Yes, you’re right, but as r approaches zero, which would be the usual null hypothesis, the Fisher transformation approaches the identity, so the practical difference is negligible.