Am Dienstag, dem 10.09.2024 um 20:36 +0200 schrieb Jakub Jelinek:
> On Tue, Sep 10, 2024 at 06:31:23PM +0000, Qing Zhao wrote:
> >
> >
> > > On Sep 10, 2024, at 14:09, Jakub Jelinek <ja...@redhat.com> wrote:
> > >
> > > On Tue, Sep 10, 2024 at 06:02:45PM +0000, Qing Zhao wrote:
> > > > > #define alloc(P, FAM, COUNT) ({ \
> > > > > __auto_type __p = &(P); \
> > > > > __auto_type __c = (COUNT); \
> > > > > size_t __size = sizeof(*(*__p)) + sizeof(*(*__p)->FAM) * __c; \
> > >
> > > Shouldn't that be
> > > size_t __size = offsetof(__typeof(*__p), FAM) + sizeof(*(*__p)->FAM) *
> > > __c; \
> > > ?
> >
> > Yeah, I think that the correct size computation should be:
> >
> > #define MAX(A, B) (A > B) ? (A) : (B)
> > size_t __size = MAX (sizeof (*(*__p)), offsetof(__typeof(*__p), FAM) +
> > sizeof(*(*__p)->FAM) * __c); \
>
> No, why? sizeof (*(*__p)) should be always >= offsetof(__typeof(*__p), FAM),
> you can't have an offset outside of a structure (ok, except doing something
> like use fld[100] as FAM). offsetof + sizeof (elt) * count is the actually
> needed size, say if it is
(offset + sizeof * c) could be smaller than sizeof (*(*__p))).
Martin
> struct S { size_t a; char b; __attribute__((counted_by (a))) char c[]; };
> then you don't really need 2 * sizeof (size_t) + N size of N elements
> in the flexible array, just sizeof (size_t) + 1 + N is enough.
>
> Or is counted_by attribute handling it in some weird way?
>
> Jakub
>
