This is an idea that’s been going around for a while, but I’m not sure if there’s been a proper discussion thread on it, so here it is.
A global constant is a michelson constant accessible from any contract and registered in the context. Most of the discussions around the topic have assumed that the constant is a lambda expression, but this needs not be the case. However, if the constant is not a lambda expression, we may want to be careful to ensure it is not (or does not contain) a ticket handle, which could introduce mutation, though storing tickets wouldn’t be an issue per se, just pointless.
It could work as follow:
- a new manager operation:
register_constanttakes a Michelson constant and writes it in the context and assigns an address to it (typically this could be the hash of the packed representation of the value, but it could be prudent to not guarantee this property). The manager pays the associated storage burn.
- a new michelson opcode:
REGISTER_CONTANTemits an operation which does the same thing
- a new michelson type:
constant_addressrepresents the address of a constant that has been registered in the context.
- a new michelson type constructor:
Constant_address "CST1x2344..."which constructs a lambda address from a string representing a valid address
- a new michelson opcode
LOAD_CONSTANTparametrized with a type
twhich takes a
constant_addressand returns an
t option, depending on whether or not that address is inhabited.
This is useful for lambdas
This is primarily useful for lambdas because it allows the creation of libraries. It also means that contracts no longer need to be fully deserialized when called. A contract could register each of its entrypoint as a constant and load them on demand.
Since many contracts are bound to repeat code (e.g. token standard code), a cache of the most commonly called function can offer much lower gas and even introduce heavy optimizations by precompiling the functions being called.
Non lambda uses
It’s not super clear that this is useful beyond lambdas, but we could imagine it being helpful for some big_map containining useful data. There is also no particularly good reason to restrict this to lambdas, even if they seem like the primary use case.