[jmandel]

A $expand whose SNOMED CT filter names a concept that the edition does not have now answers 400 (error / invalid / tx-issue-type vs-invalid). Before this change it answered 500 (exception / invalid-data). Examples are concept in 41000147108 (a refset that only exists in another edition) and concept is-a 584021000052104. The value set is invalid, so this is a client error, not a server failure.

curl -s -H 'Content-Type: application/fhir+json' 'http://localhost:3000/r4/ValueSet/$expand' -d '{"resourceType":"Parameters","parameter":[{"name":"valueSet","resource":{"resourceType":"ValueSet","status":"active","compose":{"include":[{"system":"http://snomed.info/sct","filter":[{"property":"concept","op":"in","value":"41000147108"}]}]}}}]}'

• Before: HTTP 500, {"severity":"error","code":"exception","details":{"text":"The SNOMED CT Concept 41000147108 is not known","coding":[{"system":".../tx-issue-type","code":"invalid-data"}]},"diagnostics":"..."}
• After: HTTP 400, {"severity":"error","code":"invalid","details":{"text":"The SNOMED CT Concept 41000147108 is not known","coding":[{"system":".../tx-issue-type","code":"vs-invalid"}]}}

This matches what an unknown concept in ECL already gets: the same Issue, error / invalid / vs-invalid, handleAsOO(400) (tx-ecosystem snomed-expand-ecl-unknown-concept). The text stays "The SNOMED CT Concept X is not known". That is the text ECL wraps in INVALID_ECL, and there is no ECL expression to quote here, so the issue carries no message id.

Change. In tx/cs/cs-snomed.js, the concept filter helpers throw this issue instead of a plain Error. That covers =, is-a, descendent-of, child-of, generalizes, descendent-leaf, is-not-a and in, for both the concept and code properties. ECL evaluation also calls these helpers, and its result does not change: it re-wraps the error as INVALID_ECL as before.

Tests. 13 new tests, all of which fail on main:

• tests/cs/cs-snomed.test.js: 9 tests, one per operator, plus in with a known concept followed by an unknown one. Each test checks for a 400 issue with this code, tx-issue-type and text.
• tests/tx/expand-snomed-whole.test.js: 4 tests. Each runs $expand over HTTP for in (an unknown refset), is-a, = and descendent-of, expects 400, and checks that the issue matches the one << X in ECL gives.

Results with this change:

• tests/tx without test-cases: 1356 passed.
• tests/tx/test-cases.test.js: 5182 passed.
• tests/cs: 1483 passed. 4 failed, and the same 4 fail on unchanged main in the same environment (3 LOINC external-mapping tests from a stale local test database, and 1 RXNATOMARCHIVE read-only test).

🤖 Generated with Claude Code

[grahamegrieve]

How I feel about this depends on https://chat.fhir.org/#narrow/channel/437028-Terminology-Service-Test-Cases/topic/Filter.20on.20an.20invalid.20Conceptwhat/with/638487832.

There is not consistency on this form FHIRsmith or in the test cases, and there should be, but there's two possible outcomes: silently drop the invalid code wherever it is found, or make it an error.

I favour silently dropping it, but will see based on the discussion

[grahamegrieve]

I think this is superceded by #396 now