Computation codes in network information theory are designed for scenarios where the decoder is not interested in recovering the information sources themselves, but only a function thereof. Körner and Marton showed for distributed source coding (DSC) that such function decoding can be achieved more efficiently than decoding the full information sources. Compute–forward has shown that function decoding, in combination with network coding ideas, is a useful building block for end-to-end communication over a network. In both cases, good computation codes are the key component in the coding schemes. Could these same codes simultaneously also enable full message decoding over a sufficiently strong multiple-access channel (MAC)? This work establishes a partial negative answer and converse result. Specifically, for any code that is known to be a good computation code for some MAC, we characterize a class of MACs for which that code cannot enable full message decoding (and vice versa). Finally, an analogous duality result is established for a related DSC problem.