test: remove shared-DB and timer flakiness

[coleleavitt]

Summary

• set plugin test timeout to 30s so SQLite's 5s busy-timeout reports the actual lock stack instead of Bun killing the test first
• batch the 2,005 embedding-measurement setup inserts in one SQLite transaction
• return real completion promises from message-index schedulers and await them with a fake boot-quiet clock
• warm @opentui/core before parallel runtime-specifier imports to remove the TreeSitterClient TDZ race exposed by CI

Addresses #312 (improves lock-failure observability; per-file DB isolation remains open)
Fixes #330
Fixes #333

Verification

• targeted storage/scheduler scope: 16 pass
• two concurrent targeted runners: both exit 0
• 20 repeated scheduler runs: pass
• 20 repeated TUI runtime-import runs: pass
• bun run lint
• bun run typecheck
• plugin build

Assertions are unchanged; no SQLite timeout was hidden or extended.

Greptile Summary

The PR reduces SQLite and scheduler test flakiness without changing production database timeout behavior.

• Extends the plugin test timeout to 30 seconds so SQLite lock failures surface before the test runner terminates them.
• Batches embedding-measurement fixture inserts in one transaction.
• Exposes scheduler completion promises and uses a fake boot-quiet clock in the overlap test.
• Serializes the TUI core/testing module initialization path.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
packages/plugin/package.json	Raises the package test timeout so SQLite's configured busy timeout can produce a useful failure first.
packages/plugin/src/features/magic-context/message-index-async.ts	Returns deduplicated completion promises from reconciliation and clear/reindex scheduling while preserving per-session serialization.
packages/plugin/src/features/magic-context/message-index-async.test.ts	Replaces wall-clock waiting with fake timers and awaits both scheduler completion promises.
packages/plugin/src/features/magic-context/storage-embedding-measurements.test.ts	Batches the large measurement-corpus fixture setup in one SQLite transaction.
packages/plugin/src/tui/tui-compiled-runtime-imports.test.ts	Initializes the TUI core module before parallel runtime imports to avoid a module-initialization TDZ.

Sequence Diagram

sequenceDiagram
    participant Test
    participant BootQuiet as Boot-quiet scheduler
    participant Lock as Session lock
    participant Index as Message index
    Test->>BootQuiet: schedule reconciliation
    Test->>BootQuiet: schedule clear and reindex
    Test->>BootQuiet: advance fake clock
    BootQuiet->>Lock: queue reconciliation
    BootQuiet->>Lock: queue clear and reindex
    Lock->>Index: reconcile existing messages
    Lock->>Index: clear indexed session
    Lock->>Index: rebuild from surviving messages
    Index-->>Test: resolve completion promises

Loading

_{Reviews (3): Last reviewed commit: "test(tui): stabilize runtime import orde..." | Re-trigger Greptile}

Context used:

• Knowledge Base — Plugin memory, notes, and context tools
• Knowledge Base — Magic Context installation and release flow

[cubic-dev-ai]

No issues found across 4 files

_{Re-trigger cubic}

[cubic-dev-ai]

2 issues found across 1 file (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="packages/plugin/package.json">

<violation number="1" location="packages/plugin/package.json:42">
P2: The PR's stated first goal is "Enable Bun test-file isolation for plugin tests," but this change removes the `--isolate` flag that was added earlier in this PR, leaving the test command with no isolation flag (only `--timeout 30000`). If Bun does not isolate test files into separate processes by default, the shared-state/shared-DB flakiness the PR exists to fix is not actually removed. Either restore `--isolate` (combined with the timeout) or confirm in the PR description that Bun isolates files by default and that isolation is intentionally dropped.</violation>

<violation number="2" location="packages/plugin/package.json:42">
P3: The new `--timeout 30000` raises the global per-test timeout from Bun's 5000ms default to 30s for every test file, which contradicts the PR's stated verification that "no global timeouts increased" and masks genuinely hanging tests in CI for 30s each. Set only the specific tests that actually need the longer budget instead of raising the global default, or document which tests justify 30s.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: The PR's stated first goal is "Enable Bun test-file isolation for plugin tests," but this change removes the --isolate flag that was added earlier in this PR, leaving the test command with no isolation flag (only --timeout 30000). If Bun does not isolate test files into separate processes by default, the shared-state/shared-DB flakiness the PR exists to fix is not actually removed. Either restore --isolate (combined with the timeout) or confirm in the PR description that Bun isolates files by default and that isolation is intentionally dropped.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/plugin/package.json, line 42:

<comment>The PR's stated first goal is "Enable Bun test-file isolation for plugin tests," but this change removes the `--isolate` flag that was added earlier in this PR, leaving the test command with no isolation flag (only `--timeout 30000`). If Bun does not isolate test files into separate processes by default, the shared-state/shared-DB flakiness the PR exists to fix is not actually removed. Either restore `--isolate` (combined with the timeout) or confirm in the PR description that Bun isolates files by default and that isolation is intentionally dropped.</comment>

<file context>
@@ -39,7 +39,7 @@
     "check:tui-compiled": "bun run build:tui && git diff --exit-code -- src/tui-compiled && test -z \"$(git status --porcelain -- src/tui-compiled)\"",
     "typecheck": "tsc -p ../retina-local-fs/tsconfig.build.json && tsc --noEmit && tsc -p tsconfig.scripts.json",
-    "test": "bun test --isolate",
+    "test": "bun test --timeout 30000",
     "lint": "biome check .",
     "lint:fix": "biome check --write .",
</file context>

Suggested change

	"test": "bun test --timeout 30000",
	"test": "bun test --isolate --timeout 30000",

[cubic-dev-ai]

P3: The new --timeout 30000 raises the global per-test timeout from Bun's 5000ms default to 30s for every test file, which contradicts the PR's stated verification that "no global timeouts increased" and masks genuinely hanging tests in CI for 30s each. Set only the specific tests that actually need the longer budget instead of raising the global default, or document which tests justify 30s.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At packages/plugin/package.json, line 42:

<comment>The new `--timeout 30000` raises the global per-test timeout from Bun's 5000ms default to 30s for every test file, which contradicts the PR's stated verification that "no global timeouts increased" and masks genuinely hanging tests in CI for 30s each. Set only the specific tests that actually need the longer budget instead of raising the global default, or document which tests justify 30s.</comment>

<file context>
@@ -39,7 +39,7 @@
     "check:tui-compiled": "bun run build:tui && git diff --exit-code -- src/tui-compiled && test -z \"$(git status --porcelain -- src/tui-compiled)\"",
     "typecheck": "tsc -p ../retina-local-fs/tsconfig.build.json && tsc --noEmit && tsc -p tsconfig.scripts.json",
-    "test": "bun test --isolate",
+    "test": "bun test --timeout 30000",
     "lint": "biome check .",
     "lint:fix": "biome check --write .",
</file context>

Suggested change

	"test": "bun test --timeout 30000",
	"test": "bun test"

[alfonso-magic-context]

Thank you, Cole — filed before our independent #330/#333 changes landed, and your read of the timing signature was useful, especially the #312 point that Bun's 5s test timeout and SQLite's 5s busy_timeout hide lock errors behind the same wall-clock cutoff.

In parallel, master landed different implementations: #330 now exercises the corpus-eviction invariant with an injectable small cap instead of 2,005 disk-backed inserts, and #333 waits for the observable rebuild state instead of a fixed sleep — so we don't want to merge a second implementation of those two.

Two pieces of this PR are independently useful and we're extracting them with credit: bun test --timeout 30000 so a SQLite lock surfaces as a lock error before the runner kills the test, and the @opentui/core warm-up before parallel runtime-specifier imports (a real TDZ race our suite hasn't fixed). To be explicit: the timeout change is diagnostic — it doesn't isolate per-file databases or change test busy_timeout, so #312 stays open for the real contention fix. Closing this PR as superseded; the extractions will land crediting you.