Attempt to deal with connection race conditions about p2p-project

[LivInTheLookingGlass]

First attempt: Before sending a recovery request, wait 0-0.25s

This may give enough delay to work correctly.

[LivInTheLookingGlass]

This seems to work (on localhost) ~2/3 times on python3.5.

Not good enough.

Next approach:

If connection was outgoing, wait 0.25s to send recovery request.

[LivInTheLookingGlass]

Again, seems to work 2/3 times.

Next approach:

Only send recovery request if your connection was outgoing. You accept incoming connections indiscriminately, so it's up the one reaching out to repair things.

[LivInTheLookingGlass]

This fails on SSL and Plaintext. Not a good solution. Strangely seems to work under plain disconnect, though, which seems to indicate that it matters whether it's in a routing table.

Pastebin of output in case this becomes relevant again

[LivInTheLookingGlass]

Rerunning the test, this time flagging both outputs, crash test succeeds, disconnect test fails. This is odd.

[LivInTheLookingGlass]

Moving this flag check makes one or the other fail, every time, it seems. So the question becomes, what's the root cause of SSL failing but not Plaintext?

[LivInTheLookingGlass]

To be clear about this, the method that I'm changing is mesh_daemon.mainloop

[LivInTheLookingGlass]

Possibly fixed in bbaaed5, though for the life of me I don't know why.

[LivInTheLookingGlass]

Yeah, it didn't fix it on any system but my own, and I don't understand why it worked on mine.