Multiple sidebar bugs after workspace directory migration: "56 years ago" timestamps, missing threads, threads vanish on click (macOS) #17317

Activity

1. github-actions

   added

   bugSomething isn't workingSomething isn't working

   appIssues related to the Codex desktop appIssues related to the Codex desktop app

   on Apr 10, 2026
2. github-actions commented on Apr 10, 2026

   

   github-actionsbot

   on Apr 10, 2026 – with GitHub Actions

   Contributor

   More actions

   Potential duplicates detected. Please review them and close your issue if it is a duplicate.

   • Desktop project sidebar hides active threads after state DB migration drift and stale temporary sessions #17304
   • Desktop thread panel hides existing workspace threads after workspace restructuring; backend pagination still returns them #16901
   • Codex Mac Desktop App: existing threads don't load after app restart #16817
   • Codex threads disappear after a few days on macOS #16549

   Powered by Codex Action
3. Aley3567 commented on Aug 16, 2026

   

   Aley3567

   on Aug 16, 2026 – with ChatGPT Codex Connector

   More actions

   Confirmed independently on macOS with ChatGPT/Codex Desktop 26.810.52044 (6662). This appears to be a mixed timestamp-unit bug and is not specific to a workspace-path migration.

   Evidence

   The thread data exposed to the Desktop renderer contains mixed units:

   • a normal local Codex thread can have updatedAt: 1786849794 (Unix seconds);
   • other entries, including some pinned state, can use a millisecond value such as 1783825615000.

   Passing the first value directly to JavaScript reproduces the symptom:

   new Date(1786849794).toISOString()
   // 1970-01-21T...
   
   new Date(1786849794 * 1000).toISOString()
   // 2026-08-16T...

   In the bundled renderer (webview/assets/app-initial-BqZ9AFkF.js) I found two relevant direct constructions:

   • new Date(q).toISOString() in the hover-card path;
   • new Date(f).toISOString() in the sidebar-row path.

   The same bundle already contains a timestamp normalizer (minified as K8c) that multiplies second-scale values by 1000, but these two paths bypass it. The installed application was not modified; its signature still verifies.

   Suggested fix

   The durable fix should normalize timestamps once at the thread DTO boundary, rather than teaching every component about seconds vs. milliseconds:

   function toEpochMilliseconds(value: unknown): number | null {
     const timestamp = Number(value);
     if (!Number.isFinite(timestamp)) return null;
   
     // Current Unix milliseconds are ~1e12; seconds are ~1e9.
     return Math.abs(timestamp) < 100_000_000_000
       ? timestamp * 1_000
       : timestamp;
   }

   Then keep the renderer's internal contract as epoch milliseconds (or Date) only.

   For a minimal hotfix, route both direct call sites through the existing normalizer:

   new Date(K8c(q)).toISOString()
   new Date(K8c(f)).toISOString()

   Regression coverage should include integer seconds, fractional seconds, milliseconds, missing/invalid values, hover cards, and sidebar rows. That would fix the displayed age without rewriting any persisted timestamps.
4. Aley3567 commented on Aug 17, 2026

   

   Aley3567

   on Aug 17, 2026

   More actions

   Follow-up: the timestamp bug reproduced again in ChatGPT/Codex Desktop 26.810.52044 (6662).

   I validated the minimal renderer hotfix locally by routing the two bypassing call sites through the bundle's existing seconds/milliseconds normalizer:

   new Date(K8c(q)).toISOString()
   new Date(K8c(f)).toISOString()

   Verification results:

   • both unnormalized call sites are absent after patching;
   • both normalized call sites occur exactly once;
   • ASAR entry count remains unchanged (8779 before and after);
   • the ASAR integrity hash was updated to the actual archive SHA-256;
   • strict deep code-sign verification passes after re-signing;
   • the app's entitlements are byte-for-byte unchanged.

   This confirms that timestamp-unit normalization is sufficient for these two UI paths. The local patch is only a diagnostic/workaround: an app update or repair can replace app.asar, so the durable upstream fix should still normalize updatedAt once at the thread DTO boundary and expose a milliseconds-only internal contract. Component-level normalization is useful as defense in depth, but should not be the primary contract.