In the ElGamal signature scheme we calculate a signature for a (hashed) message $m$ as
$r \equiv_p g^k$,
$s \equiv_{p-1} (m - xr) k^{-1}$
and verify it by checking
$g^m \equiv_p y^r r^s$.
Here $r$ is a group element which is in some cases used as an exponent. This is not a problem, as long as the group elements are just natural numbers, but using a different group, this causes complications (like in EC-DSA, where we have to use just one coordinate of $r$ when it ought to be an exponent).
Even though this is admittedly not a big problem, my question is: Why is it even necessary to include $r$ as an exponent in $s$? Would changing the calculation of $s$ to just
$s \equiv_{p-1} (m - x) k^{-1}$
and the verification condition to
$g^m \equiv_p y r^s$
be insecure in some way? So far I couldn't find a way to forge a signature or extract $x$ or $k$ from this modified scheme.