rpc: Register calls where they are defined #7766

Split out registration of methods to every module, apart from 'help' and 'stop' which are implemented in rpcserver.cpp itself.

• This makes it easier to add or remove RPC commands - no longer everything that includes rpcserver.h has to be rebuilt when there's a change there.
• Cleans up rpc/server.h by getting rid of the huge cluttered list of function definitions.
• Removes most of the bitcoin-specific code from rpcserver.cpp and .h.

Continues #7307 for the non-wallet.

Nice! Concept ACK.

Concept ACK. Would it make sense to get rid of the duplicate code in void Register*()?

Would it make sense to get rid of the duplicate code in void Register*()?

Maybe. I thought about it, but it's only a simple loop of four lines, sometimes factoring something so simple out to a function makes things harder to understand, not easier. Another option would be to export a 'descriptor structure' with a name, pointer to the array, and size of the array. E.g. like Linux modules. But simply having Register* function leaves it up to the module as to how to determine what to register (e.g. there may be optional features enabled on the command line).

IMO we should keep the array loop as it is (and not refactor with some hard-to-understand macro-like optimization).

Could make it two lines, even, without losing any clarity:

   for (unsigned int vcidx = 0; vcidx < ARRAYLEN(commands); vcidx++)
        tableRPC.appendCommand(commands[vcidx].name, &commands[vcidx]);

(will update the one in rpcwallet as well)

Concept ACK

Concept ACK, though I wonder if it wouldn't be cleaner to pass a pointer to tableRPC to the registration commands, so the individual RPC implementations don't even need to depend on access to the global?

pass a pointer to tableRPC to the registration commands

Awesome idea, will do that.

utACK 3be52e1

utACK 3be52e1

Squashed a20fca9 0b18725 dcb0a76 3be52e1 into fb8a8cf