test(webconfig): pure WebConfigBatch state-machine spec + host tests (Phase 6)

The WebConfig POST/result/reboot/stop batch state machine was the largest
remaining Phase 6 coverage gap (all inline in WebConfigServer.cpp, coupled to
AsyncWebServer/ArduinoJson and untestable on host). Extract its decision + timing
CORE into a pure, dependency-free spec mirroring MQTTLifecycle.h:

- src/helpers/WebConfigBatch.h: classifyPost (replay/busy/accept/no-changes with
  the DONE-vs-PENDING reqid asymmetry), drain pacing (signed 25 ms gate, sticky
  all_ok, 30 s reboot fallback), result classification + arm-once 3 s reboot,
  signed-wrap-safe reboot-due / isRebootPending, and stop gating (finalize when
  refs==0, warn-once, never force teardown). Constants verbatim from the source.
- test/test_webconfig_batch/: full host coverage incl. exact boundaries and
  millis() rollover.

Spec-first, exactly like Phase 4's MQTTLifecycle.h: this is NOT yet wired into
WebConfigServer.cpp. That server is hardware-tuned (debugged against real iOS
captive-portal + HTTP-caching + route-ordering behavior), so making the spec
load-bearing is a deliberately separate, hardware-validated follow-up.

Faithfulness independently reviewed against WebConfigServer.cpp; native suite
green (14 dirs). No production behavior change.
This commit is contained in:
agessaman
2026-07-19 05:31:00 -07:00
parent b0cf29fb33
commit 3c170eb1a8
3 changed files with 414 additions and 6 deletions
+13 -6
View File
@@ -468,12 +468,19 @@ into the pure policy seams and covered:
- `MQTTPublicationType` values frozen in `test_mqtt_topic_router`; the
bridge-side `MQTTMessageType` alignment was already a compile-time `static_assert`.
Still open (each a good follow-up PR): the **WebConfig POST/result/reboot/stop
state machine** (largest gap — all inline in `WebConfigServer.cpp`; natural to
extract a pure `WebConfigBatch` seam mirroring `MQTTLifecycle.h`) and the
**queue-orchestration** behaviors (FIFO ordering, evict/requeue-failure interplay,
the two adapters' drop-vs-keep-head divergence), which need a fake-queue harness.
The original scope list follows.
The **WebConfig POST/result/reboot/stop state machine** (the largest gap) now
has a pure, host-tested SPEC — `src/helpers/WebConfigBatch.h` +
`test/test_webconfig_batch/` on branch `phase6/webconfig-batch-seam` that
faithfully characterizes the accept/drain/result/reboot/stop decisions currently
inline in `WebConfigServer.cpp`. Like Phase 4's `MQTTLifecycle.h`, it is
**spec-first and NOT yet wired**: rewiring the hardware-tuned server to consume it
(so it becomes load-bearing and non-drifting) is a deliberately separate,
hardware-validated follow-up.
Still open (each a good follow-up PR): wiring `WebConfigServer.cpp` to the
`WebConfigBatch` spec above, and the **queue-orchestration** behaviors (FIFO
ordering, evict/requeue-failure interplay, the two adapters' drop-vs-keep-head
divergence), which need a fake-queue harness. The original scope list follows.
After lifecycle ownership is stable, broaden deterministic integration coverage:
+195
View File
@@ -0,0 +1,195 @@
#pragma once
#include <stdint.h>
// Fork-owned, dependency-free spec for the WebConfig "config batch / reboot /
// stop" decision + timing core, plus its host tests (test/test_webconfig_batch/).
//
// This is the Phase 6 counterpart of MQTTLifecycle.h: a PURE state machine that
// captures exactly what src/helpers/esp32/WebConfigServer.cpp decides today, so
// the POST-accept / drain / result-read / reboot / stop transitions can be
// exercised deterministically without AsyncWebServer, ArduinoJson, WiFi, or the
// FreeRTOS mutex/refcount.
//
// Scope boundary (spec-first, exactly like Phase 4's MQTTLifecycle.h): this
// header is the SPEC and the test seam. It is NOT yet wired into
// WebConfigServer.cpp. The batch/reboot/stop state machine there is
// hardware-tuned (it was debugged against real iOS captive-portal behavior,
// HTTP caching, and route ordering), so the production rewiring that makes these
// functions load-bearing is a deliberately separate, hardware-validated
// follow-up. Until then, keep this in sync with WebConfigServer.cpp by hand; the
// file:line references below point at the behavior each function mirrors.
//
// Behavior source (all line refs against WebConfigServer.{h,cpp} at the time of
// writing): constants at .h:90-96,155-161; POST accept at .cpp:610-719; drain at
// .cpp:289-334; result read at .cpp:721-791; reboot fire at .cpp:262-265;
// isRebootPending at .cpp:70-74; stop gating at .cpp:185-255.
namespace WebConfigBatch {
// Constants, verbatim from WebConfigServer.
static const int kMaxBatch = 24; // .h:90 MAX_BATCH
static const uint32_t kDrainPacingMs = 25; // .cpp:296 inter-command gap
static const uint32_t kRebootFallbackMs = 30000; // .cpp:331 drain-finish fallback
static const uint32_t kRebootConfirmMs = 3000; // .cpp:784 first result-read arm
static const uint32_t kStopWarnMs = 10000; // .h:95 STOP_WARN_MS
// The batch lifecycle. A fresh POST moves Idle->Pending; the drainer moves
// Pending->Done; Done stays re-readable until the next POST claims the slot;
// finalizeTeardown() resets to Idle.
enum class State : uint8_t {
Idle = 0,
Pending,
Done,
};
// millis() idioms. elapsedMs uses unsigned wraparound (correct across one 32-bit
// rollover). scheduleAt mirrors the production wrap-guard: every _reboot_at /
// _stop_warn_at assignment does `if (t == 0) t = 1;` so 0 keeps meaning
// "unscheduled" even when the deadline lands exactly on the rollover boundary.
static inline uint32_t elapsedMs(uint32_t now, uint32_t then) { return now - then; }
static inline uint32_t scheduleAt(uint32_t now, uint32_t delay) {
const uint32_t t = now + delay;
return t == 0 ? 1u : t;
}
// Signed wrap-safe "deadline reached", matching the production
// `(int32_t)(now - deadline) >= 0` comparisons.
static inline bool deadlineReached(uint32_t now, uint32_t deadline) {
return (int32_t)(now - deadline) >= 0;
}
// --------------------------------------------------------------------------
// POST accept classification (.cpp:637-718). Precedence, verbatim from the
// source: an in-flight/finished batch with the SAME reqid is an idempotent
// replay (commands are NOT re-applied); a DIFFERENT reqid while a batch is still
// PENDING is rejected as busy; otherwise a batch with no changes and no reboot
// is a no-op, and anything else is accepted. Note the asymmetry: a different
// reqid while DONE is NOT busy — the new batch overwrites the DONE slot.
//
// Assumes the request already passed reqid grammar (WebConfigKeys::wcIsValidReqId)
// and per-key allowlist/secret validation, which are covered by test_webconfig_keys.
// --------------------------------------------------------------------------
enum class PostOutcome : uint8_t {
Replay, // 202; reqid matches the current batch, commands not re-applied
Busy, // 409; a different batch is still PENDING
Accept, // 202; a new batch is accepted (from Idle, or overwriting a Done slot)
NoChanges, // 400; nothing to do (no changes and no reboot requested)
};
static inline PostOutcome classifyPost(State state, bool reqid_matches_current,
int change_count, bool reboot_after) {
if (state != State::Idle && reqid_matches_current) return PostOutcome::Replay;
if (state == State::Pending) return PostOutcome::Busy; // reqid differs (match handled above)
if (change_count <= 0 && !reboot_after) return PostOutcome::NoChanges;
return PostOutcome::Accept;
}
// The state string a Replay body reports mirrors the batch state (.cpp:640):
// "done" when the matched batch already finished, else "pending".
static inline const char* replayStateName(State state) {
return state == State::Done ? "done" : "pending";
}
// --------------------------------------------------------------------------
// Drain (.cpp:289-333). One command per tick.
// --------------------------------------------------------------------------
// The drainer waits only BETWEEN commands: never before the first (batch_next
// == 0 runs immediately and fires onConfigBatchStart), never after the last, and
// otherwise until the 25 ms pacing gap elapses. The pacing compare is SIGNED to
// mirror the source verbatim (.cpp:296 `(int32_t)(now - _batch_last_cmd) < 25`),
// matching deadlineReached()'s signedness; for all reachable inputs (elapsed
// 0..25 ms) it is identical to the unsigned form.
static inline bool drainMustWait(int batch_next, int batch_count,
uint32_t now, uint32_t last_cmd_ms) {
return batch_next > 0 && batch_next < batch_count &&
(int32_t)(now - last_cmd_ms) < (int32_t)kDrainPacingMs;
}
// all_ok is a sticky AND across command replies; a reply counts as ok iff it
// begins with "OK" (.cpp:314). Once false it stays false.
static inline bool nextAllOk(bool prev_all_ok, bool reply_is_ok) {
return prev_all_ok && reply_is_ok;
}
// The batch is finished once the post-increment drain index reaches the count
// (.cpp:319-321).
static inline bool drainFinished(int batch_next_after_increment, int batch_count) {
return batch_next_after_increment >= batch_count;
}
// On finish, a reboot-requested + all-ok batch arms the 30 s fallback deadline;
// a partially-failed batch (all_ok == false) never reboots (.cpp:324-333).
// Returns the reboot_at deadline, or 0 for "no reboot scheduled".
static inline uint32_t finishRebootAt(bool batch_reboot, bool batch_all_ok, uint32_t now) {
return (batch_reboot && batch_all_ok) ? scheduleAt(now, kRebootFallbackMs) : 0u;
}
// --------------------------------------------------------------------------
// Result read (.cpp:738-786).
// --------------------------------------------------------------------------
enum class ResultOutcome : uint8_t {
Idle, // 200 "idle" — no batch; any valid reqid is echoed
Unknown, // 404 — a batch exists but the reqid does not match it
Pending, // 200 "pending"
Done, // 200 "done" (+ per-command results)
};
static inline ResultOutcome classifyResult(State state, bool reqid_matches_current) {
if (state == State::Idle) return ResultOutcome::Idle; // no reqid check while idle
if (!reqid_matches_current) return ResultOutcome::Unknown;
return state == State::Pending ? ResultOutcome::Pending : ResultOutcome::Done;
}
// The "reboot" flag reported in a Done body (.cpp:767): only a fully-OK,
// reboot-requested batch advertises a pending reboot.
static inline bool doneReportsReboot(bool batch_reboot, bool batch_all_ok) {
return batch_reboot && batch_all_ok;
}
// The first Done read arms the confirmed (3 s) reboot exactly once (.cpp:777-786):
// the !already_armed guard makes later reads idempotent, so polling cannot push
// the deadline out. When this returns true the caller sets armed = true and
// reboot_at = confirmRebootAt(now).
static inline bool shouldArmConfirmReboot(State state, bool batch_reboot,
bool batch_all_ok, bool already_armed) {
return state == State::Done && batch_reboot && batch_all_ok && !already_armed;
}
static inline uint32_t confirmRebootAt(uint32_t now) {
return scheduleAt(now, kRebootConfirmMs);
}
// --------------------------------------------------------------------------
// Reboot fire (.cpp:262-265) and isRebootPending (.cpp:70-74).
// --------------------------------------------------------------------------
static inline bool rebootDue(uint32_t reboot_at, uint32_t now) {
return reboot_at != 0 && deadlineReached(now, reboot_at);
}
// isRebootPending() reports true only for a config-save reboot in the Done
// state, so the manual /api/reboot path (batch_reboot == false) is deliberately
// NOT reported as pending even though _reboot_at is set.
static inline bool isConfigRebootPending(uint32_t reboot_at, bool batch_reboot, State state) {
return reboot_at != 0 && batch_reboot && state == State::Done;
}
// --------------------------------------------------------------------------
// Stop gating (.cpp:244-255). Teardown waits indefinitely for in-flight async
// handlers to drain (refs == 0); the STOP_WARN_MS timer only triggers a one-time
// diagnostic — it never forces teardown.
// --------------------------------------------------------------------------
enum class StopAction : uint8_t {
Finalize, // refs == 0: finalizeTeardown() may run now
Warn, // refs > 0 and the warn deadline passed, not yet warned: log once
Wait, // refs > 0: keep the session alive and wait
};
static inline StopAction stopStep(uint32_t handler_refs, bool already_warned,
uint32_t stop_warn_at, uint32_t now) {
if (handler_refs == 0) return StopAction::Finalize;
if (!already_warned && stop_warn_at != 0 && deadlineReached(now, stop_warn_at)) {
return StopAction::Warn;
}
return StopAction::Wait;
}
} // namespace WebConfigBatch
@@ -0,0 +1,206 @@
// Host contract tests for the pure WebConfig config-batch / reboot / stop state
// machine. This spec mirrors src/helpers/esp32/WebConfigServer.cpp; it is not
// yet wired into the bridge (see WebConfigBatch.h scope note).
#include <gtest/gtest.h>
#include <stdint.h>
#include <limits>
#include "helpers/WebConfigBatch.h"
namespace Batch = WebConfigBatch;
using State = WebConfigBatch::State;
// --------------------------------------------------------------------------
// POST accept classification
// --------------------------------------------------------------------------
TEST(WebConfigBatch, IdleAcceptsANewBatchWithChangesOrRebootOnly) {
// A normal save (changes present) is accepted.
EXPECT_EQ(Batch::PostOutcome::Accept,
Batch::classifyPost(State::Idle, /*reqid_matches=*/false, /*count=*/3, false));
// A reboot-only request (no changes) is still accepted.
EXPECT_EQ(Batch::PostOutcome::Accept,
Batch::classifyPost(State::Idle, false, 0, /*reboot_after=*/true));
}
TEST(WebConfigBatch, IdleWithNothingToDoIsNoChanges) {
EXPECT_EQ(Batch::PostOutcome::NoChanges,
Batch::classifyPost(State::Idle, false, 0, false));
}
TEST(WebConfigBatch, SameReqidReplaysWithoutReapplyingWhilePendingOrDone) {
// The idempotent-replay path fires for BOTH in-flight and finished batches
// when the reqid matches; commands are never re-applied.
EXPECT_EQ(Batch::PostOutcome::Replay,
Batch::classifyPost(State::Pending, /*reqid_matches=*/true, 3, false));
EXPECT_EQ(Batch::PostOutcome::Replay,
Batch::classifyPost(State::Done, /*reqid_matches=*/true, 3, false));
// Replay wins even over a would-be no-changes request.
EXPECT_EQ(Batch::PostOutcome::Replay,
Batch::classifyPost(State::Pending, true, 0, false));
// The replay body reports the batch's own state.
EXPECT_STREQ("pending", Batch::replayStateName(State::Pending));
EXPECT_STREQ("done", Batch::replayStateName(State::Done));
}
TEST(WebConfigBatch, DifferentReqidIsBusyOnlyWhilePending) {
// A second client (different reqid) while a batch is still draining => busy.
EXPECT_EQ(Batch::PostOutcome::Busy,
Batch::classifyPost(State::Pending, /*reqid_matches=*/false, 2, false));
// But once the previous batch is DONE, a different reqid is NOT busy: the new
// batch overwrites the finished slot (asymmetry with the Pending case).
EXPECT_EQ(Batch::PostOutcome::Accept,
Batch::classifyPost(State::Done, false, 2, false));
EXPECT_EQ(Batch::PostOutcome::NoChanges,
Batch::classifyPost(State::Done, false, 0, false));
}
// --------------------------------------------------------------------------
// Drain pacing / all_ok / finish
// --------------------------------------------------------------------------
TEST(WebConfigBatch, DrainNeverWaitsBeforeTheFirstOrAfterTheLastCommand) {
// batch_next == 0: the first command runs immediately (fires onConfigBatchStart).
EXPECT_FALSE(Batch::drainMustWait(0, 5, 1000, 1000));
// batch_next >= batch_count: nothing left to pace.
EXPECT_FALSE(Batch::drainMustWait(5, 5, 1000, 1000));
// A single-command (or reboot-only) batch never paces.
EXPECT_FALSE(Batch::drainMustWait(0, 1, 1000, 1000));
EXPECT_FALSE(Batch::drainMustWait(0, 0, 1000, 1000));
}
TEST(WebConfigBatch, DrainPacesTwentyFiveMillisBetweenCommandsWithInclusiveRelease) {
const uint32_t last = 1000;
// 24 ms after the previous command: still waiting.
EXPECT_TRUE(Batch::drainMustWait(2, 5, last + 24, last));
// Exactly 25 ms: the gate releases (production uses `< 25`).
EXPECT_FALSE(Batch::drainMustWait(2, 5, last + 25, last));
EXPECT_FALSE(Batch::drainMustWait(2, 5, last + 26, last));
}
TEST(WebConfigBatch, DrainPacingSurvivesMillisRollover) {
const uint32_t last = std::numeric_limits<uint32_t>::max() - 10;
EXPECT_TRUE(Batch::drainMustWait(2, 5, last + 24, last)); // 24 ms elapsed, wrapped
EXPECT_FALSE(Batch::drainMustWait(2, 5, last + 25, last)); // 25 ms elapsed, wrapped
}
TEST(WebConfigBatch, AllOkIsAStickyAndAcrossCommandReplies) {
bool all_ok = true;
all_ok = Batch::nextAllOk(all_ok, true);
EXPECT_TRUE(all_ok);
all_ok = Batch::nextAllOk(all_ok, false); // one command failed
EXPECT_FALSE(all_ok);
all_ok = Batch::nextAllOk(all_ok, true); // stays false forever after
EXPECT_FALSE(all_ok);
}
TEST(WebConfigBatch, DrainFinishesWhenTheIndexReachesTheCount) {
EXPECT_FALSE(Batch::drainFinished(4, 5));
EXPECT_TRUE(Batch::drainFinished(5, 5));
EXPECT_TRUE(Batch::drainFinished(0, 0)); // reboot-only / empty batch
}
TEST(WebConfigBatch, FinishArmsThirtySecondFallbackOnlyForAFullyOkRebootBatch) {
const uint32_t now = 100000;
EXPECT_EQ(now + Batch::kRebootFallbackMs,
Batch::finishRebootAt(/*reboot=*/true, /*all_ok=*/true, now));
// A partially-failed batch never reboots.
EXPECT_EQ(0u, Batch::finishRebootAt(true, false, now));
// No reboot requested.
EXPECT_EQ(0u, Batch::finishRebootAt(false, true, now));
}
// --------------------------------------------------------------------------
// Result read
// --------------------------------------------------------------------------
TEST(WebConfigBatch, ResultReadClassifiesIdlePendingDoneAndUnknownReqid) {
// Idle: any valid reqid gets "idle" (no reqid check while idle).
EXPECT_EQ(Batch::ResultOutcome::Idle, Batch::classifyResult(State::Idle, false));
EXPECT_EQ(Batch::ResultOutcome::Idle, Batch::classifyResult(State::Idle, true));
// Matching reqid reflects the batch state.
EXPECT_EQ(Batch::ResultOutcome::Pending, Batch::classifyResult(State::Pending, true));
EXPECT_EQ(Batch::ResultOutcome::Done, Batch::classifyResult(State::Done, true));
// A live/finished batch with a mismatched reqid is unknown (404).
EXPECT_EQ(Batch::ResultOutcome::Unknown, Batch::classifyResult(State::Pending, false));
EXPECT_EQ(Batch::ResultOutcome::Unknown, Batch::classifyResult(State::Done, false));
}
TEST(WebConfigBatch, DoneBodyAdvertisesRebootOnlyWhenFullyOk) {
EXPECT_TRUE(Batch::doneReportsReboot(true, true));
EXPECT_FALSE(Batch::doneReportsReboot(true, false));
EXPECT_FALSE(Batch::doneReportsReboot(false, true));
}
TEST(WebConfigBatch, FirstDoneReadArmsTheThreeSecondRebootExactlyOnce) {
const uint32_t now = 500000;
// First read of a fully-OK reboot batch arms.
EXPECT_TRUE(Batch::shouldArmConfirmReboot(State::Done, true, true, /*already_armed=*/false));
EXPECT_EQ(now + Batch::kRebootConfirmMs, Batch::confirmRebootAt(now));
// Already armed => never re-arms (polling can't push the deadline out).
EXPECT_FALSE(Batch::shouldArmConfirmReboot(State::Done, true, true, /*already_armed=*/true));
// Not applicable while pending, without reboot, or after a partial failure.
EXPECT_FALSE(Batch::shouldArmConfirmReboot(State::Pending, true, true, false));
EXPECT_FALSE(Batch::shouldArmConfirmReboot(State::Done, false, true, false));
EXPECT_FALSE(Batch::shouldArmConfirmReboot(State::Done, true, false, false));
}
// --------------------------------------------------------------------------
// Reboot fire / pending
// --------------------------------------------------------------------------
TEST(WebConfigBatch, RebootFiresAtOrAfterTheDeadlineAndNeverWhenUnscheduled) {
EXPECT_FALSE(Batch::rebootDue(0, 1234567)); // 0 == unscheduled
const uint32_t at = 100000;
EXPECT_FALSE(Batch::rebootDue(at, at - 1));
EXPECT_TRUE(Batch::rebootDue(at, at)); // inclusive boundary
EXPECT_TRUE(Batch::rebootDue(at, at + 1));
}
TEST(WebConfigBatch, RebootDueSurvivesMillisRollover) {
const uint32_t at = std::numeric_limits<uint32_t>::max() - 5; // near the top
EXPECT_FALSE(Batch::rebootDue(at, at - 1));
EXPECT_TRUE(Batch::rebootDue(at, at));
EXPECT_TRUE(Batch::rebootDue(at, at + 10)); // now has wrapped past zero
}
TEST(WebConfigBatch, OnlyAConfigSaveRebootInDoneStateReportsPending) {
EXPECT_TRUE(Batch::isConfigRebootPending(/*reboot_at=*/123, /*batch_reboot=*/true, State::Done));
// Manual /api/reboot: reboot_at set but batch_reboot false => not "pending".
EXPECT_FALSE(Batch::isConfigRebootPending(123, false, State::Done));
// Not yet done, or nothing scheduled.
EXPECT_FALSE(Batch::isConfigRebootPending(123, true, State::Pending));
EXPECT_FALSE(Batch::isConfigRebootPending(0, true, State::Done));
}
// --------------------------------------------------------------------------
// Stop gating
// --------------------------------------------------------------------------
TEST(WebConfigBatch, StopFinalizesOnlyWhenNoHandlersAreInFlight) {
EXPECT_EQ(Batch::StopAction::Finalize,
Batch::stopStep(/*refs=*/0, /*warned=*/false, /*warn_at=*/50000, /*now=*/60000));
}
TEST(WebConfigBatch, StopWarnsOnceAfterTheDeadlineThenKeepsWaiting) {
const uint32_t warn_at = 50000;
// Before the warn deadline with handlers in flight: just wait.
EXPECT_EQ(Batch::StopAction::Wait, Batch::stopStep(2, false, warn_at, warn_at - 1));
// At the deadline, not yet warned: warn once.
EXPECT_EQ(Batch::StopAction::Warn, Batch::stopStep(2, false, warn_at, warn_at));
// Already warned: keep waiting (never warns again, never forces teardown).
EXPECT_EQ(Batch::StopAction::Wait, Batch::stopStep(2, true, warn_at, warn_at + 100000));
// An unscheduled warn timer never warns.
EXPECT_EQ(Batch::StopAction::Wait, Batch::stopStep(2, false, 0, 999999));
}
// --------------------------------------------------------------------------
// Wrap-around guard shared with the production _reboot_at assignments
// --------------------------------------------------------------------------
TEST(WebConfigBatch, ScheduleAtNeverReturnsTheUnscheduledSentinel) {
EXPECT_EQ(1000u + 3000u, Batch::scheduleAt(1000, 3000));
// A deadline that lands exactly on 0 is bumped to 1 so it still means "set".
const uint32_t just_below_wrap = std::numeric_limits<uint32_t>::max(); // +1 wraps to 0
EXPECT_EQ(1u, Batch::scheduleAt(just_below_wrap, 1));
}
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}