possibly restrictive bound on sp_session::SessionKeys in spawn_tasks

[kianenigma]

Is it beyond imagination for a node to not have session keys at all? What is the reasoning behind this?

For example, a runtime+node that is configured to use manual seal (eg. the smart-contract-node) must provide this useless implementation.

https://github.com/paritytech/substrate-contracts-node/blob/3419c499ed32a754df3298f9d744e0858ee451be/runtime/src/lib.rs#L484-L494

related paritytech/polkadot-sdk#186. I am trying to minimize the noise+boilerplate needed to setup a runtime+node that has the least number of pallets, and works with manual seal.

Similar example with the Metadata api: #11743

[bkchr]

The node doesn't know the format of the session keys nor what kind of session keys exists. So, it needs to call in the runtime to get this info.

The generate method is for example used for the CLI --alice etc. The decode is required when you want to check if particular session keys are registered at the node.

paritytech/polkadot-sdk#27 this will remove the compile time bound and make it a runtime error. Then we can make it optional and complain to users if they want to use something, but they don't have the runtime api not implemented.

[kianenigma]

The node doesn't know the format of the session keys nor what kind of session keys exists. So, it needs to call in the runtime to get this info.

If a node is using manual seal, or potentially other consensus primitives that don't require a session key, why should it need that?

paritytech/polkadot-sdk#27 this will remove the compile time bound and make it a runtime error. Then we can make it optional and complain to users if they want to use something, but they don't have the runtime api not implemented.

Okay, this sounds better.

All in all, it is unclear to me what are our assumptions around session keys. Is that a mandatory, substrate-wide concept? Or is it optional, based on some consensus primitives? the code is the former, but the latter seems more correct to me.

[bkchr]

Or is it optional, based on some consensus primitives?

Partly optional, but not dependent on consensus primitives. Stuff like im-online or authority discovery have their own session keys as well. So, it is a substrate-wide concept. Nevertheless, it should be optional.

[kianenigma]

I will close this in favor of paritytech/polkadot-sdk#27