[PIsberg]

Closes #1215 (first bullet; the second is out of scope, see below).

Why

analysisSurvivesAStateRegisteredButNotYetWritten builds a fresh state through newState and runs analyze() on it. That catches, every run instead of once in many, an analyze() that throws on a state published but not yet written to. That's the #1212 bug class, where ifIssue then drops the detector's whole report.

It could not reach detectors that register only through stateFor's factory overload. Their state is built by a lambda at the record site, so the test cannot make one without running the record path.

What

• Test hook in AbstractInstanceDetector.register. A private static volatile Consumer, set through a package-private setter, runs right after computeIfAbsent publishes a state. When unset, it costs one volatile read, on the registration path only, never on stateFor's get fast path. RunnerAllocationBudgetTest still passes.
• New test analysisSurvivesARecordPathBetweenRegistrationAndItsFirstWrite.
  • It installs the hook to run analyze() at that moment and calls every recording method of each per-instance detector once, each with a fresh first argument so that every call registers a new state.
  • Two detectors need a hand-written driver: ScopeResultEscapeDetector needs a scope opened first, and VirtualThreadPoolingDetector needs a real ThreadPoolExecutor.
  • It reaches all 18 factory-only detectors (floor 18), and 68 of the 75 per-instance detectors in all.
• AbstractInstanceDetectorTest: the hook fires once a first sighting is published, not on a lookup, and not once it's removed.
• docs/architecture/adding-a-detector.md: analyze() must survive a half-written state, and this test checks it.

It found no live bug.

How it was verified

• The test can fail. With ThisEscapeDetector's empty-escapes guard replaced by escapes.iterator().next(), only the new test went red, naming ThisEscapeDetector: inside recordConstructorEscape: java.util.NoSuchElementException. The two existing tests stayed green. The change was reverted afterwards.
• Tests: DetectorAnalysisUnderConcurrentWritesTest 3/3, AbstractInstanceDetectorTest 10/10 and RunnerAllocationBudgetTest 2/2 (-P e2e) all pass.
• Static gates: install with PMD, checkstyle and errorprone passes.

Not done

• The ~73 detectors that don't extend AbstractInstanceDetector (the issue's second bullet). Each keeps its state its own way, so there is no common seam to hook.
• Generic arguments. The drivers pass generic arguments, so a record path that needs specific earlier state is reached only when it's the first one called.

Checklist

• A behaviour change ships with a test that fails without the fix (test-only change; proven able to fail, above)
• Docs the change makes wrong are updated in this same change
• Nothing was hand-edited between VIBETAGS-START / VIBETAGS-END markers
• No public API change (the hook is package-private)
• Commits an agent authored carry the Co-Authored-By trailer
• No new dependency
• CI is green

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!