From: Mark Borgerding (mborgerding_at_[hidden])
Date: 2000-04-10 21:06:27
Dave Abrahams wrote:
> > Allowing exceptions to be thrown
> >
> > ... from a smart pointer destructor is another arguable choice.
>
> I don't agree with this. Under the current language rules a destructor which
> throws an exception doesn't complete. It is unclear what is to be done with
> a partially-destroyed object. That almost makes it nonsense. This is totally
> apart from the fact that program termination is likely.
>
You are right. I thought the behavior was undefined, but the standard dictates (
15.5.1 ) that terminate should be called
" -- when the destruction of an object during stack unwinding exits using an
exception."
So by the time a smart pointer destructor might try to catch(...) the exception
during a stack unwind, it is already too late -- the program has exited.
BTW, I think MSVC allows this to slip by, but I'm not 100% sure.
>
> > This can be
> > problematic in situations where the destructor might be called because of
> > another exception being thrown. The behavior of std::auto_ptr is to allow any
> > exceptions out from the destructor.
>
> No, std::auto_ptr's destructor has an empty exception-specification, so no
> exceptions will propagate out of it, ever.
>
I guess I was remembering the CD2 behavior. I just looked again and realized that
the definition had changed in the final standard.
So if ~auto_ptr has an empty excep. spec. , why does boost::shared_ptr allow
exceptions out?
Shouldn't we change this to make it as consistent as possible with the auto_ptr
interface?
> > Allowing such exceptions can reduce the
> > safety of using smart pointers. On the other hand, allowing such exceptions
> > makes smart pointers behave more like automatic objects.
>
> What does this mean? Can't a smart pointer be an automatic object?
I didn't explain myself very well.
What I meant was that allowing excpetions out of the d'tor makes a smart pointer to
a Foo act like an automatic Foo object.
i.e. If Foo::~Foo can throw an exception , so can smartptr<Foo>::~smartptr<Foo>
Thanks for the comments. I will try to get an updated copy of the document in the
vault sometime tomorrow.
Mark