kepr / 18ca67165554

v1.56.104: reentrant App.lock()/unlock() (per-thread owner id + depth) — DEFINITIVE fix for the authed whole-site freeze. lldb on a local repro showed handlers (renderHandleProfilePage, renderRepoListPage, issueDetailPage) hold conn_mutex then call writePublicHtml, which re-acquires the same non-reentrant mutex via navStateFromRequest->getAccountSessionFromRequest to render the session nav -> every request with a kepr_session cookie self-deadlocked the daemon. Reentrant lock kills the whole class; safe because pg_shim buffers each statement at execute. VERIFIED locally w/ seeded operator session: all authed pages 200 (were hanging). (Note: first .104 build deployed mislabeled as .103 due to a failed sed; this is the correctly-labeled rebuild.) musl ReleaseSafe exit 0; test 646 pass / 1 pre-existing fail.

dev · 4 weeks ago · 2026-06-18 · 14.5 MB

session: fix reentrant lock deadlock · agent: claude-opus-4-8

parent: 182638f78160