Cross-store event sourcing compliance library (#5111, #5112, #5113, #5116)

[jeremydmiller]

Phase 3 of the test-harness standardization program (#5119, epic #5110). Closes #5111, #5112, #5113, #5116.

Phases 1–2 (#5098–#5107, merged 2026-08-01) standardized Marten's own harnesses. This lifts behavioral event sourcing tests out of Marten's test project entirely and into suites written once against a store-neutral seam, so Marten, Polecat and the planned Sqlite minimal store assert the same behavior instead of hand-mirroring each other's files.

The seam (#5111)

EventStoreComplianceFixture<TOperations, TQuerySession>, mirroring JasperFx's own IEventStore<,> closure — Marten closes it <IDocumentOperations, IQuerySession>, Polecat <IDocumentSession, IQuerySession>, and convergence stays a non-goal. Everything portable flows through the shared JasperFx surfaces; the fixture only absorbs what no shared interface declares: store construction, session acquisition, SaveChangesAsync, document load-back, batched DCB queries, daemon plumbing, teardown.

Marten's side is MartenComplianceFixture (~120 lines) plus a ComplianceQuerySession global alias, and enrolling a suite is a one-line empty subclass.

The suites (#5112, #5113)

Four suites, 42 tests: SelfAggregatingEvolveCompliance, DcbTagQueryAndConsistencyCompliance, AssignTagWhereCompliance, AsyncDaemonCompliance.

Strongest assertion wins where the Marten and Polecat copies had drifted:

• Inline snapshot cases assert the persisted document through LoadDocumentAsync. The Polecat mirror read back via AggregateStreamAsync, so it would have passed with a broken inline projection.
• assign_tag_where_by_stream_id keeps Marten's e.StreamId == stream1 assertion rather than Polecat's weakened Data-type check.
• Adopts two tests Marten never had, both from the Polecat side and both green here on day one: can_query_events_across_distinct_tag_types_with_or (the OR-across-distinct-tag-types INNER JOIN regression guard) and fetch_for_writing_appends_to_existing_tag_derived_stream_without_collision.

That last assertion is not academic — it caught a real Polecat bug on the first run (polecat#392: QueryByTagsAsync never mapped stream_id onto the event envelope, so every DCB tag query returned StreamId == Guid.Empty).

Extraction (#5116)

The seam and suites live in the jasperfx repo and arrive here as the source-only NuGet JasperFx.Events.ComplianceTests (jasperfx#607). Source-only is forced, not preferred: JasperFx.Events.SourceGenerator emits aggregate dispatchers per consuming assembly and binds each product's own session type, so the shared aggregates must compile inside each consumer.

Both Marten.Testing and EventSourcingTests reference the package — the harness assembly compiles MartenComplianceFixture, and the test assembly needs the suites compiled into it for the source generator to bind Marten's session types.

Retirements

self_aggregating_evolve_method.cs and assign_tag_where_tests.cs are gone. dcb_tag_query_and_consistency_tests.cs is reduced to dcb_documentation_samples.cs, which keeps the eleven sample_marten_dcb_* snippet blocks docs/events/dcb.md pulls from plus the tag/event/aggregate types the HStore-specific fixtures share. Doc snippets stay repo-owned deliberately — after extraction the suite sources no longer live in this repo, so mdsnippets could not reach them.

The two upstream additives are not being done

Both were investigated against real usage and written up on their issues:

• [jasperfx] Additive IEventRegistry generic DIMs: AddEventType<T>() and EventMappingFor<T>() (P3.2a) #5114 (IEventRegistry generic DIMs) cannot land as specified — both proposed DIMs collide with existing EventGraph members that have incompatible return types (CS0738), and making room means a source-breaking change to AddEventType<T>()'s public return type. The seam needs neither; it asserts through EventMappingFor(Type) with zero InternalsVisibleTo.
• [jasperfx] Shared batch-DCB query interface + Marten/Polecat implementations (P3.2b) #5115 (shared batch-DCB interface) would not remove IComplianceBatch, because Execute(ct) belongs to the batch rather than to the event queries — the open question that issue asked P3.1b to settle. Deferred; the adapter is 10 lines per store.

Verified

EventSourcingTests net9.0: 1613 passed, 0 failed, 7 skipped, identical before and after the switch to the package. Also builds clean on net10.0.

Note

Pinned to the published JasperFx.Events.ComplianceTests 2.37.1 (jasperfx#607, merged and shipped). The rest of the JasperFx packages move to 2.37.1 with it so the dependency graph stays on one line. Re-verified against the published packages, not a local feed.

🤖 Generated with Claude Code