Windows app unlaunchable ("This app can't open") - MSIX ships a file at a 275+ char path; 26.721 also re-flags itself in use

Open 💬 2 comments Opened Jul 29, 2026 by jaythejet
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

The ChatGPT desktop app for Windows (Store packages OpenAI.Codex 26.721.4979.0 and OpenAI.CodexBeta 26.715.3651.0) became permanently unlaunchable on my Windows 11 Home PC (build 26200, default settings): the window closes instantly, then Windows shows "This app can't open." Reinstalling, repairing, and resetting never helped.

Root cause (diagnosed with full event-log analysis and AppxBlockMap hash verification):

  • The package ships this file at a 275-character absolute path (279 in the Beta):

C:\Program Files\WindowsApps\OpenAI.Codex_26.721.4979.0_x64__2p2nqsd0c76g0\app\resources\app.asar.unpacked\node_modules\@worklouder\device-kit-oai\node_modules\@worklouder\wl-device-kit\node_modules\serialport\node_modules\@serialport\bindings-cpp\build\release\bindings.node

  • That exceeds Windows' default 260-char MAX_PATH (LongPathsEnabled=0 is the default).
  • Windows' package-health check can't see the file, so it marks the package "Modified, NeedsRemediation." Every file hash-verifies perfectly against the AppxBlockMap - nothing is actually modified.
  • Launches then fail, and even while the app is running it dies the moment it respawns a child process: AppModel-Runtime event ID 6, error 0x3CFC (ERROR_NEEDS_REMEDIATION), "error encountered while checking the machine-level package status."
  • Because every fresh install ships the same path, reinstalling gets re-flagged within minutes - users are stuck with no way out through normal means.

Workaround attempted:

  1. Set HKLM\SYSTEM\CurrentControlSet\Control\FileSystem\LongPathsEnabled = 1
  2. Reboot (services cache the old value)
  3. Fully remove and reinstall the app to clear the stuck machine-level verdict

UPDATE - the workaround only partially holds, and the difference between your two builds is diagnostic:

  • Beta build 26.715.3651.0 (same >260-char path): stays Status=Ok and runs fine after the workaround, including through active use.
  • Stable build 26.721.4979.0: freshly reinstalled with Status=Ok and stable while idle (verified 3+ min with child-process churn), but within ~1 minute of ACTIVE use it hits 0x3CFC on a child-process spawn again (post-reboot, LongPathsEnabled=1 in every process; observed 2026-07-28 23:43:36 local) and the package is re-marked "Modified, NeedsRemediation." So something 26.721 does at runtime that 26.715 does not (updater? elevated helper spawn?) re-triggers the machine-level status failure independent of the path-length trigger.

Suggested fixes: (1) flatten or shorten the dependency chain under app.asar.unpacked (the five nested node_modules levels under @worklouder/device-kit-oai) so the longest shipped path stays safely under 260 characters from the WindowsApps package root - every default-configured Windows machine is exposed; and (2) investigate what child-process spawn 26.721 performs during active use that fails the machine-level package status check even on a long-path-enabled, freshly-rebooted system where 26.715 passes.

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 1 month ago

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

  • #35737

Powered by Codex Action

jaythejet · 1 month ago

UPDATE / correction to the report above: the claim that Beta 26.715 stays healthy turned out to be wrong — it just fails more slowly. On the same machine, OpenAI.CodexBeta 26.715.3651.0 survived ~1h45m of active use and then failed identically (AppModel-Runtime Id=6, 0x3CFC x5 at 2026-07-29 01:02:46 local; AppX container destroyed; package re-flagged Modified, NeedsRemediation). So BOTH current builds self-corrupt during active use — 26.721 within ~1 minute, 26.715 within a couple of hours.

This matches the deeper investigation in #35737 (sandbox/ACL pass appearing to modify the package's own install folder during use). Worth repeating from my verification: all 9,544 package files hash-verify clean against the AppxBlockMap while the package is flagged, so the corruption is security-descriptor/verdict state, not file contents.

Practical workaround until a fixed build ships: full Remove-AppxPackage + Store reinstall restores a working session each time (minutes-to-hours of active use per cycle). LongPathsEnabled=1 remains necessary on default machines for the separate MAX_PATH issue in the OP, but does not prevent the self-corruption.