[Sukairo-02]

v1.0.0-rc.6

• Breaking changes:

  • Drivers no longer configure output decoding in drivers' configurations - following configurations set by drizzle constructors were removed:

    • client.options.bigint = true from bun-sql/postgres, bun-sql/mysql
    • transparent decoders from postgres-js
    • global types.setTypeParser(...) from neon-http, netlify-db
    • per-query parser overrides from node-postgres, node-postgres/dsql, vercel-postgres, neon-serverless, netlify-db, cockroach, pglite
    • per-query typeCast from mysql2

    ❗postgres-js still overrides client's param serializers (client-wide, replacing user-configured ones) for timestamptz (1184), timestamp (1114), date (1082), time (1083) and their arrays:

    • Date values are sent as value.toISOString() - same as postgres.js' default for timestamptz, timestamp, date
    • strings are sent as is to be parsed by the server - postgres.js' default re-encodes them via new Date(value).toISOString(), which reads offset-less timestamps in process' local time, truncates microseconds and throws on values JS can't parse ('infinity', '12:00:00')
    • time switches Date processing from date.toString() to ISO string

    Driver config overrides were replaced with cast codecs for types:

  • pg, cockroach: bigint, bigserial, numeric, date, time, timestamp, interval, json, jsonb and their array versions

  • mysql, singlestore: decimal, date, datetime, timestamp, time, json

    sqlite: integer-based codecs (integer, integer:boolean, integer:timestamp, integer:timestamp_ms) accept number, bigint and integer text, which makes them independent of safe integer modes (better-sqlite3's defaultSafeIntegers, bun:sqlite's safeIntegers, node:sqlite's readBigInts, libsql's intMode)

    ❗ Results of raw queries (db.execute(sql...), sql fields without a column's decoder) now honor driver's configuration

    pg-proxy, mysql-proxy, singlestore-proxy servers no longer need to replicate drizzle's driver result type configurations (e.g. returning timestamp, date, interval as strings via types.setTypeParser(...) / typeCast)

    ⚠️ Codecs only cover client options with predictable output (bigint, safeIntegers, intMode, decimalNumbers, jsonStrings, ...). Custom parsing functions (pg's types.setTypeParser(...) / types.getTypeParser, postgres' types, mysql2's typeCast) that change output of types drizzle doesn't cast still require custom codecs, e.g.:

    import { refineCodecs } from 'drizzle-orm/codecs';
    import { drizzle, mysql2Codecs } from 'drizzle-orm/mysql2';
    import { createPool } from 'mysql2/promise';
    
    const client = createPool({
      uri: process.env.DATABASE_URL,
      typeCast: (field, next) => {
        // `tinyint(1)` is returned as `boolean` instead of `number`
        if (field.type === 'TINY' && field.length === 1) return field.string() === '1';
        // `BLOB`-typed fields (`text`, `blob` columns) are returned as `Buffer` instead of `string`
        if (field.type === 'BLOB') return field.buffer();
        return next();
      },
    });
    
    const fromBuffer = (value: Buffer) => value.toString('utf8');
    
    const db = drizzle({
      client,
      codecs: refineCodecs(mysql2Codecs, {
        // default `boolean` codec expects `0` / `1` - client's output is already final
        boolean: { normalize: undefined },
        // text codecs have no decoders by default - client's output is expected to be a string
        text: { normalize: fromBuffer },
        mediumtext: { normalize: fromBuffer },
      }),
    });

    import { refineCodecs } from 'drizzle-orm/codecs';
    import { arrayCompatNormalize } from 'drizzle-orm/pg-core';
    import { drizzle, postgresJsCodecs } from 'drizzle-orm/postgres-js';
    import postgres from 'postgres';
    
    type Line = [a: number, b: number, c: number];
    
    const client = postgres(process.env.DATABASE_URL, {
      types: {
        // `point` (and `point[]`) is returned as `{ x, y }` instead of `(x,y)` text
        point: {
          to: 600,
          from: [600],
          // drizzle already encodes points to `(x,y)` text
          serialize: (value: string) => value,
          parse: (value: string) => {
            const [x, y] = value.slice(1, -1).split(',').map(Number);
            return { x, y };
          },
        },
        // `line` (and `line[]`) is returned as `[a, b, c]` instead of `{a,b,c}` text, params are expected as `[a, b, c]`
        line: {
          to: 628,
          from: [628],
          serialize: ([a, b, c]: Line) => `{${a},${b},${c}}`,
          parse: (value: string) => value.slice(1, -1).split(',').map(Number) as Line,
        },
      },
    });
    
    const toABC = ([a, b, c]: Line) => ({ a, b, c });
    
    const db = drizzle({
      client,
      codecs: refineCodecs(postgresJsCodecs(client), {
        // default `point` codecs parse `(x,y)` text - client's output is already final
        point: { normalize: undefined, normalizeArray: undefined },
        // `line({ mode: 'abc' })` columns
        line: {
          normalize: toABC,
          normalizeArray: arrayCompatNormalize(toABC),
          // drizzle encodes `{ a, b, c }` to `{a,b,c}` text - array params stay array literals, which `postgres.js` sends as is
          normalizeParam: (value: string) => value.slice(1, -1).split(',').map(Number) as Line,
        },
      }),
    });

  • Changed postgresJsCodecs (drizzle-orm/postgres-js, drizzle-orm/postgres-js/dsql), bunSqlPgCodecs (drizzle-orm/bun-sql/postgres) from plain objects to (client) => codecs functions due to json param wrapper functions only being accessible from drivers via client.json(param)

  • Fixed bun-sql/mysql datetime in string returning trailing zeroes for empty second fractions

  • Bumped minimal supported version of @types/mssql peer dependency from 9.1.4 to 9.1.11 to match declarations of mssql@11.0.1; changed mssql, @types/mssql peer dependency version requirements from ^ to>=

  • Changed shapes of dialect-specific concat, substring expressions (drizzle-orm/<dialect>-core/expressions) to match their dialects' functions:

    • mysql, singlestore, mssql: concat(a, b, ...rest) now accepts any number of values (string, number, bigint, Placeholder, SQLWrapper) and emits concat(...) instead of ||, which is a logical OR in mysql, singlestore and is unsupported by mssql before SQL Server 2025
    • mysql, singlestore: substring(column, { from, for? }) - from is now required (substring(x for n) is a syntax error); from, for accept bigint
    • pg, cockroach: substring(column, { from?, for? }) now requires at least one of from, for (substring(x) is an error); from is cast to integer (pg) / int8 (cockroach) due to substring(text from pattern) overload being picked by default; cockroach positions accept bigint
    • pg, cockroach: added regular expression substring overloads:
      • substring(column, { fromPattern }) - POSIX regular expression (substring(x from pattern))
      • pg: substring(column, { similar, escape }) - SQL regular expression (substring(x similar pattern escape escape))
      • cockroach: substring(column, { fromPattern, escape }) - POSIX regular expression with escape in place of \ (substring(input, regex, escape_char))
    • sqlite: substring(column, { from?, for? }) -> substring(column, start, length?), emits substr(x, start[, length]) instead of unsupported substring(x from ... for ...); start, length accept bigint
    • mssql: substring(column, { from?, for? }) -> substring(column, start, length), emits substring(x, start, length) instead of unsupported substring(x from ... for ...)
    • concat values and substring arguments are no longer encoded with the column's encoder - they are not expected to share the column's type

  • Fixed subqueries selected as fields being inferred as non-nullable - a subquery can return no rows, so their values are now typed as | null. This also applies to such fields re-selected from views and subqueries built from these selections:

  const res = await db.select({
    id: users.id,
    // before: number, now: number | null
    lastPostId: db.select({ id: posts.id }).from(posts).where(eq(posts.authorId, users.id)).orderBy(desc(posts.id)).limit(1).as('last_post_id'),
  }).from(users);

• Fixed effect-postgres failing to encode empty array params (Cannot infer the type of an empty array) of text, varchar, char, bool, smallint, int, bigint, serial types, float4, float8, bytea columns (fixes #6566)

• Fixed effect-postgres bigint, bigserial columns in bigint mode returning strings when PgClient decodes int8 as text (fixes #6566)

• Fixed bun-sql postgres bigint, bigserial columns in bigint mode returning strings since bun started ignoring options.bigint set after client's construction

• Fixed bun-sql postgres empty multidimensional integer arrays being returned as Int32Array

• Fixed is() narrowing columns to never:

  • pg: text, varchar, char with enum (is(column, PgText)), plain varchar, char
  • cockroach: cockroachEnum columns (is(column, CockroachEnumColumn))
  • mysql: blob, tinyblob, mediumblob, longblob in buffer mode (is(column, MySqlBufferBlob))

• Fixed mysql, singlestore enum columns narrowed by is() having column.enumValues typed as | undefined

• Fixed cockroach cockroachEnum columns being typed with string dataType instead of string enum, which made validators infer string instead of the enum's values

• Fixed mysql blob, tinyblob, mediumblob, longblob in buffer mode having string dataType instead of object buffer, which made validators generate string schemas for Buffer values

• Fixed validators inferring string instead of enum members for columns of enums made from TS enums (pgEnum, cockroachEnum, mysqlEnum)

• Fixed drizzle-orm/zod createSchemaFactory without coerce option breaking refine callbacks' parameter type inference (Parameter 'schema' implicitly has an 'any' type) (fixes #5968)

• Fixed postgres-js errors on json, jsonb columns' params with boolean, Date, Buffer, Uint8Array values and arrays starting with them (column "..." is of type jsonb but expression is of type ...) (fixes [BUG]: postgres-js: top-level boolean/Date/bigint values fail to insert into jsonb columns in 1.0.0-rc.5 #6582)

• Fixed bun-sql/postgres errors on json, jsonb columns' params with top-level number, boolean, Buffer, Uint8Array values (fixes [BUG]: bun-sql: top-level number/boolean values fail to insert into jsonb columns in 1.0.0-rc.5 #6591)

• Fixed bun-sql/postgres with prepare: false errors on json, jsonb columns' params (fixes #6132)

• Fixed node-postgres, neon-serverless, vercel-postgres, netlify-db errors on json, jsonb columns' params with Date, Buffer, Uint8Array values (invalid input syntax for type json)

• Fixed type errors in emitted declarations with skipLibCheck: false (fixes #5187):

  • Cannot find module 'mysql2' in projects without mysql2 installed - general modules (drizzle-orm, drizzle-orm/pg-core, ...) no longer reference driver packages' types
  • Class 'PgRole' incorrectly implements interface 'PgRoleConfig' (same for PgPolicy, CockroachRole, CockroachPolicy) with exactOptionalPropertyTypes
  • Type 'Buffer' is not generic on TypeScript 5.6 and older in mysql-core blob codecs and drizzle-orm/arktype's bufferSchema

• Fixed: [BUG]: PostgreSQL: after drizzle-kit pull, generate and push emit SET DEFAULT for a numeric column with an expression default #6495
  Thanks to @pintu544 for opening this pull request - [Pg-kit]: Do not quote expression defaults of numeric columns on introspect #6516

• Fixed: [BUG]: drizzle-kit pull writes as: "function toLowerCase() { [native code] }" for a RESTRICTIVE policy on CockroachDB #6492
  Thanks to @shaurya703 for opening this pull request - [Kit]: Fixed drizzle-kit pull writing a function for a RESTRICTIVE CockroachDB policy #6518

• Fixed: [BUG]: SQLite generate ignores an index changing between uniqueIndex and index #6557
  Thanks to @tlq5l for opening this pull request - Drop and recreate SQLite indexes that change between index and uniqueIndex #6561

• Fixed: [BUG]: Dropping a column writes a failing migration when a check or unique constraint naming it changes in the same diff #6472
  Thanks to @Adnan-Husayn for opening this pull request with the proposed solution - [Pg-kit]: Drop an altered check or unique constraint before its column #6483

• Fixed: [BUG]: drizzle-kit pull (MySQL) fails with "Check constraint conflict" when another database on the same server has a CHECK constraint with the same name #6447
  Thanks to @Saadanjum0, @drago1520 and @shreyanshj10 for opening those pull requests - [MySQL-kit] Fix check constraint introspection matching across databases #6450, [drizzle-kit] fix: MySQL pull fails with "Check constraint conflict" when another database has a CHECK with the same name #6449, [Mysql-kit]: Scope the check constraints join to the pulled schema #6448

• Fixed: [BUG]: SQL Server diizzle-kit pull error when IDENTITY seed = 0 #6423
  Thanks to @yousefissa for opening this pull request - https://github.com/drizzle-team/drizzle-orm/pull/6432


• Fixed: [BUG]: drizzle-kit push fails to introspect MSSQL schemas owned by Microsoft Entra users #6357
  Thanks to @zhamann for opening this pull request - fix(drizzle-kit): introspect Entra-owned MSSQL schemas #6358

• Fixed: [BUG]: drizzle-kit for postgres generates incorrect migration when changing IDENTITY column to uuid #4178
  Thanks to @yyuneu for opening this pull request - [Pg-kit]: Drop identity before changing a column's type #6226

• Fixed: fix(pg): fix sparsevec_l2_ops typo and test composite index opclasses #6188. Thanks to @webdevsamran for opening this pull request

• Fixed: [BUG]: drizzle-kit pull writes an expression default as plain code or as a string for several column types (PostgreSQL, CockroachDB, SQLite, MySQL) #6493
  Thanks to @pintu544 and @shaurya703 for opening these pull requests - [Pg-kit]: Emit expression defaults as sql templates in toTs instead of broken literals #6521, Fixed drizzle-kit pull writing expression defaults as code or strings #6519

• Fixed query buidlers dropping offset with 0 value (fixes [BUG]: .offset(0) writes no offset, so a subquery fenced with offset 0 is merged into its outer query #6599)

• Fixed [BUG]: node-postgres transaction() still returns the client to the pool when BEGIN times out #6622

• Fixed [BUG]: durable-sqlite .get() leaves the statement open when several rows match, so a later DROP INDEX fails with SQLITE_LOCKED #6550

• Fixed effect-sqlite-do, effect-sqlite-bun, effect-sqlite-wasm .run() result type: [] -> readonly Record<string, unknown>[]

• Fixed [BUG]: SQLite onConflictDoNothing places the target WHERE after DO NOTHING (partial unique indexes) #6531

• mysql, singlestore: $returningId can no longer be combined with onDuplicateKeyUpdate as the result can't report updated rows' ids

• Fixed [BUG]: pg timestamp: a Date past the year 9999 is sent as "+010000-…", which PostgreSQL refuses (22009) #6552:

  • years past 9999 (+010000-01-01T00:00:00.000Z) are sent without the expanded-year sign: 10000-01-01T00:00:00.000Z
  • years before 1 AD (0000-01-01T00:00:00.000Z, -000001-...) are sent with the BC suffix: 0001-01-01T00:00:00.000Z BC

• Fixed pg, cockroach timestamp, date columns in date mode decoding values that new Date(text) misreads:

  • years 1 - 99 were read as two-digit years (0099-12-31 23:59:59+00 -> 1999)
  • date values past year 9999 were read in process' local time
  • values with the BC suffix and with seconds-precision UTC offsets were decoded as Invalid Date

• Fixed pg timestamp, date columns in date mode being decoded as wrong dates or Invalid Date in sessions with non-ISO DateStyle (fixes [BUG]: node-postgres: timestamptz read as Invalid Date or a wrong date when Postgres prints a seconds offset (LMT) or a non-ISO DateStyle #6496)

• Fixed aws-data-api/pg time columns losing fractional seconds past milliseconds and timestamp columns (date, string modes) losing the BC suffix of values with fractional seconds: they are now selected as text instead of Data API's own formatting

• Fixed postgres/aurora failing queries (HTTP 500 on the AWS Data API emulator) selecting timestamptz values with years past 9999 or seconds-precision UTC offsets, and the time, timestamp formatting losses above: time, timestamp, timestamptz columns are now selected as text

• Fixed [BUG]: not() returns SQL | undefined even for a defined condition since 1.0.0-rc.5 #6625

• Fixed [BUG]: MySQL generate stops when a new foreign key points to a table that is not exported from the schema #6534

• Fixed [BUG]: drizzle-kit pull crashes on MSSQL when tablesFilter excludes referenced FK table #5992

• Fixed [BUG]: SQLite pull reports every error after the connection as "Failed to connect to or introspect the SQLite database", dropping the cause #6501

• Fixed [BUG]: drizzle-kit ignores tsconfig paths mapping "*" under Node (only trailing /* keys supported) #6593

• Fixed [BUG]: drizzle-kit's tsconfig path-alias resolver does not follow "extends" #5712

• Fixed pg transaction config allowing snapshot without isolationLevel: 'repeatable read' | 'serializable'

  // before: compiled, failed at runtime; now: type error
  await db.transaction(async (tx) => { /* ... */ }, { snapshot });
  // ok
  await db.transaction(async (tx) => { /* ... */ }, { isolationLevel: 'repeatable read', snapshot });

• Fixed arrayCompatNormalize (drizzle-orm/pg-core) passing null array elements to normalizer instead of skipping them

[pkg-pr-new]

Open in StackBlitz

bun add https://pkg.pr.new/@drizzle-team/postgres@6588

npm i https://pkg.pr.new/@drizzle-team/postgres@6588

commit: f4225dc