[jeremydmiller]

Follow-up to #1105 for #1101.

#1105 made slice.PublishMessage(message) carry the correlation id that all of a slice's events agree on, and treats a null on an appended event as no agreement. The methods that clone an event dropped CorrelationId and CausationId, so any slice built by fan out or ReplaceEvent held nulls and quietly kept the original bug.

There were two clones, not one

Clone	Reached by
CloneEventWithNewData	a slicer's fan out rules (FanOutOperator → JasperFx.Events.Grouping.EventListExtensions.FanOut), both ReplaceEvent overloads
WithData	EventSlice.FanOut (→ JasperFx.Events.EventListExtensions.FanOut), so SliceGroup.FanOutOnEach, and custom groupers, as in Marten's sample

The two had identical bodies, as do the two FanOut extensions. CloneEventWithNewData now copies both ids, and WithData delegates to it so they cannot drift apart again.

Behaviour change

Nothing a store writes changes. This is from reading Marten 2f2769bca, Polecat 85452490c, Fisher 3f4c4bebd and Weasel at their heads. None of it was run against a store.

	Marten	Polecat	Fisher	InMemory
StoreProjection reads from the last event	Version, Sequence	nothing	Version, Sequence	nothing
EF Core StoreProjection reads	nothing	nothing	nothing	n/a
Document correlation_id / causation_id, ITracked	session	session	session	session

No store sets a session's ids from an event, and no store overrides ApplyMetadata. Clones are never persisted either, because BuildOperations writes only RaisedEvents, which are new events.

What does change:

Where	Before	After
PublishMessage(message) on a fan out or enriched slice whose events agree	tenant-only call, so the bus's id for the whole daemon page	carries the agreed correlation id, like any other slice
MessageMetadata.From(clone)	null correlation	the source's correlation
User projection code reading CorrelationId / CausationId off a clone in Apply, Create, ShouldDelete, Evolve or ApplyMetadata	null	the source's value

The last row can change stored document bodies, but only in projections that copy these ids onto their own documents, which is what Marten documents ApplyMetadata for. Today doc.X = e.CorrelationId in ApplyMetadata ends on null whenever a clone is last in the slice, because the hook runs for every event. Documents already stored keep their value until their next event or a rebuild.

Causation stays off messages: neither slice.PublishMessage nor MessageMetadata.From copies it, and a test pins that for clones. Slices with two flows still fall back to the tenant, since a clone only repeats its source's id and so can never break or create a false agreement.

Not changed

UserName, IsArchived, IsSkipped, AggregateTypeName, EventTypeName, DotNetTypeName and tags are still not copied, so both doc comments still overclaim. UserName would need the same trace, since ApplyMetadata exists for "LastModifiedBy".

Testing

• ./build.sh Test --root "$PWD": every target succeeded, on net9.0 and net10.0 where the project targets both. EventTests 1735 passed, EventStoreTests 350 passed with 23 skipped.
• With EventExtensions.cs and EventTypeExtensions.cs back at main, 14 EventTests cases fail, including fan out through both FanOut extensions, ReplaceEvent, the async runner and the inline path in both fan out modes. Two guard cases for slices that don't agree pass either way, as they should.
• The stores were not rebuilt or run against this change.

🤖 Generated with Claude Code