This argument is valid... the flaw in the author's logic is rather the assumption that encryption is a function, i.e. that it is deterministic. In fact decryption is deterministic, but encryption is not, and that's what prevents this kind of attack. Yes, you can obtain an encryption of any given ASCII character using a constant-valued function "inside the encryption", but the important point is that it's just an encryption, not the encryption, of that character. The decryption function, which you do not have access to, will send all possible encrypted representations back to the same plaintext.

You, as the person/computer performing the computation, do not get to write your own functions to run "inside the encryption".