ScopeDescendants asserts the shape, not an exact emission count (#946) by rbuergi · Pull Request #955 · Systemorph/MeshWeaver

rbuergi · 2026-08-08T10:03:23Z

Fixes #946.

The flake, and why it was the pipeline's top blocker today

ScopeDescendants_NotifiesOnAllDescendantChanges waited for >= 3 distinct Added paths, then asserted HaveCount(3) exactly. scope:descendants matches anything created under the root — so one extra descendant landing inside that window (a satellite the framework writes for the new nodes, timing-dependent under load) turns a correct system into a red build.

It hit four unrelated PRs today — #927, #940, #942, #947 — always shard 1, always passing locally. Each occurrence blocked a green PR and burned a CI cycle.

Why this is the test's bug, not the product's

AGENTS.md already forbids this exact shape:

Never assert "exactly N change events" on a stream backed by pg_notify or any change feed that can race the initial-snapshot path. Filter on the emission shape, not the count.

The test now asserts what it means: each of the three created descendants emitted an Added change. Extra descendants are explicitly allowed, because the scope legitimately matches them.

Deliberately not changed

Sibling assertions in the same class count emissions to prove a negative ("sibling create must not emit for ancestors scope") — there a bound is the assertion, and none of them has flaked. Changing passing tests without evidence is its own risk; the commit notes the pattern so it's visible if one of them starts.

ObservableQueryIntegrationTests 9/9 locally. No What's New — test-only.

🤖 Generated with Claude Code

Fixes #946. ScopeDescendants_NotifiesOnAllDescendantChanges waited for >= 3 distinct Added paths and then asserted HaveCount(3) EXACTLY. `scope:descendants` matches anything created under the root, so one extra descendant landing inside that window — a satellite the framework writes for the new nodes, which is timing-dependent under load — turned a correct system into a red build. That made it the top flake of 2026-08-08: four unrelated PRs (#927, #940, #942, #947), always shard 1, always passing locally. AGENTS.md already forbids exactly this ("Never assert 'exactly N change events' on a stream backed by a change feed that can race the initial-snapshot path — filter on the emission shape"). The test now asserts what it actually means: each of the three created descendants emitted an Added change. Extra descendants are explicitly allowed, because the scope legitimately matches them. Left alone deliberately: the sibling assertions in this class that count emissions to prove a NEGATIVE ("sibling create must not emit for ancestors scope") — a bound is the point there, and none of them has flaked. Changing passing tests without evidence is its own risk; noted here so the pattern is visible if one of them starts. ObservableQueryIntegrationTests 9/9. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Copilot

Pull request overview

Updates the ScopeDescendants_NotifiesOnAllDescendantChanges integration test to assert the presence of the three expected descendant Added paths rather than asserting an exact emission/item count, eliminating a load-dependent CI flake while preserving the intended liveness contract of scope:descendants.

Changes:

Replace exact-count assertions with set-based containment checks for the three explicitly created descendant paths.
Make the wait condition in the test predicate depend on “all expected paths observed” instead of “>= N emissions then count == N”.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

github-actions · 2026-08-08T10:19:52Z