relationships in validations as string instead of list has questionable ergonomics

[hermannloose]

We really like the ability to "unit-test" more complex permissioning schemes with validation files, but the fact that relationships is a string instead of a list makes it unnecessarily cumbersome to document the interesting bits for a particular scenario, as the whole set of relationships tends to contain a lot of boilerplate as well, to create users, roles, policies etc. If relationships was a list, it would be easy to break it up into logical groups with comments, as well as explain the "why" for any non-obvious parts, e.g. "we are sharing this with a user outside the org, which should never happen through app logic, but in case a relation like this ever appears, the & operation in foo->view protects against it" etc.

Making it a list could also open up avenues for reusing relationships from other files, similar to how schemaFile: schema.zed avoids having to repeat the schema everywhere. Think for example:

relationships:
  - all_from: organizations.yaml
  - from: users.yaml
    key: basic # e.g. to include only relationships under a particular dictionary key in the other file
  - "foo_resource:123#organization@organization:1"
  ...

The complexity of this shouldn't snowball of course, but some support for composing lists would really help with keeping test relationships realistic yet their size in a particular validation file manageable.

[josephschorr]

See #1437 for the current proposal on how to address this issue; changing the YAML format would be non-trivial due to its wide use in various tools

[vroldanbet]

@hermannloose please share your feedback on the proposal in #1437