On 3/15/10, Paolo Bonzini <pbonz...@redhat.com> wrote:
> >
> > > I'd consider not changing assert(0)->abort()
> > > if there is code after the assert that looks like an attempt at
> recovering.
> > > Example:
> > >
> > > if (!p) {
> > > printf ("the impossible has happened!");
> > > assert (0);
> > > }
> > >
> > > return p->q;
> > >
> > > should be changed to abort, while
> > >
> > > if (!p) {
> > > printf ("the impossible has happened!");
> > > assert (0);
> > > return 0;
> > > }
> > >
> > > return p->q;
> > >
> > > should not.
> > >
> >
> > Why not? According to manual page, assert(x) is equal to if (!x) abort().
> > As I mentioned earlier, system emulators don't handle SIGABRT
> >
>
> ... which won't be generated if !NDEBUG. Only if the recovery code makes
> sense, of course. However, my point was that those cases where there is
> recovery code are not no-brainers.
Except that compiling with -DNDEBUG was broken and fixed only recently
with a6c6f76ceb95a0986fd1a36cc30f8241734d20c3. Thus I suspect nobody
uses -DNDEBUG for production builds and the code paths after assert(0)
are untested.
