access type to a_task.


My preference would be c), although you could argue you shouldn't need
to use or manipulate access types here.


gcc-3.4 also has the same problem:

[519] [snoopy:bam] ~/abc >gnatmake test
gcc-3.4 -c test.adb
parent-child.ads:9:09: "derived_access" conflicts with declaration at line 5
gnatmake: "test.adb" compilation error
-- 
Brian May <bam@snoopy.apana.org.au>

Re: Odd error with access to derived types

[Stephen Leake]

Brian May <bam@snoopy.apana.org.au> writes:

> >>>>> "Stephen" == Stephen Leake <stephen_leake@acm.org> writes:
> 
>     Stephen> A.Imran <me5@privacy.net> writes:
>     >> Thank you for looking at the code.  Here is the error message:
>     >> 
>     >> ~> gnatmake test gcc -c test.adb parent-child.ads:10:09:
>     >> "derived_access" conflicts with declaration at line 6 gnatmake:
>     >> "test.adb" compilation error
> 
>     Stephen> That is certainly a confusing error message. I'd call it
>     Stephen> a compiler bug; submit this to report@gnat.com.
> 
> Was the previous code valid? If not, why not?

I don't know. 

> <snip>
> 
> It would appear that derived_access needs the full definition of
> derived_type in order to work because it contains a task. I don't
> understand why.

This probably has something to do with "freezing rules", which are
notoriously tricky.

-- 
-- Stephe

Re: Odd error with access to derived types

[A. Imran]

Thank you all for your time and looking at the code.

As a workaround, I had implemented choice (c) in my code,
which was to use an access to task type.

I'll submit this to GNAT.