Be a state machine. Just like js-ipfs :)

[daviddias]

[daviddias]

@pgte is this still something you would like to pursue?

[pgte]

Yes.

What does this entail? Is it something that other libp2p impls should be doing too?

[daviddias]

@lgierth not at all. It is just how the libp2p Node in JS changes its state.

[daviddias]

@pgte you have a ton of experience on this and have pointed out multiple issues with the current state-machine in js-ipfs. Do you have any recommendations? Modules to use, examples to take inspiration from so that we don't fall in the same thing? I want to implement it asap for #159

[pgte]

@diasdavid I think it can be boiled down to this:

• define a FSM (states and transitions) for a libp2p node
• let the user commands drive transitions of the state machine
• act on state changes instead of user commands

Some states transitions are initiated by the machine itself. Apply the same principle: do not act directly, only start transitions, decoupling state transition from handling the state transition. This only should allow you to escape some nasty bugs and require less verification and synchronisation.

As for the library, I like https://www.npmjs.com/package/fsm-event 's API.

Any more pointers you would like?

[pgte]

Ha, also another thing, to your consideration:
(This doesn't apply directly to libp2p, I think, but can perhaps be translated to IPFS more directly.)

Allow operations while offline. If the operation requires the node to be online, wait until it is inline instead of failing.
For instance, pubsub publish can perhaps fail silently if the node is offline (since the API offers no delivery guarantees).

While, OTOH, the file.get(cid) can be postponed to when the node is online.

This would allow a more friendly programming style, similar to many database drivers out there, where we don't need to wait until we connect before issuing commands or queries.

[daviddias]

Thank you! We use fsm-event in js-ipfs https://github.com/ipfs/js-ipfs/blob/master/src/core/state.js#L7 but I know you had some issues with how it was implemented. Could you elaborate on those?