chore: tranche 2 — nine more test suites leave core, and two dead ones are deleted by rbuergi · Pull Request #2968 · Systemorph/MeshWeaver

rbuergi · 2026-09-01T12:10:20Z

🚦 Stacked on #2847, and gated on two merges

Base is chore/drop-migrated-tests, not main. Merge order:

MeshWeaver.Plugins#1097 — the nine suites build and RUN there
chore: 17 mesh test suites move to MeshWeaver.Plugins — core stops building them #2847 — tranche 1
this PR

Not armed. Deleting before #1097 lands would leave the nine running nowhere.

What moves

Nine of the smallest surviving leaves — nothing in core's tree references any of them — chosen so
the dual-path support-lib shape is proven at 20 test files rather than discovered at
Graph.Test's 165. They cover all three dependency shapes:

Shape	Suites
none	`PluginImage`, `Testcontainers`, `Kernel`, `Json`
`Fixture`	`Reactive.Assertions`, `Search`, `Serialization`
`Monolith.TestBase`	`Todo`
both	`FutuRe`

What is DELETED outright — two dead projects

Project	Evidence
`MeshWeaver.Hub.Fixture`	Zero `ProjectReference`s in core, zero source mentions, zero across all three satellites. Only `MeshWeaver.slnx` named it — built and referenced by nothing.
`MeshWeaver.TestDomain`	Zero consumers; its only mentions are its own files. (A Plugins skill doc cites `CashflowElement.cs` as a modelling example by path — prose, not a build edge; being updated separately.)

These are not migrations. Nothing consumes them, so nothing needs them anywhere.

Coverage is CHECKED, not asserted

plugins#1095's drift gate reports OK
for all nine — every [Fact]/[Theory] in core's copy has a counterpart in Plugins:

FutuRe 59 · Json 60 · Reactive.Assertions 23 · Serialization 14 · Kernel 11 · Todo 6 ·
PluginImage 4 · Testcontainers 4 · Search 3

In Plugins they were executed, not just compiled: 8 suites green (487 tests); PluginImage's
4 skip cleanly because MW_TEST_IMAGE is unset locally — CI sets it — and every suite emitted a
.trx, so none is the silent no-verdict shape.

Safety, re-verified on the merged tree

No surviving project has a ProjectReference to any of the eleven. After the deletion neither
MeshWeaver.slnx nor shard-assign.sh names one.

Numbers that moved because the tree shrank — each re-derived by probing at 0

Guard	Before	After	Why
`TestTimeoutLiteralRatchetGuard.Baseline`	401	386	the nine carried 15 literals
`BlockingBridgeInTestRatchetGuard.TotalBudget`	26	26	neither `.allow` file names any of the nine
`RouterAsTestRequestOriginRatchetGuard.TotalBudget`	3	3	same
`ShardWeightHeadroomGuard` parse floor	20	10	see below
`shard-assign.sh`	24 entries / 731 s	14 / 658 s	10 weight rows removed

🚨 The parse floor is a NON-VACUITY check, not a policy floor — it exists to catch a parse that
returned nothing, so it has to track the table's real size: 41 entries → 24 (#2847) → 14 here. A
floor left above the real count fails every run while claiming the parse broke.
Hosting.Orleans.Test (269 s) remains the long pole and still clears the headroom trigger (60 % of
the 480 s cap = 288 s).

Verification actually executed

MW_ALLOW_SOLUTION_BUILD=1 dotnet build MeshWeaver.slnx -c Release -warnaserror --no-incremental
→ 59/59 projects, 0 warnings, 0 errors
dotnet test test/MeshWeaver.Documentation.Test -c Release → 150/150 passed

🤖 Generated with Claude Code

…s are deleted Stacked on #2847 (tranche 1, 17 suites). The directive is that all tests move out of core and are REMOVED from it; this is the second tranche. MOVED to MeshWeaver.Plugins (plugins#1097), where they build and RUN: PluginImage, Testcontainers, Kernel, Json, Reactive.Assertions, Search, Serialization, Todo, FutuRe. Nine of the smallest surviving leaves, chosen so the dual-path support-lib shape is proven at 20 test files rather than discovered at Graph.Test's 165. They cover all three dependency shapes: none, Fixture, Monolith.TestBase, and both. DELETED outright, because nothing anywhere uses them: MeshWeaver.Hub.Fixture — zero ProjectReferences in core, zero source mentions, zero in all three satellite repos. Only MeshWeaver.slnx named it, so it was built and referenced by nothing. MeshWeaver.TestDomain — zero consumers; its only mentions are its own files. (A Plugins skill doc quotes CashflowElement.cs as a modelling example by PATH; that citation is prose, not a build edge, and is being updated separately.) Safety, re-verified on the merged tree: no surviving project has a ProjectReference to any of the eleven, and after the deletion neither MeshWeaver.slnx nor shard-assign.sh names any of them. Coverage is not lost, and that is now CHECKED rather than asserted: plugins#1095's drift gate reports OK for all nine — every [Fact]/[Theory] in core's copy has a counterpart in Plugins (FutuRe 59, Json 60, Reactive.Assertions 23, Serialization 14, Kernel 11, Todo 6, PluginImage 4, Testcontainers 4, Search 3). Numbers that moved because the tree shrank, each re-derived by probing at 0: * TestTimeoutLiteralRatchetGuard.Baseline 401 -> 386 (the nine carried 15 literals). * BlockingBridgeInTestRatchetGuard.TotalBudget stays 26 and RouterAsTestRequestOriginRatchetGuard.TotalBudget stays 3 — neither .allow file names any of the nine, so neither budget moves. * ShardWeightHeadroomGuard's parse floor 20 -> 10. That is a NON-VACUITY floor — it catches a parse that returned nothing — so it has to track the table's real size: 41 entries -> 24 (#2847) -> 14 here. A floor left above the real count fails every run while claiming the parse broke. * shard-assign.sh: 10 weight rows removed, 24 entries/731 s -> 14/658 s. Hosting.Orleans.Test (269 s) remains the long pole and still clears the headroom trigger (60% of the 480 s cap = 288 s). Verified: full solution Release build with -warnaserror -> 59/59 projects, 0 warnings, 0 errors; dotnet test test/MeshWeaver.Documentation.Test -c Release -> 150/150 passed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

github-actions · 2026-09-01T12:25:32Z