Bug in stream handling in AbstractRepository

[alkampfergit]

[This happens only if you remove snapshot cache from repository, but since snapshot cache is probably something that should be removed, this is an issue to take care of.].

Method OpenStream of AbstractRepository stores opened stream in a dictionary whose key is obtained by StreamId and BucketId and this generates error with snapshot.

Consider this sequence of operation on an aggregate that is at version 10

• repository.GetById(ID); //no snapshot on disk, Repository opens the stream from 0 to 10
• aggregate events are applied, aggregate is at version 10.
• a method of aggregate is called, another event is generated.
• Save is called and aggregate new event is persisted.

At this moment, since the aggregate is in memory, we could get a snapshot to store in cache. Snapshot is at version 11.

Now with the same instance of the repository issue the GetById(ID) command again and this happens

• Repository load the snapshot (
• Method OpenStream finds the stream in _streams dictionary cache, so no stream is loaded.
• ApplyEventsToAggregate is called, but it uses incorrectly the same stream instance from before, opened at version 0, so all the events are applied to aggregate, causing errors.

A proposed fix is avoiding caching stream inside _stream dictionary with a string key, but using the instance of the aggregate as a key. A stream cached is valid only if the very same instance of aggregate is reused.