Add Mockito-based unit tests for the four ServiceImpl classes that the
TokenServiceImpl pipeline transitively trusts. Other ServiceImpls in
dc3-common-auth follow the same MyBatis-Plus CRUD shape and will be
covered as Testcontainers PG slice tests in a later stage where the
real schema can validate the lambda query wrappers end-to-end.
- TenantServiceImplTest (16): save / remove / update / selectById /
selectByCode / selectByPage. Covers DuplicateException on save,
NotFoundException on remove and update, AddException / DeleteException
/ UpdateException when the manager returns false, the same-row
duplicate check during update, and Pages defaulting / preservation.
- UserLoginServiceImplTest (12): the standard CRUD trio plus the four
selectByLoginName branches (blank + throw / blank + silent /
not-found + throw / happy path) and the three checkLoginNameValid
outcomes (unknown / disabled / enabled).
- UserPasswordServiceImplTest (9): the CRUD trio plus the password
hashing assertion via DecodeUtil.md5 (save and update both rehash
before persistence). Reset-password is asserted to set
AlgorithmConstant.DEFAULT_PASSWORD before delegating to update.
- TenantBindServiceImplTest (8): CRUD trio plus the
selectByTenantIdAndUserId branches and the null-id short circuit on
listUserIdsByTenantId. The happy path of listUserIdsByTenantId
requires mybatis-plus lambda metadata not available without a real
ORM context and is deferred to the slice test stage.
Verified locally: 45 new cases green; dc3-common-auth now ships 69
unit tests across biz + service tiers.
Add the first batch of dc3-common-auth coverage targeting the security
sensitive biz/impl tier:
- TokenServiceImplTest covers the full credential pipeline:
generateSalt rejects unknown tenants; generateToken rejects unknown
tenant, unknown login, unbound user, missing password row, blank salt
and mismatched md5; the happy path produces a JWT that parses with
the same salt + tenant. checkValid covers the unknown-tenant throw,
blank-token short-circuit, unknown-login + unbound-user invalid
branches, the happy path that returns valid + expireTime, and the
catch-all that swallows parser exceptions and returns invalid. 15
cases in total — every UnAuthorizedException site has at least one
dedicated test.
- DictionaryForAuthServiceImplTest covers the empty-tenant case and
the label/value mapping path used by the admin UI.
- ResourceRegistrySyncServiceImplTest is intentionally narrow: it
fails fast on null command, blank serviceName, blank or unsupported
method, blank path; verifies removeMenuResource and syncMenuResource
are null-safe and never touch the resource manager. Behavioural
coverage of the full reconcile flow (insert / update / delete /
grouping nodes) requires a real PostgreSQL schema and lands as a
Testcontainers slice test in a later stage.
Verified locally: 24 cases green via mvn -pl dc3-common-auth test.
Add a single TestFactory-driven contract test under dc3-common-constant
that exercises every enum following the standard
"Byte index + getIndex() + ofIndex(Byte)" convention. For each constant
in each enum the contract asserts:
- getIndex() is non-null and unique within the enum;
- ofIndex(constant.getIndex()) round-trips back to the same constant;
- ofIndex of an unused byte value returns null (the contract used by
the metadata cache to detect unknown rows).
31 enums are exercised in one place; adding a new enum that follows
the convention only requires adding its class literal to the
INDEXED_ENUMS list. TimeRangeKeyEnum is intentionally excluded because
it keys off a string code; its contract is exercised in
TimeRangeUtilTest.
MapStruct builder behaviour tests are intentionally deferred to the
business stages (S3 manager, S4 data, S5 driver, etc.) where the
builder lives next to its BO/VO/DO trio, so the tests can use the
real conversion fixtures without dragging cross-module dependencies
into dc3-common-constant.
Verified locally: 275 dynamic test cases across the 31 enums,
all green via mvn -pl dc3-common-constant test.
Add unit tests for dc3-common-driver/CodecUtil, the byte-level helper
shared by every protocol driver:
- Base64 encode/decode round-trip across ASCII, UTF-8 and binary
payloads.
- BCD <-> decimal-string round-trip including the odd-length leading
zero trim path.
- bytesToHex emits two characters per byte and preserves leading zero.
- bytesToInt / bytesToIntLE compare big-endian and little-endian
decoding plus the short-array zero-padding fallback.
- bytesToAscii uses ISO-8859-1 so non-printing bytes survive
round-trip.
- byteReverse, mergerBytes, xorBytes and sumBytes assert the array
manipulation invariants used by Modbus / S7 framing.
Note: dc3-common-exception coverage (CustomExceptionContractTest,
ExceptionMessageFormatterTest, ExceptionUtilTest) was committed in
parallel by Copilot in 85b9bd5f0; my locally written copies were
byte-identical so this commit only ships the driver codec piece
that is genuinely new.
Verified locally: 18 cases green via mvn -pl dc3-common-driver test.
Round out the dc3-common-public coverage with the small wrapper types
that the rest of the platform relies on for null-safe field updates and
unified API envelopes:
- StringOptional / IntegerOptional / LongOptional / JsonOptional /
CollectionOptional / EnableOptional tests assert that ifPresent and
ifPresentOrElse fire (or do not fire) for present, null, empty,
non-positive, invalid-JSON and unknown-enum-index inputs respectively.
These guards are what stops MapStruct builders from clobbering
existing rows with null/zero defaults during partial updates.
- R<T> tests cover all eight static factories (ok / fail with no args,
message, ResponseEnum, enum + message, data, data + message) and the
resulting envelope state. The data-overload cases use Integer payloads
to disambiguate from the String-message overloads -- a subtle Java
overload resolution corner that future contributors should not have
to rediscover.
- Pages / TreeNode / Keys / RequestHeader cover their constructors,
setters and the utility-class guards on the enclosing types.
Verified locally with `make test`: 52 new test cases green; total in
dc3-common-public is now 263 from this work-in-progress (an unrelated
contract test in the user's tree currently throws a regex error and is
not part of this commit).
Continue the dc3-common-public coverage with the remaining stateless
helpers used across the platform:
- TimeUtilTest / LocalDateTimeUtilTest / ZoneDateTimeUtilTest cover
format <-> parse round-trips, expireTime arithmetic and the null
fallback path used when callers feed unparseable input.
- TimeRangeUtilTest covers the rolling-window resolver: TODAY / 24h /
7d / 30d enum branches, the legacy hours/days fallback and the
null-when-unbounded escape hatch, plus the wire-format code overload.
- RegexUtilTest exercises every matcher (numeric, name, phone, email,
password, IPv4 host, driver port range) with parameterised positive
and negative inputs.
- PageUtilTest validates Pages -> MyBatis-Plus Page coercion: nulls,
current=0, size cap at MAX_PAGE_SIZE, default create_time order
injection, idempotent create_time, copyPage records mapping plus the
null-source short-circuit.
- TreeUtilTest covers buildByLoop / buildByRecursive / findChildren on
empty input, single-root chains, multi-root and no-matching-root.
- MapStructUtilTest, FieldUtilTest, EnvironmentUtilTest, CodeUtilTest
and HostUtilTest cover the smaller helpers used by builders, mapper
lambdas and node-id generation.
Verified locally: 134 new test cases (212 total in dc3-common-public),
all green via `make test`.
Add the first batch of unit tests under dc3-common-public covering the
core "stateless function" surface used everywhere else in the platform:
- ArithmeticUtilTest: precise add/subtract/multiply/divide/round across
BigDecimal/float/double overloads, plus scale-bound rejections and
division by zero.
- DecodeUtilTest: UTF-8 round-trips, MD5 stability and salting,
Base64 byte/string round-trips and hex en/decoding.
- HmacAuthSignerTest: enabled/disabled state derivation from the secret,
hex digest stability, mismatched-payload + mismatched-secret behavior,
null-payload guards and constant-time verification on length mismatch.
- JsonUtilTest: parseObject/parseArray from String/byte[]/InputStream/
TypeReference, isJson validation, toJsonString/Bytes round-trip and
unknown-property tolerance.
- KeyUtilTest: AES and RSA encrypt/decrypt round-trips, JWT signing
followed by salt/subject/tenant rejection paths.
The Makefile test/test-it/test-e2e/coverage targets get an explicit
-Dmaven.test.skip=false override because .mvn/maven.config sets the
flag globally so `mvn package` does not run tests; without the override
the new targets would silently produce empty reports.
dc3-common-public/pom.xml declares spring-boot-starter-test test-scope.
Verified locally: 77 new test cases run via `make test`, all green.
Maven dependencyManagement does not flow across the iot-dc3 root reactor
into dc3-common because dc3-common inherits from the externally
published dc3-parent. As a result, dc3-common-test could not resolve
JUnit 5 / Testcontainers / Instancio dependencies and dc3-e2e could not
resolve dc3-common-test itself.
Mirror the relevant slice of test-toolchain version management into
dc3-common/pom.xml (junit-bom + testcontainers-bom imports plus pinned
versions for AssertJ, Awaitility, Instancio, REST-assured and WireMock)
and add the missing dc3-common-test version entry to the iot-dc3 root
dependencyManagement so the e2e module can import it.
Verified locally with mvn -B clean compile -pl dc3-common/dc3-common-test -am,
which now resolves all 9 source files and produces
dc3-common-test-2026.5.5.jar.
Introduce dc3/doc/TESTING.md as the canonical reference for the iot-dc3
test pyramid, naming conventions, time/randomness isolation rules,
Testcontainers conventions pinned to the production timescaledb-ha:pg18
image, fixture data strategy, contract base classes shipped by
dc3-common-test, the make commands developers run locally and the
ci.yml/test.yml/e2e.yml workflow contract.
Update AGENTS.md so:
- The Validation Checklist references make test, make test-it,
make coverage and proto contract tests; coverage regressions over 1%
block PRs.
- The Documentation Rules section now lists TESTING.md alongside the
other domain-specific docs to keep the guide current as the harness
evolves.
This closes the S0 stage of the test build-out: every subsequent stage
(S1..S8) will deliver business test cases under the layout defined here.
Wire the test toolchain into developer-facing entry points and CI:
- Makefile gains test, test-it, test-e2e and coverage targets; the help
banner advertises them alongside the existing build commands so
contributors do not need to memorise the underlying mvn invocations.
- A new Test workflow runs three sequenced jobs: unit (mvn test, skipping
the integration phase), integration (mvn verify, skipping unit phase,
running with TESTCONTAINERS_REUSE_ENABLE=true), and a final coverage
job that downloads both jacoco .exec artifacts and rebuilds the
aggregate report from dc3-coverage.
- A new E2E workflow runs make test-e2e on develop/release/main pushes
and on manual dispatch, with a 90 minute timeout that accommodates
cold compose start-up plus the full critical path battery.
- The existing CI workflow keeps its fast-fail compile job unchanged so
PR feedback latency stays low; tests are now owned exclusively by
the new Test workflow.
Introduce two reactor modules that exist purely to host repository-wide
test concerns:
- dc3-coverage (packaging=pom) runs jacoco:report-aggregate against all
contributing modules and applies a tree-wide jacoco:check rule. The
aggregator excludes generated sources, deployable Application classes,
POJO entity packages and the third-party modbus4j sero utilities so
the percentage stays anchored on real business logic. Coverage gates
start at 0% and tighten in later stages as more tests land.
- dc3-e2e (packaging=jar) reserves a place for the docker-compose driven
end-to-end suite delivered in stage S8. Unit and integration phases
are skipped by default and only the e2e profile turns them on.
Both modules are wired into the root reactor.
Centralise test toolchain wiring at the iot-dc3 root pom so child modules
do not need to repeat versions or plugin definitions:
- Import the JUnit 5 and Testcontainers BOMs and pin AssertJ, Awaitility,
Instancio, REST-assured and WireMock versions.
- Add Surefire 3.5.2 and Failsafe 3.5.2 plugin management with the
expected include patterns (*Test.java for unit, *IT.java for
integration) plus argLine wiring for the Jacoco agents.
- Add Jacoco 0.8.12 plugin management with both unit and integration
agent + report executions; integration coverage lands in jacoco-it.exec
while unit coverage stays in jacoco.exec.
- Surface skip.unit.tests and skip.integration.tests properties so CI and
Makefile targets can flip phases without rewriting plugin configuration.
The new plugins are enabled at this layer; subsequent commits add the
aggregator coverage module that reads both .exec files and the workflows
that exercise them.
Bootstrap dc3-common-test, a new infrastructure-only Maven module under
dc3-common that ships:
- Reusable Testcontainers wrappers for the production-aligned
timescale/timescaledb-ha:pg18 image, RabbitMQ with publisher confirms
and Eclipse Mosquitto MQTT broker.
- JUnit 5 support beans: a fixed Clock configuration, an in-process gRPC
extension, an Awaitility-backed Rabbit harness and JSON assertion
helpers.
- Reusable contract base classes for domain enum index round-trips and
for asserting that secret-bearing fields never leak through Lombok
toString() output.
- A test profile (application-test.yml) and logback config aligned with
the production RabbitMQ confirm/return semantics.
Module-specific fixtures that depend on business BO/VO/DO types are
deliberately deferred to their owning modules' src/test trees to avoid
introducing reverse dependencies into dc3-common-test.
This commit only adds the new module; no other module consumes it yet.
The test BOM and plugin wiring that exposes it as a test-scope artifact
to the rest of the tree lands in a follow-up commit.
Drop the stale `test-results/.last-run.json` Playwright artifact at the
backend repo root and extend `.gitignore` to cover surefire, failsafe and
jacoco report directories plus testcontainers state, in preparation for
the upcoming test infrastructure bootstrap.
- Change ModelConfigRequest/ModelProviderRequest defaultFlag and
enableFlag from Byte to DefaultFlagEnum/EnableFlagEnum, aligning
with Jackson's default enum name serialization.
- Remove redundant @AfterMapping enum conversion in builders since
Jackson now handles enum deserialization natively.
- Add @Primary ChatClient.Builder bean to resolve Spring AI
ChatModel ambiguity when multiple providers are configured.
- Replace hardcoded DEFAULT_MODEL ("dc3-agentic") with DB-driven
resolution via ChatClientFactory.resolveModel(), falling back to
the defaultFlag=DEFAULT model config when no model is specified.
Add @Mapping ignores for tenantId/userId on VO→BO builders and
groupId/labelId on query builders — these fields are resolved by
the service layer, not by MapStruct.
Expose AGENTIC_MEMORY_ENABLED, AGENTIC_TOOL_CALLING_ENABLED,
AGENTIC_MEMORY_MAX_MESSAGES, and AGENTIC_ATTACHMENT_STORAGE_PATH
in development and production compose files, with defaults in dev.env.
Also mark OpenAI base-url config as deprecated in application-agentic.yml.
Replace inline `provider` and `baseUrl` string fields with a `providerId`
reference to ModelProvider. Add provider existence validation on save/update.
Thread user identity (creator/operator) into ModelConfig save and update
operations via UserHeader.
Also switch ModelController add/update endpoints from synchronous to
reactive async with UserHeader resolution.
Each model provider (OpenAI, Anthropic, etc.) needs its own ChatClient
instance with provider-specific credentials. Extract ChatClient creation
into ChatClientFactory and resolve the client dynamically at request time.
Also reduce ChatClientConfig scope to only ChatMemory/JDBC setup.
Introduce ModelProvider as a standalone entity so model configurations
reference a provider by foreign key instead of inline provider/baseUrl
strings. Includes entity, mapper, manager, service, and controller.
Spring Boot 4.0.6 still loads EnvironmentPostProcessor implementations
via SpringFactoriesLoader from META-INF/spring.factories, not from the
new .imports file. The .imports mechanism only applies to
AutoConfiguration classes.
The previous commit incorrectly deleted all spring.factories files,
which broke profile activation for every service (auth, data, manager,
gateway, etc.). Restore them with only the EnvironmentPostProcessor
entries (EnableAutoConfiguration entries remain removed since those
use AutoConfiguration.imports).
Rename the generic DriverCustomServiceImpl in each driver module to a
protocol-specific name (e.g., ModbusTcpDriverCustomServiceImpl) so
that static analysis tools can scan every driver module without
filtering duplicate fully-qualified class names. Update
DRIVER-AUTHORING.md to reflect the new convention.
Vickey