(MODULES-1738) Don't modify the global seed in fqdn_rotate()

[elyscape]

MODULES-1738 is similar to Redmine-15791, which was addressed in puppetlabs/puppet@292233c. The current implementation of fqdn_rotate() leaves the global seed in a deterministic state, which is bad. Puppet::Util.deterministic_rand() exists to avoid running into this issue, but is only present starting in Puppet 3.2.0.

An alternative (and, IMO, superior) way to handle this would be to instead replace lines 34-37 of fqdn_rotate.rb with:

Puppet::Parser::Functions.function('fqdn_rand')
function_fqdn_rand([elements, arguments]).to_i.times {
   result.push result.shift
}

This would have the benefits of being much cleaner and not having duplicated code from the Puppet codebase that would have to be maintained if for some reason it changed there. On the other hand, it would not prevent fqdn_rotate() from poisoning the global seed in versions of Puppet prior to 3.2.0. That being said, since fqdn_rand() already poisons the global seed in versions of Puppet prior to 3.2.0, maybe it doesn't matter if fqdn_rotate() does the same in those versions.

[elyscape]

In the event that we change fqdn_rotate() so that it uses fqdn_rand() to generate numbers in order to avoid duplicating Puppet::Util.deterministic_rand, we'll need to modify this test to only execute on Puppet >=3.2.0. One way to do that would be to extract the if Puppet::Util.respond_to?(:deterministic_rand) from the preceding test and have it wrap both the preceding test and this one.

[elyscape]

Rebased onto master, so tests should pass soon.

[cmurphy]

I think this implementation is fine, and it makes it easy to remove the reimplemented code when it's appropriate. Thanks!