Version 1.11 of ai05s/ai05-0215-1.txt

!topic Missing rules in Implemented definition?
!reference 9.5(10-18)
!from Adam Beneschan 10-03-26
!discussion


(1) I don't see any rule that says that you can't write two
    Implemented pragmas for the same subprogram.  (If the RM said that
    Implemented were a representation pragma, as it does for
    No_Return, I think this would follow automatically; but I don't
    think it says that anywhere.)

(2) The pragma doesn't make sense unless Implemented pragmas
    explicitly specified for subprograms apply to inherited
    subprograms.  However, I haven't found anywhere in the RM that
    says that pragmas that apply to subprograms automatically apply to
    inherited subprograms.  Nothing in 3.4 (which is the main place
    subprogram inheritance is discussed) says anything about
    subprograms inheriting anything other than the name and profile.
    In 3.9.2(10), it appears that it was deemed necessary to say
    explicitly that the convention was inherited for some inherited
    subprograms.  So I think that somewhere, either 9.5 should say
    that if Implemented applies to a subprogram it applies to an
    inherited subprogram (and 6.5.1 needs to say something similar
    about No_Return), or there needs to be a more general statement
    about what properties or characteristics are inherited by an
    inherited subprogram.

****************************************************************

!topic Errors in AI05-0030-2
!reference 9.5.4
!from Adam Beneschan 10-03-30
!discussion

There appear to be some errors in the changes made for AI05-0030-2.

9.5.4(3/3): "The procedure_or_entry_name of a requeue_statement shall resolve
to denote a procedure or an entry (the requeue target) that either has no
parameters, or that has a profile that is type conformant (see 6.3.1) with
the profile of the innermost enclosing entry_body or accept_statement."

First of all, I note that in the online version
(www.adaic.org/standards/1zaarm/html/AA-9-5-4.html), the paragraph didn't
get updated correctly; it says "denote an entry" rather than "denote a
procedure or entry" which of course defeats the whole purpose.  But this
error isn't worth changing since the whole paragraph is wrong anyway.  If
the procedure_or_entry_name denotes a procedure, doesn't it have to denote
a procedure whose first parameter is a synchronized interface type (or access-to-)?
If that's the case, than it certainly can't resolve to a procedure with
no parameters.
Also, if it's a procedure, its profile can't be type conformant with that
of the enclosing entry; rather, isn't it the procedure's *prefixed view
profile* that has to be type conformant?  So I think this just needs to
be redone.  I'm not sure exactly how, since I don't know what you want to
put in a Name Resolution Rule.

9.5.4(5/3) has the same problem: "If the requeue target has parameters, then
its profile shall be subtype conformant with the profile of then innermost
enclosing callable construct"---again, if the requeue target is a procedure,
the prefixed view profile is what has to be subtype conformant.  Also, if
the requeue target is a procedure with *one* parameter, then this rule
wouldn't apply---the one parameter is the controlling synchronized interface
parameter, but otherwise it should be treated like a parameterless entry.
So maybe this needs to be something like

    If the requeue target is an entry that has parameters, then its
    profile shall be subtype conformant with the profile of the
    innermost enclosing callable construct.  If the requeue target is
    a prefixed view of primitive subprogram [procedure] of a
    synchronized interface, and if the subprogram has more than one
    parameter, then its prefixed view profile shall be subtype
    conformant with the profile of the innermost enclosing callable
    construct.

****************************************************************

From: Bob Duff
Sent: Sunday, January 23, 2011  2:35 PM

...
>     * AI05-0215-1 Suggest a better aspect name than "Implemented" [Nov.
> phone meeting]

I don't see anything about this in those minutes.
But anyway, here's my suggested name changes (for the pragma and it allowed
arguments):

    Implemented --> Synchronized_Implementation
    By_Entry --> Entry
    By_Protected_Procedure --> Protected_Procedure
    By_Any --> Any

Or maybe just change the pragma name.

****************************************************************

From: John Barnes
Sent: Sunday, January 23, 2011  4:15 PM

Seems preferable as it is to me.

****************************************************************

From: Robert Dewar
Sent: Sunday, January 23, 2011  5:37 PM

>      Implemented -->  Synchronized_Implementation

I find this much too long

****************************************************************

From: Bob Duff
Sent: Sunday, January 23, 2011  5:46 PM

Do you have a better idea?

Does anybody remember why we wanted to change this name?
I think it's because (1) "Implemented" is way to general a term for what this
thing does, and (2) it doesn't read well as an aspect.  But I don't really
remember the discussion. And I'm still ambivalent about shoe-horning everything
under the Sun into "aspect" syntax.  The minutes don't clue me in.

Anyway, I don't buy "too long", because it's rarely used.
It will increase the total size of all Ada programs world wide by perhaps
0.000_001%".  ;-)

****************************************************************

From: Robert Dewar
Sent: Sunday, January 23, 2011  5:52 PM

> Does anybody remember why we wanted to change this name?
> I think it's because (1) "Implemented" is way to general a term for
> what this thing does, and (2) it doesn't read well as an aspect.  But
> I don't really remember the discussion.
> And I'm still ambivalent about shoe-horning everything under the Sun
> into "aspect" syntax.  The minutes don't clue me in.

I think the current proposal is better than your proposed substitute.

> Anyway, I don't buy "too long", because it's rarely used.
> It will increase the total size of all Ada programs world wide by
> perhaps 0.000_001%".  ;-)

The fact that a nasty long name is not going to be used often is not a good
enough justification for me.

****************************************************************

From: Randy Brukardt
Sent: Monday, January 24, 2011  9:36 PM

...
> Does anybody remember why we wanted to change this name?

My recollection is that you were the one that wanted to change it, so I would
have hoped you remembered why. ;-)

> I think it's because (1) "Implemented" is way to general a term for
> what this thing does, and (2) it doesn't read well as an aspect.  But
> I don't really remember the discussion.
> And I'm still ambivalent about shoe-horning everything under the Sun
> into "aspect" syntax.  The minutes don't clue me in.

I think that sounds about right. "Implemented" doesn't seem to indicate at all
what it is about. It's really "the overriding or implemented-by routine must be
of this kind". "Implemented by" only is true for task and protected operations,
and this definition is wider than that anyway.

I think we were hoping that you would have a wonderful idea that's nice and
short and descriptive. Part of the problem is that *Ada* doesn't have such a
term, so it's asking a lot for something meaningful.

Perhaps someone will have a better idea from these musings.

****************************************************************

From: Brad Moore
Sent: Tuesday, January 25, 2011  8:33 AM

> Perhaps someone will have a better idea from these musings.

How about just Synchronized?  The word Implemented doesn't really add anything.
Also, this allows us to get rid of  By_Any. If the pragma doesn't have an argument, that would be the By_Any case.

     Implemented -->  Synchronized
     By_Entry -->  Entry
     By_Protected_Procedure -->  Protected_Procedure
     By_Any -->

****************************************************************

From: Bob Duff
Sent: Tuesday, January 25, 2011  8:45 AM

> How about just Synchronized?  The word Implemented doesn't really add
> anything.

I like it.

> Also, this allows us to get rid of  By_Any. If the pragma doesn't have
> an argument, that would be the By_Any case.

I'm not sure about that.  It's not intuitively obvious that "pragma
Synchronized;" means "by-any".

>      Implemented -->  Synchronized
>      By_Entry -->  Entry
>      By_Protected_Procedure -->  Protected_Procedure

Agreed.

>      By_Any -->

I think I prefer "By_Any --> Any".

OTOH, I'd be happy to keep the "By_" prefixes, if folks like them.  The main
thing is to rename the pragma/aspect itself.

****************************************************************

From: Tucker Taft
Sent: Tuesday, January 25, 2011  8:49 AM

Unfortunately, both "synchronized" and "entry" are reserved words.

Perhaps:

Is_Synchronized => By_Entry
Is_Synchronized => By_Protected_Procedure
Is_Synchronized [=> True]
Is_Synchronized => False

****************************************************************

From: Bob Duff
Sent: Tuesday, January 25, 2011  8:59 AM

> Unfortunately, both "synchronized" and "entry" are reserved words.

I was thinking that's OK for pragmas.  But now I realize that if you want to
make it an aspect, reserved words are a no-no.

> Perhaps:
>
> Is_Synchronized => By_Entry
> Is_Synchronized => By_Protected_Procedure Is_Synchronized [=> True]
> Is_Synchronized => False

Looks good.

****************************************************************

From: Brad Moore
Sent: Tuesday, January 25, 2011  9:42 AM

I suppose "synchronous" could be another alternative, that isn't a reserved
word. If that were the case, is it just the aspect name that causes a problem?
Or would the aspect qualifiers also present a problem if they conflict
with reserved words.

ie
   pragma Synchronous (Entry);
   pragma Synchronous (Protected_Procedure);

     Synchronous => Entry
     Synchronous => Protected_Procedure
     Synchronous [=> True]
     Syncrhonous => False;

In any case, I'm not sure which I like better. They both look good to
me, and an improvement over the pragma Implemented.

****************************************************************

From: John Barnes
Sent: Tuesday, January 25, 2011  11:23 AM

I am happy with Is_Synchronized and keeping By. I much prefer By_Any to True or
False.  Having them all start By_ has a nice rhythm to it.

By_Whatever, By_Any_Old_How, By_George...

****************************************************************

From: John Barnes
Sent: Tuesday, January 25, 2011  2:02 PM

I should have added By_Hook_Or_By_Crook.

****************************************************************

From: Edmond Schonberg
Sent: Tuesday, January 25, 2011  3:36 PM

Not to mention By_Mistake and By_Pure_Luck.

****************************************************************

From: Edmond Schonberg
Sent: Tuesday, March 15, 2011  2:31 PM

> AI05-0215-1: (see minutes of meeting #42)

There is no mention of 0215 in the October minutes. Bob had the same reaction in
January, and then went on to propose different values for the corresponding
aspect. Given that this is a correction of another AI, should we just withdraw
AI05-030 (even though WG9 approved)  and rewrite it without the Implemented
pragma (no one uses it yet) and with the new Aspect Implemented_By?

****************************************************************

From: Randy Brukardt
Sent: Tuesday, March 15, 2011  2:50 PM

> There is no mention of 0215 in the October minutes.

Sorry, it was the November phone meeting. But that doesn't say much:

Ed will take this one, and convert it to an aspect, removing the pragma.
[Editor's note: change the subject when this is done, and add a question to this
effect.] Is the name "Implemented" good? We don't have a better idea, and
typical formatting puts the aspect by itself on the left. Bob thinks the name is
too general; he will take this off-line and try to come with a better name.
"Implemented_By_Entry" as a boolean aspect is a thought. "Implemented_By_Entry
=> False" is really "maybe", not "no", so that isn't perfect. We'll let Bob
think about this.

[End of November minutes.]

> Bob had
> the same reaction in January, and then went on to propose different
> values for the corresponding aspect.

That was Bob responding to the action item above.

> Given that
> this is a correction of another AI, should we just withdraw AI05-030
> (even though WG9 approved)  and rewrite it without the Implemented
> pragma (no one uses it yet) and with the new Aspect Implemented_By?

I don't think we want to withdraw it completely, since quite a bit of it will be
unchanged. (Certainly all of the stuff about Requeue has no reason to change.)

So I think you should just write up this AI saying that it is replacing
paragraphs whatever to whenever. But we're not going to "withdraw" a WG 9
approved AI (even if this one ends up rewriting 90% of the paragraphs -- which I
doubt).

****************************************************************

From: Bob Duff
Sent: Tuesday, March 15, 2011  3:51 PM

> > Bob had
> > the same reaction in January, and then went on to propose different
> > values for the corresponding aspect.
>
> That was Bob responding to the action item above.

I still pretty-much hate the name Implemented (or Implemented_By), for the
reasons I already mentioned.

Tucker's suggestion ("Tuesday, January 25, 2011  8:49 AM", in the !appendix)
seems perfect to me:

Is_Synchronized => By_Entry
Is_Synchronized => By_Protected_Procedure
Is_Synchronized [=> True]
Is_Synchronized => False

****************************************************************

From: Edmond Schonberg
Sent: Tuesday, March 15, 2011  4:20 PM

I agree with this choice, that should be the name and the legal values of the
new aspect.

****************************************************************

From: Randy Brukardt
Sent: Tuesday, March 15, 2011  4:31 PM

Fine by me, and since you are writing it up, you get to make the recommended
choice anyway. If someone wants a choice other than the one you make, *they* can
write it up. That should minimize the alternatives proposed. ;-)

****************************************************************

From: Bob Duff
Sent: Tuesday, March 15, 2011  4:41 PM

> > Is_Synchronized => By_Entry
> > Is_Synchronized => By_Protected_Procedure Is_Synchronized [=> True]
> > Is_Synchronized => False
>
> I agree with this choice, that should be the name and the legal values
> of the new aspect.

Note that True and False here, are not values of type Boolean.
They are "identifiers specific to an aspect".  That's not a problem, just a
language-lawyerly oddity.

It means you can't say:

    Is_Synchronized => Boolean'(False)
    Is_Synchronized => Standard.False

    X : constant Boolean := False;
    ...
    Is_Synchronized => X

If you don't like that, I suppose you could define an enumeration type
(By_Entry, By_Protected_Procedure, True, False), or perhaps (By_Entry,
By_Protected_Procedure, Yes, No). But I wouldn't bother with that.

****************************************************************

From: Edmond Schonberg
Sent: Wednesday, March 16, 2011  3:55 PM

Here is an aspect-centric version of this AI.  The new aspect is
Is_Synchronized, and the aspect definition is a synchronization_kind:

      Synchronization_Kind ::= By_Entry | By_Protected_Procedure | By_Any.

I did not see the need to have an additional value to specify a default, given
the stated inheritance rules. I guess Bob will move the skeleton of the previous
version of the AI to annex J.

[This is version /02 of the AI - Editor.]

****************************************************************

From: Tucker Taft
Sent: Wednesday, March 16, 2011  4:06 PM

Can't say I love "Is_Synchronized => By_Any"
Doesn't make a whole lot of sense to me.
"Yes" or "True" would read better, IMHO.

****************************************************************

From: Edmond Schonberg
Sent: Wednesday, March 16, 2011  4:14 PM

"True" seems confusing because this is not a boolean aspect. I guess
Is_Synchronized => Yes is acceptable, to indicate that some subsequent
overriding operation will be a legal target of a requeue operation.

****************************************************************

From: Bob Duff
Sent: Wednesday, March 16, 2011  5:01 PM

> I did not see the need to have an additional value to specify a
> default, = given the stated inheritance rules.

OK.

> I guess Bob will move the skeleton of the previous version of the AI
> to = annex J.

I think that's Randy's job.

****************************************************************

From: Randy Brukardt
Sent: Wednesday, March 16, 2011  9:46 PM

> I guess Bob will move the skeleton of the previous version of the AI
> to annex J.

I thought that we were not even going to define a pragma for this one (unlike
CPU). Pragmas are nasty for subprograms, and these are subprograms. So no need
to put anything in Annex J.

****************************************************************

From: Randy Brukardt
Sent: Wednesday, March 16, 2011  9:53 PM

> Can't say I love "Is_Synchronized => By_Any"
> Doesn't make a whole lot of sense to me.
> "Yes" or "True" would read better, IMHO.

Except that it really means that it is not necessarily synchronized (it can be a
normal subprogram, which is not synchronized by any imagination), so it is more
like "Maybe_Not".

     Is_Synchronized => Maybe_Not

Perfect. ;-)

     Is_Synchronized => Probably_Not

     Is_Synchronized => Hardly_Ever

     Is_Synchronized => Not_Specified

(but specifying something to be Not_Specified seems like a tautology :-)

     Is_Synchronized => Not_Required

     Is_Synchronized => Not_Necessarily_Synchronized

Fill in your own ideas here.

****************************************************************

From: Randy Brukardt
Sent: Wednesday, March 16, 2011  10:14 PM

> Here is an aspect-centric version of this AI.

Your version seem to have lost the revised 9.5(18/3) that defined the
inheritance of the aspects. (This was part of the fixes of this AI, not
something that previously existed - there are no rules in the Standard that
described inheritance for aspects of subprograms - of course, Tucker might be
writing those as I write this.) Anyway, I put it back, pending whatever Tucker
does in 13.1.

Modify 9.5(18/3):

  [A pragma Implemented is said to *apply* to the procedure denoted
  by its *procedure_*local_name.] {Inherited subprograms inherit the
  Implemented aspect, if any, from the corresponding subprogram of the
  parent or progenitor type.}  If an overriding operation does not have
  a [pragma] {directly specified} Implemented {aspect} then [any pragma
  Implemented applying to] {Implemented aspect of} the inherited
  operation [applies to] {is inherited by} the overriding operation.

The last paragraph doesn't mark the changes properly (since
"implementation-kind" and "Implemented" were changed without markings):

Modify 9.5.4(5.1/3):

  If the target is a procedure, the name shall denote a rename of an
  entry, or shall denote {a view or} a prefixed view of a primitive
  subprogram of a synchronized interface, where the first parameter of
  the unprefixed view of the primitive subprogram shall be a controlling
  parameter, and [an Implemented pragma] {the Is_Synchronized aspect shall be
  specified} with [implementation_kind]{Synchronization_Kind} By_Entry [shall
  apply to] {for} the primitive subprogram.

I've corrected the AI with these changes.

****************************************************************

From: Edmond Schonberg
Sent: Wednesday, March 16, 2011  10:20 PM

> Your version seem to have lost the revised 9.5(18/3) that defined the
> inheritance of the aspects. (This was part of the fixes of this AI,
> not something that previously existed - there are no rules in the
> Standard that described inheritance for aspects of subprograms

I thought (naively no doubt) that the rules for pre- and post-conditions are
precisely inheritance rules for aspects of subprograms.

****************************************************************

From: Randy Brukardt
Sent: Wednesday, March 16, 2011  10:33 PM

...
> > Your version seem to have lost the revised 9.5(18/3) that defined
> > the inheritance of the aspects. (This was part of the fixes of this
> > AI, not something that previously existed - there are no rules in
> > the Standard that described inheritance for aspects of subprograms
>
> I thought (naively no doubt) that the rules for pre- and
> post-conditions are precisely inheritance rules for aspects of
> subprograms.

I don't think Pre or Post are inherited at all. There is a rule for Pre'Class
and Post'Class (13.3.2(10/3)) that specifically defines that they "apply" to
descendants, but formally they aren't inherited either (which is why you can
give a new one for them on the descendant).

So that isn't much help. ;-)

****************************************************************

From: Jean-Pierre Rosen
Sent: Thursday, March 17, 2011  4:13 AM

> Fill in your own ideas here.

Is_Synchronized => Optional

****************************************************************

From: Gary Dismukes
Sent: Wednesday, April  6, 2011  7:10 PM

This is a wording nit for AI05-0215.

In the !corrigendum, the third instance of "Is_Synchronized aspect"
should be preceded by 'the':

  Modify 9.5(18/3) as follows:

    [A pragma Implemented is said to *apply* to the procedure denoted
    by its *procedure_*local_name.] {Inherited subprograms inherit the
    Is_Synchronized aspect, if any, from the corresponding subprogram of the
    parent or progenitor type.}  If an overriding operation does not have
    a [pragma] {directly specified} Is_Synchronized {aspect} then [any pragma
    Implemented applying to] {Is_Synchronized aspect of} the inherited
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^

{Is_Synchronized aspect of} => {the Is_Synchronized aspect of}

    operation [applies to] {is inherited by} the overriding operation.


And I may as well mention a really small nit wrt the !corrigendum wording for
9.5.4(5.1/3):

  Modify 9.5.4(5.1/3) as follows:

    If the target is a procedure, the name shall denote a rename of an entry,...

I suggest changing "a rename of" to "a renaming of".

While there are a few places in the RM that use the wording "rename of", the
majority say "renaming of", and personally I think the latter sounds better.  I
suppose all the various instances of this wording should be made consistent.  (I'll
understand if you choose to simply ignore this suggestion. :)

****************************************************************

From: John Barnes
Sent: Thursday, April 7, 2011  6:20 AM

In the following note some typos

Modify 9.5(13/3 - 16/3) as follows:


  For the declaration of a primitive procedure of a synchronized tagged type
  the following language-defined representation aspect may be specified with
  an aspect_specification:
    Is_Synchronized.
      If specified, the aspect definition shall be a Synchronization_Kind.

  The Synchronization_Kind By_Protected_Procedure shall not be applied to
  a primitive procedure of a task interface.

  A procedure for which the specified Synchronization_Kind is By_Entry shall
  be implemented by [a] an entry. A procedure for which the
  specified  Synchronization_Kind is [as] By_Protected_Procedure shall be
  implemented by a protected procedure.

  If a primitive procedure overrides an inherited operation for which the
  Is_Synchronized aspect has been specified, then any specification of the
  aspect Is_Synchronized applied to the overriding operation shall have the
  same Synchronization_Kind, or the inherited Synchronization_Kind shall be
  Optional.


For completeness in the para starting "A procedure" in the above. I would add
something about the poor lonely Optional. eg

A procedure for which the the specified Synchronization_Kind is Optional can 
be implemented by an entry or by a procedure (including a protected 
procedure).

Maybe it is somthing to do with the battle of Agincourt but I don't like 
Optional but much prefer By_Any.  It just has to start By_

****************************************************************

From: Randy Brukardt
Sent: Thursday, April 7, 2011  9:20 PM

...
> Modify 9.5(13/3 - 16/3) as follows:
...
>   A procedure for which the specified Synchronization_Kind is By_Entry shall
>   be implemented by [a] an entry. A procedure for which the
>   specified  Synchronization_Kind is [as] By_Protected_Procedure shall be
>   implemented by a protected procedure.
...
> For completeness in the para starting "A procedure" in the above. I 
> would add something about the poor lonely Optional. eg
> 
> A procedure for which the the specified Synchronization_Kind is 
> Optional can be implemented by an entry or by a procedure (including a
> protected procedure).

This is a fine idea, but "can" is not allowed in International Standards (by the
drafting rules). We have to use "may" here.

We can use "may", but may not use "can". Similarly, we must use "shall", we shall
not use "must". But we cannot use "may not", and we may not use "cannot".

Easy to remember, right? ;-)

****************************************************************