mirror of
https://github.com/MoonshotAI/kimi-code.git
synced 2026-08-01 20:44:53 +00:00
248e47fa24
Net -40% PR test lines (7504 -> 4508) while keeping a verified probe for every failure mode the PR introduces: - delete harness self-tests and the subprocess e2e harness (klient dual-instance, spawn machinery) whose behavior is pinned by real scenario tests; cover holder-death lock release with a cheap SIGKILL unit test in kernel-file-lock instead - drop duplicate verdict-matrix variants, trivially-true assertions, and re-proven endpoint checks across fencing, ledger, journal, broadcaster, fs-watch, config, and web suites - compact shared rigs (sessionLease helpers, fileFencing/fileLedger slim DI wiring, broadcaster it.each + shared sessionEvents helper, ws-lifecycle shared setup) - sacrifice a documented set of sole-probe tests for volume (fsService gitignore cache, skill-listing reminder, kap-server ownership e2e, journal LAST-header/malformed-middle, cross-process wait/deadline paths); each is recoverable from history if the behavior regresses - add a held_by=peer ownership-join probe to kap-server sessions tests |
||
|---|---|---|
| .. | ||
| kimi-code | ||
| kimi-inspect | ||
| kimi-web | ||
| vis | ||
| vscode | ||