[emrecpp]

Problem

On Windows, with "load at startup" on, the engine sometimes stays stopped after logon: the UI and the hook are both running, but the mouse is not hooked until a manual Start.

The rolling daemon event trace (daemon-events.log) shows the pattern across several logons on one machine:

Logon	Events after Connected	Result
A	Stopped 8 s later, then Loaded, Running	hooked
B	Stopped 8 ms later, then nothing	stopped
C	Stopped 2.6 s later, then Loaded only at the manual Start ~90 min later	stopped
D	Stopped 0.5 s later, then Loaded only at the manual Start ~80 min later	stopped

On the bad logons Current.xml was not written either, so no Load/Run was sent at all.

Cause

The UI answers a daemon's first Stopped after Connected by sending it the layout, but only if (MonitorsLayout is not null && …). At a cold logon the daemon is launched as soon as the services are resolved (MainBootloader construction → MainService → LittleBigMouseClientService.Listen), while the bootloaders are still running on the same dispatcher as the daemon's events. A daemon that connects and reports Stopped before MainBootloader.LoadAsync reaches UpdateLayout() gets no answer, and nothing asks again.

The logs fit this: the logon that worked is the one where Stopped came late. I could not prove it outright, because the 5.6.0 build on that machine logs handler failures with Debug.WriteLine, so nothing reaches ui.log in Release.

Fix

• The post-connect Stopped is kept pending while there is no layout yet (StartIfJustConnectedAsync).
• The first UpdateLayout() answers it, on the dispatcher, when the daemon still reports Stopped and the layout is enabled. Later rebuilds keep going through the display-change path (ReconcileFreshLayoutAsync), so nothing gets started twice.
• MainService's post to the UI thread can be swapped (internal PostToUi) so tests can run the handlers inline: Avalonia's UI dispatcher belongs to whichever thread touched it first, which made a Dispatcher.UIThread.RunJobs()-based test order-dependent.

Testing

• 3 new tests in MainServiceLifecycleTests:
  • a daemon that connects before the first layout is started once the layout is built (fails without the fix, passes with it), and a later rebuild does not start it again;
  • a daemon that connects after the first layout is started exactly once;
  • a layout the user turned off is not started.
• The whole LittleBigMouse.Ui.Avalonia.Tests suite passes. One unrelated test, LocationControlViewModelTests.AfterDisposalTheLayoutNoLongerReachesTheViewModel, fails intermittently on master too (2 runs in 10 without this change).
• I have not yet logged in again with a build carrying this fix.

🤖 Generated with Claude Code