Linux is not liking something about the virtual window_init and window_settings. Was all good on my local...
Looking into it.

Calling virtual methods from a class constructor looks like grey and complex territory.

I have reverted back to non-virtual methods - we should just keep in mind not to diverge the constructor functions across backends.

@SteffenL could you nudge the two timed-out Windows tests

Done.

As for calling virtual methods from a constructor:

As long as it's called from the derived class then I don't believe it's that bad. If someone were to derive from that class again and also override the same methods then that would be more concerning.

Since we know hopefully know what we're doing then in my opinion that's just fine, but I'm aware Clang will emit a warning as I discovered myself the other day... We should eliminate warnings, but sometimes it can make sense to suppress them if possible.

I wonder if the warning will disappear if we make the derived class final so that nobody can derive from it. We currently derive from it in tests, and for now need it to not be final, but I don't think it makes much sense to let others derive from it.

I am happy to let go of the derived constructor methods in this PR as it's primary purpose is to tidy up your refactoring requirements, which it does.