I suggest we change the return type from void to std::error_code, the start can return an error if something does wrong :) Then the callee can check it and report a proper error :)

I need some help on this one...

If something fails return an error (the ec value), if the function succeeds return no error: return {};

In the start function the boost functions will return the error code, we can just return that if it is an error.

// some boost function with an ec param
if(ec)
{
  return ec;
}

The caller can check if the start() succeeds:

if(auto ec = sim.start())
{
  // error
}

Traintastic server uses this in the Train::acquire function. This is really powerful as you can also return the reason why it failed.

Ok but the logic is run in the worker thread so returned error would be async.
What should I do?

worker thread should exit on error, the main thread then needs to check if the simulator is started or has an error. Those variables are accessed by two threads, so best is to use std::atomic<>:

std::atomic<bool> m_started = false;
std::atomic<std::error_code> m_error;

When start() set m_started to true else set m_error.

The main thread then can check (m_started || m_error), after the startup one of them must be true.

I'd like to have discovery optional, we can add a bool discoverable = false parameter to the start function for that. In most cases discovery isn't needed I'd expect.

Done. Now we should add console argument to turn it on/off. For now it's still forced on because it needs testing.
Also another argument for localhost only server.

yes, command line options to control those would be great.

Should we use boost to parse command line?

Oh I see there is already a parseCommandLine() function in main.cpp
I'll use that!