Applying JSON schema results in errors

[imiljkovic]

After generating JSON schema with the following command:

atlas schema inspect -u "sqlserver://sa:Pass123@localhost:1433?database=test"  --format '{{ json . }}' > schema.json

I tried applying it to the different database:

schema apply -u "sqlserver://sa:Pass123@localhost:1433?database=test2" --to file://schema.json --dev-url "sqlserver://sa:Sifra123%23@localhost:1433?database=test2"

resulting with:

Error: unknown schema file: "schema.json"

HCL schema works as expected.

I use the latest version (0.15.0) with SQL Server beta driver.

Is this a bug, or working with JSON is a one-way street? If so, are there any plans for adding full JSON support?

[a8m]

Hey @imiljkovic 👋

JSON is just one of the options for formatting the inspected schema, just like the{{ mermaid . }} option. There are no plans to support reading JSON schemas natively in Atlas as some basic features cannot be implemented with JSON.

However, you can write an external schema loader (data source) for it if you're interested. See this doc as an example: https://atlasgo.io/atlas-schema/external

Feel free to join our Discord server and message me if you need help with it.

[folliehiyuki]

Hi @a8m ! Can you give more details on which basic features of Atlas DDL that can't be expressed in JSON?

In my opinion, being able to have JSON schema input has these benefits:

• Easier to parse for various machine tasks (.e.g data validation). Yes, these tasks can use the inspected JSON schema, but I actually want to "write" the schema input in JSON (see the below point).
• I can use another language to generate final schema input (.e.g. Jsonnet, CUE, Starlark, or even a real programming language). This gives flexibility on how I can structure the configuration.

For context, Terraform supports *.tf.json as input. This is used for the 2nd use case I list above, for example, in their terraform-cdk project.