Mutations

@dbsp/core provides four mutation builders — insert, update, delete, and upsert. All require an explicit .execute() call; none run implicitly. Every builder also supports .dump() for SQL inspection without hitting the database.

---

Insert

values() checks column keys and value types, but it cannot yet derive which columns are database-generated or have defaults. Required-column enforcement therefore remains with the database; an empty row is rejected before SQL is compiled.

// Insert a single row
orm.insert('users')
  .values({ name: 'Alice', email: 'alice@example.com' })
  .dump();

Bulk Insert

Pass an array to .values() to insert multiple rows in a single statement:

orm.insert('users')
  .values([
    { name: 'Alice', email: 'alice@example.com' },
    { name: 'Bob',   email: 'bob@example.com'   },
  ])
  .dump();

RETURNING

Use .returning() to get back specific columns from the inserted rows:

const { sql: newUserSql } = orm.insert('users')
  .values({ name: 'Alice', email: 'alice@example.com' })
  .returning(['id', 'name', 'createdAt'])
  .dump();

---

Update

const userId = 1;

import { eq } from '@dbsp/core';

orm.update('users')
  .set({ name: 'Alice Smith' })
  .where(eq('id', userId))
  .dump();

update() requires a .where() clause. Omitting it throws UnsafeOperationError — this is a safety guard against accidental full-table updates.

Update All Rows (Intentional Full-Table)

When you genuinely need to update every row, use updateAll():

orm.updateAll('users')
  .set({ active: false })
  .dump();

Update with RETURNING

import { eq } from '@dbsp/core';

const { sql: updatedSql } = orm.update('users')
  .set({ active: true })
  .where(eq('email', 'alice@example.com'))
  .returning(['id', 'name', 'active'])
  .dump();

---

Delete

import { eq } from '@dbsp/core';

orm.delete('posts')
  .where(eq('published', false))
  .dump();

Like update(), delete() requires a .where() clause. Use deleteAll() when you intend to remove every row:

orm.deleteAll('users').dump();

Delete with RETURNING

import { eq } from '@dbsp/core';

const { sql: removedSql } = orm.delete('posts')
  .where(eq('published', false))
  .returning(['id', 'title'])
  .dump();

Relation guards with exists()

Mutation WHERE clauses can use relation-aware exists() guards. The relation must be declared in the schema; the compiler emits the foreign-key correlation and your extra relation filter. .dump() compiles the mutation without opening a database connection.

const __mutationGuardDb = schema({
  posts: {
    id: { type: 'integer', primaryKey: true },
    title: 'string',
    archived: 'boolean',
  },
  comments: {
    id: { type: 'integer', primaryKey: true },
    postId: ref('posts'),
    flagged: 'boolean',
  },
} as const);

const __mutationGuardOrm = createOrm({
  schema: __mutationGuardDb,
  adapter: createPgsqlCompileOnlyAdapter({ model: __mutationGuardDb.model }),
});

const __mutationGuardDump = __mutationGuardOrm.update('posts')
  .set({ archived: true })
  .where(exists('comments', { where: eq('flagged', true) }))
  .dump();

if (
  !__mutationGuardDump.sql.includes('WHERE EXISTS') ||
  !__mutationGuardDump.sql.includes('comments_exists_0.flagged = $2') ||
  __mutationGuardDump.parameters.length !== 2 ||
  __mutationGuardDump.parameters[0] !== true ||
  __mutationGuardDump.parameters[1] !== true
) {
  throw new Error('mutation exists() relation guard did not compile as expected');
}

---

Upsert

Insert a row, or update it on conflict:

// Auto-update all non-conflict columns
orm.upsert('users')
  .values({ name: 'Alice', email: 'alice@example.com', active: true })
  .onConflict(['email'])
  .doUpdate()
  .dump();

// Update only specific columns on conflict
orm.upsert('users')
  .values({ name: 'Alice', email: 'alice@example.com', active: true })
  .onConflict(['email'])
  .doUpdate({ name: 'Alice Updated', active: true })
  .dump();

// Skip the row silently on conflict
orm.upsert('users')
  .values({ name: 'Alice', email: 'alice@example.com' })
  .onConflict(['email'])
  .doNothing()
  .dump();

// Conflict by constraint name instead of columns
orm.upsert('users')
  .values({ name: 'Alice', email: 'alice@example.com' })
  .onConflictConstraint('users_email_unique')
  .doNothing()
  .dump();

// With RETURNING
const { sql: resultSql } = orm.upsert('users')
  .values({ name: 'Alice', email: 'alice@example.com' })
  .onConflict(['email'])
  .doUpdate()
  .returning(['id', 'name'])
  .dump();

Advanced ON CONFLICT patterns

The basic upsert patterns above cover most use cases. The following patterns handle edge cases you will encounter in production schemas.

Selecting a specific constraint

When a table has multiple unique constraints, specify which one governs conflict detection. Use .onConflictConstraint(name) to target a named constraint instead of listing columns:

// doctest: skip — constraint names are DB-specific
orm.upsert('users')
  .values({ name: 'Alice', email: 'alice@example.com', externalId: 'ext-001' })
  .onConflictConstraint('users_email_unique')
  .doUpdate({ name: 'Alice Updated' })
  .dump();
// SQL: INSERT INTO "users" (...) VALUES ($1, $2, $3)
// ON CONFLICT ON CONSTRAINT "users_email_unique"
// DO UPDATE SET "name" = $4

Source: onConflictConstraint(constraintName: string) in packages/core/src/dx/mutation-builders.ts.

Partial-index conflict targets with a WHERE clause

PostgreSQL partial indexes restrict conflict detection to rows that satisfy a condition. The fluent .onConflict() API accepts only a column array and cannot express the partial-index predicate, so do not use it to target a partial unique index. Use a non-partial unique constraint with the fluent builder, or an explicit SQL escape hatch when you must target a partial index.

// A non-partial unique index: CREATE UNIQUE INDEX ON "products" ("sku")
orm.upsert('products')
  .values({ sku: 'ABC', price: 99.99, active: true })
  .onConflict(['sku'])
  .doUpdate({ price: 99.99 })
  .dump();
// SQL: INSERT INTO "products" ("sku", "price", "active") VALUES ($1, $2, $3)
// ON CONFLICT ("sku")
// DO UPDATE SET "price" = $4

Note: The WHERE passed as the second argument to .doUpdate() applies to the UPDATE action, not to conflict-target inference.

Conditional DO UPDATE predicates

Use the second argument to .doUpdate(set, where) when conflict detection should happen normally, but the conflicting row should only be updated if the existing target row matches a predicate:

import { eq } from '@dbsp/core';

orm.upsert('products')
  .values({ sku: 'ABC', price: 99.99, active: true })
  .onConflict(['sku'])
  .doUpdate({ price: 99.99 }, eq('active', true))
  .dump();
// SQL: INSERT INTO "products" ("sku", "price", "active") VALUES ($1, $2, $3)
// ON CONFLICT ("sku") DO UPDATE SET "price" = excluded."price"
// WHERE "products"."active" = $4

This WHERE belongs to the DO UPDATE action. If it evaluates to false, PostgreSQL leaves the existing conflicting row unchanged.

Multi-column conflict targets

List all columns that compose the unique constraint when the conflict target spans multiple columns:

orm.upsert('user_roles')
  .values({ userId: 1, roleId: 3, grantedAt: new Date() })
  .onConflict(['userId', 'roleId'])
  .doUpdate({ grantedAt: new Date() })
  .dump();
// SQL: INSERT INTO "user_roles" (...) VALUES ($1, $2, $3)
// ON CONFLICT ("userId", "roleId")
// DO UPDATE SET "grantedAt" = $4

DO NOTHING vs DO UPDATE — when to use each

Need	Use
Idempotent insert: ignore if already exists	.doNothing()
Upsert: create if new, update if exists	.doUpdate() (no args = auto-update all non-conflict columns)
Selective upsert: update only specific fields	.doUpdate({ col: value })
Conditional upsert: only update if a condition holds	.doUpdate({ col: value }, whereCondition)

.doNothing() compiles to ON CONFLICT DO NOTHING — the entire row is left unchanged on a conflict. It is the correct choice for "insert if not exists" patterns where updating the existing row would be incorrect (e.g., first-write-wins semantics).

---

Safety Rules

Builder	Safety rule
orm.update()	Requires .where() — throws UnsafeOperationError without it
orm.delete()	Requires .where() — throws UnsafeOperationError without it
orm.updateAll()	No WHERE required — explicit opt-in for full-table update
orm.deleteAll()	No WHERE required — explicit opt-in for full-table delete
values() / doUpdate() / set()	Column keys and value types are checked; INSERT requiredness remains database-enforced until default/serial metadata is modeled

These rules prevent silent data loss from forgotten filter conditions.

---

Observability with dump()

All mutation builders support .dump() — inspect the SQL and parameters without executing:

const { sql, parameters } = orm.insert('users')
  .values({ name: 'Alice', email: 'alice@example.com' })
  .dump();

console.log(sql);
// INSERT INTO "users" ("name", "email") VALUES ($1, $2)

console.log(parameters);
// ['Alice', 'alice@example.com']

This works even without a database connection (compile-only mode). See Observability for more on dump().