test: add TDD bug-capture test for #992 — DI container audit cache failure

Add a dedicated Behave bug-capture scenario that reproduces the get_container() audit subscriber initialization caching defect from #992. The steps patch container construction so audit subscriber initialization fails once then succeeds, call get_container() twice, and assert a retry should occur on the second access.

The assertion intentionally fails against current behavior (only one initialization attempt), and the @tdd_expected_fail + @tdd_bug + @tdd_bug_992 tags invert that failure to a CI pass until the production fix lands. This creates a precise regression guard that will only pass without inversion when retry semantics are implemented.

ISSUES CLOSED: #1096

features/steps/tdd_di_audit_cache_failure_steps.py

@@ -0,0 +1,70 @@
+"""Step definitions for tdd_di_audit_cache_failure.feature.
+
+This test captures bug #992: ``get_container()`` caches a container even when
+eager audit subscriber initialization fails once during startup, and it does
+not retry on subsequent ``get_container()`` calls.
+
+The scenario is tagged ``@tdd_expected_fail`` so the expected assertion failure
+(which proves the bug still exists) is inverted to a CI pass until the fix for
+#992 lands. After the fix, remove ``@tdd_expected_fail``.
+"""
+
+from __future__ import annotations
+
+from typing import Any
+
+from behave import given, then, when
+from behave.runner import Context
+
+
+@given("g992- a container whose audit subscriber init fails once then succeeds")
+def step_given_container_with_transient_audit_init_failure(context: Context) -> None:
+    """Patch container factory so audit subscriber fails once, then succeeds."""
+    from cleveragents.application import container as container_module
+
+    class FakeContainer:
+        """Minimal fake container to track audit subscriber initialization calls."""
+
+        def __init__(self) -> None:
+            self.audit_init_attempts = 0
+
+        def audit_event_subscriber(self) -> object:
+            """Fail first call (simulating DB not ready), then succeed."""
+            self.audit_init_attempts += 1
+            if self.audit_init_attempts == 1:
+                raise RuntimeError("database not ready yet")
+            return object()
+
+    context.g992_original_container_class = container_module.Container
+    container_module.Container = FakeContainer
+
+    def _cleanup() -> None:
+        container_module.Container = context.g992_original_container_class
+        container_module.reset_container()
+
+    context.add_cleanup(_cleanup)
+    container_module.reset_container()
+
+
+@when("g992- I call get_container twice")
+def step_when_call_get_container_twice(context: Context) -> None:
+    """Call get_container twice and keep references for assertions."""
+    from cleveragents.application.container import get_container
+
+    context.g992_first_container = get_container()
+    context.g992_second_container = get_container()
+
+
+@then("g992- audit subscriber init should have been attempted twice")
+def step_then_audit_subscriber_should_be_retried(context: Context) -> None:
+    """Expect retry on second access after first initialization failure."""
+    first = context.g992_first_container
+    second = context.g992_second_container
+    assert first is second, "Expected get_container() to return a singleton instance"
+
+    attempts: Any = getattr(first, "audit_init_attempts", None)
+    assert attempts == 2, (
+        "Bug #992: get_container() did not retry audit subscriber initialization "
+        "after the first failure. Expected 2 attempts after two get_container() "
+        f"calls, but observed {attempts}."
+    )

features/tdd_di_audit_cache_failure.feature

@@ -0,0 +1,21 @@
+@tdd_expected_fail @tdd_issue @tdd_issue_992
+Feature: TDD Bug #992 — get_container() caches failed audit subscriber initialization
+  As a long-lived process using the shared DI container
+  I want get_container() to retry audit subscriber initialization after an initial failure
+  So that audit subscriptions are eventually registered once prerequisites become available
+
+  This feature captures bug #992. Today, get_container() eagerly calls
+  audit_event_subscriber() only during first container creation. If that first call
+  fails, the exception is logged and swallowed, but the global _container remains set.
+  Subsequent get_container() calls return the cached container without retrying the
+  subscriber initialization.
+
+  This scenario is intentionally tagged with @tdd_expected_fail until bug #992 is
+  fixed. The underlying assertion fails (proving the bug exists) and the tag inversion
+  mechanism converts that failure to a CI pass. Remove @tdd_expected_fail when #992 is
+  fixed so this runs as a normal regression test.
+
+  Scenario: Bug #992 — get_container retries audit subscriber initialization after first failure
+    Given g992- a container whose audit subscriber init fails once then succeeds
+    When g992- I call get_container twice
+    Then g992- audit subscriber init should have been attempted twice