Type-Safety System

MountSQLI derives row types from your schema with phantom types and inference helpers. No generators, no decorators.

The trick: phantom type params

Each ColumnBuilder carries a phantom type param for its SQL type and nullability:

const users = defineTable("users", {
  id: int().pk(),          // ColumnBuilder<"int", false>
  email: text().notNull(), // ColumnBuilder<"text", false>
  age: int(),              // ColumnBuilder<"int", true>  (nullable)
});

The phantom params are erased at runtime but drive inference at compile time.

InferTable maps them

InferTable uses a TypeMap to turn each column’s phantom type into a TS type, and nullability decides | null:

type User = InferTable<typeof users>;
// { id: number; email: string; age: number | null }

The TypeMap

SQL type	TS type
int	number
text	string
bool	boolean
blob	Uint8Array
json	unknown
timestamp	Date

Why no codegen?

Because the type lives in the column builders themselves, inference is immediate and always in sync. Change a column, and every query result type updates on the next tsc — there’s nothing to regenerate.

Best practices

• Export type User = InferTable<typeof users> next to each table.
• Use DbFromConfig<typeof config> for the app-wide Db.
• Let type errors catch bad column names before runtime.

Common mistakes

• Hand-writing row interfaces that drift from the table.
• Using any and losing the inference.

Related

• InferTable — the user-facing API.
• Defining Tables — the column builders.