`to_tsvector(NULL)` returns NULL and `ts_rank(NULL, ...)` therefore
returns NULL — under Postgres's default `NULLS FIRST` for `ORDER BY DESC`
this would surface NULL-valued rows ahead of any matching ones, an
asymmetry with the multi-field path where `concat_ws(' ', ...)` already
skips NULLs and yields a 0.0 rank. Coalescing the field to `''` aligns
the two paths.
Adds `subtitle: String? @fullText` to the test fixture and a regression
test that orders a NULL-subtitle row against a matching one — without
the fix, the NULL row ranks first under DESC.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Introduces a Prisma-style full-text search capability gated by a new
field-level `@fullText` ZModel attribute. PostgreSQL only — MySQL/SQLite
throw NotSupported. Mirrors the existing `@fuzzy` design.
- Filter operator: `where: { title: { fts: { search, config? } } }`
emits `to_tsvector(field) @@ to_tsquery(query)` (or with a `::regconfig`
cast when `config` is provided; otherwise Postgres uses the database's
`default_text_search_config`).
- OrderBy operator: `_ftsRelevance: { fields, search, config?, sort }`
emits a single `ts_rank(...)`. Multi-field combines fields with
`concat_ws(' ', ...)` so AND queries match terms across fields
(matches Prisma's behavior).
- Type-level gating: the `fts` operator and `_ftsRelevance` orderBy
appear only on String fields annotated with `@fullText` and only when
the schema's provider is `postgresql`. Slicing's `'FullText'` filter
kind controls availability of the runtime operator.
- Cursor pagination is rejected when combined with `_ftsRelevance`
(parallel to `_fuzzyRelevance`).
Also refactors `buildOrderBy` to dispatch to small per-branch helpers
(`applyScalarOrderBy`, `applyAggregationOrderBy`, `applyRelationOrderBy`,
`applyFuzzyRelevanceOrderBy`, `applyFtsRelevanceOrderBy`).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The generated schema previously used `as const` on entire model/enum/typeDef
objects, causing TypeScript to deeply infer literal types for all nested
properties. This is unnecessary for `attributes`, `default`, and `foreignKeyFor`
which are only used at runtime, not in CRUD type computations.
Changes:
- Add type assertions (`as readonly AttributeApplication[]`, `as FieldDefault`,
`as readonly string[]`) to prevent deep const inference on these properties
- Extract `FieldDefault` type alias from `FieldDef` for cleaner generated code
- Change `FieldHasDefault` to use key existence check (`'default' extends keyof`)
instead of value type check, enabling the `default` widening
- Conditionally import `AttributeApplication` and `FieldDefault` only when used
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Introduces a new `$diagnostics()` method on ZenStackClient that returns
Zod schema cache statistics and slow query information, helping users
monitor and debug ORM performance.
- Add `diagnostics` option to `ClientOptions` with `slowQueryThresholdMs`
and `slowQueryMaxRecords` settings
- Track slow queries in `ZenStackQueryExecutor` when diagnostics is enabled
- Share slow query collection across derived clients (via $setAuth,
$setOptions, $use, transactions, etc.)
- Cap slow query records with an eviction policy that keeps the slowest
queries (default max: 100)
- Validate diagnostics config with Zod in ClientImpl constructor
- Add `Diagnostics`, `QueryInfo`, and `ZodCacheStats` types
- Add e2e tests covering all diagnostics features
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
