apt: Fix all strict variable cases.

[daenney]

A few of these fixes are absolutely horrendous but we have no choice as we need to stay current- and future-parser compatible for now.

Once we can go Puppet 4 only we can use the $facts hash lookup instead which will return undef/nil for things that aren't set instead of them not being defined at all.

[daenney]

@hlindberg I've had to do a few horrible things here for strict variables, mainly not being able to bind fact values on parameters because it might not be set.

This wouldn't be a problem if we didn't explicitly have test cases that check if for example $release is nil when we don't set lsbdistcodename in our tests but by not doing so we can't access the variable and everything blows up.

Do we have any way to make this better except for using the $facts hash which we can't just yet?

[hlindberg]

Since there is no way to enforce that the $facts hash is available on 3x from within a module (and users would probably not be happy about it since it may run into problems if users are using $facts as a regular variable).

defined("$name") is really the only way from within Puppet Code to check if a variable is defined or not.

I can however imagine writing a compatibility function that is invoked early in the process that adds global variables with the value 'undef' - the Puppet Code can then continue to look at the variables since they will then not be missing. Then at least, you do not have to have the defined?("$name") all over the place.

[hlindberg]

Hm, then there is $caller_module_name

[hlindberg]

yet another alternative would be to write a function that does the same as $facts. That would require squirreling away the data that is put into $facts to allow this function to find it later.

[hlindberg]

I logged this https://tickets.puppetlabs.com/browse/PUP-4066 for the built in variables case

[daenney]

Thanks @hlindberg!

[daenney]

Right, so using defined()? bound in a parameter causes puppet-lint to freak out.

[underscorgan]

Shouldn't this be $xfacts['lsbdistcodename']?

[daenney]

Damn it.

[underscorgan]

thanks @daenney !

[hlindberg]

good work - looks less horrible now

[daenney]

Yes, and with #450 merged we'll have a full set of 'guarded' facts. I wanted to write it as a function but I didn't really feel like figuring out how to get to those variables from Puppet. If someone would like to add that to stdlib, much obliged.

[hlindberg]

see the file runtime3_support.rb method get_variable_value(name, o, scope). And remember when in a 3x function, self is an instance of Scope. The --strict_variables will raise a ruby exception (the symbo :undefined_variable) - you understand when you look at the code.

Here is the method in question

  # Returns the value of the variable (nil is returned if variable has no value, or if variable does not exist)
  #
  def get_variable_value(name, o, scope)
    # Puppet 3x stores all variables as strings (then converts them back to numeric with a regexp... to see if it is a match variable)
    # Not ideal, scope should support numeric lookup directly instead.
    # TODO: consider fixing scope
    catch(:undefined_variable) {
      x = scope.lookupvar(name.to_s)
      # Must convert :undef back to nil - this can happen when an undefined variable is used in a
      # parameter's default value expression - there nil must be :undef to work with the rest of 3x.
      # Now that the value comes back to 4x it is changed to nil.
      return (x == :undef) ? nil : x
    }
    # It is always ok to reference numeric variables even if they are not assigned. They are always undef
    # if not set by a match expression.
    #
    unless name =~ Puppet::Pops::Patterns::NUMERIC_VAR_NAME
      fail(Puppet::Pops::Issues::UNKNOWN_VARIABLE, o, {:name => name})
    end
  end