- Re-implement resetStateMachineReactively as fully reactive
chain by following same logic as in blocking version.
- New interface ReactiveStateMachineAccess as we need to
know if one need to call reactive method.
- Fixes#911
- Now doing triggerSink.emitNext with retry instead
of parking which caused blocking errors from blockhound.
- Retry is now hardcoded and settings for it could be
exposed later.
- As a bad behaviour we relied on reactor throwing
this particular exception to re-throw it to propagate
exception order to deny event.
- Now trying to use callback to do a same thing as
newer reactor versions don't throw anymore.
- Original preStateChange and postStateChange are now deprecated
and once those are removed we don't need to do calls to both
in StateMachineInterceptorList.
- Relates to #871
- As it's currently expected that pre/post interceptor are called
for both preStateChange and postStateChange, fix
AbstractPersistingStateMachineInterceptor so that actual persist
doesn't happen twice.
- Eventually original preStateChange and postStateChange methods
will get removed after deprecation(which should have happened
a long time ago).
- Forward port #871
- Before attempting to add better level of error handling,
discard all errors what might come from action execution.
- Handle some cases where event deny is reported back as accepted.
- Fix reactor error for double subcription.
- Relates #821
- Now starting machine outside of a sycn block with
acquire vs. release. This should allow long running
start in a machine to allow next acquire to work faster.
- Forwardport #770
- Relates #777
- For Guard and Transition change call stach to be fully
reactive from executor. Some changed signatures similarly
what was needed for reactive Actions.
- Disabling one smoke test to get figure out later as
something is broken somewhere, possible reactor bug...
- Relates #791
- First changes to introduce ReactiveGuard similar to work done
for ReactiveAction. User level interface is ReactiveGuard but internally
it is handled as its super type Function<StateContext<S, E>, Mono<Boolean>>.
- This change still keeps some calls as blocking which will get changed
in futher commits to get a full reactive chain.
- Touching only Transition to change guard signature, other use of blocking guard
internally need to get similar changes in next commits.
- Baby steps...
- Relates #791
- remove junit4 dependecies
- refactor the unit tests by junit5
- enhance the spring-statemachine-test and spring-statemachine-build-tests so they can work together with junit4/junit5
- relates #779
- Polish previous change to AbstractState by using a shared function
completionStateListenerSink to track completion sink as functionality
between entries to submachine/regions should be similar.
- Mostly relates to #743
- This commit changes a way how triggerless transitions are executed by
going via new doOnComplete method in StateListener which returns
Mono<Void>. This used to be a simple fire and forget subscribe via listener
and now fully handled via reactive chain when state is complete. Rest of
a changes are to tweak state actions to run parallel to be able to cancel
those and then follow and track when triggerless transitions need to be
executed.
- AbstractState still have some work to do for disposing things around
submachines which currently seem to break thins if handleStateDoOnComplete
is disposed when submachine state is exited. We'll leave this to get
fixed later.
- Add tag handling for junit5 which can be set via gradle build properties
statemachineIncludeTags and statemachineExcludeTags.
- Add BlockHound to build which can be activated via gradle build
property statemachineBlockHound.
- Add org.awaitility:awaitility to various test deps.
- Mostly relates to #734
- Move to use reactive methods and for now block
as concept is really blocking. Full reactive stuff
needs bigger overhaul.
- Fix event issues in DistributedStateMachine as these
caused issues when StateMachineTestPlan were changed to
send events via reactive methods.
- Relates #744
- Relates #750
