A recent change to link_connection_do_initiate() seems to prevent ORBit2 from contacting other hosts using IPv4 completely. (This is just my impression from reading the code, but the author of the change confirms that it indeed looks so. No hard verification, though.)

Here is our mail exchange:

> Your recent change to linc-protocols.c in ORBit2 caused a slight
> problem for me on Win32. On Win32, link_protocol_get_sockinfo_ipv4()
> makes sure the string "127.0.0.1" is always what's used for the local
> host. Not, for instance "localhost" or what gethostname() returns. We
> want to be 100% sure we always use the loopback interface.
> 
> I worked around it by adding an explicit check for "127.0.0.1":

[...]

> This made the problem go away for me.
> 
> But anyway, looking at what this code you added does, doesn't it
> prevent contacting other hosts? Is this really the intention? (If I
> have misunderstood the code snippet, stop reading now, and tell me ;)

Yes, I think you might be right here.  I think the right approach might
be to back out this patch.  Could you file a bug about this issue at
bugzilla.gnome.org in the ORBit2 category and cc: me on the bug report?

> I know that this restriction is good enough for presumably the vast
> majority of use cases (heck, in GNOME all use of ORBit2 even uses Unix
> domain sockets, doesn't it?), but should it now then be set in stone
> like this? If yes, isn't there then lots of code left elsewhere [in ORBit2] that
> can be cleaned away, and why not then just use the loopback address
> everywhere in the IPv4 cases? (Hmm, does it then make any sense to
> have the IPv6 support in ORBit2, why would one want to allow
> connecting other hosts with IPv6 but not IPv4?)

Yes, I think you are right and my change should be backed out.  Thanks
for catching this.