fix(77): restore the working exit route, ship ameba v1.3.2

Re-includes swing in the remote-activity exit check on both halves. Removing it was a
regression I introduced while fixing the self-cancel.

What happened: the entry gesture is "Horizon Airflow x6" -- the swing button -- so swing changes
kept arriving for a beat after the window opened and instantly cancelled it (open 95424 ms,
killed 95674 ms, every attempt). That got fixed TWO ways at once: exclude swing, and add a 6 s
settling grace. The grace alone is sufficient.

Excluding swing on top of it removed the exit route that was actually doing the work. Pressing
the pattern again is a SWING press, and this A/C appears to emit its 0x20 pulse only on ENTRY --
so the "toggle" path never fires and, with swing excluded, nothing got out of "77" except the
180 s expiry. Confirmed on both nodes by the user.

Note the earlier "all the exits work" bench result was therefore measuring the swing path, not
the toggle. The toggle stays in (harmless, and correct if a unit does pulse on exit) but must not
be relied on.

Ameba 1.3.1 -> 1.3.2 (int 10302), flashed to node 14 over Matter OTA and verified; esp32 rebuilt
and flashed over USB. Both nodes avail=True, endpoints {0..10}.

Known remaining oddity, filed separately rather than patched late: pressing the swing button to
exit briefly shows the setpoint and then returns to "77". Most likely our exit fires on the swing
change while the A/C emits another 0x20 in the same interaction, so we immediately re-open. Power
and setpoint buttons exit cleanly.

Assisted-by: AI

Author: AndrewDemsDS

firmware/esp32-matter/main/app_main.cpp

@@ -525,19 +525,22 @@ static void on_status(const HisenseState *st)
      * the unit from Matter -- and if something is, the user is plainly not mid-pairing. */
     if (recommission_window_is_open() && s_status.valid &&
         esp_timer_get_time() > s_recommission_grace_us) {
-        /* SWING IS EXCLUDED ON PURPOSE. The gesture that ENTERS "77" is "Horizon Airflow x6" --
-         * i.e. the swing button. Including vswing/hswing here made entering the mode instantly
-         * exit it: measured window-open at 95424 ms, killed at 95674 ms, every single attempt.
-         * The entry gesture can never be the exit signal.
+        /* Swing IS included, and the grace period above is what makes that safe.
          *
-         * The grace period covers the same trap from the other side: the swing changes from the
-         * final presses keep arriving for a beat after the window opens, and any of the other
-         * fields could be mid-settle too. */
+         * History worth keeping: the entry gesture is "Horizon Airflow x6" -- the swing button --
+         * so swing changes arrive for a beat after the window opens and briefly self-cancelled it
+         * (open 95424 ms, killed 95674 ms). The first fix excluded swing AND added the grace. The
+         * grace alone covers the settle; excluding swing on top of it removed the exit route that
+         * was actually working, because pressing the pattern again is a SWING press and this A/C
+         * appears to emit its 0x20 pulse only on ENTRY. Result: nothing exited "77" but expiry.
+         *
+         * So: keep swing, rely on the grace. Stock exits on any button, and swing is a button. */
         const bool user_touched =
             st->power_on   != s_status.power_on   || st->mode      != s_status.mode      ||
             st->setpoint_c != s_status.setpoint_c || st->fan_raw   != s_status.fan_raw   ||
             st->eco_on     != s_status.eco_on     || st->turbo_on  != s_status.turbo_on  ||
-            st->mute_on    != s_status.mute_on    || st->sleep_raw != s_status.sleep_raw;
+            st->mute_on    != s_status.mute_on    || st->sleep_raw != s_status.sleep_raw ||
+            st->vswing_on  != s_status.vswing_on  || st->hswing_on != s_status.hswing_on;
         if (user_touched) {
             ESP_LOGW(TAG, "A/C driven from the remote during the \"77\" window -> treating as EXIT");
             chip::DeviceLayer::PlatformMgr().ScheduleWork(recommission_user_cancel, 0);

firmware/src/sdk-edits/matter_drivers.cpp

@@ -191,16 +191,20 @@ static void hisense_on_status(const HisenseState *state)
     /* Stock behaviour: ANY remote button exits "77", not just pressing the pattern again. We
      * cannot see IR, but every press lands on the bus as a change to a user-settable field.
      *
-     * SWING IS EXCLUDED ON PURPOSE -- the gesture that ENTERS "77" is "Horizon Airflow x6", i.e.
-     * the swing button, so including vswing/hswing makes entering the mode instantly exit it
-     * (measured on the esp32 half: window open at 95424 ms, killed at 95674 ms, every attempt).
-     * Temperatures / compressor / current are excluded too: they drift every frame. */
+     * Swing IS included, and the grace period is what makes that safe. The entry gesture is
+     * "Horizon Airflow x6" -- the swing button -- so its changes arrive for a beat after the
+     * window opens and briefly self-cancelled it on the esp32 half (open 95424 ms, killed 95674
+     * ms). Excluding swing on top of the grace removed the exit route that was actually working:
+     * pressing the pattern again is a SWING press, and this A/C appears to emit its 0x20 pulse
+     * only on ENTRY, so the toggle never fires and nothing but expiry got out of "77".
+     * Temperatures / compressor / current stay excluded: they drift every frame. */
     if (recommission_window_is_open() && s_status.valid && recommission_grace_expired()) {
         const bool user_touched =
             state->power_on   != s_status.power_on   || state->mode      != s_status.mode      ||
             state->setpoint_c != s_status.setpoint_c || state->fan_raw   != s_status.fan_raw   ||
             state->eco_on     != s_status.eco_on     || state->turbo_on  != s_status.turbo_on  ||
-            state->mute_on    != s_status.mute_on    || state->sleep_raw != s_status.sleep_raw;
+            state->mute_on    != s_status.mute_on    || state->sleep_raw != s_status.sleep_raw ||
+            state->vswing_on  != s_status.vswing_on  || state->hswing_on != s_status.hswing_on;
         if (user_touched) {
             ChipLogProgress(DeviceLayer, "A/C driven from the remote during \"77\" -> EXIT");
             chip::DeviceLayer::PlatformMgr().ScheduleWork(recommission_user_cancel, 0);

firmware/src/version.txt

@@ -1 +1 @@
-1.3.1
+1.3.2