* Sam Hartman <hartm...@debian.org> [240614 22:52]: > >>>>> "Chris" == Chris Hofstaedtler <z...@debian.org> writes: > > Chris> When building krb5 with sbuild, configured to use the unshare > Chris> backend, the t_iprop.py test apparently times out without any > Chris> output. > > I'm guessing, but have not confirmed that sbuild unshare is setting up a > network namespace with nothing but lo in it.
Yeah. For details, you can check: https://sources.debian.org/src/sbuild/0.85.10/lib/Sbuild/ChrootUnshare.pm/#L320 > I've certainly seen the same behavior you see in such a configuration. > What's happening is that get_wildcard_address in kpropd.c passes > AI_ADDRCONFIG into getaddrinfo. Since loopback addresses are ignored by > AI_ADDRCONFIG, there is no valid v4 or v6 address and the routine fails > to find a wildcard address to bind to. > > I'm mistified why the daemon doesn't just fail at thait point, but I > have not looked into that. Having it stuck there is certainly annoying. > I did modify get_wildcard_address to prefer AI_ADDRCONFIG but to fall > back to calling getaddrinfo without AI_ADDRCONFIG if that's all that > will work. > > That's enough for t_kpropd.py to succeed. > I'm working through a few other FTBFS issues in a container, some of > which apparently didn't show up for sbuild unshare (they happen earlier > in the build) but which are getting in my way. > I hope to have 1.21.2 uploaded and building by end of weekend. Great. Thanks for your work! Chris
