Extract rule: template-sort-invocations

[NullVoxPopuli]

Split from #2371.

[NullVoxPopuli-ai-agent]

Review: template-sort-invocations

Compared against ember-template-lint sort-invocations.js.

General Correctness

1. reported flag limits to one error per file: The rule uses a let reported = false flag that prevents reporting more than one sorting violation per file. The original ETL rule does NOT have this behavior -- it reports all violations found across all nodes. This means if a file has multiple unsorted elements, the ESLint rule will only report the first one. This is a behavioral difference from the original. Was this intentional? If so, it should be documented. If not, the reported guard should be removed.

2. No autofix: The original ETL rule has full fix mode support (sorting attributes, modifiers, hash pairs, and reordering splattributes). The ESLint port sets fixable: null. This is a significant feature loss. The original marks all violations as isFixable: true. Consider implementing autofix or at least noting this limitation in the docs.

3. localeCompare vs simple string comparison: The ESLint port uses nameA.localeCompare(nameB) while the original uses simple > / < string comparison. This could produce different sort orders for non-ASCII attribute names. The original's behavior is more predictable (pure lexicographic). This is a minor behavioral difference but could cause inconsistencies between the two tools.

4. GlimmerStringLiteral field name: In compareModifiers, the port accesses eventA.value for GlimmerStringLiteral nodes, while the original accesses eventA.original for StringLiteral nodes. Verify that value is the correct property on GlimmerStringLiteral nodes in the ember-eslint-parser -- it might be original or value depending on the parser.

5. canSkipSplattributesLast logic difference: The original checks attributes.at(-1) and modifiers.at(-1) without first checking array lengths. The ESLint port adds extra guards for empty arrays, which is good defensive coding. However, the original also doesn't have the attributes.length === 1 special-case message logic that the port adds. The port reports attributeOrder when ...attributes is the only attribute, vs. splattributesOrder otherwise. This is a nice UX improvement.

6. Comprehensive test coverage: The tests cover the main scenarios well. Good to see both gjs and hbs tests.

Scope Analysis (gjs/gts)

This rule only checks structural properties: attribute names, attribute ordering, modifier ordering, hash pair ordering. It does not match any helper/component names by string. No scope analysis needed.

---

🤖 Automated review comparing with ember-template-lint source