[bryantaustin13]

uses codes from HL7/cql#103 (comment)

Answers this issue
#81 -- Timezoneoffset policy tests

Use this version of cql-tests-runner:
cqframework/cql-tests-runner#77 (comment)

[raleigh-g-thompson]

1. Incorrect outputs on the three "NullWhen..." tests in the "No default offset" group — shouldn't these be null rather than true?
2. Missing version attribute — why is there no version attribute on the Policy: Default to server offset (normalize offset-less to server) group?
3. {{SERVER_OFFSET_ISO}} placeholder — could you add a comment explaining the runtime substitution mechanism to help future maintainers?
4. New capability codes — are timezone-offset-policy.no-default-offset and timezone-offset-policy.default-server-offset documented somewhere, or is that planned as a follow-up?

[bryantaustin13]

@brynrhodes @raleigh-g-thompson I've reworked this PR around Bryn's comment. When a DateTime has no offset, it takes the evaluation request's offset (Author's Guide, Date, DateTime, and Time). Extracting the offset from such a value returns that offset, "not null" (Extracting Date and Time Components). So a "no default offset" policy isn't something an engine can choose, and the policy framing is gone:

• I removed both policy groups and the timezone-offset-policy.* capability codes. In their place, the "Default timezone offset" group and its tests are tagged default-timezone-offset-to-request, the single capability Bryn suggested on Add the default-timezone-offset-to-request language capability HL7/cql#103. I've reworked that PR to add just that code to the capabilities CodeSystem (Timezoneoffset policy tests #81).
• I removed the {{SERVER_OFFSET_ISO}} placeholder, so the runner doesn't need to substitute anything.
• The explicit-offset tests are now one group instead of two copies.
• A new "Default timezone offset" group compares offset-less values against Now(). Now() returns the evaluation request timestamp, so it carries the same offset, and the expected results hold whatever offset the request uses:
  • (timezoneoffset from @2012-04-01T00:00) = (timezoneoffset from Now()) → true
  • (timezoneoffset from DateTime(2012, 4, 1, 0, 0)) = (timezoneoffset from Now()) → true
  • @2012-04-01T00:00:00.000 = DateTime(2012, 4, 1, 0, 0, 0, 0, timezoneoffset from Now()) → true, plus < in both directions
• I reverted the DateTimeDifferenceWeeks3 change. Weeks count Sunday boundaries (difference). 2012-03-10 and 2012-03-24 are both Saturdays, so the operator crosses Mar 11 and Mar 18, and the existing 2 is correct.

Raleigh, this also covers your points. The NullWhen… tests, the server-offset group, the placeholder and the policy capability codes are all gone.

Checked on the cqframework Java engine (CQL engine 5.5.0 via clinical-reasoning 4.13.0 on HAPI FHIR 8.12.0): all 23 tests pass at an evaluation offset of -06:00, and again after adding the capability tags.

Checked on cql-execution 4.0.0-beta.1 (test-server): all 23 tests pass at an evaluation offset of +00:00. With the test server's evaluation offset set to +05:30, three of the default-offset tests fail. The engine gives offset-less values +05:00. src/elm/datetime.ts passes the context offset as an Integer literal, so the half hour is dropped. I filed that as cqframework/cql-execution#395, and it's fixed by cqframework/cql-execution#396, which has since merged.