https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127559
--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Actually, I'm afraid the above is not sufficient. I think
template <typename T>
struct S {
T x;
int y;
};
void
foo (int x)
{
auto b = [&] { return x; };
S <decltype (b)> c = { b, 42 };
#pragma omp target
if (c.x () != 42)
__builtin_abort ();
}
int
main ()
{
foo (42);
}
is a valid testcase, so we can't just record in lambda_objects_accessed
vars/parms/results, but on the other side, remembering there random expressions
which never can't be mapped seems wrong too. If it is a simple chain of
COMPONENT_REFs/ARRAY_REFs (one can also do S <decltype (b)> d[2] = { { b, 42 },
{ b, 43 } }; and use
d[i].x ();
etc.), one can check if the the get_base_address of it is a local_decl or not.
What about pointers? One can have above auto *p = &c; etc. before the target
region.
