Re: Ensuring postconditions in the face of exceptions

[Robert A Duff <bobduff@shell01.TheWorld.com>]

"Randy Brukardt" <randy@rrsoftware.com> writes:

> "Robert A Duff" <bobduff@shell01.TheWorld.com> wrote in message 
> news:wcc7hpgckyy.fsf@shell01.TheWorld.com...
>> "Randy Brukardt" <randy@rrsoftware.com> writes:
>>
>>> For the record, I think you mean "limited record types". There are no
>>> requirements on how limited private types or record types that happen to 
>>> be
>>> limited because of a limited component are passed.
>>
>> Well, that needs some clarification:  for example, a record containing
>> a task is passed by reference, even if the record doesn't explicitly
>> say "limited".
>>
>> The point is, the compiler looks at the full types of everything
>> involved when deciding whether it must pass something by reference.
>> So "limited private" is irrelevant.
>
> I missed a word: I meant "explicitly limited record type" (that is one that 
> contains the word "limited" in the record declaration). That and "tagged" 
> are the only ways to *guarantee* that a record is passed by reference.

That's still not quite right.  A record that contains a task, protected,
or a tagged component is passed by ref, even if the record
is not explicitly limited.

See RM-6.2.

> Jeff Carter writes:
>>Really? ARM 6.2 seems to me to say that any limited type (a type whose full 
>>view
>>is limited) is passed by reference. This explicitly includes a composite 
>>type
>>with a component of a limited type.

...with a component of a by-ref type.

> You suddenly added "full view" to this, and that's the crux of the problem: 
> "limited" is a property of a view, while "by-reference" is a property of a 
> type. (BTW, 6.2(7) is subtly broken, see AI05-0096-1 for details if you're 
> Adam or want to know every possible detail. :-) The point is that there are 
> many limited types that are not required to be passed by reference, 
> including some record types that are limited.

Right.  Example: record containing a component of a limited private
type, whose full type is scalar.

> Also note that your statement here is also subtly wrong: a record type has 
> to be passed by reference if it has a by-reference component -- not a 
> limited component! A component of a limited private type would make the 
> record limited but not necessarily by-reference (read the rules carefully 
> again).
>
> Repeat after me: "limited" is a property of a view! It's not constant for a 
> particular type! That's a common mistake that even those of us on the ARG 
> make from time-to-time.

I don't understand why the definition of "by reference" is subtly
different from the definition of "immutably limited".

- Bob