https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98830
--- Comment #3 from Martin Sebor <msebor at gcc dot gnu.org> ---
The warning in pr98646 was determined to be a false positive at least in part
because the cast was guarded by a test for the result of the call being
nonnull, like so:
if (p->f ())
static_cast<C*>(p->f ())->g (); // bogus -Wanalyzer-null-argument
The analyzer diagnoses this too despite the test. (Admittedly it's possible
for first call to f() to return nonnull and the second one null.)
But I would (and in pr96003.C did) consider it a false positive even without
the guard, because -Wnonnull isn't meant to trigger solely on the basis that a
pointer could be null (it has to be known to be null). For example, both the
result of f() below may be null as well as A::g, but neither -Wnonnull nor
-Wanalyzer-null-argument triggers:
$ cat a.C && gcc -O2 -S -Wall -fanalyzer a.C
struct A { void (*g)(void); };
A* f (void);
void g (void)
{
f ()->g ();
}
If diagnosing the original test case is appropriate because p->f() may be null
then I would expect a warning for the call to f()->g() for the same reason.
