Convert exchange and keystore gen_fsms to gen_statem

[hmmr]

Two instances of gen_fsm (aae_keystore and aae_exchange) have been converted to gen_statem.

[martinsumner]

Just wanted to confirm my understanding why it is necessary to include the handling of an info message {gen_event , Reply}.

When the exchange is initiated it is passed an AAE send fun (to be used by remote processes to reply to the exchange), within Riak this is generated from:

https://github.com/OpenRiak/riak_kv/blob/openriak-3.4/src/riak_kv_vnode.erl#L619-632

When this is called by the exchange, it creates the sender to be used for response messages as {fsm, undefined, self()}. So the exchange is identified as an FSM (not a STATEM) within Riak.

When the vnode processes an AAE message (e.g. a request to fetch_root), it will response with riak_core_vnode:reply/2 :

https://github.com/OpenRiak/riak_core/blob/openriak-3.4/src/riak_core_vnode.erl#L1134-L1135

This will use https://github.com/OpenRiak/riak_core/blob/openriak-3.4/src/riak_core_send_msg.erl#L50-L56.

So when used in Riak the responses that are expected in the waiting_all_results state will be received as messages to a gen_fsm not a gen_statem. Hence they arrive as standard process info messages, which then need to be considered as cast responses used when kv_index_tictcatree is tested as a standalone application.

Is that the correct understanding?

Although this is a bit hacky ... riak_core doesn't have message handling for gen_statem at present; and also during transition an exchange could be either a gen_fsm or a gen_statem. So it probably makes sense to use this workaround for now. Although we should add a test within this repo where gen_fsm style replies are used to make sure we have this approach covered within this repo - not just in the riak_test tests.

Does this seem reasonable?