Here's some cleanups to the netlink monitor that are enough to fix #1518 (closed).
A theory that I have, that I haven't tested, is that the relevant part of e0b120cc that fixes this bug is passing a GError into the initable_init. This means that the bug I'm fixing in 986bb037 was fixed by accident in the 2.58 cycle.
The first two commits (986bb037 and b206ce7e) are bug fixes that are needed here to make the test pass reliably. Not sure why we don't see b206ce7e more often.
I could split 2b0336b4 and 3524817d into another MR if you want. They are for debuggability - it was good when fixing this to be able to see what GIO was doing.