Flaky test: concurrent CLI redownload times out waiting for the progress log to go away

[EhabY]

test/unit/core/cliManager.concurrent.test.ts fails intermittently on CI. It passed on attempt 1 of run 35406584668 and failed on attempt 2 of the same commit, in Unit Test (Linux, Electron 37).

FAIL test/unit/core/cliManager.concurrent.test.ts > CliManager Concurrent Downloads > redownloads when version mismatch is detected concurrently
AssertionError: promise resolved "undefined" instead of rejecting
- Expected: Error { "message": "rejected promise" }
+ Received: undefined
 ❯ test/unit/core/cliManager.concurrent.test.ts:102:56

Line 102 is the progress-log half of expectLockFilesRemoved:

await vi.waitFor(async () => {
  await expect(fs.access(binaryPath + ".lock")).rejects.toThrow();
  await expect(fs.access(binaryPath + ".progress.log")).rejects.toThrow();
});

fs.access resolved, so <binary>.progress.log was still on disk when the poll gave up. The test starts three fetchBinary calls, two of which detect the version mismatch and redownload, and Promise.all resolves before every peer has finished releasing its progress log. vi.waitFor defaults to a 1s timeout at a 50ms interval, which a loaded runner can outlast.

#1043 added that polling for this same flake, so the window helps but is not always enough.

Worth considering:

• Give vi.waitFor an explicit timeout, which is the cheap fix but still time-based.
• Make the cleanup observable and await it, rather than polling the filesystem.
• Decide whether the progress log belongs in the assertion at all after a concurrent redownload, since the lock is the part the contract depends on.

[linear-code]

VSC-21