[jeremydmiller]

What was wrong

It was not the generator. A JasperFx.Events.SourceGenerator at another version than the JasperFx.Events runtime generates normally on its own: 2.100.0 against a 2.100.1 runtime emits the evolver and the marker.

The cause, reproduced with Polecat 5.49.0 and a direct JasperFx.Events.SourceGenerator 2.100.0 (Wolverine's PolecatTests shape):

1. The SDK drops the pinned copy. Its package file conflict resolution (_HandlePackageFileConflicts) sees two analyzers named JasperFx.Events.SourceGenerator.dll and keeps only the higher assembly version. -v:diag shows it: "Choosing 'Analyzer:polecat/5.49.0/…' because AssemblyVersion '2.100.1.0' is greater than '2.100.0.0'".
2. A project target removes the bundled copy. Wolverine's DropDuplicateBundledEventSourceGenerator drops the copy under /polecat/. It was written for the equal-version case, where both copies survive to that point.
3. Nothing reaches the compiler. With the bundled copy gone there is no generator left: no evolvers, no marker, no warning. The first sign is No source-generated dispatcher found at run time.

At equal versions the SDK cannot pick a winner, both copies survive, and the project target or JasperFx's own dedupe (#891) leaves one. That is why pinning everything to 2.100.1 fixed wolverine#5394.

What changed

• JFXEVT1138, a build-time warning in buildTransitive/JasperFx.Events.targets.
  • When it fires: the SDK dropped a copy of the generator (it is in _ConflictPackageFiles) and no copy reached the compiler.
  • What it names: the dropped copy, and the version of the JasperFx.Events runtime.
  • What it suggests: pin the generator to that version, or remove the target that drops the bundled copy.
  • When it runs: AfterTargets="CoreCompile", so it reads exactly what the compiler was given, whatever order the BeforeTargets hooks ran in.
  • Kept quiet otherwise: a version difference on its own stays silent, because the higher copy runs.
  • Opt-out: JasperFxEventsReportLostSourceGenerator=false.
• The run-time message for a project with no marker (the "left no trace" one) now names a generator pinned to another version than the runtime as a cause, with the runtime's version and JFXEVT1138.

How it was tested

• Smoke test: SmokeTestBuildTargets has four new arrangements, 10 in all, and passes. The refused answer is a_generator_at_another_version_never_produces_nothing_without_saying_so. The other three are a dropped lower copy with the higher copy compiled, no generator anywhere, and the opt-out.
• Unit test: SourceGeneratorMarkerTests.an_unconfirmed_assembly_names_a_generator_pinned_to_another_version_than_the_runtime. All 1,832 EventTests pass.
• Real packages: checked by hand against Polecat 5.49.0 with generator 2.100.0 or 2.100.1, with and without Wolverine's drop target. JFXEVT1138 appears only in the 2.100.0 + drop case, the one that generated nothing.

Not done

• Transitive pinning: not done. The runtime package can't flow its matching generator without becoming an analyzer carrier itself, and Two copies of the analyzer generate identical file paths, so file-scoped evolvers collide with CS0433 #891 / marten#4557 already settled how the analyzer flows.
• Follow-up for Wolverine: DropDuplicateBundledEventSourceGenerator in PolecatTests (and any copies of it) can be deleted, since JasperFx.Events has deduped equal copies since Two copies of the analyzer generate identical file paths, so file-scoped evolvers collide with CS0433 #891.

Fixes #1138

🤖 Generated with Claude Code