Repository navigation
Description
What happens
On desktop nightly 0.0.46-nightly.20261012.3021, the sidebar's "update everything" button (#18053) updates desktop-app servers but never includes a CLI server. The connected server is a launcher-managed t3 __service-launcher systemd service on 0.0.46-nightly.20261012.3010, which advertises serverSelfUpdate: "boot-service". That server's row in Settings → Connections still shows its update icon, and updating from there works.
npm already has the target version (nightly dist-tag = 0.0.46-nightly.20261012.3021), and both versions are on the nightly channel, so the server should be included.
Cause
useUpdateEverything keeps non-desktop-managed servers only after npm confirms the target version:
apps/web/src/components/sidebar/useUpdateEverything.ts:56-58filters withuseNpmReleasePublished(targetVersion)apps/web/src/state/npmRelease.ts:14checks with Effect'sHttpClient.get+FetchHttpClient.layer
Effect's HttpClient adds traceparent and b3 headers to every request by default (HttpClient.TracerPropagationEnabled defaults to true). Those headers make the browser send a CORS preflight, and registry.npmjs.org answers OPTIONS with 404 and no Access-Control-Allow-Origin. The fetch is blocked, the atom fails, AsyncResult.getOrElse(result, () => false) (npmRelease.ts:40) reports "not published", and every npm/CLI server is dropped from the run. This affects the desktop renderer and app.t3.codes alike. The Connections "Update all" and per-row update don't check npm, which is why they still work.
Repro / evidence
Preflight is rejected:
$ curl -sI -X OPTIONS -H 'Origin: https://app.t3.codes' -H 'Access-Control-Request-Method: GET' \
https://registry.npmjs.org/t3/0.0.46-nightly.20261012.3021
HTTP/2 404 # no access-control-allow-origin
What the effect 4.0.1 the web app ships sends for HttpClient.get(...) with FetchHttpClient.layer:
{"b3":"d25f…-a0cb…-1","traceparent":"00-d25f…-a0cb…-01"}
The same fetch inside the repo's Electron 44 (Chromium), run from an https:// page:
plain GET: ok 200
with trace headers: FAILED Failed to fetch
Expected
"Update everything" includes boot-service CLI servers once npm has the target version.
Suggested fix
Turn off trace propagation for this cross-origin request in npmReleasePublishedAtom (apps/web/src/state/npmRelease.ts):
}).pipe(
Effect.provideService(HttpClient.TracerPropagationEnabled, false),
Effect.provide(FetchHttpClient.layer),
)With that change no extra headers are sent and the request stays a simple CORS GET (checked with the same probe). A test on the fetch-failure path would also help: right now a failed check looks the same as "not published", which hid this.
Activity
juliusmarminge commentedon Oct 12, 2026
Note
Grok responding on behalf of Julius.
Confirmed bug. On main,
npmReleasePublishedAtom(apps/web/src/state/npmRelease.ts) callsHttpClient.getthroughFetchHttpClient.layerwith trace propagation on. That adds thetraceparent/b3headers, and those headers make the browser send a preflight.registry.npmjs.organswersOPTIONSwith a 404 and noAccess-Control-Allow-Origin, but a plain GET returns 200 withaccess-control-allow-origin: *. I checked both with curl. The failed fetch then hitsAsyncResult.getOrElse(result, () => false), souseUpdateEverythingdrops every server that isn't desktop-managed.Workaround: update CLI/npm servers from their own row in Settings → Connections, or use "Update all" there. Neither one checks npm.
The likely fix is to turn off
HttpClient.TracerPropagationEnabledfor this request. It would also help to stop treating a failed check as "not published".