Skip to main content

Object & CRUD Generation

Steps:

  1. DMMF taken from options.dmmf when Prisma supplies it, avoiding a redundant re-parse of the schema; getDMMF({ datamodel, previewFeatures }) is only used as a fallback.
  2. Input object & model operation lists cloned.
  3. Hidden models/fields resolved from model comments.
  4. Missing input object types added (legacy support) before filtering.
  5. Object schemas generated if emit.objects and not pure-only heuristics; each object pre-filtered with filterFields.
  6. CRUD operation schemas assembled (model args, aggregate support) if emit.crud.
  7. Objects index synthesized (for integration consistency).

isObjectSchemaEnabled checks model enablement + required operations + minimal mode pruning heuristics.

Aggregate result schemas produced via generateResultSchemas (if results not forcibly disabled).

Generated Shapes​

A few details of the assembled schemas are worth knowing when you read the output:

  • Select schemas accept nested relation args. A relation field in a select is a union, not a bare boolean, so you can pass query arguments through it (here for a Form model with a layouts relation):

    layouts: z.union([z.boolean(), z.lazy(() => LayoutFindManySchema)]).optional(),
    _count: z.union([z.boolean(), z.lazy(() => FormCountOutputTypeArgsObjectSchema)]).optional(),

  • Relation fields in Create* and Unchecked*Input keep DMMF optionality. schema.inputObjectTypes is the source of truth, because that is what the Prisma Client input types are generated from. The datamodel recompute may only relax requiredness (a field with a database default becomes optional); it never promotes a DMMF-optional field to required. List fields are always treated as optional, matching the client's input types.

  • Result schemas reference pure models for relations. When pureModels is enabled, a relation field is typed as z.array(<Model>Schema).optional() or <Model>Schema.optional() and imported from ../models/. Self-relations, runs without pure models, and single-file mode fall back to z.array(z.unknown()).optional() / z.unknown().optional(). Relation fields are always .optional() because they are only present when the query includes them.

  • Prisma is only imported when it is used. import type { Prisma } is emitted only when the file's typed export actually references Prisma, so files that do not need it no longer carry an unused import. In decimalMode: "decimal" it becomes a value import instead, since Prisma.Decimal.isDecimal checks and Decimal defaults need the runtime value.