Extension for a throw-like C++ qualifier

[Sergio Giro]

On 3/30/07, Mike Stump <mrs@apple.com> wrote:
> ?  Just what did you want that isn't in the standard again?  Is the
> feature you want just static checking for exception specifications at
> compile time?
>
Yes, it is. Please read "compile time" when it says "runtime". The
errors mentioned are compile errors, not runtime errors. I don't know
if it is exactly "static checking", because some programs yielding
compile errors with this feature should not yield runtime errors
without this feature.
As an example,
int TheClass::wrongMethod() _throw () {
  if(0)
    throw TheException();
}
would yield a compile error, but using the standard throw does not
generate a runtime error.
In addition, I think that it is not convenient to perform the check as
a feature for the existing throw. I think that it is convenient to
create a new keyword (which I oh-so-cleverly named _throw) and allow
such an extension for the language which something like -fallow_throw.
    Cheers,
       Sergio


On 3/30/07, Mike Stump <mrs@apple.com> wrote:
> On Mar 30, 2007, at 11:05 AM, Sergio Giro wrote:
> > int TheClass::exceptMethod() _throw TheException {
> >   throw TheException();
> > }
> >
> > In this case, the gcc would check at runtime that the only exception
> > the method exceptMethod may throw is TheException.
>
> It does.
>
> > Moreover
> >
> > int TheClass::wrongMethod() _throw () {
> >   exceptMethod();
> > }
> >
> > should yield an error,
>
> It does, through, at runtime.
>
> > should correct, since the exception is catched.
>
> It is.
>
> > I think that, if it is pessimistically supposed that every exception
> > not enclosed into a try { } catch { } block may be thrown, the check
> > can be done at runtime.
>
> It is.
>
> > The overall idea is, of course, to give information to the caller
> > about the exceptions it must handle.
>
> Right, exception specifications do.
>
> > I think that this extension would worth the while. Any ideas?
>
> ?  Just what did you want that isn't in the standard again?  Is the
> feature you want just static checking for exception specifications at
> compile time?
>