Linter: Future Enhancements and follow up tasks

[Karan-Palan]

This issue tracks leftover tasks or future enhancements for the JSON Schema linter and canonicalizer. These were identified during GSoC work and should be considered for future development.

1. Documentation

• Move existing rule documentations to the JSON Schema CLI repo for now, so they are more visible and semi-official until we upstream them properly.

2. Reference Handling

• Break schemas when necessary and use rereference() in rules that transform applicators, to ensure $ref pointers stay correct.

3. Non-auto-fixable Rules

Implement rules that cannot be safely auto-fixed, but should instead produce a clear user-friendly lint message.

• conflicting_readonly_writeonly.json
• const_and_enum_conflict.json
• examples_must_validate.json
• forbid_empty_enum.json
• format_type_mismatch.json
• incoherent_exclusive_limits.json
• incoherent_inclusive_limits.json
• mincontains_greater_than_maxcontains.json
• mincontains_zero_without_maxcontains.json
• no_actionable_text_in_title_description.json
• oneof_min_branches.json
• require_absolute_draft_uri.json
• require_description_title.json
• require_schema_declaration.json
• top_level_id_must_be_absolute.json
• unbounded_collection_limits.json
• unique_branches_in_oneof.json
• unresolved_ref_pointer.json
• $schema sibling to $ref in JSON Schema Draft 7 and older (we can't just remove it otherwise the linter evaluation breaks given it doesn't know the dialect?)
• Only use $anchor in a saved definition (not randomly)
• $ref must not point to a random schema location, but to $defs, or an anchor, or a schema resource

5. Autofixable rules

• Remove unused non-referenced definitions
• Flag references to non-existent locations (remove the $ref)
• Keywords in allOf branches that already exist outside are duplicating work and can be removed
• properties in an allOf branch while the outer schema has additionalProperties: false and no properties/patternProperties can be removed. This is one case where we know keywords would never match. Like dead code?
• Delete any nested $schema that doe not have $id/id (maybe just check if the parent location is a resource?)

6. Additional Rule Enhancements

• Extend unnecessary_allof_wrapper_properties to work on older dialects, not just modern ones.
• Improve diagnostics in duplicate_anyof_branches by highlighting which specific branches are duplicated.
• Improve diagnostics in duplicate_enum_values by highlighting which specific enum entries are duplicated.
• Do not warn about contentMediaType or contentSchema without contentEncoding if contentEncoding is available through a $ref
• Prevent pattern alongside enum or const
• Lint that pattern and patternProperties only have ECMA-262 compliant regexes

6. TODOs Inside Existing Rules

Specifically in UnnecessaryAllOfWrapperModern:

• Add logic to check for type intersections in type arrays, not just direct equality.
• Create a separate rule that removes a keyword if it is exactly equal to an instance of the same keyword already present outside the wrapper.