Release v16.16.0

Caution

Known broken package — Cratis.Chronicle.Testing in this release bakes in Cratis.Screenplay 1.x types, so consumer specs fail against Screenplay 2.0+ (#3598). Use v16.19.3 or later.

Summary

Defects and missing seams found running Chronicle in a large production application.

Added

• IEventStore.Registration reports what became of each declared artifact when registration ran, and what stopped the run if it failed, with WaitForRegistration() to await it
• IProjections.GetStateForModel<TReadModel>() and GetFailedPartitionsForModel<TReadModel>() reach a projection by the read model it maintains, which for a model-bound projection is the only handle there is
• ReadModelScenario exposes the client artifacts registry, so a spec can ask which artifacts were registered
• [ReservedProtoFields] keeps retired proto field numbers reserved across regeneration
• CHR0039 now covers every awaitable and the integration testing surface

Changed

• A unique constraint violation message no longer includes the offending value; it moved to the violation details, where the application chooses whether to render it
• Client EventSequenceNumber.Max is now ulong.MaxValue - 1, matching the kernel; code that captured the old value sees it as an ordinary sequence number
• Subscribing the same handler to IConnectionLifecycle.OnConnected or OnDisconnected twice now subscribes it once; removing is unchanged
• IEventStore gains a member, so any implementation outside this repository must add it

Fixed

• The pre-16.12 constraint-store upgrade shipped but never ran, because its serializer was never registered
• One unregistered event type aborted projection discovery for every projection, taking the whole read side down
• A projection replay rewrote every stored child whole, defaulting every member the projection did not set
• Reading a read model turned an absent collection into null instead of an empty one
• The compliance release pass blanked a value that had never been encrypted
• A client concurrency scope meaning "no expected sequence number" arrived at the kernel as an ordinary number, so the kernel ran a check that could only pass
• A skipped concurrency check was attributed to the caller
• GetStateFor<TProjection>() looked in the read model dictionary with a projection type, so it could only ever throw
• A failed projection registration named every identifier in the batch instead of the definition that failed
• A registration that threw left its outcome looking unfinished, so a consumer waiting on it timed out instead of being told what failed
• Two projections maintaining the same read model resolved silently by declaration order
• Two dependency injection containers in one process sharing a namespace were handed the same event store
• Concurrent callers asking for the same event store each built their own and ran discovery twice
• An event store that could not be reached on its first resolution stayed unreachable for the lifetime of the process
• CHR0002 demanded [EventType] on generic arguments that are not events
• CHR0004 and CHR0005 analysed a reactor's private helper methods as if they were handlers
• CHR0010 counted the absence of [EventStore] as a distinct store, so a single-store projection reported as multi-store
• CHR0025 reported a collision an event named by [ChildrenFrom] cannot cause
• The code analysis rule index was missing its CHR0039 row
• [NoAutoMap] was ignored on [ChildrenFrom] children and [Nested] members, and a nested object used the root's exclusions instead of its own
• Two [SetFromContext<T>] for the same event type on one member silently discarded all but the last
• A message: on a property-level [Unique] was silently discarded
• Three attribute docstrings promised that projections filter on event metadata
• Cratis.Chronicle.XUnit.Integration omitted the embedded kernel's runtime dependencies from its package
• A unique constraint definition compared its covered events by reference, so every re-registration looked like a change and asked for a reindex