[DYH1319]

Problem

With the DotNet-first startup and SingleInstance enabled, launching the app a second time leaves the second .NET process hanging forever.

The second Electron process loses the single-instance lock and quits right after printing its socket port, before the socket bridge has connected. What happens next:

1. ElectronProcessActive transitions to Stopped → RuntimeControllerDotNetFirst.HandleStopped() sees the socket bridge is not stopped and calls socketBridge.Stop().
2. SocketBridgeService.StopCore() disposes the socket, and LifetimeServiceBase.Stop() moves the bridge to Stopping.
3. The bridge only ever reaches Stopped through BridgeDisconnected, but a socket that never connected never raises it.
4. The controller therefore never reaches Stopped. WaitReadyTask / WaitStoppedTask never complete, so the host's Main waits forever.

The same applies whenever Electron exits before the bridge connects (e.g. a crash during startup).

Fix

SocketBridgeService overrides Stop(): after the base implementation (dispose + Stopping), it transitions to Stopped if the disconnect event has not already done so. A connected socket that raises BridgeDisconnected behaves as before. The state is only checked after base.Stop() returns, so the Stopping transition inside base.Stop() can't run after Stopped and throw for an invalid backward transition.

The controller then reaches Stopped. Because Stopped > Ready, WaitReadyTask completes too, so the host can check State and exit.

Testing

Added lifecycle tests to ElectronNET.IntegrationTests. Neither class needs an Electron runtime:

• SocketBridgeServiceLifecycleTests starts a bridge against a closed loopback port, so it can never connect, and checks that after Stop() it:

  • reaches Stopped and completes WaitStoppedTask;
  • raises Stopping then Stopped, exactly once;
  • stays in Stopped after the pending connect attempt gives up.

  Without the fix, the first two fail with the bridge stuck in Stopping and Stopped never raised. The third one also fails, because the bridge never got to Stopped in the first place.

• LifetimeServiceBaseTests covers the state-machine rules the fix and the controller rely on:

  • Start / Stop transitions;
  • Stopped before Ready completing every earlier wait task, including WaitReadyTask, so a host awaiting readiness can observe the stopped runtime and exit;
  • a transition to the current state being ignored;
  • backward transitions, and Stop() after Stopped, throwing.

Results on Linux:

• The new tests pass consistently over 5 runs.
• Full integration suite: 111 passed, 18 skipped, and 1 failure. The failure is GlobalShortcutTests.Can_register_and_unregister, which also fails on its own on this machine: Wayland/Hyprland doesn't allow Electron global shortcuts.
• In a downstream app (Linux, Electron 43, DotNet-first, packaged), a second launch used to hang until killed. With the fix it exits in about 2 seconds.

🤖 Generated with Claude Code

[FlorianRappl]

Create the tests for the lifecycle. Rest looks fine to me.

[DYH1319]

Thanks! Added lifecycle tests for SocketBridgeService (the never-connected stop path, which fails without the fix) and LifetimeServiceBase in 1a0f287; neither needs an Electron runtime.

[FlorianRappl]

Right now almost all tests are failing. Not sure if this is related but beforehand tests worked and this is not just one faulty machine but across all runners.

[DYH1319]

Thanks for flagging this. I looked into it, and the failures come from develop rather than from this PR.

Electron crashes at startup with Cannot find module 'inherits', so ElectronFixture times out and every test that uses Electron fails. The develop push run for 05aa92e fails the same way (run 40534971218), and that was before this PR was opened.

The cause is the Host lockfile. It's copied into .electron/ next to the generated package.json, which still requests image-size ^1.2.1, while the lockfile pins 2.0.3. After the #1118 merge it also contains queue without its dependency inherits, so npm never installs inherits.

I've opened #1125 to repair the lockfile. Locally on Linux (Electron 38.2.2), that fix with this PR merged on top gives 112 passed, 18 skipped, 0 failed. Once it's merged, I'll update this branch from develop to re-run CI.

[github-actions]

pr-comment: Run #161

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Pending ⏳	Other ❓	Flaky 🍂	Duration ⏱️
1820	1658	0	0	0	162	0	3m 51s

🎉 All tests passed!

Github Test Reporter by CTRF 💚