What mikro-orm/mikro-orm shipped

New releases 7.2.7 and 7.2.8 are out. upsert now bumps the version column and merges every composite FK column on conflict, extra composite FK columns are filled correctly, and clone and insertFrom skip generated columns. mssql now waits 60s by default.

Two patch releases landed this week. Version 7.2.7 and 7.2.8 ship the fixes that came in over the past days. em.upsert() and em.upsertMany() now bump the version property when an existing row is updated on conflict, and the conflict merge covers every column of a composite foreign key instead of just the first. Extra columns of a composite foreign key referencing non-PK target columns are now written correctly, em.clone() and qb.insertFrom() skip generated columns, and a partial refresh keeps the older snapshot data instead of writing it back. The mssql driver now defaults its request timeout to 60 seconds, matching the other drivers more closely, and mongo partial indexes map property names to field names in the where clause.

MikroORM v7.2.6 is out. This week also fixed MSSQL date dirtiness, escaped string defaults on MySQL/MariaDB, Unicode JSON writes on MSSQL, joined loading column selection and populate with partial fields, plus faster metadata discovery.

MikroORM v7.2.6 is out. This week's changes focus on correctness across supported databases. String defaults are now escaped properly so schema:update stops producing permanent diffs and invalid DDL on MySQL and MariaDB. MSSQL writes JSON as Unicode literals and maps date results to strings, so loaded entities are no longer dirty. Joined loading picks the right column when a property shares a name with another column, populate handles nested partial fields through references, and query comments reject forbidden sequences again. Metadata discovery no longer does quadratic lookups, and MySQL can set the session time zone.

This week: PostgreSQL commits answered with ROLLBACK are now rejected, failed savepoint rollbacks block the outer commit, MySQL mixed batch inserts map generated ids correctly, and M:N collections with nested composite keys load right.

This week's fixes focus on cases where the database and the ORM could disagree. A PostgreSQL COMMIT that the server answers with ROLLBACK is now rejected instead of resolving, and a failed rollback to a savepoint prevents the outer transaction from committing. MySQL batch inserts mixing explicit and generated primary keys now map generated ids correctly, M:N collections with a nested composite primary key load correctly in 6.x, and custom types are no longer converted twice in collection operator subqueries. Schema-side fixes keep property-level index and unique options, use persisted JSON keys for indexes on object embeddable properties, and match PostgreSQL index columns as whole items.

This week: MSSQL composite FK introspection, PostgreSQL pooled client release on dropped transactions, escaped property names in hydration code, schema diffing for index columns without options, and refresh fixes. Docs corrected on loading strategies and concurrent index migratio

This week in MikroORM: MSSQL introspection now keeps composite foreign key columns paired. PostgreSQL releases the pooled client when a connection drops mid-transaction. Core escapes entity property names in generated hydration and comparison code. Schema diffing treats index columns without options as default options. em.refresh() skips scalars of uninitialized references. On the docs side, the relationship loading strategies page now correctly introduces balanced as the default for SQL drivers, there is an explanation that create index concurrently migrations need allOrNothing false, and a note that Opt and Hidden intersections need explicit types with reflect-metadata.

7.2.5 is out: fixes for em.refresh with custom types, STI embeddable hydration, null values in upsert results, plus STI join matching, TPT flush ordering, formula selection in joined queries, and glob paths in project directories.

Release 7.2.5 is available with fixes for custom types and serializedName in em.refresh, STI object embeddable hydration, and null custom type values in upsert results. Alongside the release, this week brings fixes for metadata discovery with same-named properties across polymorphic embeddable variants, joins into non-abstract STI entities, formulas in joined queries, TPT flush ordering and descendant formulas, MongoDB upserts without a unique value, schema down migrations for routines and triggers, and project paths containing glob characters. Performance work also reduces reload cost in upsertMany and groups MongoDB M:N population in fewer passes.

7.2.3 and 6.6.17 released. Fixes for decimal precision, versioned batch updates, TPT identity reseeding, and Oracle triggers. Autoincrement columns now support sequence start/increment options.

Mikro-orm 7.2.3 and 6.6.17 are out. These releases address critical schema and data handling issues: decimal string precision detection beyond double precision, versioned entity batch updates with race condition prevention, TPT identity reseeding on MSSQL, and Oracle trigger support. New feature: autoincrement columns can now specify sequence start and increment values. Schema diffs improved for PostgreSQL identity options, SQLite primary key changes, and SQLite foreign key constraints.

This week: filter and order by properties of any polymorphic relation target, cache Kysely result maps, add per-entity forceUndefined, fix STI and upsert consistency issues, respect returning hints on updates.

Mikro-orm 6.4 week: polymorphic relations now let you filter and order by properties across all target types, not just the first. Improved Kysely performance with result map caching. Added per-entity forceUndefined options. Fixed data consistency for STI subtypes and heterogeneous upserts. Better handling of returning hints on updates and upserts.

V7.2.1 is out. Fixes for property escaping in generated code, MSSQL legacy column type mapping, returning clause handling, and many-to-many reference preservation.

V7.2.1 released with several important fixes: property names are now properly escaped in generated hydrators and comparators, MSSQL legacy ntext and image types are correctly mapped, INSERT returning clauses properly sync entities with database values, raw query parameters with special characters no longer cause parsing errors, and filtered many-to-many references are preserved correctly.