fix(test): restore env per-test, not once at the end

[tranthanhnhatkhoa]

The bug

Two describe blocks in test/gstack-memory-helpers.test.ts save env in beforeEach and restore it in afterAll:

beforeEach(() => {
  savedRealHome = process.env.HOME;   // re-captured every test
  process.env.HOME = testHome;
});

afterAll(() => {
  else process.env.HOME = savedRealHome;   // runs once, at the end
});

beforeEach re-captures on every test, so after the first one savedRealHome no longer holds the real HOME — it holds the previous test's temp dir. The afterAll then "restores" HOME to that stale fixture.

That fixture contains a .gbrain/config.json with engine=postgres, so the next file in the run reads it as if it were the user's real config:

test/gbrain-detect-install.test.ts:107
  expect(j.gbrain_engine).toBe('pglite')
  Expected: "pglite"   Received: "postgres"

That file passes in isolation (16/16) and fails in the suite — which is what makes this read as flakiness rather than a leak. It also takes out gbrain-sync's two split-engine cases for the same reason.

The fix

Pair the hooks. beforeEach saves, afterEach restores, so every capture is of a genuinely un-mutated value. No change to what the tests assert.

Verified

polluter alone       29 pass  0 fail   (unchanged)
polluter + victim    45 pass  0 fail   (was 42 pass, 3 fail)
full suite           17 fails -> 12    (3 gbrain-detect + 2 gbrain-sync gone)

Environment: bun 1.3.14, Ubuntu 24.04.4 LTS, main @ 960c3a8d (v1.60.2.0).

Related

Independent of #2480, which hardens gstack-config.test.ts against a different leak (module-scope GSTACK_HOME assignments in five suites). That one makes a victim immune; this one removes a polluter. Both are needed — they're separate bugs with the same symptom class, and the suite has enough cross-file env mutation that it's worth a broader audit than either PR does.

Worth noting the two are hard to spot because bun shares one process across files and the suite's exit code is unreliable (#2435), so a leak like this shows up as "those tests are flaky" rather than as a failure anyone chases.

🤖 Generated with Claude Code

https://claude.ai/code/session_01H3KRWmyJnSvzzxeooDxuU3

[trunk-io]

Merging to main in this repository is managed by Trunk.

• To merge this pull request, check the box to the left or comment /trunk merge below.

After your PR is submitted to the merge queue, this comment will be automatically updated with its status. If the PR fails, failure details will also be posted here