Schema generator generating invalid js names

[lckrugel]

I have a MySQL table that has some columns with names like:

  `2g` tinyint(1) DEFAULT '0',
  `3g` tinyint(1) DEFAULT '0',
  `4g` tinyint(1) DEFAULT '0',

And Lucid generates the schema like:

  @column()
  declare 2G: boolean | null
  @column()
  declare 3G: boolean | null
  @column()
  declare 4G: boolean | null

Which is invalid property name in javascript.

Bad naming practices and table structure aside, is there any way I can configure a custom rule to rename this column so Lucid generates a valid schema? Even skipping auto generating this table would be fine.

I do not "own" this table, so changing the column name is out of question, but I do need to generate a schema for the database where it is from.

For now I'm having to resort to disabling schema generation for this connection.

[manimovassagh]

This is a known limitation of the schema generator — it does not handle column names that start with digits or contain characters invalid for JS identifiers.

A few workarounds:

1. Use the columnName decorator option to map to a valid property name:

After the schema is generated, manually fix the problematic columns:

@column({ columnName: "2g" })
declare twoG: boolean | null

@column({ columnName: "3g" })
declare threeG: boolean | null

@column({ columnName: "4g" })
declare fourG: boolean | null

The columnName option tells Lucid which actual DB column to map to, while the property name can be any valid JS identifier. This is the intended escape hatch for columns with non-standard names.

2. Exclude the table from auto-generation and write the model manually:

If you have many problematic columns, it might be easier to skip auto-generation for that table and write the model by hand. You can create the model file manually in app/models/ and use columnName for every column that needs renaming.

3. Post-process the generated schema:

If you want to keep auto-generation for the other tables, you could write a small script that runs after generation and fixes the output — replacing declare 2G with a valid identifier and adding the columnName mapping.

This is definitely worth reporting as a bug/enhancement though — the generator should either escape these names automatically (e.g. prefix with underscore or use a naming convention like _2g) or output bracket notation. Column names starting with digits are valid in MySQL/PostgreSQL, so the generator should handle them gracefully.

[lckrugel]

I might report that on Lucid's repo as a bug.

Besides that, how do I exclude a table from the auto-generation? Docs only talk about excluding the whole connection.

[rayfadh]

You can just override the column name with model level overrides. Just use the model class since it extends the schema class.

[lckrugel]

But in this case, when importing the schema, the application will throw an error due to the invalid identifier.

I've found no practical solution to that besides ignoring the schema generation altogether for that particular connection.

[rayfadh]

You could still use manimovassagh's solution regarding the invalid identifier for now.

I might even try implementing that later this week.

Cool 🙂

[lckrugel]

I've opened an issue for this on Lucid's repo as it do seem like an unexpected behavior from the schema generator.

The solution of prefixing the attributes with an underscore seems pretty simple and unoffensive, I might even try implementing that later this week.

[lckrugel]

A fix to this was merged into Lucid: e150d15

Property names will be prefixed with an underscore and we'll also be able to exclude tables from schema generation with custom rules.