Export instances of procedures

Export instances of procedures

[Gautier]

Hi. GNAT 3.10-3.12 doesn't seem to want to export directly,
i.e. without "renames", instances of procedures out of a package.
Is there an Ada rule behind it ? No explanation found in the Manual.

Example:

1) This works:

     package DMA is
48:    procedure Read_8bit_AutoInit( linearAddressOfBuffer: unsigned_32;
                                     length: integer );
     end DMA;

  package body DMA is
    generic
      dmaCommand: Unsigned_8;
    procedure Program_DMA( linearAddressOfBuffer: unsigned_32;
                           length: integer );

    procedure Program_DMA( linearAddressOfBuffer: unsigned_32;
                           length: integer ) is
...
    end Program_DMA;

    procedure R8AI is new Program_DMA( 16#58# );

    procedure Read_8bit_AutoInit( linearAddressOfBuffer: unsigned_32;
                                  length: integer )
      renames R8AI;
  end DMA;

2) But, if I put: "procedure Read_8bit_AutoInit is new Program_DMA( 16#58# );"
   I get: "Read_8bit_Autoinit" conflicts with declaration at line 48
   and of course: missing body for "Read_8bit_AutoInit" declared at line 48

Strange, isn't it ?

______________________________________________________
Gautier  --  http://members.xoom.com/gdemont/gsoft.htm

Re: Export instances of procedures

[Vincent Marciante]

Gautier wrote:
> 
> Hi. GNAT 3.10-3.12 doesn't seem to want to export directly,
> i.e. without "renames", instances of procedures out of a package.
> Is there an Ada rule behind it ? No explanation found in the Manual.
> 
> Example:
> 
> 1) This works:
> 
>      package DMA is
> 48:    procedure Read_8bit_AutoInit( linearAddressOfBuffer: unsigned_32;
>                                      length: integer );
>      end DMA;

At this point, Read_8bit_AutoInt(...) requires a completion.
A subprogram renaming declaration (RM 8.5.4) can act as a completion but
an instantiation can't.

Other valid completions are: a body, a body stub and an import pragma.
I don't remember where in the RM a list of all valid completions is
listed.

> 
>   package body DMA is
>     generic
>       dmaCommand: Unsigned_8;
>     procedure Program_DMA( linearAddressOfBuffer: unsigned_32;
>                            length: integer );
> 
>     procedure Program_DMA( linearAddressOfBuffer: unsigned_32;
>                            length: integer ) is
> ...
>     end Program_DMA;
> 
>     procedure R8AI is new Program_DMA( 16#58# );
> 
>     procedure Read_8bit_AutoInit( linearAddressOfBuffer: unsigned_32;
>                                   length: integer )
>       renames R8AI;
>   end DMA;
> 
> 2) But, if I put: "procedure Read_8bit_AutoInit is new Program_DMA( 16#58# );"
>    I get: "Read_8bit_Autoinit" conflicts with declaration at line 48
>    and of course: missing body for "Read_8bit_AutoInit" declared at line 48
> 
> Strange, isn't it ?
> 
> ______________________________________________________
> Gautier  --  http://members.xoom.com/gdemont/gsoft.htm

Re: Export instances of procedures

[Gautier]

Vincent Marciante wrote:

> At this point, Read_8bit_AutoInt(...) requires a completion.
> A subprogram renaming declaration (RM 8.5.4) can act as a completion but
> an instantiation can't.

Sorry for insisting on this question, but _why_ cannot the instance
be considered as complete ? Well, the generic body could depend on an
uncompletely defined type, hence the instance too. But the instance
could also be clearly complete, couldn't it ? In my code, it is!
In that case what does the "renames" bring ? Is it purely formal
or does it reflect some indirections (indirect access to the procedure) ?

NB: I have seen in the Manual many topics about completion (3.10.1, 3.11.1,
7.6.1, 9.5.2) and the one that seems the most relevant here, 8.5.4(5) but
I didn't guess the motivations behind these rules...
Maybe some wise one knows them or can decipher the Runes ? ;-)
______________________________________________________
Gautier  --  http://members.xoom.com/gdemont/gsoft.htm

Re: Export instances of procedures

[Robert Dewar]

In article <3905CCCB.C48B9113@maths.unine.ch>,
  Gautier <gautier.demontmollin@maths.unine.ch> wrote:
> Vincent Marciante wrote:
>
> > At this point, Read_8bit_AutoInt(...) requires a completion.
> > A subprogram renaming declaration (RM 8.5.4) can act as a
completion but
> > an instantiation can't.
>
> Sorry for insisting on this question, but _why_ cannot the
instance
> be considered as complete ?

Because that's the rule! It is absolutely clear in the RM that
an instantiation cannot act as a completion. Could the language
be designed to allow this? Sure! Is it designed that way? No!
Why, because this was considered an unnecessary complexity.


Sent via Deja.com http://www.deja.com/
Before you buy.

Re: Export instances of procedures

[Gautier]

Robert Dewar: <<Because that's the rule!>>

OK, OK, thank you! Now I can sleep well...

_____________________________________________
Gautier  --  http://members.xoom.com/gdemont/

Re: Export instances of procedures

[Vincent Marciante]

Gautier wrote:
> 
> Vincent Marciante wrote:
> 
> > At this point, Read_8bit_AutoInt(...) requires a completion.
> > A subprogram renaming declaration (RM 8.5.4) can act as a completion but
> > an instantiation can't.
> 
> Sorry for insisting on this question, but _why_ cannot the instance
> be considered as complete ? Well, the generic body could depend on an
> uncompletely defined type, hence the instance too. But the instance
> could also be clearly complete, couldn't it ? In my code, it is!
> In that case what does the "renames" bring ? Is it purely formal
> or does it reflect some indirections (indirect access to the procedure) ?

I only believe that I know the answer to the last question.  Some
indirection _is_ (/can be?) involved.  The kind of type conformance
called "subtype conformance" is related to "renaming as body".  It lets
the renaming as body be implemented by a compiler as a simple jump
instruction.  I read this description in section 6.2 on page 6-4 of the
Ada 95 Rationale.   

> 
> NB: I have seen in the Manual many topics about completion (3.10.1, 3.11.1,
> 7.6.1, 9.5.2) and the one that seems the most relevant here, 8.5.4(5) but

The Rationale section that I mentioned also refers to the RM Section on
Conformance rules (RM95 6.3.1). 

> I didn't guess the motivations behind these rules...
> Maybe some wise one knows them or can decipher the Runes ? ;-)

Yes, I do not know the underlying reasons, I just remembered from
reading the RM.

> ______________________________________________________
> Gautier  --  http://members.xoom.com/gdemont/gsoft.htm

Re: Export instances of procedures

[Tucker Taft]

Gautier wrote:
> 
> Vincent Marciante wrote:
> 
> > At this point, Read_8bit_AutoInt(...) requires a completion.
> > A subprogram renaming declaration (RM 8.5.4) can act as a completion but
> > an instantiation can't.
> 
> Sorry for insisting on this question, but _why_ cannot the instance
> be considered as complete ?

This was explicitly considered during the Ada 9X design process.
The conclusion was that it might create readability problems.  An
instantiation does not repeat the formal parameter part, so
matching a potentially overloaded spec with the appropriate generic
instantiation might be difficult for the human reader.

I remember there being a rather heated discussion about this issue,
but the readability-over-writability philosophy won the day,
given the availability of using renaming as a body.

> ...
> Maybe some wise one knows them or can decipher the Runes ? ;-)

You might find a discussion in minutes of some Ada 9X design
meeting, if you were able to locate them...

> ______________________________________________________
> Gautier  --  http://members.xoom.com/gdemont/gsoft.htm

-- 
-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: Export instances of procedures

[Jean-Pierre Rosen]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 833 bytes --]


Gautier <gautier.demontmollin@maths.unine.ch> a �crit dans le message :
38FCBA00.992AA7A2@maths.unine.ch...
> Hi. GNAT 3.10-3.12 doesn't seem to want to export directly,
> i.e. without "renames", instances of procedures out of a package.
> Is there an Ada rule behind it ? No explanation found in the Manual.
>
> 2) But, if I put: "procedure Read_8bit_AutoInit is new
Program_DMA( 16#58# );"
>    I get: "Read_8bit_Autoinit" conflicts with declaration at line 48
>    and of course: missing body for "Read_8bit_AutoInit" declared at line
48
>
> Strange, isn't it ?
An instantiation is both a spec AND a body.  Just remove the explicit spec,
and it will work.

--
---------------------------------------------------------
           J-P. Rosen (Rosen.Adalog@wanadoo.fr)
Visit Adalog's web site at http://pro.wanadoo.fr/adalog

Re: Export instances of procedures

[Pascal Obry]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 787 bytes --]


Jean-Pierre Rosen a �crit dans le message <8djpm3$qef$2@wanadoo.fr>...
>An instantiation is both a spec AND a body.  Just remove the explicit spec,
>and it will work.


This is not possible in his example. The spec is in a package spec (to
be exported) and he wants the body to be an instance of a generic (all
this defined in a package body).

There is no other choice than using another name and doing a "rename" I
would say.

Pascal.

--|------------------------------------------------------
--| Pascal Obry                           Team-Ada Member
--| 45, rue Gabriel Peri - 78114 Magny Les Hameaux FRANCE
--|------------------------------------------------------
--|         http://perso.wanadoo.fr/pascal.obry
--|
--| "The best way to travel is by means of imagination"

Re: Export instances of procedures

[Jean-Pierre Rosen]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1079 bytes --]


Pascal Obry <p.obry@wanadoo.fr> a �crit dans le message :
8djth8$du2$1@wanadoo.fr...
>
> Jean-Pierre Rosen a �crit dans le message <8djpm3$qef$2@wanadoo.fr>...
> >An instantiation is both a spec AND a body.  Just remove the explicit
spec,
> >and it will work.
>
>
> This is not possible in his example. The spec is in a package spec (to
> be exported) and he wants the body to be an instance of a generic (all
> this defined in a package body).
>
??? You can instantiate anything in a package spec. And yes, it will have
the effect of putting a body inside a spec. For example, try to instantiate
in a spec a generic package whose body has statements, like Put_Line; the
body will be executed when it is elaborated, i.e. from the middle of the
specification.

When you are a user, you can think that a spec cannot contain a body; those
who have approached a compiler would not make such a statement ;-)

--
---------------------------------------------------------
           J-P. Rosen (Rosen.Adalog@wanadoo.fr)
Visit Adalog's web site at http://pro.wanadoo.fr/adalog

Re: Export instances of procedures

[Pascal Obry]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1284 bytes --]


Jean-Pierre Rosen a �crit dans le message <8dkjcq$skf$1@wanadoo.fr>...
>??? You can instantiate anything in a package spec. And yes, it will have
>the effect of putting a body inside a spec. For example, try to instantiate
>in a spec a generic package whose body has statements, like Put_Line; the
>body will be executed when it is elaborated, i.e. from the middle of the
>specification.
>
>When you are a user, you can think that a spec cannot contain a body; those
>who have approached a compiler would not make such a statement ;-)


You are right but my point was that this was not possible with the *supplied
code*
sample. But as you point it is always possible to have this working but here
the Program_DMA generic unit must be moved to the spec (or another spec
package). But here you have a side effect. The generic unit is now visible
to
users ! Maybe there is a nice way to handle that with some child
libraries...

Pascal.

--|------------------------------------------------------
--| Pascal Obry                           Team-Ada Member
--| 45, rue Gabriel Peri - 78114 Magny Les Hameaux FRANCE
--|------------------------------------------------------
--|         http://perso.wanadoo.fr/pascal.obry
--|
--| "The best way to travel is by means of imagination"

Re: Export instances of procedures

[Gautier]

Pascal Obry:

> You are right but my point was that this was not possible with the *supplied
> code*
> sample. But as you point it is always possible to have this working but here
> the Program_DMA generic unit must be moved to the spec (or another spec
> package). But here you have a side effect. The generic unit is now visible
> to
> users ! Maybe there is a nice way to handle that with some child
> libraries...

You guessed well that the generic "Program_DMA" is rather of internal use!
Anyway in this context putting it into the spec. is not too dramatic : the full
DMA package (spec. & body) has now found its place inside another one.
The simple solution of putting the generic in the spec. had just not hit me...
Thank you (J.-P.) for the idea and debate!
The case appears in other pieces of code :-)

_____________________________________________
Gautier  --  http://members.xoom.com/gdemont/