coordinator/tributary was tributary-chain. This crate has been renamed
tributary-sdk and moved to coordinator/tributary-sdk.
coordinator/src/tributary was our instantion of a Tributary, the Transaction
type and scan task. This has been moved to coordinator/tributary.
The main reason for this was due to coordinator/main.rs becoming untidy. There
is now a collection of clean, independent APIs present in the codebase.
coordinator/main.rs is to compose them. Sometimes, these compositions are a bit
silly (reading from a channel just to forward the message to a distinct
channel). That's more than fine as the code is still readable and the value
from the cleanliness of the APIs composed far exceeds the nits from having
these odd compositions.
This breaks down a bit as we now define a global database, and have some APIs
interact with multiple other APIs.
coordinator/src/tributary was a self-contained, clean API. The recently added
task present in coordinator/tributary/mod.rs, which bound it to the rest of the
Coordinator, wasn't.
Now, coordinator/src is solely the API compositions, and all self-contained
APIs are their own crates.
Makes `Clone` calls significantly cheaper as now only the outer Arc is cloned
(the inner ones have been removed). Also wraps uses of Serai in an Arc as we
shouldn't actually need/want multiple caller connection pools.
Decoding slash evidence requires specifying the instantiated generic
`TendermintNetwork`. While irrelevant, that generic includes a type satisfying
`tributary::P2p`. It was only possible to route now that we've redone the P2P
API.
The check in validators prevented connections from non-validators.
Non-validators could still participate in the network if they laundered their
connection through a malicious validator. allow_block_list ensures that peers,
not connections, are explicitly limited to validators.
Moves UpdateSharedValidatorsTask to validators.rs. While prior planned to
re-use a validators object across connecting and peer state management, the
current plan is to use an independent validators object for each to minimize
any contention. They should be built infrequently enough, and cheap enough to
update in the majority case (due to quickly checking if an update is needed),
that this is fine.
TaskHandle::close meant run_now may panic if the task was closed. Now, tasks
are only closed when all handles are dropped, causing all handles to point to
running tasks (ensuring run_now won't panic).
