Rpc Error handling

[sander2]

#634 generalized the code over rpc implementations. This PR changed the subxt::Error::Rpc error to a generalized one. First to a string, later to a dyn std::Error:

subxt/subxt/src/error.rs

Lines 111 to 118 in 15ffbb6

	pub enum RpcError {
	// Dev note: We need the error to be safely sent between threads
	// for `subscribe_to_block_headers_filling_in_gaps` and friends.
	/// Error related to the RPC client.
	ClientError(Box<dyn std::error::Error + Send + Sync + 'static>),
	/// The RPC subscription dropped.
	SubscriptionDropped,
	}

The problem is that this removes our ability to react to different rpc errors. For some errors we can recover, such as the POOL_TOO_LOW_PRIORITY error that we get when two connections use the same nonce at the same time. Other errors are unrecoverable, such as JsonRpseeError::RestartNeeded. Would it be possible to make subxt::Error generic over the rpc error type so that we can use proper pattern matching in our client?

[sander2]

Currently using a fork that makes the error concrete again. Obviously this will be a problem for non-jsonrpc transport implementations though. Ideally, the ClientT should take an associated Error type, and the global subxt::Error struct should be generic over the TransportError. That's quite a refactor, though..

[niklasad1]

@sander2

The current approach is that you have to downcast the ClientError(Box<dyn std::error::Error + Send + Sync + 'static> to jsonrpsee::Error such as this which is not super nice but that's intention IIRC.

or am I misunderstanding something?

[jsdw]

Currently using a fork that makes the error concrete again. Obviously this will be a problem for non-jsonrpc transport implementations though. Ideally, the ClientT should take an associated Error type, and the global subxt::Error struct should be generic over the TransportError. That's quite a refactor, though..

Yeah; It was so nice to be able to remove a generic param from the error type (which infests and complicates everything else) that I'd be somewhat reluctant to put it back!

[sander2]

Thanks, I wasn't aware of the downcast function - that'll solve my issue. Sorry for the noise!