On Wed, 2015-01-28 at 16:09 -0500, Alan Conway wrote: > On Tue, 2015-01-27 at 09:46 -0500, Rafael Schloming wrote: > > Do you have tracing turned on when they time out? With the protocol trace > > enabled the tests slow down enough that some of them don't finish before > > the various timeouts kick in. > > Much more insidious. It turns out that if you use 0.0.0.0 as a connect > address AND you have a VPN running AND you are using jython then things > hang (at least on fedora 20 in my house in January.) > > Looking up the ip(7) man page it seems like 0.0.0.0 is only supposed to > be used as a bind address, not a connect address, so I switched the > tests in question to use 127.0.0.1 and everything works (it also works > if I turn off my VPN or use proper python) > > I didn't switch everything in python to use 127.0.0.1 for connect (NOTE > 0.0.0.0 is still correct for listening) in case there was some hidden > disaster there. Probably at least the examples and maybe the default > port for class Url should be changed, any dissent?
I'm curious to know where 0.0.0.0 as a default connect address came from because it should never have worked at all, unless it is handled specially somewhere. Recently I found a problem with other tests that sometimes (depending on machine configuration) would listen on IPv4 and connect on IPv6 (or maybe it was the reverse) - remember that Rafi? That would be a similar sort of problem. Forcing IPv4 (127.0.0.1) may or may not be what you want to do depending on whether you want to allow IPv6 as well (::1). Andrew
