[Bug]: Desktop renderer polls getLocalEnvironmentBootstraps via sync IPC ~2x/s, flooding desktop traces #19505

Before submitting

• I searched existing issues and did not find a duplicate.
• I included enough detail to reproduce or investigate the problem.

Area

apps/desktop

Steps to reproduce

1. Launch T3 Code Desktop (Linux, local primary backend only, no WSL).
2. Leave it open with the sidebar visible; no interaction needed.
3. Count spans in ~/.t3/userdata/logs/desktop.trace.ndjson:

   jq -r .name ~/.t3/userdata/logs/desktop.trace.ndjson* | sort | uniq -c | sort -rn | head

Expected behavior

The renderer learns about desktop-local backend topology changes when they happen (or polls rarely), and an idle app writes little to the desktop trace.

Actual behavior

useDesktopLocalBootstraps (apps/web/src/connection/useDesktopLocalBootstraps.ts) polls desktopBridge.getLocalEnvironmentBootstraps() every 2 s per mounted consumer (LegacySidebar, CommandPalette). That bridge call is a synchronous IPC (ipc.handleSync) and each call runs desktop.backendConfiguration.resolvePrimaryLabel, which re-reads persisted settings just to build the label.

Each poll emits four spans (desktop.ipc.invokeSync → desktop.ipc.method → desktop.ipc.window.getLocalEnvironmentBootstraps → desktop.backendConfiguration.resolvePrimaryLabel), and getLocalEnvironmentEnabled is polled alongside it. On an idle machine this is the bulk of the desktop trace:

265437 spans total (2026-10-12T02:19Z → 2026-10-14T14:30Z)
 69900 desktop.ipc.method
 45380 desktop.ipc.invokeSync
 29963 desktop.ipc.window.getLocalEnvironmentBootstraps
 29963 desktop.backendConfiguration.resolvePrimaryLabel
 24520 desktop.ipc.invoke
 15411 desktop.ipc.localEnvironment.getEnabled

Per-second histogram of resolvePrimaryLabel calls: 14,858 one-second buckets with exactly 2 calls, i.e. a steady ~2 calls/s for as long as the window is open. The topology never changed during that time (single primary backend, no WSL).

Effects:

• Rotated trace files fill with polling noise, crowding out useful history when debugging other problems (I hit this while investigating websocket drops).
• Sync IPC from the renderer blocks its main thread for each round trip; cheap individually (~0.1 ms handler time) but continuous and needless.

The hook's doc comment already notes the cause: "The bridge exposes no change event, so we re-read on an interval".

Suggested fix: have the desktop push a topology-changed event (the pool already knows when instances register/unregister or bootstrap tokens rotate) and let the hook subscribe, dropping the interval. Alternatively, share one poller across consumers and/or stop tracing this hot path.

Impact

Minor bug or occasional failure

Version or commit

0.0.46-nightly.20261014.3105 (also present on main @ 7ac952c)

Environment

NixOS (Linux 7.2), T3 Code Nightly desktop, Electron 44.5.1 / Chrome 152

Logs or stack traces

$ jq -r 'select(.name=="desktop.backendConfiguration.resolvePrimaryLabel") | (.startTimeUnixNano|tonumber/1e9|floor)' desktop.trace.ndjson* \
    | sort | uniq -c | awk '{print $1}' | sort -n | uniq -c | sort -rn
  14858 2
     76 1
     37 3
     15 4