Record aggregate question (language lawyer needed!)

Record aggregate question (language lawyer needed!)

[mcriley on BIX]

The following test program is accepted without question by one Ada
compiler and executes without any qualms, while another warns of
a constraint_error at runtime.  Which is correct, or, is the exact
implementation left up to the compiler implementor?

procedure Rec_Test is

   type Task_Record_Type is
	record
	    Originator : String (1 .. 20);
	    Header_Dtg : String (1 .. 12);
	end record;
	
    Task_Record : Task_Record_Type := (others => (others => ' '));

begin
  null;
end rec_test;

The relevant part of the LRM is 4.3.1[1]: "A component association
with the choice others [...] is only allowed if the represented
components are all of the same type."  Well the components are
both of the same _type_, but obviously the subtype constraints vary.
The nature of the warning produced by the one compiler reflects
that: "Subtype_Match_Error".

Any input on this would be most appreciated.

Marc A. Criley

Re: Record aggregate question (language lawyer needed!)

[Norman H. Cohen]

In article <mcriley.835361851@BIX.com>, mcriley@BIX.com (mcriley on BIX) writes: 
|>
|> The following test program is accepted without question by one Ada
|> compiler and executes without any qualms, while another warns of
|> a constraint_error at runtime.  Which is correct, or, is the exact
|> implementation left up to the compiler implementor?
|>
|> procedure Rec_Test is
|>
|>    type Task_Record_Type is
|>      record
|>          Originator : String (1 .. 20);
|>          Header_Dtg : String (1 .. 12);
|>      end record;
|>
|>     Task_Record : Task_Record_Type := (others => (others => ' '));
|>
|> begin
|>   null;
|> end rec_test;
|>
|> The relevant part of the LRM is 4.3.1[1]: "A component association
|> with the choice others [...] is only allowed if the represented
|> components are all of the same type."  Well the components are
|> both of the same _type_, but obviously the subtype constraints vary.

Since you quote the Ada-83 RM, Ada-83 AI-00244 applies.  The summary of
that AI is: 

   In a record aggregate, a component association having multiple choices
   denoting components of the same type is considered equivalent to a
   sequence of single choice component associations representing the same
   components.

Thus, what you have written is equivalent to

   Task_Record:  Task_Record_Type :=
                    (Originator => (others => ' '),
                     Header_Dtg => (others => ' ') );

It should work without problem.  Perhaps the compiler that raises
Constraint_Error (or threatens to) predates AI-00244, which received
final approval in November 1986.

The URL for the full text of AI83-00244 is: 

ftp://sw-eng.falls-church.va.us/public/AdaIC/standards/83com/ai-00244-bi.wj

(By the way, the Ada-95 equivalent of the rule in RM83-4.3.1(1) is in the
last sentence of RM95-4.3.1(16).  The effect of AI83-00244 is captured in
RM95-4.3.1(20).)

--
Norman H. Cohen    ncohen@watson.ibm.com

Re: Record aggregate question (language lawyer needed!)

[Robert A Duff]

In article <mcriley.835361851@BIX.com>, mcriley on BIX <mcriley@BIX.com> wrote:
>   type Task_Record_Type is
>	record
>	    Originator : String (1 .. 20);
>	    Header_Dtg : String (1 .. 12);
>	end record;
>	
>    Task_Record : Task_Record_Type := (others => (others => ' '));

This is fine -- no exception should be raised.  The inner aggregate
needds to get evaluated twice, and will have different bounds those two
times.

IMHO, "others" should not have been allowed in record aggregates -- only
in array aggregates.  It's just too weird, and it causes an unreasonable
amount of implementation difficulty.

- Bob

Re: Record aggregate question (language lawyer needed!)

[Theodore E. Dennison]

mcriley on BIX wrote:
>    type Task_Record_Type is
>         record
>             Originator : String (1 .. 20);
>             Header_Dtg : String (1 .. 12);
>         end record;
> 
>     Task_Record : Task_Record_Type := (others => (others => ' '));
> 
> The relevant part of the LRM is 4.3.1[1]: "A component association
> with the choice others [...] is only allowed if the represented
> components are all of the same type."  Well the components are

Technically, I think they are of two different (anonymous) types,
both of which happen to be subtypes of STRING.


-- 
T.E.D.          
                |  Work - mailto:dennison@escmail.orl.mmc.com  |
                |  Home - mailto:dennison@iag.net              |
                |  URL  - http://www.iag.net/~dennison         |

Re: Record aggregate question (language lawyer needed!)

[Robert A Duff]

In article <31CAEC7B.446B9B3D@escmail.orl.mmc.com>,
Theodore E. Dennison <dennison@escmail.orl.mmc.com> wrote:
>mcriley on BIX wrote:
>> The relevant part of the LRM is 4.3.1[1]: "A component association
>> with the choice others [...] is only allowed if the represented
>> components are all of the same type."  Well the components are
>
>Technically, I think they are of two different (anonymous) types,
>both of which happen to be subtypes of STRING.

No.  Mcriley is correct -- they are of the same type (type String), but
different subtypes.

- Bob

Re: Record aggregate question (language lawyer needed!)

[Norman H. Cohen]

In article <31CAEC7B.446B9B3D@escmail.orl.mmc.com>, "Theodore E. Dennison"
<dennison@escmail.orl.mmc.com> writes: 

|> Technically, I think they are of two different (anonymous) types,
|> both of which happen to be subtypes of STRING.

Newcomers trying to learn Ada are urged to ignore this misstatement,
which can only serve to confuse.  It reflects a fundamental
misunderstanding of the relationships among types and subtypes.

All subtypes of String are of the same type, namely type String!

--
Norman H. Cohen    ncohen@watson.ibm.com

Re: Record aggregate question (language lawyer needed!)

[Philip Brashear]

In article <31CAEC7B.446B9B3D@escmail.orl.mmc.com>,
Theodore E. Dennison <dennison@escmail.orl.mmc.com> wrote:
>Technically, I think they are of two different (anonymous) types,
>both of which happen to be subtypes of STRING.
>
>
>-- 
>T.E.D.          
>                |  Work - mailto:dennison@escmail.orl.mmc.com  |
>                |  Home - mailto:dennison@iag.net              |
>                |  URL  - http://www.iag.net/~dennison         |


'Fraid not, T.E.D.  As you (almost) stated yourself, they're both of type
STRING.  It's type that counts here, not subtype.  Note the component
declarations themselves: they include a type name (STRING), plus a constraint.
They're not anonymous.

(This whole question of "others" in aggregates has always been a tough one
to explain.)

Phil Brashear

Re: Record aggregate question (language lawyer needed!)

[Theodore E. Dennison]

Philip Brashear wrote:
> 
> In article <31CAEC7B.446B9B3D@escmail.orl.mmc.com>,
> Theodore E. Dennison <dennison@escmail.orl.mmc.com> wrote:
> >Technically, I think they are of two different (anonymous) types,
> >both of which happen to be subtypes of STRING.
> >
> 'Fraid not, T.E.D.  As you (almost) stated yourself, they're both of type
> STRING.  It's type that counts here, not subtype.  Note the component
> declarations themselves: they include a type name (STRING), plus a constraint.
> They're not anonymous.
> 
> (This whole question of "others" in aggregates has always been a tough one
> to explain.)


...Tough to figure out too.  
:-)



-- 
T.E.D.          
                |  Work - mailto:dennison@escmail.orl.mmc.com  |
                |  Home - mailto:dennison@iag.net              |
                |  URL  - http://www.iag.net/~dennison         |