[nineinchnick]

This is the second pass of #355. That pass skipped the test files that open PRs were changing. Those PRs are merged now, so this pass reviews those files: TestOpenApiSpec, TestOpenApiConfig, TestOpenApiClientRequests, TestOpenApiClient, TestOpenApiServers, TestSpecLocation, TestCachingHttpClient, TestAuthenticatingHttpClient, and TestKeycloakAuthentication. It also reviews test files added since then, like TestUserCredentials and TestOpenApiSchemaFromTags. A repeated test is more code to keep in sync, and it covers nothing new. Repeated loads of a big spec make the suite slower for no gain.

Changes

Change	Why
TestOpenApiSpec: the five getDatadog* tests share one parsed spec	Each test parsed specs/datadog.yaml with the same config. The five tests took 4.6 s before and 1.6 s after. The trimmed-prefix test still parses its own spec, because it uses a different config
TestOpenApiConfig.testApiKeysCombinedWith*: four tests become one	They checked one validation and differed only in the property name and value. The setter order test stays separate, because it checks isApiKeysValid() directly
Removed TestOpenApiClientRequests.testLeadingPathParameter	testNotFoundWithPathParametersIsEmpty reads the same table of the same spec, with the same path parameter, and checks the same request URI and empty result

Coverage of src/main (JaCoCo, mvnd clean verify -Dair.check.skip-jacoco=false)

Counter	Before	After
Lines	4981 / 5276	4982 / 5276
Branches	2388 / 2787	2389 / 2787
Instructions	27858 / 29515	27864 / 29515

I compared the covered lines and branches one by one. No line or branch lost coverage. The one gained line is ProbeLimit.toString() with a set limit (line 52). As #355 found, that line runs only as the session property encoder, so its coverage depends on timing.

Tests: 485 before, 481 after.

Not touched

Open PR #356 changes authentication/TestAuthentication.java, so I left it alone.

Possibly redundant, but kept

• TestCachingHttpClient.testRequestCanForceRevalidation: testCacheDirectivesDontSelectStoredResponse sends the same first two requests and makes the same final assertions. But it checks only the state after a third Pragma request, so it can't tell which request revalidated.
• TestCachingHttpClient.testCookieParametersSelectStoredResponse: HttpCache puts every header in the key, so this runs the same code as testHeaderParametersSelectStoredResponse. It still protects against Cookie being left out of the key later.
• TestAuthenticatingHttpClient.testRejectedTokenOfClientSecretBasicIsRefreshed: its assertions are a subset of testRejectedTokenIsRefreshed. But it is the only fast test of a refresh with client_secret_basic. The Keycloak test needs a container.
• TestCachingHttpClient.testVary: every header is already in the cache key, so its assertions would still pass if Vary matching broke. A stronger test would be better than removing it.
• TestOpenApiSpec.testResultsPathToArrayOrObject and testResultsPathToUnion: their first assertions repeat testPaginatedResultsNullability and testUnionOfObjectAndArrayResponses. Their other assertions are unique.
• TestOpenApiSpec.testTablesWithoutColumns: its /names part repeats testPlainValueResponses. The /count part is unique.
• TestOpenApiSpec.getGithubPaginatedTables and getGithubPatchedOperations overlap, but each checks things the other doesn't.
• TestOpenApiConfig.testTrimmedPathPrefixes: its first assertion repeats testExplicitPropertyMappings. Its error message is unique.
• TestOpenApiClientRequests.testInvalidJsonIsExternalError and testReadsResponsesWithLoggedBodies have the same message, but the second one runs with body logging on.