The fix in #635 is fairly subtle and it could easily happen for new code that would pick the wrong way to compare network paths. But we currently don't have any tests that catch this, only specific machine setups in iroh's CI caught this.
I think it might be possible to test, but we'll need some test infrastructure improvements for it: we have to be able for the PairRoutingTable to allow a packet through but not provide a source_ip. That way we could construct a test that has a transport without source_ip. But even once we have that we kind of need to be able to run all (or most) existing tests in that mode... which I don't even have any idea for.
And we need those fixes for the release so I'm creating this issue for the testing so to not further block the release.
The fix in #635 is fairly subtle and it could easily happen for new code that would pick the wrong way to compare network paths. But we currently don't have any tests that catch this, only specific machine setups in iroh's CI caught this.
I think it might be possible to test, but we'll need some test infrastructure improvements for it: we have to be able for the PairRoutingTable to allow a packet through but not provide a source_ip. That way we could construct a test that has a transport without source_ip. But even once we have that we kind of need to be able to run all (or most) existing tests in that mode... which I don't even have any idea for.
And we need those fixes for the release so I'm creating this issue for the testing so to not further block the release.