[finding] `os generate migration` emits no declared index at all — a generated table carries none of the object's `unique` constraints, while driver-sql creates them

[os-litant]

Found while closing the keyed half of #16091 (PR #16298). Not fixed there: that card is about the COLUMN, this is about the INDEX.

The measurement

Live PostgreSQL 16.13, one object driven through all three producers — driver-sql via initObjects, os generate migration --format sql via db.raw, and the typescript format by importing the emitted module and calling up(db). Object:

{ name: 'probe', fields: { keyed_unique: { type: 'text', unique: true, maxLength: 100 } } }

pg_indexes for each producer's schema afterwards:

driver   probe_pkey, uniq_probe_keyed_unique
sql gen  probe_pkey
ts gen   probe_pkey

The column itself now agrees in all three (character varying(100), after #16298). The CONSTRAINT does not exist in either generated table.

Why it matters, and why it is sharper now

Two rows with the same keyed_unique value are refused by the platform's table and accepted by both generated ones. A scaffold that creates the table for an object silently drops a uniqueness guarantee the object declares, and nothing reports it — the same shape as #16091's own hard-failure class, one declaration over.

It is sharper after #16298 because the generators now READ field.unique and indexes[] to size the column (indexedKeyColumns' key set is what the driver's text-family arm branches on). The key set is therefore already computed in generate.ts; the index it implies is still not emitted. driver-sql's own normalizers are uniqueIndexesFromFields (field-level unique, ADR-0120 D1/D3 scoping) and normalizeDeclaredIndex (object-level indexes[]), both exported from @objectstack/driver-sql.

Not free to answer, which is why this is filed rather than fixed

• The ADR-0120 organization-scoped form is COALESCE(organization_id, '__global__') as a key part — an expression index. Whether the SQL format should emit that expression, and whether the typescript format can express it through knex at all, is a decision.
• packages/cli production modules may not statically value-import a driver (objectstack dev 在工作区未构建时刷 12 段无关命令的 MODULE_NOT_FOUND，唯一可执行的那条却指向错误修法 #5726), so the naming and scoping rules would be transcribed and pinned, as the widths already are.
• --format sql is PostgreSQL-only (os generate migration's audit-stamp columns diverge from driver-sql — the generators emit NOT NULL where the driver emits nullable, and the SQL format emits TIMESTAMP where both knex paths yield timestamptz #15521); index DDL is more dialect-divergent than column DDL.

Reproduce

Drive the object above into three PostgreSQL schemas, one per producer, and read pg_indexes for each.