Repository navigation
Unique Spring test contexts still boot a full application #315
Description
What is wrong
The 2026-10-21 H2 run spent about 211s between JUnit suite timestamps. That gap is context startup (maxParallelForks = 1). These classes each force a context the shared @SpringBootTest + @AutoConfigureMockMvc cache cannot reuse. Suite method time for each one is under a second.
| Class | Mechanism | Gap before the suite |
|---|---|---|
ResetPasswordControllerTest |
Profiles.MAIL and @Import(MockMailConfig.class) |
16.43s |
TemplateMoreInfoAdvancedRenderSecurityTest |
@MockitoBean TemplateExecutionService |
15.79s |
UserPositionRepositoryDataRestTest |
@AutoConfigureWebMvc and @AutoConfigureTestDatabase(NONE) while the other Data REST tests use @AutoConfigureMockMvc |
14.73s |
JsonWebTokenServiceTest |
@AdditiveActiveProfiles(Profiles.TEST) and @Import(FixedClockConfig.class). The tests call JsonWebTokenService only |
13.91s |
SpatialReferenceSystemTest |
extends BaseTest and @MockitoBean CodeListValueRepository |
12.01s |
UserChecksServiceTest |
@SpringBootTest plus mocked repositories, then userChecksService.getWarnings |
11.30s |
UserPositionBusinessLogicTest |
nested @Import of the business-logic class |
6.16s |
DefaultLanguageChangeControllerTest |
extends BaseTest and @MockitoBean DefaultLanguageChangeService |
4.93s |
TaskLiteralEnrollmentDataRestTest |
@MockitoSpyBean LiteralTranslationEnsureService |
4.83s |
DatabaseConnectionControllerTest |
extends BaseTest and @MockitoBean DatabaseConnectionRepository |
4.67s |
CodeListTest is the same shape as SpatialReferenceSystemTest and paid 10.17s. Leave it on BaseTest.
Why this is not a test-only edit
A @WebMvcTest or plain unit move changes which beans and filters run. The mock, profile, or spy is the cache key and also the behavior under test. TaskLiteralEnrollmentDataRestTest needs the spy to throw on one method and the real enroll path on the other.
Done when
Each class either shares an existing context or documents why its context must stay unique. A wrong observable result still fails the test.
Activity
fjlopez commentedon Oct 21, 2026
A later H2 run on the same day, 08:04:19Z to 08:06:49Z, measured these gaps again after the empty i18n nested configurations were hoisted. Method time for the ten classes is still under a second. The gaps are smaller than the 211s total in the issue, and these classes still do not share the warm
BaseTestcontext.CodeListTest(5.00s) andSpatialReferenceSystemTest(6.12s) extendBaseTestand add@MockitoBean CodeListValueRepository. The field names differ (codeListValueRepositoryandrepository), so they are two cache keys.HttpURLTest,MinMaxTest, andBoundingBoxTestreuseBaseTestat about 0s.The seed already has the answers those stubs force.
STM_CODELIST.csvlegend types areLINK,LEGENDGRAPHIC, andCAPABILITIES, soWRONG VALUEis absent.service.typeincludesWMS.@Srschecks theEPSG:pattern inSrsSystemListValidator, not the code list. Dropping the mock keeps both classes onBaseTest. That is not a@WebMvcTestmove.CodeListTeststaying onBaseTeststill holds.The other classes in this issue still need the mock, profile, or spy that the assertion reads. Do not fold them into this comment's smaller edit.
fjlopez commentedon Oct 21, 2026
The working tree removes
@DirtiesContextfromUserPositionBusinessLogicTestandUserPositionRepositoryDataRestTest. Both classes still have the cache key named in this issue.CodeListTeststill has@MockitoBean CodeListValueRepository. The pass method that expected no$.errors[?(@.property)]match is gone. The remaining method still stubs that repository, so the 5.00s gap is unchanged by that deletion.fjlopez commentedon Oct 22, 2026
Fixed on dev by #351.