[Desktop] Fatal renderer error: "Stale read from <Show>" when opening a project folder

[PaysNatal]

Environment

• App: OpenCode Desktop 1.18.10 (packaged, oc://renderer/)
• OS: macOS, arm64
• Runtime (from debug bundle manifest): Electron 42.3.3, Chrome 148.0.7778.218, Node 24.15.0

Error

Logged as fatal renderer error in renderer.log (url: oc://renderer/index.html, platform: desktop):

Error: Stale read from <Show>.
    at castError (oc://renderer/assets/main-DX5ErpZp.js:735:10)
    at handleError (oc://renderer/assets/main-DX5ErpZp.js:748:18)
    at runComputation (oc://renderer/assets/main-DX5ErpZp.js:501:12)
    at updateComputation (oc://renderer/assets/main-DX5ErpZp.js:470:3)
    at runTop (oc://renderer/assets/main-DX5ErpZp.js:566:7)
    at runQueue (oc://renderer/assets/main-DX5ErpZp.js:638:45)
    at completeUpdates (oc://renderer/assets/main-DX5ErpZp.js:594:5)
    at runUpdates (oc://renderer/assets/main-DX5ErpZp.js:584:5)
    at oc://renderer/assets/main-DX5ErpZp.js:473:7

Repro context

Triggered when opening/switching to a project folder in the desktop app.

Timeline from the debug bundle session logs:

[2026-08-01 10:18:32.859] [error] ResizeObserver loop completed with undelivered notifications.
[2026-08-01 10:18:32.885] [error] ResizeObserver loop completed with undelivered notifications.
[2026-08-01 10:18:47.937] [error] (renderer)   fatal renderer error { ... }

The renderer.log for the failing session is very short (only these lines) — the renderer appears to die during initial/first-screen render after the project switch. Server layer kept running normally (no server-side errors in server.log).

Observations

• Intermittent: several earlier sessions (previous day and same day) were completely clean — no renderer errors. Only this one session hit the fatal error.
• Renderer-only failure: the underlying server/agent layer was unaffected.
• The app auto-generated a debug bundle (opencode-debug-20260801T021852) containing renderer.log, main.log, server.log, crash.log, network logs and manifest.json — available if you need more details.

Initial assessment (not a diagnosis)

Looks like a SolidJS stale-signal race in the renderer: <Show> read a signal that had already been disposed/replaced during a batched update run triggered by the project switch (stack goes through runUpdates -> runQueue -> completeUpdates). The two ResizeObserver loop warnings right before it suggest layout churn around the same moment, possibly related.

Happy to share the full debug bundle or run extra logging on request — otherwise leaving the deep diagnosis to you.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Desktop crashes with stale Show read during cross-project session navigation #39840: Same "stale read from <Show>" renderer crash triggered by opening/switching to a project folder in Desktop 1.18.10. The symptoms and error signatures are very similar.

[david1gp]

Also hitting this on 1.18.10, but in the browser web UI against a headless server — no Electron involved. That may help narrow the cause, since this issue and #39840 are both Desktop/renderer reports.

Environment

• opencode 1.18.10 (opencode-ai from npm, latest tag), Linux x64
• Run as a systemd user service: opencode serve, bound 0.0.0.0:4096
• Accessed over the network in a normal browser (not the desktop app)
• Served bundle: assets/index-BBRqEIzV.js

Error

Full-page "Something went wrong / An error occurred while loading the application":

Error: Stale read from <Show>.
    at h2 (http://localhost:4096/assets/index-BBRqEIzV.js:2:10382)
    at C1 (http://localhost:4096/assets/index-BBRqEIzV.js:2:10580)
    at TD (http://localhost:4096/assets/index-BBRqEIzV.js:2:7024)
    at Cg (http://localhost:4096/assets/index-BBRqEIzV.js:2:6602)
    at tw (http://localhost:4096/assets/index-BBRqEIzV.js:2:8043)
    at AV (http://localhost:4096/assets/index-BBRqEIzV.js:2:8926)
    at lse (http://localhost:4096/assets/index-BBRqEIzV.js:2:8298)
    at Jo (http://localhost:4096/assets/index-BBRqEIzV.js:2:8228)
    at http://localhost:4096/assets/index-BBRqEIzV.js:2:6717

Same shape as the Desktop reports (castError → handleError → runComputation → updateComputation → runTop → runQueue), just minified differently.

Notes

• Not Electron-specific. Reproduces in a plain browser, so the renderer/packaging layer can be ruled out.
• Server side is clean. Nothing in journalctl or ~/.local/share/opencode/log/ at the time of the crash — the failure is entirely client-side.
• Restarting the service and hard-reloading the browser does not help; the same bundle hash is served before and after.
• Possibly closer to Web UI crashes with stale <Show> after /global/event network error behind base-path reverse proxy #28928 (stale <Show> after a /global/event network error) than to the Desktop folder-switching reports, given this is a remote server over the network where the event stream can drop. I did see an unrelated upstream socket drop in my logs during the same period, though I have not confirmed any connection.
• The UI reports its version as 1.18.9 while the binary is 1.18.10 — that appears to be Web UI version number lags: v1.17.17 is displayed when running v1.17.18, it is suspected that the embedded front-end build version is not synchronized #36232. Mentioning it only so the version string isn't confusing if it shows up in screenshots.

Trigger

I don't yet have a reliable reproduction — I hit this during normal use and haven't isolated the exact action, so I can't say whether it's fresh page load, session switching, or an event-stream drop. I'll follow up with the browser devtools console (including anything preceding the error) once I catch it again with devtools open.

[UltimateCedarFox]

+1