Repository navigation
Test hygiene: no 15 s sleep in the sync test, bounded polls instead of fixed sleeps, test doubles out of the package - #336
Conversation
…f fixed sleeps, test doubles out of src - syscall_cli.cmd_sync polls on a module constant SYNC_POLL_S (still 15 s); test_sync_cli_waits_for_the_done_marker shrinks it instead of paying it (15.01 s -> 0.31 s). CLI behaviour unchanged. - tests/helpers.py: wait_until(predicate, timeout, interval). Fixed sleeps and hand-rolled deadline loops around real subprocesses now poll the condition they wait for: test_sweep_git_locks (git visible in the process table as the sweep sees it), test_harness and test_local_compute (killed child gone), test_syscall_cli (request marker seen), test_watcher (done marker written). - tests/fakes.py: FakeHarness and RecordingDispatcher moved out of src/outerloop/harness.py and src/outerloop/tick.py; nothing in src or scripts used them. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
There was a problem hiding this comment.
Round 1 — reviewed head 02e4627a — reviewer summarizer:hermes/gpt-5.6-terra over coverage+credentials+deployment+general+lifecycle+prose.
terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.
Verdict: 1 blocking, 2 advisory.
3 findings attached to the lines below.
Verdict: blocking — the new live-process readiness wait times out in two serial sweep tests, causing them to fail before exercising the sweep. Also, the sync test lacks coverage proving the poll interval is configurable. Rejected findings: none; the six reports of the two serial test failures were merged because they make the same claim about the same readiness assertion.
…sweep does; the sync test times itself - _live no longer depends on pgrep plus ps agreeing: it checks the child is alive and reads its cmdline the way the sweep does (/proc on Linux, ps on a Mac), so the test's readiness and the sweep's detection never diverge. - the sync test asserts it returns well under one default interval, proving the shrunk SYNC_POLL_S is actually used and not just that a marker arrived. - wait_until's docstring says a timeout fails the test. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
1 participant
Digest #332, grouping 3 (test hygiene), the mechanical part. No behaviour change in the kernel.
cmd_syncpolled with a hardtime.sleep(15); the interval is now the module constantSYNC_POLL_Sand the test shrinks it. 15.01 s → 0.31 s.tests/helpers.py::wait_until): the sweep tests poll for the git process being live the way the sweep script itself looks for it; the harness and local-compute tests poll for the child to be gone; the syscall-CLI and watcher tests poll for their markers.FakeHarnessandRecordingDispatcher, move totests/fakes.py; nothing undersrc/orscripts/used them.Left for later, deliberately: consolidating the per-file
FakePopen/FakeMeasurer/_githelpers (a larger refactor with more room for semantic drift).🤖 Generated with Claude Code