[bryantaustin13]

Fixes #151.

Problem

A library-style test marked invalid, with no <output>, was always skipped with "Library-style test output has no 'name' attribute naming the define to evaluate". So a negative library test could never pass or fail. cqframework/cql-tests#142's AmbiguousOverload is one: an overload call that should fail to translate.

Change

src/shared/results-shared.ts:

• Result constructor: a library test with no <output> no longer skips in the library branch. It falls through to the shared check after the branch, which already allows a missing <output> when invalid is true or semantic, and skips with "No output specified" otherwise. That's the same rule expression tests follow.
• generateParametersResource: when no define is named, the expression parameter is left out, so Library/$evaluate evaluates the whole library.

Scoring is unchanged. run-test-core.ts already scores invalid true/semantic as pass only when the response indicates an error.

Tests

• test/library-style-tests.test.ts:
  • a negative test with no output runs, with expression set to '';
  • a test with no output that doesn't expect an error still skips;
  • the parameters for a negative test contain only url.
• test/run-tests.test.ts: a negative library test passes when Library/$evaluate returns an error, and fails when the library evaluates.

npx vitest run test/: 240 passed (235 on main). tsc --noEmit reports the same two errors in src/server/rest-routes.ts as on main, which #139 addresses. I left test/run-tests.test.ts in its existing two-space style, which isn't Prettier-clean on main either.

Live run

Using cqframework/cql-tests#142's CqlOverloadMatching.xml against a local HAPI FHIR 8.12.0 server, with conf/localhost-evaluate.json:

Test	main	this branch
AmbiguousOverload	skip ("Library-style test output has no 'name' attribute…")	pass: "Library AmbiguousOverload-1.0.0 loaded, but had errors: Call to operator A(System.Integer) is ambiguous with: …"
the other five overload tests	pass	pass

The server accepted the PUT of the untranslatable library and reported the translation error at $evaluate, which settles #151's open question for HAPI. The published library was removed afterwards. Other servers may behave differently.

cql-execution couldn't be checked. Its test server only provides POST /fhir/$cql for a single expression, with no Library/$evaluate and no Library upload. The runner skips every library-style test on a $cql server before sending a request ("Library-style tests require a server configured for the Library/$evaluate operation"), on main and on this branch alike.

Related

• Handle every error kind the test schema's invalid attribute defines #131 extends the recognised error kinds to syntax and execution. This change uses the existing true/semantic check, so it picks those up once Handle every error kind the test schema's invalid attribute defines #131 merges.
• Send an evaluation request timestamp, defaulting to UTC, with every test #142 also edits results-shared.ts and run-test-core.ts. This branch merges cleanly with both Handle every error kind the test schema's invalid attribute defines #131 and Send an evaluation request timestamp, defaulting to UTC, with every test #142.

🤖 Generated with Claude Code