Extract rule: template-require-context-role

[NullVoxPopuli]

Split from #2371.

[NullVoxPopuli-ai-agent]

Review: template-require-context-role (PR #2601)

Comparison with ember-template-lint require-context-role.js

General correctness:

1. Role mapping: The original builds its role mapping from a ROOTS object (parent → children) and inverts it to get VALID_PARENTS_FOR_CHILD. The PR defines ROLES_REQUIRING_CONTEXT directly (child → valid parents). Comparing the mappings:

   • Original's inverted mapping produces:
     • cell → [row] — Missing from the PR. The original's ROOTS.row includes 'cell' but the PR's ROLES_REQUIRING_CONTEXT does not have a cell entry.
     • listitem → [group, list] — PR has [list, group, directory]. The PR adds directory which is NOT in the original. directory is a deprecated ARIA role, so this is a debatable addition.
     • option → [listbox] — PR has [listbox, group]. The PR adds group which is NOT in the original.
     • rowgroup → [grid, table, treegrid] — PR matches (has all three).

   So there are discrepancies in both directions. cell role is completely missing from the PR's map — this means <div role="cell"> inside a <div> (without role="row") will NOT be flagged. The original would flag it.

2. directory role: The PR includes directory as a valid parent for listitem. This role is deprecated in ARIA 1.2. Not wrong per se but deviates from the original.

3. option with group parent: The PR allows role="option" inside role="group", but the original does not. According to WAI-ARIA spec, option requires listbox context. group is allowed for grouping options but only within a listbox. This is an over-relaxation.

4. Parent traversal approach: The original walks up using path.parent (ember-template-lint's AST path) and stops at the first non-presentation parent element. The PR uses an elementStack maintained via enter/exit visitors. This is the standard ESLint approach and works correctly.

5. aria-hidden handling: The original checks aria-hidden on the current node AND ancestor nodes. The PR checks aria-hidden only on ancestors (via isInsideAriaHidden). However, the original also returns early if the current element has aria-hidden, essentially skipping all children. The PR's approach of checking ancestors seems functionally similar since child elements would find the aria-hidden ancestor.

   Actually, there's a difference: the original checks aria-hidden on the element with the context-requiring role itself and returns early. The PR only checks ancestors via isInsideAriaHidden (which excludes the current element at elementStack.length - 1). So if you have <div aria-hidden="true" role="listitem">, the original would skip it, but the PR would still flag it (since it only checks ancestors). This is a behavioral difference — consider also checking the current node for aria-hidden.

6. Root-level elements: The PR skips reporting when elementStack.length <= 1 (no parent elements) or when getAccessibleParentRole returns undefined. The original's while loop naturally handles this. The PR's approach is correct — you can't validate context when there's no parent.

7. Template tag transparency: The PR treats <template> tag as transparent in getAccessibleParentRole. This is correct for the gjs/gts case where <template> is a wrapper, not a semantic HTML element.

8. Error message format: Different from original but conveys the same information. The original includes a W3C reference URL; the PR does not. Consider adding the reference URL.

Scope analysis (gjs/gts):

This rule only checks role attribute values on GlimmerElementNode. Pure structural/attribute checking. No scope analysis needed.

Overall: The migration has some mapping discrepancies:

• Missing: cell → [row] role context (present in original, absent in PR)
• Added: directory as valid parent for listitem (not in original)
• Added: group as valid parent for option (not in original)
• Behavioral: aria-hidden on the current element itself is not checked

These should be reviewed to ensure they are intentional deviations or corrected to match the original.

🤖 Automated review comparing with ember-template-lint source