fix(ash_ui validations): trust custom:* opaque widget types in :list binding allow-list

[ty13r]

Summary

Allows :list binding_type on "custom:*" opaque widget types in validate_element_binding_locality! at lib/ash_ui/resources/validations/authoring.ex:373.

(Scope narrowed in r2 per Codex P2 #2 + #3 review.) Original draft also added "artifact_row" to @list_widgets; Codex correctly pointed out that :list_repeat is the canonical destination wrapper for collection hydration (per the Phase31RepeatListElement test fixture), NOT :artifact_row. Row-template primitives like :artifact_row continue to belong INSIDE a :list_repeat parent that owns the list binding.

Why

The validator currently raises ArgumentError "does not expose collection semantics" for any :list-typed binding on Elements whose canonical widget_type isn't in @list_widgets ("list", "table", "info_list", "select", "list_repeat").

"custom:*" opaque widget types are application-extension primitives outside the canonical catalog (per Storage.canonical_widget_type at lib/ash_ui/dsl/storage.ex:162). The validator can't know their collection semantics and shouldn't gate the :list binding allow-list on them; trust the authoring intent.

Per Codex P2 #3 the bypass is tightened to ONLY match "custom:*" binary strings; nil widget_type (ui_element block missing entirely) still raises through downstream missing-element error paths.

Triggering precedent

Wave AshUI-3.9 Item 2 dispatch (2026-05-17/18). Ariston-ui attempted to add ui_bindings do binding_type(:list) end blocks to 11 destination Element modules per docs/iur-missing-list-binding-sizing.md. The sizing investigation got the architectural premise wrong (proposed :artifact_row as destination); the correct pattern is :list_repeat wrapping the row template (per Phase31RepeatListElement). Ariston-side fix becomes a structural restructure rather than a flat ui_bindings add. This PR's custom:* bypass is the only ash_ui-side gap that's truly Pascal-roadmap material.

Per umbrella CLAUDE.md "substrate gaps become our roadmap, not asks-of-Pascal" — the custom:* bypass DOES belong here since the canonical catalog has no opinion on opaque application types.

Files

• lib/ash_ui/resources/validations/authoring.ex — validate_element_binding_locality! gets a custom_widget? bypass for "custom:*" widget types (binary-prefix check only)
• test/ash_ui/resource/locality_test.exs — 1 new positive test for "custom:thread_card"; helper assert_validator_accepts_list_binding/1 for the validator-passed-but-downstream-domain-error pattern

Test plan

• mix test test/ash_ui/resource/locality_test.exs — 5/0 pass (4 original + 1 new)
• Existing negative tests still pass (regression check that allow-list still gates unknown-bad types like :text with :list)
• New positive test: "custom:thread_card" widget compiles with binding_type :list

Cross-references

• ariston-ui Wave AshUI-3.9 Item 2 (sizing doc): docs/iur-missing-list-binding-sizing.md (PR #385); doc being updated in a follow-up to reflect the architectural premise correction
• Issue IUR hydration: expand_list_repeat doesn't expand repeat-marked children for :artifact_row or custom:* widget types #114 — IUR hydration gap; reshaping after this PR lands and the ariston-side :list_repeat restructure verifies whether the gap is moot or narrower (custom:*-only)

Scope notes

• Narrower than r1. artifact_row extension dropped; custom:* bypass tightened
• Validator scope extension only. No change to runtime hydration; no change to widget-component catalog; no change to how Storage.canonical_widget_type resolves types

🤖 Generated with Claude Code

[ty13r]

Codex r2 disposition — false-positive on the _build/ finding

Codex r2 raised one P2: "Remove generated build artifacts from the patch — _build/dev/lib/ash/ebin/ash.app:1-1 and many BEAM/manifests are also modified or deleted."

Verified false-positive. The PR diff is exactly 2 files:

$ gh pr view 113 --repo The-Metagraph/ash_ui --json files --jq '.files[].path'
lib/ash_ui/resources/validations/authoring.ex
test/ash_ui/resource/locality_test.exs

$ git diff --stat origin/main..HEAD
 lib/ash_ui/resources/validations/authoring.ex | 13 +++++-
 test/ash_ui/resource/locality_test.exs        | 65 +++++++++++++++++++++++++++
 2 files changed, 77 insertions(+), 1 deletion(-)

Codex appears to be reading the local working tree's _build/ state (which contains compilation byproducts from running mix test) rather than the actual PR diff against main. Same false-positive surfaced in r1.

Substantive correctness concerns from r1 (P2 #2 + P2 #3) were addressed in r2:

• P2 Fix: Correct repository file paths #2 (:artifact_row shouldn't be in @list_widgets) — accepted; r2 narrows scope to custom:*-only bypass per the canonical Phase31RepeatListElement pattern
• P2 Integrate Ash UI with unified-ui ecosystem #3 (nil widget_type exemption was too permissive) — accepted; r2 tightens to is_binary(widget_type) and String.starts_with?(widget_type, "custom:") only

The architectural pivot is documented in ariston-ui PR #387 (sizing-doc update) and Issue #114 comment.

Merging on the substantive Codex-clean verdict (the _build/ false-positive doesn't block).