Re: Forcing Exception Handling

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Tue, 1 Mar 2011 07:23:56 -0800 (PST), Shark8 wrote:

> On Mar 1, 1:56�am, "Dmitry A. Kazakov" <mail...@dmitry-kazakov.de>
> wrote:
>>
>> Also consider this:
>>
>> � �declare
>> � � � Foo : exception:
>> � �begin
>> � � � ...
>> � � � raise Foo;
>> � �end; �-- Foo propagates out its scope!!
>>
> 
> What's wrong with handling it with OTHERS?

You have an object of non-existent type:

   when Error : others =>

Assuming that Error is of Root_Exception_Type'Class, what should Error'Tag
return?

>> A possibility to handle this is exception promotion, e.g. Foo to
>> Program_Error, when Foo gets finalized.
> 
> A third is to convert Exceptions to a hierarchy as stated in my last
> post.

No, that does not work, because the exception type would be dead in the
handler. C++ model cannot work in Ada 2005+, because in Ada tagged types
can be local.

Further problems are overlapping choices in the exception clause. If the
semantics of

   exception
      when X =>
      when Y =>

is dispatching on the tag of the current exception, which people suggesting
class-wide exceptions have in mind. Then that would be incompatible with
the choice

      when others =>

which has the semantics of a class-wide call. If you choose that semantics
instead, then you would have to write

   exception
      when X'Class =>
      when Y'Class =>

and lose the property that alternatives do not overlap. That is way
unacceptable to me.

These are the reasons why I don't think that exception -> type mapping is
such a good idea. I prefer exception -> value mapping + contracts.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de