On Thu, Oct 01, 2015 at 04:00:09PM -0400, Russell Bryant wrote: > On 10/01/2015 03:54 PM, Ben Pfaff wrote: > > On Thu, Oct 01, 2015 at 03:50:35PM -0400, Russell Bryant wrote: > >> Commit e91b927d8966bfcb9768225392324dde4fd7d7f6 added optional usage of > >> the libcap-ng library. It's packaged in Fedora, so go ahead and added > >> it to the Fedora spec file. > >> > >> Our default systemd unit files don't make use of the --user option that > >> requires this library, but conceivably someone may want to customize > >> them and use this option. > >> > >> Signed-off-by: Russell Bryant <rbry...@redhat.com> > > > > I'd consider also adding --with-libcapng=yes to the configure command > > line. That way, a build without libcap-ng installed will fail. That > > will both root out problems in whatever RPM autobuilders happen to be > > running against OVS and make sure that every RPM build supports > > libcap-ng. > > > > The rpm build would fail anyway if it's not installed at all. I tested > that.
That's because you are adding Requires and BuildRequires. The BuildRequires enforces that the libpcap-ng is installed during the RPM build. > It would not catch possible failures due to distro or package version > differences. For example, if the header file was installed was > installed in a different location and the configure script didn't find > it, the package would still happily build now, but without libcap-ng > support. --with-libcapng=yes would catch that, at least. The only issue I see is to have one more dependency to carry on which for some use cases might be a problem. Since this is an optional feature, worth to look at the %bcond_with and %bcond_without macros. We already use that to enable/disable running the testsuite. We could use the same approach to build with (default) or without libpcap-ng. Does that make sense? fbl > Thanks for the suggestion. I'll send out a v2. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
