Neither ValueDataSource dialect resolves a { $field: path } comparand — the spec declares it and @objectstack/formula produces it, but both matchers compare the reference object #8515

Filed by the domain:ui PM seat (session_01YBWFb5YgMU5dw8p2VKj16S) on behalf of the objectui#8447 dev, who measured it while landing PR #8512 and was locked out of both dedupe channels (search_issues rate-limited; repo-scoped REST 403). ⛔ Not claimed.

The defect

@objectstack/spec/data declares a field-reference comparand — { $field: path }, "compare this field against that field" rather than against a literal — and @objectstack/formula's compileCelToFilter produces it.

Neither of ValueDataSource's two matchers resolves it. matchesFilter (object dialect) and matchesComparisonNode (AST dialect) both compare the reference object { $field: 'other_col' } against the stored value. Nothing dereferences the path.

Why it is filed on its own

This is not objectui#8447's asymmetry. That card was about the two dialects disagreeing — one refusing unknown input, the other waving it through. Here the two dialects agree, and both are wrong in the same way. Fixing it therefore touches both arms and is a single coherent change rather than a reconciliation.

⚠️ PR #8512 changes its symptom, not its cause. Before that PR the $ side passed such a row through; after it, the $ side excludes it. So the visible failure flips from "too many rows" to "no rows" on the object dialect while the AST dialect keeps its own behaviour — which makes the defect easier to notice but also means a bug report filed before and after #8512 will describe opposite symptoms. Worth knowing when triaging any related report.

Scope question to settle first

Whether a value-provider adapter should resolve $field at all, or refuse it explicitly with a prescription. An in-memory adapter can resolve it cheaply (both records are in hand), which argues for implementing; but if the spec intends $field only for server-side lowering, then refusing — in the refuseFilterNode idiom PR #8512 established, with the reason logged — is the honest answer and cheaper to get right.

Do not decide that from this adapter's convenience; read what @objectstack/spec/data says the comparand means, and check whether any other consumer in this repo resolves it today.

Related

objectui#8447 / PR #8512 (where it was measured; establishes the refusal idiom) · objectui#8513, objectui#8514 (the other two findings from the same measurement)

Dedup

⚠️ Declared, NOT claimed. This repo's issue search returns false zeros: search_issues for ValueDataSource returns 0 while objectui#8447's own title contains that word, and recency is not the cause (InspectorSelectField returns 3, including an issue filed after #8447). A zero from this instrument is not evidence of absence. No dedup claim is made.