valkyrjaio / valkyrja-java / 30328670610

100%

push

github

[Log] Resolve the log provider publishers through the container to fix a flaky test (#59)

# Description

Fixes a flaky test that fails roughly half of all runs, on `26.x` and on
every
open PR branch.


`LogServiceProviderTest::everyPublisherBindsItsServiceThroughTheContainer`
drove
the publishers by iterating the provider's own map:

```java
provider.publishers().forEach((id, publisher) -> publisher.accept(container));
```

`publishers()` returns a `Map.of(...)`, and **the JVM deliberately
randomizes
`Map.of` iteration order on every run** (an immutability-contract
safeguard, so
callers cannot depend on an order that was never specified).
`publishLogger`
reads the `FileLogger` singleton that `publishFileLogger` binds:

```java
container.setSingleton(LoggerContract.class, container.getSingleton(FileLogger.class));
```

So whenever the randomized order happened to reach `LoggerContract`
before
`FileLogger`, the publisher ran against a container that had no
`FileLogger` yet
and threw `ContainerInvalidReferenceException: Service with
io.valkyrja.log.logger.FileLogger not found`.

**This is a test defect, not a production one.**
`Container.getSingleton` calls
`publishUnpublishedDeferred(id)` before resolving, so once the provider
is
registered, requesting `LoggerContract` publishes `FileLogger` on demand
—
whatever order the publishers are stored in. The test was the only
caller
invoking publishers directly on a bare container, bypassing the very
mechanism
that makes ordering irrelevant. No `src/` change is warranted, and none
is made
here: hardening `publishLogger` against an unbound `FileLogger` would be
a branch
the container guarantees can never be taken.

The test now registers the provider and resolves through the container —
the path
production takes — and a second test pins that the deferred logger is
the *same
instance* as the file logger, which is the actual contract
`publishLogger`
implements.

Verified 6 consecutive `--rerun-tasks` runs green,... (continued)

1904 of 1906 branches covered (99.9%)

Branch coverage included in aggregate %.

6958 of 6958 relevant lines covered (100.0%)

4.34 hits per line