Fix build-ordering race: track javaagent config as run-task input

[ryandens]

Problem

JavaagentApplicationRunPlugin wires the run task's -javaagent arguments through a CommandLineArgumentProvider (in JavaForkOptionsConfigurer) that resolves the javaagent configuration's files at execution time. That provider is not a tracked task input, so Gradle established no dependency on the tasks that build the agent jars.

Under parallel execution the run task could launch with -javaagent:<jar> before that jar had been produced:

Error opening zip file or JAR manifest missing : .../simple-agent.jar

Reproducible 100% locally and intermittently on CI:

./gradlew :plugin:functionalTest --tests "*can attach to application run task*" --rerun-tasks

Fix

Register the javaagent configuration as a tracked input of the run task:

it.inputs.files(javaagentConfiguration)

A Configuration is Buildable, so input-tracking makes the run task depend on the tasks that build the agent jars — the same way JavaagentApplicationDistributionPlugin already wires the start-script task.

Downstream fix in the otel modification plugin

Input-tracking requires every artifact added to the javaagent configuration to carry its producing-task dependency (builtBy). JavaagentOTelModificationPlugin added the extendedAgent output via extendedAgent.outputs.files.singleFile, which resolves to a plain File and drops that provenance — so tracking the configuration as a run-task input tripped Gradle's strict implicit-dependency validation:

A problem was found with the configuration of task ':app:run' (JavaExec).
  Task ':app:run' uses this output of task ':app:extendedAgent' without
  declaring an explicit or implicit dependency.

Fixed at the source by adding it via the task provider, which carries the producing task as a builtBy dependency:

project.dependencies.add("javaagent", project.files(extendedAgent))

Consumers that track the configuration now establish the :extendedAgent dependency automatically.

Tests

• Added a deterministic regression test that asserts, via --dry-run, that :simple-agent:jar is scheduled before :hello-world:run. It fails without this fix and passes with it — independent of task-scheduling timing.
• The existing can attach to application run task test also reproduces the race (verified: 3/3 failures without the fix, 3/3 passes with it under --rerun-tasks).

Verification

• ./gradlew :plugin:functionalTest --tests "*can attach to application run task*" --rerun-tasks — green, run repeatedly
• ./gradlew :otel:functionalTest --rerun-tasks — green (implicit-dependency validation no longer tripped; reproduced the failure first, then confirmed the fix)
• ./gradlew build — green (including spotless formatting)

🤖 Generated with Claude Code

Note

Fix build-ordering race by tracking javaagent config as a run task input

• Registers the javaagent configuration as an explicit input on the run task in JavaagentApplicationRunPlugin, so Gradle orders agent-building tasks before the run task.
• Fixes JavaagentOTelModificationPlugin to add the extendedAgent artifact via project.files(extendedAgent) instead of resolving it to a plain File, preserving TaskProvider provenance and its builtBy dependency.
• Adds a functional test that uses --dry-run to assert :simple-agent:jar is scheduled before :hello-world:run.

^{Macroscope summarized 4dafe32.}

[ryandens]

Addressed in 4b96150. Switched the run task to inputs.files(javaagentConfiguration) (matching the distribution plugin's start-script wiring) and fixed the downstream otel error at its source: JavaagentOTelModificationPlugin now adds the extendedAgent output via project.files(extendedAgent) instead of extendedAgent.outputs.files.singleFile, so the producing task's builtBy is carried into the javaagent configuration. Reproduced the :app:run implicit-dependency validation failure first, then confirmed :otel:functionalTest --rerun-tasks is green with the fix.