I've removed the dmidecode patch because it no longer cleanly applies and doesn't seem to be getting tracked upstream anymore. Needs more investigation and then potentially porting the patch to the 2.x series.
When run as a non-root user, facter would always spew "/dev/mem:
Permission denied" because it called dmidecode without redirecting
stderr.
This was filed upstream as FACT-86¹. It's unclear what is insufficient
about this method, but failing tests that are not visible aren't as much
of a concern to me as clearly useless output on every run of the
command.
¹ https://tickets.puppetlabs.com/browse/FACT-86
The previous patch caused spurious warnings from facter when the
virt-what command produced no output, as it would if run by root on a
physical host:
$ sudo facter virtual
Could not retrieve virtual: private method `gsub' called for nil:NilClass
physical
Josh Cooper from Puppet Labs caught and fixed this.
With the additional requirement of virt-what that was added in
facter-1.6.18-1, the virtual fact was incorrect when facter was not run
by root:
$ facter | egrep '^(facterversion|virtual)'
facterversion => 1.6.18
virtual => virt-what: this script must be run as root
This is upstream ticket #19989¹. It was applied to the 1.7.x branch².
¹ https://projects.puppetlabs.com/issues/19989
² https://github.com/puppetlabs/facter/commit/b7784a3
For whatever reason, the install.rb sript doesn't install the prebuilt
man page that is in the tarball when rdoc is not present. It simply
skips the man page.
Prior to 1.6.14, facter did not reliably return when a host was running
under KVM (and RHEV). The upstream fix depends on the virt-what
utility. This is a small shell script with no dependencies other than
dmidecode, which we already require.
This was also discussed in upstream issue #12879¹. No clear solution
was found, so a condrestart of puppet seems like the best band-aid for
the time being.
¹ http://projects.puppetlabs.com/issues/12879
This is a rebase to the upstream of 1.6.15. This should be a backward
compatible release with what is currently in Fedora/EPEL. The known
issue of openstack/ec2 fact exception handling has been resolved in
1.6.15.
This commit also adds the .asc file back as it used by facter
maintainers during package creation.
Signed-off-by: Michael Stahnke <stahnma@puppetlabs.com>
This is a patch to rebase facter to 1.6.14 for rawhide (and should be a
clean change in F17). Note I had to add a requires on rpsec (not just
rspec-core) to make this build cleanly in the %%check phase.
This patch also removes the asc file, as it is never (and has never
been) used in the process of making the facter package.
rpmlint looks reasonable to me upon build.
Bugzilla: 871211
Signed-off-by: Michael Stahnke <stahnma@puppetlabs.com>
Older releases fail with:
undefined method `unstub' for #<Mocha::ClassMethods::AnyInstance:0x7fe0dcca8388>
This may be due to an older version of mocha, but isn't something I'm
going to track down today.
The move to ruby-1.9 in Feodra 17 means we have to take different
installation locations into account. The ruby(abi) requirement also
needs to be changed.
Also drop %%has_ruby_{abi,noarch} macros. All supported releases have
these.
The virtual fact makes use of lspci from pciutils in addition to
dmidecode. Various facts use commands from the net-tools package (arp,
dnsdomainname, hostname, and ifconfig).
Several facts imlicitly relied upon dmidecode without checking that it
was available. Ensure that these facts work as intended.
Refer to http://projects.puppetlabs.com/issues/11041 for details.
Michael Stahnke
Peter Robinson
Dennis Gilmore
Lubomir Rintel
Sam Kottler
Todd Zullinger
Sam Kottler
Vít Ondruch
Jeroen van Meeuwen (Ergo Project)