Footnote in section on Address-Of Operator

[JoJoModding]

Hi all,

in the paragraph on address and indirection operators (6.5.3.2 in the C23 draft N3047), there is a footnote (footnote 117 in that draft), which says that

&*E is equivalent to E (even if E is a null pointer)

This seems to imply that sizeof(&*E) == sizeof(E), which is unexpected if E is an array. For example, on most C compilers, we have sizeof("foo") == strlen("foo")+1, while sizeof(&*"foo") == sizeof(char*). So clearly they are not equivalent.

Further, we have that

If an invalid value has been assigned to the pointer, the behavior of the unary * operator is undefined.

However the footnote says that &*E is equivalent to E, so if E is an invalid pointer value, *E would be undefined behavior, but &*E is not? (I'm aware that compilers may treat them equivalent, but this is precisely because *E has undefined behavior even if the immediate child of an & operator.) The sentence quoted above is not part of the constraints, so &* does still remove UB even though "the constraints on the operators still apply"?

How is one to read this footnote, and the paragraph in general? Why does it try to say that things are "equivalent" that sometimes are not?

Thanks,
Johannes

[Ben Bacarisse]

Post by JoJoModding
in the paragraph on address and indirection operators (6.5.3.2 in the
C23 draft N3047), there is a footnote (footnote 117 in that draft),
which says that

&*E is equivalent to E (even if E is a null pointer)

This seems to be a case where a footnote might add confusion rather than
clarity. The normative text makes it clear that &*E can't be equivalent
to E in every way because &*E is not an lvalue. And &*E has type
constraints that E does not have.

Post by JoJoModding
This seems to imply that sizeof(&*E) == sizeof(E), which is unexpected
if E is an array.

There are much simpler examples if the apparent non-equivalence. If p
is a pointer object, p can be assigned to by &*p can't be. And due to
the clause about constraints. &*(void *)0 is a constraint violation,
but (void *)0 is obviously fine.

I say "apparent" because equivalence is a slippery term. It does not
mean "exactly the same as" but something much less specific so it may
have been chosen for this very reason.

Post by JoJoModding
Further, we have that

If an invalid value has been assigned to the pointer, the behavior of
the unary * operator is undefined.

However the footnote says that &*E is equivalent to E, so if E is an
invalid pointer value, *E would be undefined behavior, but &*E is not?

Yes &*p is fine even if p is an invalid pointer because undefined
behaviour only exists if *p is evaluated, and nether the * nor the & are
evaluated in &*p.

Post by JoJoModding
... &* does still remove UB even though "the constraints on the
operators still apply"?

It removes some but not all. &*0 is a constraint violation (and hence
UB), but &*(int *)0 is not. De-referencing a null pointer is not a
constraint violation.

Post by JoJoModding
How is one to read this footnote, and the paragraph in general? Why
does it try to say that things are "equivalent" that sometimes are
not?

Well, I think it means equivalent in some ways and not in others. But
I'm not sure it adds any clarity to the normative wording that has been
around for many years.

[Tim Rentsch]

Post by Ben Bacarisse

Post by JoJoModding
in the paragraph on address and indirection operators (6.5.3.2 in the
C23 draft N3047), there is a footnote (footnote 117 in that draft),
which says that

&*E is equivalent to E (even if E is a null pointer)

This seems to be a case where a footnote might add confusion rather than
clarity. The normative text makes it clear that &*E can't be equivalent
to E in every way because &*E is not an lvalue. And &*E has type
constraints that E does not have.

Post by JoJoModding
This seems to imply that sizeof(&*E) == sizeof(E), which is unexpected
if E is an array.

There are much simpler examples if the apparent non-equivalence. If p
is a pointer object, p can be assigned to by &*p can't be. And due to
the clause about constraints. &*(void *)0 is a constraint violation,
but (void *)0 is obviously fine.

What makes you say &*(void*)0 is a constraint violation? I
don't see any constraints that are violated.

[Ben Bacarisse]

Post by Tim Rentsch

Post by Ben Bacarisse

Post by JoJoModding
in the paragraph on address and indirection operators (6.5.3.2 in the
C23 draft N3047), there is a footnote (footnote 117 in that draft),
which says that

&*E is equivalent to E (even if E is a null pointer)

This seems to be a case where a footnote might add confusion rather than
clarity. The normative text makes it clear that &*E can't be equivalent
to E in every way because &*E is not an lvalue. And &*E has type
constraints that E does not have.

Post by JoJoModding
This seems to imply that sizeof(&*E) == sizeof(E), which is unexpected
if E is an array.

There are much simpler examples if the apparent non-equivalence. If p
is a pointer object, p can be assigned to by &*p can't be. And due to
the clause about constraints. &*(void *)0 is a constraint violation,
but (void *)0 is obviously fine.

What makes you say &*(void*)0 is a constraint violation? I
don't see any constraints that are violated.

Isn't * applied to a pointer to an incomplete type a constraint? Let me
go look... ah, no it is not! Thanks.