init: change refclocks data type to Hash

[bmagistro]

This updates the refclock parameter data types to be a hash which matches the examples and updates the template to properly generate multiple refclock entries.

This looks to have broken via a combination of #141 and #79. Rather than restore the full variations that were supported prior to #79 a simplified hash structure was chosen.

Fixes #189

[bmagistro]

need guidance on how to regenerate ereference.md

[benjunmun]

need guidance on how to regenerate ereference.md

So I'm not a maintainer, but I believe the right thing to use is puppet strings. Specifically puppet strings generate --format markdown

[smortex]

Update REFERENCE.md with

$ bundle exec rake strings:generate:reference

Add it to the PR and CI should continue.

Can you also update the test suite (in spec/classes/chrony_spec.rb) to make sure this work as expected and we don't break it again in the future?

[kenyon]

I wonder if this is the change we should make. The refclocks parameter has always been an array since it was added in #17. We could just update the documentation to avoid the breaking change in data type.

Here is how I am configuring refclocks, for example:

• https://github.com/kenyon/puppet/blob/445cd668cbc117eac264561fa50159eeca96b682/data/virtualization/kvm.yaml#L13-L14
• https://github.com/kenyon/puppet/blob/445cd668cbc117eac264561fa50159eeca96b682/data/nodes/router.kenyonralph.com.yaml#L57-L59

[smortex]

Following on @kenyon notes, and as someone who never used this module, the parameter looks a bit scary. My guess is that a proper usage of Puppet's Struct/Tuple parameter may make it easier to understand what is expected?

Switching to Struct is necessarily a breaking change, but with a Variant of String and Tuple, I think we can have a backwards-compatible way of making the data validated (untested):

Array[
  Variant[
    String[1], # for backwards compatibility
    Tuple[
      Enum['PHC', 'PPS', 'SOCK'],
      Stdlib::Absolutepath,
      Pattern['\A[^ ]+ [^ ]+\z'],
      2,
      default,
    ],
    Tuple[
      Enum['SHM'],
      Integer[0],
      Pattern['\A[^ ]+ [^ ]+\z'],
      2,
      default,
    ],
  ],
]

No strong opinion regarding the above, but it feels cheaper to fix the doc so that it match the code than adjusting the code so that it match the doc.

[bmagistro]

I can try reverting my changes and see if it generates correctly. I vaguely remember trying an array at one point and not being able to get the multiple entries I needed.