Function signature literals

[pirapira]

Out of a gitter conversation there came an idea of looking up authorization_table[msg.sig][msg.caller]. For setting up such a table, it would be convenient to have function signature literals signature_hash(f(void)) that represents the signature hash.

contract C {
    mapping (bytes4 => mapping (address => bool)) authorization_table;
    modifier auth() {
        if (!authorization_table[msg.sig][msg.caller]) throw;
        _;
    }

    function interface() auth;
    function allow_interface(address _somebody) auth {
        authorization_table[signature_hash(interface(void))][_somebody] = true;
    }
}

[VoR0220]

I think this falls under Christians group modifier proposal.

[pirapira]

@VoR0220 the proposal #623 does not use msg.sig.

Since msg.sig is a part of the language, the language should also support interpreting the value of msg.sig in a convenient manner. For interpreting msg.sig, the language needs a way to associate the signature hashes and the function signatures.

[VoR0220]

It could very well use msg.sig "under the hood" is my point. This is a potential way to implement it is my point.

[VoR0220]

But i think maybe you're looking for something else is your point.

[axic]

You can already create function signatures with bytes4(keccak256("interface()")).

However since we have function types, we could introduce .sig member on them to return this: interface.sig.

[frangio]

Manually creating function signatures seems error prone. A .sig member sounds great.