FHIR Chat · Implementation of Choice Data Types · implementers

Stream: implementers

Topic: Implementation of Choice Data Types

Joe Paquette (May 23 2019 at 13:57):

Example: Patient properties deceased[x] and multipleBirth[x]

My reading of http://build.fhir.org/formats.html#choice [ref: "Elements that have a choice of data type cannot repeat - they must have a maximum cardinality of 1. When constructing an instance of an element with a choice of types, the authoring system must create a single element with a data type chosen from among the list of permitted data types."] implies to me that the implementation of choice data types should result in only one choice property (e.g., single polymorphic deceased property as opposed to the multiple deceasedBoolean and deceasedDataTime properties). Am I reading this correctly? Should the implementation be:

• a single deceased property that can support either a boolean data type or a dateTime data type
• two properties of deceasedBoolean and deceasedDataTime where only one of these fields can be used at a time
• something else that I'm missing...

Looking at the generated fhir.schema.json from the downloads (JSON SCHEMA), the schema definition for "Patient" specifically defines the following (ignoring the "underscored prperties"): deceasedBoolean, deceasedDateTime. If the requirement is for a single polymorphic property, the "Patient" JSON Schema should have a single deceased property having a oneOf for type definitions for boolean and dataTime (string).

If the intent is to implement choice fields as one per data type (e.g., deceasedBoolean, deceasedDateTime), then based on the requirement to allow only one of these, the JSON Schema should have a "oneOf" for deceasedBoolean or deceasedDateTime.

I'd appreciate any insight on the underlying intent and how other folks are handling this in real implementations.

Thanks!

Jean Duteau (May 23 2019 at 13:58):

Example: Patient properties deceased[x] and multipleBirth[x]

• two properties of deceasedBoolean and deceasedDataTime where only one of these fields can be used at a time

This.

Joe Paquette (May 23 2019 at 14:12):

OK, thanks...
So, shouldn't the JSON Schema enforce this by using "oneOf" for deceasedBoolean and deceasedDateTime? Or is the intent that an implementer must add appropriate logic on how to handle the specific uses?

Lloyd McKenzie (May 23 2019 at 14:49):

@Grahame Grieve

nicola (RIO/SS) (May 28 2019 at 18:14):

and again ;) choice types in fhir json are represented in a wrong way - you will have combinatorial problems with more than one union in resource, you could not have collection of union attributes, search problems etc. I will bomb until we fix it!

Grahame Grieve (May 28 2019 at 20:38):

the intent that an implementer must add appropriate logic on how to handle the specific uses

Json schema can't represent all the rules. So we add the ones to schema that people find useful, based on feedback. There's a tradeoff for different uses of the schema . for XML, we publish 2 different schema, for different purposes.

Last updated: Apr 12 2022 at 19:14 UTC