Naming errors in Entity Generator

[jefmenegazzo]

Describe the bug

There are two minor bugs about naming in the Entity Generator.

1. Class names that not follows Pascal Case

In entities with more than one underscore between words, pascal case is not applied correctly.

@Entity({ tableName: "wac_tela_principal_wac___ident" })
export class WacTelaPrincipalWac_dent { }

@Entity({ tableName: "sft_tqm___ocorrencia_envolvido" })
export class SftTqm_correnciaEnvolvido { }

@Entity({ tableName: "sft_tela_principal_sft___ident" })
export class SftTelaPrincipalSft_dent { }

As a temporary workaround, I'm using lodash as follows:

onInitialMetadata: (metadata: EntityMetadata<any>[], platform) => {

	for (const entity of metadata) {
		entity.className = lodash.upperFirst(lodash.camelCase(entity.tableName));
	}
},

2. When using scalarPropertiesForRelations: 'always', some relations are not formatted correctly

@ManyToOne({ entity: () => FrUsuario, ref: true, fieldName: "usr_codigo_app" })
public.frUsuario!: Ref<FrUsuario>;

@ManyToOne({ entity: () => SftSubProduto, ref: true, fieldName: "subpro_id_subtituto", nullable: true })
public.sftSubProduto?: Ref<SftSubProduto>;

@OneToOne({ entity: () => SftContatoWhatsOptIn, ref: true, mappedBy: "public.sftContato" })
public.sftContatoInverse?: Ref<SftContatoWhatsOptIn>;

It appears to only happen with ManyToOne and OneToOne relationships. In this case, 'public' is the schema name.

Reproduction

async function getDatabaseMetadata(connection: DatabaseConnection) {

	const orm = await MikroORM.init({
		discovery: {
			warnWhenNoEntities: false,
		},
		extensions: [EntityGenerator],
		driver: PostgreSqlDriver,
		schema: connection.schema,
		dbName: connection.dbName,
		host: connection.host,
		port: connection.port,
		user: connection.user,
		password: connection.password,
		logger: logger.log,
		// debug: true,
	});

	let initialMetadata: EntityMetadata<any>[] = [];
	let processedMetadata: EntityMetadata<any>[] = [];

	const entitiesSource = await orm.entityGenerator.generate({
		bidirectionalRelations: true,
		identifiedReferences: true,
		scalarTypeInDecorator: true,
		scalarPropertiesForRelations: 'always',
		onlyPurePivotTables: false,
		fileName: (className: string) => lodash.kebabCase(className),
		onInitialMetadata: (metadata: EntityMetadata<any>[], platform) => {

			// WORKAROUND
			for (const entity of metadata) {
				entity.className = lodash.upperFirst(lodash.camelCase(entity.tableName));
			}

			initialMetadata = metadata;
		},
		onProcessedMetadata: (metadata, platform) => { processedMetadata = metadata; },
	});

	await orm.close(true);

	return { entitiesSource, initialMetadata, processedMetadata };
}

What driver are you using?

@mikro-orm/postgresql

MikroORM version

6.1.12

Node.js version

v20.11.1, typescript 5.3.3

Operating system

Windows 11

Validations

• Read the Contributing Guidelines.
• Read the docs.
• Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
• Check that this is a concrete bug. For Q&A open a GitHub Discussion or join our Slack.
• The provided reproduction is a minimal reproducible example of the bug.

[boenrobot]

About 1, you can also use a custom naming strategy as a more "legit" workaround.

Extend the underscore one, and override the getEntityName() method. This way, you'll ensure that any references to these tables are also updated properly. You can do that in the hooks too, sure, but it's more work for the hook.

[boenrobot]

About 2... can you share the database's DDL? Not the full... just a reduced subset, including just the tables involved above.

I was recently tackling weird edge cases like this, with the results so far being in #5359, but as I mention even there, there is one edge case that's not covered there, and inherently also in the current version... identifiers with dot in the name. I suspect you have something (table, column, index name, FK constraint name...) that has a dot in its identifier, and that ends up screwing the whole thing.

Or perhaps this is a problem with cross schema FKs... hard to tell without a DDL.

Oh, and btw... you could work around that via a custom naming strategy too... override columnNameToProperty().

[jefmenegazzo]

@boenrobot About 1, I hadn't seen about Naming Strategy, it really is enough for the problem :)
About 2, I have a simple DDL where the problem also occurs: dump.zip

[boenrobot]

I've added a fix for the first problem in #5359, so that you wouldn't need to override getEntityName().

The same PR also already addressed having invalid characters as identifiers by quoting, meaning that your first example would become

@ManyToOne({ entity: () => FrUsuario, ref: true, fieldName: "usr_codigo_app" })
'public.frUsuario'!: Ref<FrUsuario>;

@ManyToOne({ entity: () => SftSubProduto, ref: true, fieldName: "subpro_id_subtituto", nullable: true })
'public.sftSubProduto'?: Ref<SftSubProduto>;

@OneToOne({ entity: () => SftContatoWhatsOptIn, ref: true, mappedBy: "public.sftContato" })
'public.sftContatoInverse'?: Ref<SftContatoWhatsOptIn>;

which is a valid class definition, where you can access the properties via f.e. (in the case of the first)
entityWithFrUsario['public.frUsuario']

Fixing the "true" problem of the schema name being at all present in the FK name will require more significant refactoring, which I already plan to tackle for the sake of supporting "." in identifiers.

As already mentioned, you can workaround this in the meantime by overring columnNameToProperty() in the naming strategy.

The problem is that when generating a persist: false for a prop that already exists as FK, the FK is prefixed with the name of the table being referenced if possible (i.e. when the FK is the only reference from the current table to the referenced table), or the name of the constraint if not. However, with Postgresql, table names also feature the schema name. I think the problem above, along with supporting "." in table/index/fk names can be "easily" solved by storing the schema in a separate prop, but the hard part is that all related code needs to be updated to handle cross schema references properly after that split.

[B4nan]

This is getting quite stale, are we missing some fix, a reproduction, or can we close?

[boenrobot]

I suppose you could rename it, or maybe create a new issue linked to this one...

This was two issues in one, and only one of them is already solved.

What is not solved yet is that table names in Postgresql feature the schema name, so whenever a FK is named after the referenced table, we have "schema_name.entity_name" get turned into "schemaName.EntityName" instead of just "EntityName", like in other drivers.

The dump above is sufficient reproduction DDL. scalarPropertiesForRelations: 'always' can be added to make the problem even more obvious, since conflicts between FKs and columns named after FKs is inevitable in that scenario, so table names are always used for the FKs.

Simply stripping out the schema name blindly isn't a good fix, because a FK may reference a different schema - we should check if the FK is to the same schema as the entity, and only strip it out then, or better yet, not feature it in the table name in the first place, when that's the case... Or ideally, never feature it in the table name, and have a separate prop for it.

[B4nan]

Simply stripping out the schema name blindly isn't a good fix, because a FK may reference a different schema

Yes, but the FK property should never contain the schema prefix IMO, it feels quite weird to me, even if the schema is not the same. I can understand that only if there is a naming conflict, and even then, it can't contain the dot separator.