mongodb: how to describe a json field which can have different types in _schema

[Ognian]

I'm using trino with mongodb. I'm trying to write a _schema entry for a mongodb collection. The entries in this collection are described with json schema:

...
  "delta": {
    "＄schema": "http://json-schema.org/draft-04/schema#",
    "type": "array",
    "items": {
      "type": "object",
      "properties": {
        "op": { "type": "string" },
        "path": { "type": "string" },
        "value": {
          "anyOf": [
            { "type": "string" },
            { "type": "number" },
            { "type": "integer" },
            { "type": "boolean" },
            { "type": "object" },
            { "type": "array" }
          ],
        }
      },
      "additionalProperties": false
    }
  }
...

the entry would look like

          {
            "name": "delta",
            "type": "array(row(\"op\" varchar,\"path\" varchar,\"value\" json))",
            "hidden": false
          }

So the interesting point here is that the field value can have on of the types listed below.
The only way to express this was to use the json type, BUT this is the error I got:

[65536] Query failed (#20210621_212459_01468_5jhjw): Unhandled type for Slice: json io.trino.spi.TrinoException: Unhandled type for Slice: json

The question is how to define a field which can have more than one type?
Funny enough if I change from json to varchar it looks like it works, but it makes no sense to me...

Thanks
Ognian

[ebyhr]

As far as I know, there is no simple solution except for your varchar workaround. Let me label as enhancement.

[Ognian]

@ebyhr it gets even funnier:
when using superset as the frontend and having an address field with subfields street, zip, city, country.
If I define it in _schema of type: row("street" varchar,"zip" varchar,"city" varchar,"country" varchar)
the it is displayed as an array:

BUT If I define it in _schema of type: varchar

which display's it correct
BUT of course this is a superset error since if I try to select address.zip it errors...
apache/superset#15364

[academy-codex]

Looks interesting. @ebyhr If you're not taking it up I can have a look.

[ebyhr]

@academy-codex I implemented half of this enhancement locally, but please feel free to take it. MongoPageSource.writeSlice() is the relevant method. The point is json generated by Document.toJson() is little different from Trino json response.

edit: As we talked offline on Jul 17, l will send a PR.