diff --git a/‎.changeset/20099-having-where-doors.md‎
Lines changed: 33 additions & 0 deletions b/‎.changeset/20099-having-where-doors.md‎
Lines changed: 33 additions & 0 deletions
@@ -0,0 +1,33 @@
+---
+"@objectstack/objectql": minor
+---
+
+fix(objectql)!: `engine.aggregate({ having })` takes the rest of the filter doors `where` takes — the comparand-type door, a check that `having` is a filter condition at all, refusals that no longer depend on the rows, and `{ $field }` references resolved against the aggregated row (#20099)
+
+Clause-②: no (narrowing)
+
+<!-- adr-0087: not-required (already-registered filter-between-field-reference-endpoint-refused, filter-icontains-comparand-refused-at-parse, filter-regex-options-retired) this change adds no new transition. The first id's replacement prescribes the two-bound { $field } spelling and states that a reference is legal as the WHOLE comparand of $eq / $ne / $gt / $gte / $lt / $lte; `having` now evaluates exactly that, and refuses the list-member position the same ruling removed. The other two register the $icontains comparand and the retired $regex / $options refusals, which `having` already made on a populated grouped set and now makes whatever the rows. The comparand-type set was ruled at the shared face with no ledger entry; this change runs that face, unmodified, on one more position. The remaining refusals are of inputs `having` never had a meaning for: a value that is not a filter condition (FilterConditionSchema, the declared type of `having`, refuses it), an operator outside the vocabulary, and a reference that names no column. `having` is a request-only key: no metadata type stores it, so there is no stored document for `objectstack migrate meta` to rewrite. The table below is the author-facing remedy for each row, not a mechanical rewrite. -->
+
+**BREAKING**: this narrows what `having` accepts on `engine.aggregate` (and on the REST aggregate query that forwards it there). Every refusal below is `INVALID_FILTER` / 400, raised once per query before any driver is asked for a row, on both the native `driver.aggregate()` path and the in-memory fallback. It ships as `minor` under the launch-window convention for accept-set narrowings.
+
+`having` took one of `where`'s filter doors, the comparand-shape face. The engine evaluates `having` itself, once per aggregated row, and that walker answered the shapes the other doors refuse — usually with no group and no error. Measured on the base through `engine.aggregate` on `driver-memory` and `driver-sqlite-wasm`, both paths, over groups with totals 500, 900 and 20 and a `max_cap` of 50, 5000 and 20:
+
+| you wrote in `having` | what it did before | write instead |
+|:--|:--|:--|
+| `{ total: { $eq: { v: 1 } } }`, `{ total: undefined }`, a `Map`, a function, a `Symbol`, a bigint beyond 2^53, or `{ $field: 5 }` as a comparand | depended on the operator. On the numeric `total`: in the implicit slot (the non-object values) and under `$eq`, `$gt` and `$gte` it kept no group; under `$ne` it kept every group; under `$lt` / `$lte` it kept no group, except the bigint, which kept every group. A `Symbol` under an ordering operator threw a raw `TypeError` with no `code` and no `status` on a populated grouped set. The same comparand in `where` is refused by the comparand-type door, and `having` now gets that door's refusal, with the path rooted at `having` | a string, number, bigint, boolean, `null` or `Date`; `{ total: { $eq: null } }` for "has no value" |
+| `[['total', '>', 100]]`, `['total', '>', 100]` or `['and', …]` | kept no group: the array's index keys were read as column names | `{ total: { $gt: 100 } }`. The array form is input-only sugar declared on `where` alone, and `having` is declared as a filter condition object |
+| `[]`, a string such as `'total > 100'`, a number, a boolean, a `Map` or a `Date` | kept every group, as if there were no `having` | the object form, or no `having` |
+| an unknown or retired operator (`$median`, `$regex`), or an empty or non-string `$icontains` | refused only when a grouped row reached it: an empty grouped set, a condition on a column the row does not carry, or a `$or` whose earlier branch held all answered without an error | the operator the refusal names |
+| `{ total: { $field: 'max_cap' } }` (a reference with no operator) | refused as an unsupported operator, again only when a grouped row carried `total` | `{ total: { $eq: { $field: 'max_cap' } } }`, or `$ne` / `$gt` / `$gte` / `$lt` / `$lte` |
+| a `{ $field }` reference as an `$in` / `$nin` member, a `$contains` / `$startsWith` / `$endsWith` / `$notContains` pattern, or an `$exists` / `$null` operand | compared the reference object itself, so the answer never depended on the column it named: no group under `$in` / `$contains`, every group under `$nin` / `$notContains` / `$exists` | a literal there, or the comparison as one of the six scalar operators |
+| a `{ $field }` reference naming a column the aggregated row does not have, or carrying an `addDays` that is not an integer or a `{ $field }` | compared the reference object itself, so the answer depended on the operator: no group under `$eq`; every group under `$ne`; under an ordering operator, no group against a number column, and against a text or date column an answer that follows each value's string order against the text `[object Object]` | a groupBy projection or an aggregation alias of the same query (the refusal lists them); a whole-day `addDays` |
+
+Not refused, but answering differently:
+
+- **A `{ $field }` reference as the whole comparand of `$eq` / `$ne` / `$gt` / `$gte` / `$lt` / `$lte` is now resolved against the aggregated row.** Before, it was compared as an object: `{ total: { $gt: { $field: 'max_cap' } } }` kept no group, and the same reference under `$ne` kept every group. It now keeps the groups whose `total` exceeds their own `max_cap`. The two-bound spelling the `{ $field }` `$between` refusal prescribes, `{ total: { $gte: { $field: 'max_cap' }, $lte: 1000 } }`, now works on `having`. The reference names another column of the same aggregated row: a groupBy projection or an aggregation alias. The comparison is the one the platform's in-memory filter evaluator makes and the SQL cross-field compiler matches row for row: an ordering against a missing value is false, `$eq` holds when both sides have no value, and `addDays` adds whole days to a date column.
+- **The same resolution applies to a per-aggregation `filter`** (`aggregations[i].filter`), which the engine evaluates with the same walker against the source rows. `{ function: 'count', filter: { amount: { $gt: { $field: 'cap' } } } }` used to count no row. It now counts the rows whose `amount` exceeds their `cap`.
+- **An exact-range bigint comparand is narrowed to a number, as it is in `where`.** `{ total: { $in: [500n, 20n] } }` kept no group, because `[500n].includes(500)` is false. It now keeps the 500 and 20 groups. The caller's `having` object is not edited.
+
+Who is affected: `having` is a request-only key (`QuerySchema.having`, `EngineAggregateOptions.having`), and no metadata type stores it. Every `having` in this repository's docs and published skills is a scalar comparison against an aggregation alias (`{ order_count: { $gt: 5 } }` and the like), which answers exactly as before. Callers of `engine.aggregate` and of the REST aggregate query in a deployment were NOT measured.
+
+Not changed: scalars, `null` in the equality slot, `$in` / `$nin` lists, a two-bound `$between`, scalar ordering bounds, `{}`, `null` and an omitted `having`, on both paths. The `$like` / `$ilike` operators are still refused on `having` (they are staged out of `FILTER_OPERATORS`), and `$ne` with a list is still answered until the shared face judges it. A `having` key that names no column still keeps no group rather than being refused.