BFTNotaryServiceTests generates a master node independently of the cluster nodes, so it can put it
at the end of the list of nodes. The calling function however treats the first node in the list as
the master, not the last node. This simplifies the code while maintaining the same behaviour.
* Add information on why state machine was removed from StateMachineManager.
There are two cases: normal end of flow or error.
Return flow result as part of state machine remove data.
Make Change a sealed class with Add and Remove.
fiber.actionOnEnd takes ErrorOr<R> parameter.
* Remove unnecessary fields from StateMachineManager.Change.
DemoBench: Misc usability improvements:
- Pre-fill details for some fictional banks when Add Node is pushed.
- Make services a checkbox list rather than one where you have to know how to use the keyboard to do multi-select.
- Make web server launch button spin until server is launched to show activity.
- Suppress an exception that spams the log due to inability to load all the states. It'll get fixed as part of the vault API and serialisation work.
* Subscribe to updates for transaction and vault RPCs.
* Ensure we unsubscribe our observables at the end.
* Use Rx scheduler that can observe on FX application thread.
* Add information on who started flow on a node with name where possible.
Add sealed class holding information on different ways of starting a flow: RPC, peer, shell, scheduled.
* Remove invokeFlowAsync from ServiceHub, move it to ServiceHubInternal.
We shouldn't be able to start new state machines from inside flows.
Assuming that Autocomplete will be added, according to JIRA https://r3-cev.atlassian.net/browse/CORDA-367
Note that there are still a few duplicates in the source with slightly different coordinates. I assume the autocomplete will be able to cope, or these can be removed.
Move functions for generating transactions into OnLedgerAsset from various locations in
the code (VaultService, AbstractConserveAmount, etc.) to unify the code paths and reduce
duplication.
* Remove progress Observable from FlowHandle, unless explicitly requested.
* Refactor FlowHandle creation into FlowStateMachine.
* Prevent server-side queue subscription for dummy Observable.
* Refactor so that RPC client does not receive any unused progress Observables. This is the simplest way of ensuring we have no dangling "hot" Observables when the RPC client closes.
* Test flow has correct handle.
* Resolve some compiler warnings.
* Document how starting a flow does not involve progress tracking by default.
* Update changelog and release notes for RPC API.
* Rename new RPC API to startTrackedFlow().
* Remove optimisation because of its affect on the client-side.
* Update documentation.
