EpicenterHQ

drizzle-orm

Drizzle ORM patterns for type branding and custom types. Use when working with Drizzle column definitions, branded types, or custom type conversions.

EpicenterHQ 4,604 348 Updated 3mo ago
GitHub

Install

npx skillscat add epicenterhq/epicenter/drizzle-orm

Install via the SkillsCat registry.

SKILL.md

Drizzle ORM Guidelines

Use $type() for Branded Strings, Not customType

When you need a column with a branded TypeScript type but no actual data transformation, use $type<T>() instead of customType.

The Rule

If toDriver and fromDriver would be identity functions (x) => x, use $type<T>() instead.

Why

Even with identity functions, customType still invokes mapFromDriverValue on every row:

// drizzle-orm/src/utils.ts - runs for EVERY column of EVERY row
const rawValue = row[columnIndex]!;
const value = rawValue === null ? null : decoder.mapFromDriverValue(rawValue);

Query 1000 rows with 3 date columns = 3000 function calls doing nothing.

Bad Pattern

// Runtime overhead for identity functions
customType<{ data: DateTimeString; driverParam: DateTimeString }>({
	dataType: () => 'text',
	toDriver: (value) => value, // called on every write
	fromDriver: (value) => value, // called on every read
});

Good Pattern

// Zero runtime overhead - pure type assertion
text().$type<DateTimeString>();

$type<T>() is a compile-time-only type override:

// drizzle-orm/src/column-builder.ts
$type<TType>(): $Type<this, TType> {
  return this as $Type<this, TType>;
}

When to Use customType

Only when data genuinely transforms between app and database:

// JSON: object ↔ string - actual transformation
customType<{ data: UserPrefs; driverParam: string }>({
	toDriver: (value) => JSON.stringify(value),
	fromDriver: (value) => JSON.parse(value),
});

Keep Data in Intermediate Representation

Prefer keeping data serialized (strings) through the system, parsing only at the edges (UI components).

The principle: If data enters serialized and leaves serialized, keep it serialized in the middle. Parse at the edges where you actually need the rich representation.

Example: DateTimeString

Instead of parsing DateTimeString into Temporal.ZonedDateTime at the database layer:

// Bad: parse on every read, re-serialize at API boundaries
customType<{ data: Temporal.ZonedDateTime; driverParam: string }>({
	fromDriver: (value) => fromDateTimeString(value),
});

Keep it as a string until the UI actually needs it:

// Good: string stays string, parse only in date-picker component
text().$type<DateTimeString>();

// In UI component:
const temporal = fromDateTimeString(row.createdAt);
// After edit:
const updated = toDateTimeString(temporal);