Compare commits

...

5 Commits

Author SHA1 Message Date
Dai Ha cb4a6869b9 fleetd #677: guard exact lead tab collisions
CI / shell-tests (pull_request) Failing after 8s
CI / contract (pull_request) Successful in 46s
CI / build (pull_request) Failing after 1m56s
2026-10-03 22:10:29 +02:00
Dai Ha 209e1231ea Merge PR #682: fleetd #651 — raise turnSettleSeconds default to 300, correct the settle javadoc
CI / shell-tests (push) Failing after 6s
CI / contract (push) Successful in 55s
CI / build (push) Failing after 1m54s
The deferred roll waits for the calling lead's own turn to end. The old 20s bound
was shorter than one ordinary closing message (measured elapsed=20394ms), so a
lead that followed the handover skill's instruction to say its goodbye in the
same turn landed in the refusal branch. 300 matches idleAfterSeconds' existing
written reason: five minutes absorbs a normal pause without stalling.

The javadoc said both waits are for the pane to report an injectable state.
AgentStatus.injectable() accepts BLOCKED; the settle check requires IDLE or DONE.
The wording named the wrong predicate.
2026-10-03 21:43:25 +02:00
Dai Ha 5d8b9d365c fleetd #651: raise turnSettleSeconds default from 20 to 300
CI / shell-tests (pull_request) Failing after 7s
CI / contract (pull_request) Successful in 56s
CI / build (pull_request) Failing after 1m55s
A lead that runs the documented handover procedure writes its goodbye
message in the same turn as fleet_handover confirm. The deferred roll
then waits turnSettleSeconds for that same pane to reach IDLE or DONE.
An ordinary goodbye turn took 20394ms, so a 20s budget was too small
and the roll refused itself with TURN_NEVER_SETTLED. The refusal
branch is correct; only the budget was wrong.

300 is not derived from that single 20394ms observation. It matches
leadHeartbeat.idleAfterSeconds (also default 300), the only other
constant in this codebase answering "how long may a lead legitimately
be mid-turn", whose own javadoc reasons that 5 minutes absorbs normal
pauses without stalling. The two constants answer the same question
and should not disagree by a factor of fifteen. The wait stays bounded
on purpose: an unbounded wait would let a lead whose turn never ends
park a continuation and hold a pending token forever, which is harder
to notice than a logged refusal.

Also corrects FleetConfig's javadoc for turnSettleSeconds and
clearSettleSeconds, which described both waits as waiting for the
pane to report an "injectable" state again. The code requires IDLE or
DONE specifically and excludes BLOCKED (a live turn merely paused),
which is the exact state where sending /clear would destroy context.
LeadRollover.java's javadoc already states this correctly; only
FleetConfig.java's was wrong.

Adds FleetConfigTest coverage for the default's resolution: unset,
positive, and <= 0 all resolve as expected. Coverage for the two
turn-boundary properties (settles-in-time vs. never-settles, and
BLOCKED is not treated as settled) already existed in
LeadRolloverTest and needed no change — confirmed by mutating the
default back to 20 and the IDLE||DONE check to also accept BLOCKED;
both mutations were caught by existing or new tests, then reverted.
2026-10-03 21:35:46 +02:00
Dai Ha a6aeda39e7 Merge PR #681: fleetd #680 — pin the jar-path split, check the plist's jar, widen the locator
CI / shell-tests (push) Failing after 10s
CI / contract (push) Successful in 46s
CI / build (push) Failing after 1m50s
Verified by the lead before merge, each mutation re-run independently:
- three-dot diff: 1 commit, 3 files, nothing dragged in; clean trial merge
- suite exit 0, 134 test functions (5 new); the 3 FAIL and 5 mktemp lines are
  the documented deliberate self-test output, unchanged from baseline
- JAR moved under target/        -> RED  (the core invariant is pinned now)
- JAR moved elsewhere outside it -> PASS (not a refuse-everything test)
- PATTERN narrowed back to run/  -> RED
- plist-jar mismatch die removed -> RED
- comm=java allowlist removed    -> RED (widening PATTERN did not weaken #593)
- mvn clean install from fleetd/: Tests run: 1929, Failures: 0, 172 reports

Measured separately: during a full Maven build no extra java process matches
the widened 'fleetd.jar' pattern, so assert_single_daemon does not false-
positive while a worker builds.
2026-10-03 21:34:14 +02:00
Dai Ha 31b3c24caa fleetd #651: tell a rolling lead that surviving its goodbye means the roll refused
CI / shell-tests (push) Failing after 8s
CI / contract (push) Successful in 47s
CI / build (push) Failing after 2m9s
A lead calls fleet_handover confirm from inside its own turn, and the skill
tells it to write its goodbye in that same turn. The deferred roll then waits
turnSettleSeconds for that pane to reach IDLE or DONE. Measured on 2026-10-02,
an ordinary goodbye turn took 20394ms against a 20s budget, so the roll wrote
TURN_NEVER_SETTLED and sent no /clear. confirm() had already returned accepted,
so there was no caller left to tell, and the lead carried on believing it had
been replaced.

The refusal is the correct branch — clearing a live turn would destroy context.
The gap is that nobody is told to look afterwards.

fleet_handover{action:"status", token} already reports the outcome: FleetMcp
dispatches it to LeadRollover.status, which reads the same outcomes map the
refusal writes into. Nothing new is needed in the daemon for this half.

The signal costs nothing, because a successful roll clears the lead. A lead
that is still running after its goodbye already knows the roll failed. That is
more reliable than warning it to keep the goodbye short, which would make the
feature worst exactly when it matters most.

The live budget on this host was raised to 300s separately, in fleetd.yaml,
which is hot: the config-watcher reloaded at 21:28:00 with no "needs a restart"
clause, four seconds after the edit, with no daemon restart. The code default
is a separate change.
2026-10-03 21:29:21 +02:00
4 changed files with 153 additions and 140 deletions
+7
View File
@@ -164,6 +164,13 @@ fails.
- **`accepted` does not mean your pane has been cleared.** It means every gate passed and the roll
is scheduled to run once your current turn ends. Say your goodbye in the same turn — you will not
get another one.
- **If you are still running after that turn, the roll did not happen.** A roll that works clears
you, so surviving your own goodbye is itself the signal that it refused. Check with
`fleet_handover{action: "status", token}`, using the token you confirmed. `TURN_NEVER_SETTLED`
means your turn ran past `leadRollover.turnSettleSeconds` and **no `/clear` was ever sent**: your
context is intact and nothing was lost. Open a fresh request and retry. Never assume the roll
succeeded because `confirm` answered `accepted` — by the time it refuses, there is no caller left
to tell, so this check is the only thing that closes that gap.
- **There is no terminal or session parameter, on purpose.** The pane is always your own, resolved
from your connection, so you can only ever roll yourself.
- **`operatorConfirmed` is your report of what a human told you.** Do not pass `true` because you
@@ -1129,11 +1129,8 @@ public record FleetConfig(
* {@code tab} can never be discovered, launched or not
* @param instances how many of this lead should be live (default 1). The daemon
* launches only the shortfall, so a restart adopts rather than doubles
* @param tabPrefix no longer used to find a lead's tab — {@code tab} is matched
* exactly. Its only remaining job is the startup collision guard
* ({@link #validateLeadTabPrefixes()}), which still uses it to refuse
* a worker {@code tabLabel} template that could be misread as a lead.
* Default {@code "lead:"}
* @param tabPrefix lead-tab naming convention checked against member labels. Lead
* identity uses {@code tab}. Default {@code "lead:"}
* @param scanIntervalSeconds how long a tab scan is cached before herdr is asked again; also the
* worst case before a newly-labelled tab is recognised. Default 10
* @param kind which agent runs there ({@code claude}, {@code opencode}, …)
@@ -1236,10 +1233,7 @@ public record FleetConfig(
String tabLabel) {
/**
* Role first, so the tab bar reads as the fleet and so the label shares a namespace with a
* lead's {@code tabPrefix}. Because {@code {role}} comes from a closed enum, a generated
* member label can never begin with {@code "lead:"} — the clash that
* {@link #validateLeadTabPrefixes()} used to have to check for is unrepresentable here.
* Role first, so the tab bar identifies the member's fleet role.
*/
public static final String DEFAULT_TAB_LABEL = "{role}: {profile} #{n}";
@@ -1399,11 +1393,13 @@ public record FleetConfig(
* called FROM the calling lead's own turn, so its pane is still {@code WORKING} the instant
* {@code confirm()} validates every gate and schedules the roll. {@code
* dev.ltms.fleet.lead.LeadRollover}'s deferred continuation waits up to this many seconds for
* that SAME pane to report an injectable state again — i.e. for the calling turn to actually
* end — before it sends {@code /clear} at all. If that wait times out, no {@code /clear} is
* that SAME pane to report {@code IDLE} or {@code DONE} — i.e. for the calling turn to actually
* end — before it sends {@code /clear} at all. {@code BLOCKED} does not count: that is a live
* turn merely paused, not one that has finished. If that wait times out, no {@code /clear} is
* ever sent: a lead that never goes idle is still doing real work, and clearing it would
* destroy live context. This is a separate wait from {@code clearSettleSeconds} below, which
* bounds the SECOND wait, for the pane to re-settle AFTER {@code /clear} has already gone out.
* bounds the SECOND wait, for the pane to reach {@code IDLE} or {@code DONE} again AFTER
* {@code /clear} has already gone out.
*
* @param handoverPath required when this block is present — where the handover file a fresh
* lead session reads must live. There is no sane non-null default for an
@@ -1422,12 +1418,13 @@ public record FleetConfig(
* @param maxDocAgeSeconds default 3600 — refuse a handover file whose modified time is older
* than this many seconds, so a stale leftover from an earlier rollover
* attempt can never be mistaken for a fresh one.
* @param turnSettleSeconds default 20 — bound on how long the deferred roll waits for the
* CALLING lead's own turn to end (its pane to report injectable again)
* before sending {@code /clear} at all. See the paragraph above.
* @param turnSettleSeconds default 300 — bound on how long the deferred roll waits for the
* CALLING lead's own turn to end (its pane to report {@code IDLE} or
* {@code DONE}) before sending {@code /clear} at all. See the paragraph
* above.
* @param clearSettleSeconds default 20 — bound on how long to wait for the lead's pane to
* report an injectable state again after {@code /clear} before giving up. A
* roll that times out here never sends {@code bootstrapText}.
* report {@code IDLE} or {@code DONE} again after {@code /clear} before
* giving up. A roll that times out here never sends {@code bootstrapText}.
* @param bootstrapText default a sentence naming the RESOLVED handover path — sent to the
* lead's pane once it settles after {@code /clear}, telling the fresh
* session where to read the handover and carry on. Left {@code null} here
@@ -1444,7 +1441,7 @@ public record FleetConfig(
public LeadRollover {
requireOperatorConfirm = requireOperatorConfirm == null || requireOperatorConfirm;
maxDocAgeSeconds = (maxDocAgeSeconds == null || maxDocAgeSeconds <= 0) ? 3600 : maxDocAgeSeconds;
turnSettleSeconds = (turnSettleSeconds == null || turnSettleSeconds <= 0) ? 20 : turnSettleSeconds;
turnSettleSeconds = (turnSettleSeconds == null || turnSettleSeconds <= 0) ? 300 : turnSettleSeconds;
clearSettleSeconds = (clearSettleSeconds == null || clearSettleSeconds <= 0) ? 20 : clearSettleSeconds;
bootstrapText = (bootstrapText == null || bootstrapText.isBlank()) ? null : bootstrapText;
}
@@ -2682,28 +2679,11 @@ public record FleetConfig(
}
/**
* Reject a lead-scan convention that a worker tab would also satisfy (CB-531).
* Reject a member tab-label template that could render as a configured lead tab or match a
* lead-tab naming convention.
*
* <p>The scan reads a tab label and concludes "a lead lives here". fleetd also <em>writes</em>
* tab labels — every member gets one rendered into its tab. Choose a lead {@code tabPrefix} that
* a member template matches and the daemon starts labelling its own members as leads, promoting
* the entire fleet to {@link dev.ltms.fleet.auth.Role#PRIMARY} with no message and no diff.
* {@link #validatePanePlacementAgainstLeadTabs()} is the check that stops a pane-placed member
* from landing inside a lead's tab in the first place; this check is a second, independent
* guard that catches the hazard even when every profile places members correctly, by refusing
* a label that a scan would still misread as a lead.
*
* <p>CB-557 shrank this check rather than removing it. The default template is
* {@code "{role}: {profile} #{n}"} and {@code {role}} comes from a closed enum, so a
* <em>generated</em> label can no longer collide by construction. What remains checkable is what
* an operator still writes by hand: the {@code fleet.tabLabel} template and any per-profile
* {@code tabLabel} override.
*
* <p>Fatal rather than a warning, unlike {@link #warnUnknownTopLevelKeys}: an unknown key means
* a feature does nothing, while this means a feature does the opposite of what it says.
*
* @throws IllegalStateException when the fleet template or any profile's {@code tabLabel}
* override starts with a configured lead prefix
* @throws IllegalStateException when the fleet template or a profile {@code tabLabel} override
* can render as a configured lead tab or match a lead-tab prefix
*/
public void validateLeadTabPrefixes() {
if (fleet == null || fleet.leaders().isEmpty()) {
@@ -2714,20 +2694,30 @@ public record FleetConfig(
if (leader == null) {
return;
}
String tab = leader.tab();
String prefix = leader.tabPrefix();
// The fleet-wide template is checked once per prefix: it labels every member that has no
// override, so one bad template promotes the entire fleet, not one profile.
if (startsWithIgnoreCase(fleet.tabLabel(), prefix)) {
if (templateCanRenderAs(fleet.tabLabel(), tab)) {
bad.add("fleet.tabLabel=\"" + fleet.tabLabel() + "\" can render as the tab of "
+ "lead '" + leadName + "' (\"" + tab + "\")");
} else if (startsWithIgnoreCase(fleet.tabLabel(), prefix)) {
bad.add("fleet.tabLabel=\"" + fleet.tabLabel() + "\" starts with the tabPrefix of "
+ "lead '" + leadName + "' (\"" + prefix + "\")");
}
profiles().entrySet().stream()
.filter(e -> startsWithIgnoreCase(e.getValue().tabLabel(), prefix))
.map(Map.Entry::getKey)
.sorted()
.forEach(p -> bad.add("profile '" + p + "' overrides tabLabel with \""
+ profiles().get(p).tabLabel() + "\", which starts with the tabPrefix of "
+ "lead '" + leadName + "' (\"" + prefix + "\")"));
.forEach(p -> {
String label = profiles().get(p).tabLabel();
if (templateCanRenderAs(label, tab)) {
bad.add("profile '" + p + "' overrides tabLabel with \"" + label
+ "\", which can render as the tab of lead '" + leadName
+ "' (\"" + tab + "\")");
} else if (startsWithIgnoreCase(label, prefix)) {
bad.add("profile '" + p + "' overrides tabLabel with \"" + label
+ "\", which starts with the tabPrefix of lead '" + leadName
+ "' (\"" + prefix + "\")");
}
});
});
if (bad.isEmpty()) {
return;
@@ -2738,6 +2728,31 @@ public record FleetConfig(
+ "lead tabs cannot be confused.");
}
private static boolean templateCanRenderAs(String template, String tab) {
if (template == null || template.isBlank() || tab == null || tab.isBlank()) {
return false;
}
var placeholders = Pattern.compile("\\{(?:role|profile|model|n)}").matcher(template);
StringBuilder expression = new StringBuilder("^");
int literalStart = 0;
while (placeholders.find()) {
expression.append(Pattern.quote(template.substring(literalStart, placeholders.start())));
expression.append(".*");
literalStart = placeholders.end();
}
expression.append(Pattern.quote(template.substring(literalStart))).append("$");
return Pattern.compile(expression.toString(), Pattern.CASE_INSENSITIVE).matcher(tab).matches();
}
/** Case-insensitive prefix test that tolerates a null or blank label. */
private static boolean startsWithIgnoreCase(String label, String prefix) {
if (label == null || prefix == null || prefix.isBlank()) {
return false;
}
String stripped = label.strip();
return stripped.regionMatches(true, 0, prefix, 0, prefix.length());
}
/**
* Reject a profile that places its members by {@code "pane"} while any {@code fleet.leaders}
* entry names a {@code tab}. A pane-placed member lands inside the focused tab rather than its
@@ -2797,15 +2812,6 @@ public record FleetConfig(
}
}
/** Case-insensitive prefix test that tolerates a null/blank label. */
private static boolean startsWithIgnoreCase(String label, String prefix) {
if (label == null || prefix == null || prefix.isBlank()) {
return false;
}
String stripped = label.strip();
return stripped.regionMatches(true, 0, prefix, 0, prefix.length());
}
/**
* Reject a subscription profile whose {@code env:} block tries to reseat the Anthropic binding
* (CB-542).
@@ -676,12 +676,8 @@ class FleetConfigTest {
assertEquals(5, hb.quietNudgeCap());
}
/**
* The hazard the guard exists for: fleetd writes worker tab labels and reads lead tab labels.
* Overlap the two and every worker it spawns is read back as a lead.
*/
@Test
void aLeadPrefixThatAProfileTabLabelOverrideAlsoMatchesRefusesToStart(@TempDir Path dir)
void aProfileTabLabelOverrideMatchingALeadTabRefusesToStart(@TempDir Path dir)
throws Exception {
Path f = dir.resolve("collide.yaml");
Files.writeString(f, """
@@ -689,32 +685,33 @@ class FleetConfigTest {
port: 8080
profiles:
gx10:
tabLabel: "lead: {profile} #{n}"
tabLabel: "alpha"
fleet:
leaders:
opus:
tab: "lead: opus"
tabPrefix: "lead:"
tab: "alpha"
""");
FleetConfig cfg = FleetConfig.load(f);
IllegalStateException e =
assertThrows(IllegalStateException.class, cfg::validateLeadTabPrefixes);
assertTrue(e.getMessage().contains("gx10"), "the message must name the offending profile");
assertTrue(e.getMessage().contains("alpha"), "the message must name the offending label");
}
/** A bad fleet-wide template promotes every member, not one profile — so it is checked too. */
@Test
void aFleetTabLabelThatMatchesALeadPrefixRefusesToStart(@TempDir Path dir) throws Exception {
void aFleetTabLabelTemplateThatCanRenderAsALeadTabRefusesToStart(@TempDir Path dir) throws Exception {
Path f = dir.resolve("collide-template.yaml");
Files.writeString(f, """
bind:
port: 8080
profiles:
pha: {}
fleet:
tabLabel: "lead: {role} {profile}"
tabLabel: "al{profile}"
leaders:
opus:
tab: "lead: opus"
tab: "alpha"
""");
FleetConfig cfg = FleetConfig.load(f);
@@ -723,12 +720,28 @@ class FleetConfigTest {
assertTrue(e.getMessage().contains("fleet.tabLabel"));
}
/**
* The point of making role the label's first field: {@code {role}} comes from a closed enum, so
* a generated label cannot begin with {@code "lead:"} however the fleet is configured.
*/
@Test
void theDefaultTabLabelCannotCollideWithTheDefaultLeadPrefix(@TempDir Path dir) throws Exception {
void anExactFleetTabLabelCollisionRefusesToStart(@TempDir Path dir) throws Exception {
Path f = dir.resolve("exact-tab-collision.yaml");
Files.writeString(f, """
bind:
port: 8080
fleet:
tabLabel: "alpha"
leaders:
alpha:
tab: "alpha"
""");
IllegalStateException e = assertThrows(IllegalStateException.class,
() -> FleetConfig.load(f).validateAll());
assertTrue(e.getMessage().contains("fleet.tabLabel"),
"the message must name the offending label");
assertTrue(e.getMessage().contains("alpha"), "the message must name the colliding lead tab");
}
@Test
void aFleetTabLabelTemplateThatCannotRenderAsALeadTabIsAllowed(@TempDir Path dir) throws Exception {
Path f = dir.resolve("ok.yaml");
Files.writeString(f, """
bind:
@@ -737,17 +750,13 @@ class FleetConfigTest {
gx10:
baseUrl: http://gx00.gw:8000
fleet:
tabLabel: "worker-{profile}"
leaders:
opus:
tab: "lead: opus"
tab: "alpha"
""");
assertDoesNotThrow(() -> FleetConfig.load(f).validateLeadTabPrefixes());
for (MemberRole role : MemberRole.values()) {
assertFalse(FleetConfig.Fleet.DEFAULT_TAB_LABEL
.replace("{role}", role.wireName()).startsWith("lead:"),
"no role renders a label that reads as a lead");
}
assertDoesNotThrow(() -> FleetConfig.load(f).validateAll());
}
@Test
@@ -3175,4 +3184,41 @@ class FleetConfigTest {
FleetConfig cfg = FleetConfig.load(f);
assertTrue(cfg.models().offIds().isEmpty());
}
// ── fleetd #651: leadRollover.turnSettleSeconds default resolution ─────────────────────────
@Test
void turnSettleSecondsDefaultsTo300WhenUnset(@TempDir Path dir) throws Exception {
Path f = dir.resolve("bare-rollover.yaml");
Files.writeString(f, "bind:\n port: 8080\nleadRollover: {}\n");
FleetConfig.LeadRollover rollover = FleetConfig.load(f).leadRollover();
assertNotNull(rollover);
assertEquals(300, rollover.turnSettleSeconds());
}
@Test
void turnSettleSecondsUsesAnExplicitPositiveValue(@TempDir Path dir) throws Exception {
Path f = dir.resolve("rollover.yaml");
Files.writeString(f, """
bind:
port: 8080
leadRollover:
turnSettleSeconds: 45
""");
FleetConfig.LeadRollover rollover = FleetConfig.load(f).leadRollover();
assertEquals(45, rollover.turnSettleSeconds());
}
@Test
void turnSettleSecondsFallsBackTo300WhenZeroOrNegative(@TempDir Path dir) throws Exception {
Path zero = dir.resolve("zero.yaml");
Files.writeString(zero, "bind:\n port: 8080\nleadRollover:\n turnSettleSeconds: 0\n");
assertEquals(300, FleetConfig.load(zero).leadRollover().turnSettleSeconds());
Path negative = dir.resolve("negative.yaml");
Files.writeString(negative, "bind:\n port: 8080\nleadRollover:\n turnSettleSeconds: -5\n");
assertEquals(300, FleetConfig.load(negative).leadRollover().turnSettleSeconds());
}
}
@@ -17,63 +17,21 @@ import static org.junit.jupiter.api.Assertions.assertThrows;
import static org.junit.jupiter.api.Assertions.assertTrue;
/**
* The gap this class exists to close: mutation testing on the fleetd ticket "central allow-list
* of usable models" found that although {@link FleetConfig#validateModels()}'s own logic was well
* pinned, nothing proved either real caller ({@code Fleetd.main} and {@link ConfigRef#reload()})
* still invoked it — deleting the call site left the full suite green (1478/0/0/0). A follow-up
* measurement (same technique — remove one call site, run the suite, not read the code) found the
* SAME gap for all five of {@link FleetConfig}'s other validators at startup, and for four of the
* six inside {@link ConfigRef#reload()}. This is a class of gap, not one line's mistake: every one
* of those thirteen tests called the validator itself directly, never the real caller that was
* supposed to.
* Tests the reflective validator sweep and {@link FleetConfig#validateAll()} reachability.
*
* <p>The fix replaces the six individual {@code cfg.validateXxx()} calls at each of the two real
* call sites with one {@link FleetConfig#validateAll()}, which reaches every validator by
* reflection rather than by a hand-maintained list of names. A hand-maintained list of six names
* would have exactly the defect it replaces: the seventh validator someone adds next month has no
* reason to be added to it, and nothing would say so. This class proves TWO separate claims, and
* keeps them separate on purpose:
*
* <ol>
* <li>{@link #theSweepMechanismIsGenericNotHardcodedToFleetConfigsSixNames()} and its neighbours
* prove the reflective sweep itself ({@link FleetConfig#invokeAllValidators}) is a general
* mechanism — it runs whatever public, no-arg, void {@code validateXxx()} methods a class
* happens to declare today, including a class with more of them than {@link FleetConfig}
* has right now. This is the proof that a future, real seventh validator on {@link
* FleetConfig} would be swept automatically, without needing to add a real (unwanted)
* seventh validator just to exercise the claim.</li>
* <li>{@link #validateAllReachesEveryOneOfTodaysRealValidators()} proves {@link
* FleetConfig#validateAll()} itself is wired to that same generic mechanism and genuinely
* reaches every one of today's real validators — reusing the exact minimal failing
* configurations {@code FleetConfigTest} already established for each one directly, plus a
* dedicated fixture for {@link FleetConfig#validateLeadRollover()}, which no other test
* drives through {@code validateAll()} — so a single call to {@code validateAll()} is shown
* to reproduce every one of those failures.</li>
* </ol>
*
* <p>Together with the direct-{@code Fleetd.main}-invocation tests in {@code
* FleetdStartupValidationTest} (which prove the real startup call site still calls {@code
* validateAll()}) and the {@code ConfigRefTest} reload tests (which prove the same for {@link
* ConfigRef#reload()}), removing {@code cfg.validateAll();} from either real call site now fails
* a test in this module.
*
* <p><b>What is NOT pinned, measured rather than assumed.</b> Reverting {@link
* FleetConfig#validateAll()} to a hardcoded list of today's method calls leaves the whole
* suite green. Nothing ties {@code validateAll()} to
* the generic sweep — claim 1 proves {@link FleetConfig#invokeAllValidators} is generic, and claim
* 2 proves {@code validateAll()} reaches today's validators, and a hardcoded list satisfies both. So the
* reflective sweep is a convenience, not the guarantee. The guarantee is {@link
* #fleetConfigDeclaresExactlyTheseValidatorsToday()}: it fails the moment any validator is added
* or removed, which forces whoever changes the set to look at this file.
* <p>{@link #theSweepMechanismIsGenericNotHardcodedToFleetConfigsSixNames()} and its neighbours
* prove that {@link FleetConfig#invokeAllValidators} runs each public, no-arg, void
* {@code validateXxx()} method on its target. {@link #fleetConfigDeclaresExactlyTheseValidatorsToday()}
* is the canary for the validator set. {@link #validateAllReachesEveryOneOfTodaysRealValidators()}
* is the reachability check for that set.
*/
class FleetConfigValidateAllTest {
// ── Claim 1: the reflective sweep is a general mechanism, not six names in disguise ──────────
// ── Claim 1: the reflective sweep is a general mechanism ─────────────────────────────────────
/**
* A throwaway fixture class, unrelated to {@link FleetConfig} in every way except shape: three
* public, no-arg, void methods named {@code validateXxx}. Proves the sweep works on ANY class
* with this shape, not on something special-cased to {@link FleetConfig}.
* Fixture with public, no-arg, void methods named {@code validateXxx}. It proves the sweep uses
* the target's method shape rather than special handling for {@link FleetConfig}.
*/
static class ThreeValidators {
final List<String> ran = new ArrayList<>();
@@ -102,12 +60,8 @@ class FleetConfigValidateAllTest {
}
/**
* The core of the "self-maintaining" requirement: the exact same class shape as {@link
* ThreeValidators}, plus one more method — standing in for "a developer adds a validator next
* month". Nothing about the sweep changes to pick it up; the new method is invoked purely
* because it exists and matches the shape. This is what makes adding a seventh real validator
* to {@link FleetConfig} safe without touching {@link FleetConfig#validateAll()} or either
* call site — there is no "wire it in" step left to forget.
* Fixture with an added valid method. It proves the sweep reaches a method because it matches
* the validator shape.
*/
static class FourValidators {
final List<String> ran = new ArrayList<>();
@@ -135,7 +89,7 @@ class FleetConfigValidateAllTest {
FleetConfig.invokeAllValidators(target);
assertEquals(List.of("validateAlpha", "validateBeta", "validateDelta", "validateGamma"),
sorted(target.ran),
"the fourth method must be reached automatically — proving a class can grow the "
"the added method must be reached automatically — proving a class can grow the "
+ "set of things it validates with no change to the sweep itself");
}
@@ -297,16 +251,16 @@ class FleetConfigValidateAllTest {
port: 8765
""", "auth.mode: token");
// validateLeadTabPrefixes: a fleet-wide tabLabel that starts with a lead's own tabPrefix.
// validateLeadTabPrefixes: a fleet-wide tabLabel that equals a lead tab.
assertValidateAllRefuses(dir, "lead-tab-prefixes.yaml", """
bind:
host: 127.0.0.1
port: 8765
fleet:
tabLabel: "lead: {role} {profile}"
tabLabel: "alpha"
leaders:
opus:
tab: "lead: opus"
tab: "alpha"
""", "fleet.tabLabel");
// validateSubscriptionProfiles: subscription: true with env: reseating ANTHROPIC_BASE_URL.