Prevent default fromrepo being evaluated as string 'None'

[myii]

Discussion

@noelmcloughlin Thanks for #209; this follows-on from that.

I've been using this formula for some time with use_upstream_repo: True. Didn't have a value for fromrepo in the pillar when running this formula for a new minion. Encountered an error:

          ID: postgresql-server
    Function: pkg.installed
      Result: False
     Comment: Problem encountered installing package(s). Additional info follows:
              
              errors:
                  - Running scope as unit: run-r446ecf8982944cd4a4caa7bf54ad2e02.scope
                    E: The value 'None' is invalid for APT::Default-Release as such a release is not available in the sources

Initially tracked back to:

postgres-formula/postgres/codenamemap.yaml

Line 16 in 3d8d848

{% set fromrepo = repo.fromrepo or name + '-pgdg' %}

• This was being evaluated as 'None'

Then tracked back to:

postgres-formula/postgres/defaults.yaml

Line 60 in 3d8d848

fromrepo:

• Used debug output of defaults.postgres to establish that the bare fromrepo: above results in {..., 'fromrepo': 'None', ...}

The proposed change in this PR allows fromrepo to be evaluated as intended.

---

Further potential issues

When collecting the debug output of defaults.postgres, the 'None' strings present correlate with all of the bare defaults:

{
    ...

        'homebrew': {
            'url', 'None',
            'sum': 'None'
        },

    ...

    'fromrepo': 'None',

    ...

    'linux': {
        'altpriority': 'None'
    }
}

Appears that it would be appropriate to set default values for these as well, unless there is a specific reason for using 'None' as a string.

---

Consideration: Add fromrepo to pillar.example

It's probably useful to have fromrepo mentioned in pillar.example. Finding an appropriate example value may be difficult. Was thinking about adding fromrepo: 'bionic-pgdg' to this PR but that is Ubuntu-specific.

[noelmcloughlin]

@myii good analysis. this is approved. Would be great if you can update all None to '' & update pillar.example.

[vutny]

Good finding @myii

It seems that or comparison for non-Boolean values in Jinja are not reliable.
Although you actually fixed the issue I suggest to harden it even more and change that line:

postgres-formula/postgres/codenamemap.yaml

Line 16 in d22b66d

{% set fromrepo = repo.fromrepo or name + '-pgdg' %}

Lets use default filter like this:

 {% set fromrepo = repo.fromrepo|default(name ~ '-pgdg', true) %}

[myii]

Would be great if you can update all None to '' & update pillar.example.

@noelmcloughlin Will do that shortly. My only concern is that altpriority appears to be an integer, so the default really should be set to an integer rather than an empty string. What would be the appropriate value for that?

Lets use default filter like this:

{% set fromrepo = repo.fromrepo|default(name ~ '-pgdg', true) %}

@vutny OK, will apply that to both lines 16 and 41, which are identical.

[myii]

@noelmcloughlin @vutny Commits pushed. I'll also push pillar.example soon. In the meantime, if there is a better suggestion for altpriority, I'll modify that and rebase.

[noelmcloughlin]

@myii. Just set altpriority: '', thanks.

update: just saw your comment. I usually set this value (in other formulas) to 0 (zero).

[myii]

@noelmcloughlin OK, added pillar.example and rebased to set altpriority: 0. All done from my end.

[myii]

@noelmcloughlin Just a thought: should alternatives system be disabled by default? This won't cause regressions for users?

[noelmcloughlin]

Yes, it should be disabled by default - definitely. 8 out of 10 people hate linux alternatives ;-)

[noelmcloughlin]

Thank you, this is great work!!

[myii]

No problem, glad to have been of assistance.