Join function expectation conflicts with FHIRPath tests

[johngrimes]

We have two tests that have an expectation that an empty collection input to the join function results in an empty string:

https://github.com/FHIR/sql-on-fhir-v2/blob/master/tests/fhirpath.json#L335

    {
      "title": "string join",
      "tags": ["shareable"],
      "view": {
        "resource": "Patient",
        "status": "active",
        "select": [
          {
            "column": [
              {
                "name": "id",
                "path": "id",
                "type": "id"
              },
              {
                "name": "given",
                "path": "name.given.join(', ' )",
                "type": "string"
              }
            ]
          }
        ]
      },
      "expect": [
        {
          "id": "pt1",
          "given": "g1.1.1, g1.1.2, g1.2.1"
        },
        {
          "id": "pt2",
          "given": ""
        },
        {
          "id": "pt3",
          "given": ""
        }
      ]
    },
    {
      "title": "string join: default separator",
      "tags": ["shareable"],
      "view": {
        "resource": "Patient",
        "status": "active",
        "select": [
          {
            "column": [
              {
                "name": "id",
                "path": "id",
                "type": "id"
              },
              {
                "name": "given",
                "path": "name.given.join()",
                "type": "string"
              }
            ]
          }
        ]
      },
      "expect": [
        {
          "id": "pt1",
          "given": "g1.1.1g1.1.2g1.2.1"
        },
        {
          "id": "pt2",
          "given": ""
        },
        {
          "id": "pt3",
          "given": ""
        }
      ]
    }

This conflicts with the FHIRPath.js test suite, which expects that the join function will return an empty collection if given one as input:

https://github.com/HL7/fhirpath.js/blob/master/test/cases/5.6_string_manipulation.yaml#L370

  - desc: '** join on empty input collection'
    expression: Functions.str.empty.join(',')
    result: []

[johngrimes]

Should we clarify what the representation of an empty collection is in view results?

[johngrimes]

Let's start up a Zulip thread for this so that we can get feedback on the impact of potentially changing the expectation to be a null (empty collection).

We also need to rationalise the join tests within the SQL on FHIR test suite so that they are in one place (experimental, given the function is not yet normative).

[arjun]

NS:

• Wait until BrianP is done with refactoring FP tests
• is the join fn part of the normative release of FP? for a stable target
• part of ballot MS

[brianpos]

p.s. the spec for join clealy states empty input, empty output.
(not empty string)
It's part of the CI build and was in the ballot version we put out early last year, and we're not far from publishing, last couple of tickets almost done.
And yes, after doing a quick test, several of the existing engines aren't doing it right.
p.p.s. I have a large number of new fhirpath unit test cases coming that I'm waiting till the spec is published to ensure all align with the final published version. Including one that specifically tests this.