DevelopersPre-alpha

ORM

How core/orm turns Fields into tables and how records are read and written.

Role

core/orm owns the model registry, PostgreSQL access, schema sync from FieldDefinition, security-aware CRUD, and mutation/outbox paths.

From Fields to tables

On install and module update (-i / -u), registered models run additive schema sync from registered model metadata: new columns, indexes, and unique constraints are created. Types map Char → varchar/text, Boolean, Integer, relations, and so on.

dropStaleColumnUniques removes single-column UNIQUE constraints when a field no longer declares Unique: true — it does not drop columns or tables. There is no boot-time schema_migrate repair pass; schema changes happen during module load only.

Pre-alpha. Sync is additive. Renaming or removing fields does not automatically drop old columns. Plan manual SQL on shared databases.

Records as maps

Public APIs search/create/write with maps. Addon authors still declare typed structs for Field metadata via the SDK.

SDK boundary

Addon code should import sumeru/core/sdk (sdk.Model tags). Registration is generated by make generate. Keep direct ORM imports for core/engine work.

Computed fields and read_group

orm.RegisterCompute derives field values on read (see Models & fields). orm.ReadGroup and the read_group JSON-RPC method provide sum/count aggregation with the same ACL and record rules as search.

Compile and run

Terminal

shell

$cd sumeru$go test ./test/orm/...$cd ../sumeru_custom_addons && make run

What not to do

Do not bypass the mutation pipeline with ad-hoc SQL for business rows unless you accept outbox/event divergence.
Do not change FieldType meanings in addons. Extend models instead.

Next step

See load order in Module loader, or practical fields in Models & fields.