fleetd #537: pin CapturedLog.close()'s appender-detach and setLevel-immunity contracts

Only the level-restore half of close() was pinned before this
(WorktreeSessionManagerTest). Deleting logger.detachAppender(appender)
from close() left mvn clean install green (1701 tests, 0 failures) --
the appender-detach half of the contract was unmeasured.

Adds CapturedLogTest with three tests, each using a logger name no
production class uses:
- closeDetachesTheAppenderSoALaterLogIsNotCaptured: an event logged
  after close() must not land in events().
- closeRestoresTheLevelCapturedAtOpen: the helper's headline contract
  in one place, independent of any production class.
- setLevelDuringCaptureDoesNotChangeWhatCloseRestores: setLevel()'s
  own javadoc claim that close() always restores the level captured
  at construction, never a value set through setLevel() mid-capture.

Test-only change; CapturedLog.java itself is untouched.

.gitea/workflows/ci.yml

@@ -39,10 +39,6 @@ jobs:
         # that runs does so against the fake UDS herdr and fake ccs/claude stubs.
         run: mvn -B clean install
 
-      - name: javadoc reference lint
-        working-directory: fleetd
-        run: mvn -B -DskipTests javadoc:javadoc -Ddoclint=reference
-
       # Deliberately NOT actions/upload-artifact: this Gitea instance presents as GHES, and
       # @actions/artifact v2+ (i.e. upload-artifact@v4) refuses to run there —
       # "GHESNotSupportedError ... not currently supported on GHES", which red-Xes an otherwise

fleetd/src/main/java/dev/ltms/fleet/config/ConfigRef.java

@@ -208,7 +208,7 @@ import java.util.function.Supplier;
  * that is correctly hot and for a key nobody triaged. Three times now — {@code worktreeGroup} (#323),
  * {@code primary}/{@code configReload} (#326), and {@code fleet.leaders} sitting in the escape hatch
  * (#333) — the second kind hid among the first. A top-level coverage checker in the
- * {@code ConfigRefProfileCoverageTest} shape (one level up, over {@code FleetConfig} itself rather
+ * {@link ConfigRefProfileCoverageTest} shape (one level up, over {@code FleetConfig} itself rather
  * than {@code FleetConfig.Profile}) proves this file's four classes exhaust the record's components
  * — see {@code ConfigRefTopLevelCoverageTest}. That test proves the record's <em>shape</em> is fully
  * triaged; it does NOT prove a {@code SPLIT_KEYS}/{@code COLD_KEYS}/{@code DEFERRED_KEYS} member has

fleetd/src/main/java/dev/ltms/fleet/config/FleetConfig.java

@@ -1670,7 +1670,7 @@ public record FleetConfig(
      * <p>A herdr pane runs a login shell that re-sources the operator's own secret store, so a
      * member inherits every credential the operator's shell holds — measured at 31 names on this
      * host, of which only one ({@code GITEA_ACCESS_TOKEN}) used to be blocked, and that block was a
-     * single name hardcoded in {@link dev.ltms.fleet.member.HerdrPeerLauncher} rather than driven by config (gitea issue
+     * single name hardcoded in {@link HerdrPeerLauncher} rather than driven by config (gitea issue
      * #82). This record replaces that hardcoded shadow with a config-driven one.
      *
      * <p><b>deny-by-default, not a deny-list.</b> A deny-list (block these specific names, let

fleetd/src/main/java/dev/ltms/fleet/msg/AmqpReplyInbox.java

@@ -246,7 +246,7 @@ public final class AmqpReplyInbox implements ReplyInbox, AutoCloseable {
      * neither. So before this method existed with a requeue step, it dropped {@link #held}'s entries
      * for {@code target} while the broker still considered them outstanding: never acked, never
      * nacked, never requeued, and no longer reachable by {@link #peek} — permanently invisible. This
-     * is unlike {@link RecoveryListener#handleRecovery(Recoverable)} and {@link #close()}, whose bare {@code held.clear()} is
+     * is unlike {@link #handleRecovery} and {@link #close()}, whose bare {@code held.clear()} is
      * correct because each has already made the broker requeue (a real connection drop, or
      * {@code channel.close()} respectively) before clearing local state.
      *

fleetd/src/main/java/dev/ltms/fleet/msg/MessageService.java

@@ -713,7 +713,7 @@ public final class MessageService {
      * failure.
      *
      * <p><strong>Without {@code sweepAsking} on the release path, a target torn down while
-     * genuinely {@code ASKING} was unrecoverable.</strong> {@link Rendezvous#resolveQuestion(String, String, String)} had already
+     * genuinely {@code ASKING} was unrecoverable.</strong> {@link #resolveQuestion} had already
      * closed the forward waiter the instant the question surfaced (so the {@code waiter} branch
      * below finds nothing to fail), the {@code question == null} guard excluded the task from
      * {@code matching} (so the loop below skipped it too), and the worker's own {@code fleet_ask}

fleetd/src/main/java/dev/ltms/fleet/msg/ReplyPushLoop.java

@@ -44,7 +44,7 @@ import java.util.stream.Collectors;
  * still holds an unacked message ({@link #pendingReplies}), tickets not yet collected
  * ({@link #pendingTickets}), and open questions not yet answered or lapsed
  * ({@link #pendingQuestions}) — and sends at most one combined nudge per tick
- * ({@link #injectNudge(String, int, int, int, int, int)}). Work that arrives while the lead is busy is
+ * ({@link #injectNudge(String, int, int, int)}). Work that arrives while the lead is busy is
  * never lost: it is re-read fresh on every tick until the lead is injectable or its own reminder
  * cap ({@link #maxReminders}) is reached — each source spends from its own budget, so one source
  * exhausting its cap does not stop nudges about the others (post-CB-590 regression fix; see

fleetd/src/test/java/dev/ltms/fleet/testing/CapturedLogTest.java

@@ -0,0 +1,96 @@
+package dev.ltms.fleet.testing;
+
+import ch.qos.logback.classic.Level;
+import ch.qos.logback.classic.Logger;
+import org.junit.jupiter.api.Test;
+import org.slf4j.LoggerFactory;
+
+import static org.junit.jupiter.api.Assertions.assertEquals;
+
+/**
+ * fleetd #537: pins {@link CapturedLog#close}'s own contract — the appender detach half, the level
+ * restore half, and that {@link CapturedLog#setLevel} does not change what {@code close} restores.
+ * Before this, only the level-restore half was pinned (by {@code
+ * WorktreeSessionManagerTest.sharedSessionManagerLoggerLevelIsRestoredAfterDirtyWorktreeReleasePinsWarn}).
+ * Measured: deleting {@code logger.detachAppender(appender);} from {@code close()} still left
+ * {@code mvn clean install} green — 1701 tests, 0 failures — before this file existed.
+ *
+ * <p>Every test below uses a logger name no production class uses, and unique per test, so this
+ * file cannot become the next entry in fleetd #525's leak family: {@link CapturedLog}'s own
+ * javadoc re-measure commands (top of that file) would otherwise need to start naming this class.
+ */
+class CapturedLogTest {
+
+    /**
+     * The appender must be detached on close: an event logged through the raw logger after close
+     * must not land in {@link CapturedLog#events()}. Asserting on the observable list (rather than
+     * {@code logger.iteratorForAppenders()}) is what the ticket asked for, and it is also what a
+     * real leak would actually break — a later test's own {@code ListAppender} silently gaining
+     * events emitted by code under test that has nothing to do with it.
+     */
+    @Test
+    void closeDetachesTheAppenderSoALaterLogIsNotCaptured() {
+        String loggerName = "capturedlog-test-only.appender-detach";
+        Logger rawLogger = (Logger) LoggerFactory.getLogger(loggerName);
+
+        CapturedLog log = CapturedLog.of(loggerName);
+        rawLogger.info("while open");
+        int eventsWhileOpen = log.events().size();
+        assertEquals(1, eventsWhileOpen, "the event logged while open must be captured");
+
+        log.close();
+        rawLogger.info("after close");
+
+        assertEquals(eventsWhileOpen, log.events().size(),
+                "close() must detach the appender: an event logged after close must not be "
+                        + "captured, but the captured list grew from " + eventsWhileOpen + " to "
+                        + log.events().size());
+    }
+
+    /**
+     * The helper's own headline contract, pinned in one place independent of any production
+     * class's behaviour: {@code close()} restores the level the logger had before {@link
+     * CapturedLog#at} pinned it.
+     */
+    @Test
+    void closeRestoresTheLevelCapturedAtOpen() {
+        String loggerName = "capturedlog-test-only.level-restore";
+        Logger rawLogger = (Logger) LoggerFactory.getLogger(loggerName);
+        rawLogger.setLevel(Level.DEBUG);
+
+        CapturedLog log = CapturedLog.at(loggerName, Level.ERROR);
+        assertEquals(Level.ERROR, rawLogger.getLevel(), "the pinned level took effect while open");
+
+        log.close();
+
+        assertEquals(Level.DEBUG, rawLogger.getLevel(),
+                "close() must restore the level captured when at() was called (DEBUG), not leave "
+                        + "the pinned level (ERROR) in place");
+    }
+
+    /**
+     * {@link CapturedLog#setLevel}'s javadoc claims that re-pinning the level mid-capture does not
+     * change what {@code close()} restores — that restore always uses the level captured when the
+     * instance was created, never a value set through {@code setLevel}. Nothing checked this
+     * before: open with a pinned WARN, call {@code setLevel(TRACE)}, close, and the result must be
+     * the level from BEFORE {@code at} — neither WARN nor TRACE.
+     */
+    @Test
+    void setLevelDuringCaptureDoesNotChangeWhatCloseRestores() {
+        String loggerName = "capturedlog-test-only.setlevel-no-effect";
+        Logger rawLogger = (Logger) LoggerFactory.getLogger(loggerName);
+        rawLogger.setLevel(Level.DEBUG);
+
+        CapturedLog log = CapturedLog.at(loggerName, Level.WARN);
+        log.setLevel(Level.TRACE);
+        assertEquals(Level.TRACE, rawLogger.getLevel(), "setLevel took effect immediately");
+
+        log.close();
+
+        assertEquals(Level.DEBUG, rawLogger.getLevel(),
+                "close() must restore the level captured at open (DEBUG) regardless of any "
+                        + "later setLevel() call: it must be neither WARN (the level pinned by "
+                        + "at()) nor TRACE (the level set via setLevel() mid-capture), but got "
+                        + rawLogger.getLevel());
+    }
+}