Basically, our current invalid directory of unit tests is insufficiently smart - it will fail if a single schema item is invalid, so if we accidentally add something that is valid while believing it to be invalid, we won't be warned.
As far as I can tell, the best thing we can do to properly validate is to compare the outputs of the --all-errors flag in ajv-cli against a source-controlled list of expected errors which we'll need to manually update as we add new failing tests.
Follow on from the thread at https://github.com/dbt-labs/dbt-jsonschema/pull/117#discussion_r1465257368
Basically, our current
invaliddirectory of unit tests is insufficiently smart - it will fail if a single schema item is invalid, so if we accidentally add something that is valid while believing it to be invalid, we won't be warned.As far as I can tell, the best thing we can do to properly validate is to compare the outputs of the
--all-errorsflag in ajv-cli against a source-controlled list of expected errors which we'll need to manually update as we add new failing tests.Open to alternative approaches