Document the fact that DH JSON is a bare list and not compatible with LinkML tools as is about dataharmonizer

[pkalita-lbl]

Let me see if I understand Mark's concern correctly. If I have a schema that implement's the typical LinkML container object pattern:

id: http://example.org/test
name: test
imports:
  - linkml:types
prefixes:
  linkml: https://w3id.org/linkml/

slots:
  s1:
    range: string
  s2:
    range: string
  entries:
    range: Entry
    multivalued: true

classes:
  Entry:
    slots:
      - s1
      - s2
  EntrySet:
    tree_root: true
    slots:
      - entries

I could point DataHarmonizer to the Entry class and it would show me an interface with two columns (for s1 and s2). I could enter some data and then export that data to JSON through the interface. It would look something like:

[
  {
    "s1": "row 1 col 1",
    "s2": "row 1 col 2",
  },
  {
    "s1": "row 2 col 1",
    "s2": "row 2 col 2",
  }
]

The issue is that I can't validate that file as-is using linkml-validate or using a generic JSON Schema validator and the JSON Schema derived from the LinkML schema. That's because LinkML doesn't really have a concept of an array at the root level -- hence the container object pattern.

So what Mark is saying is that if DataHarmonizer could somehow produce JSON that instead looks like:

{
  "entries": [
    {
      "s1": "row 1 col 1",
      "s2": "row 1 col 2",
    },
    {
      "s1": "row 2 col 1",
      "s2": "row 2 col 2",
    }
  ]
}

Now we have an object at the root level. That object corresponds to the EntrySet class in the schema and could be validated as such.

I don't have an exact proposal for how to resolve the situation, but it will probably involve a combination of logic to guess at the so-called container class and index slot (presumably via teaching DataHarmonizer to understand the tree_root metaslot), as well as ways to specify them manually (see also: https://linkml.io/linkml/data/csvs.html).

[ddooley]

Looking back on this, I think DH should input/output LinkML (JSON-LD) native JSON directly via browser, so need to understand the javascript required to do so. The existing "file -> Save as > .json" could be renamed to "file -> Save as > flat .json", and we could add a "file -> Save as > LinkML .json" option for the pure version. This avoids us having to use command line python tools as intermediary step.

@pkalita-lbl for comment.

(The LinkML data inlining options will come into play here later when we add 1-many data relations.)