release: 1.17.6a-zephcore pre-release

Bump the firmware version to 1.17.6a-zephcore and move the release
notes to RELEASE_NOTES_1.17.6a-zephcore.md, which the release workflow
publishes as the release body. The workflow's prerelease switch is
already on, so the configurator keeps offering 1.17.5c.

Notes corrected against the commits since 1.17.5c: the light-sleep
section now names the boards that sleep by default, explains the
1.17.1 DIO1 regression correctly (the button's EXT1 polarity disarmed
DIO1 on the Heltecs; the XIAO and Station G2 never had light sleep),
describes the pad holds and the USB console behaviour after the
console window, and says what the powersaving migration really does.
Added what was missing: get pm, GPS UART suspend on boards without a
power line, the optional XIAO nRF54L15 GPS, poweroff/shutdown and
get pwrmgt.*, the flush before reboot, the XIAO ESP32-S3 and Station
G2 button fixes, and a pre-release notice asking Heltec users for
light-sleep reports.

releasenotes/RELEASE_NOTES_1.17.6a-zephcore.md

@@ -1,11 +1,17 @@
-# ZephCore 1.17.6-zephcore
+# ZephCore 1.17.6a-zephcore
 
 A large release: companion firmware now speaks WiFi (TCP, on boards that support it), preferences
 move to upstream's `prefs.json` format, and ESP32-S3 repeaters can finally light-sleep. Underneath,
 the mesh core, radio drivers and companion protocol handling were rebuilt against upstream MeshCore's
 text so the two codebases read the same again. Also: GPS reliability fixes, more accurate battery
-reporting, reduced flash wear, and the remote CLI (protocol v14) now executes commands from admin
-contacts, matching upstream.
+reporting, reduced flash wear, and the remote CLI (protocol v14) now executes commands sent by
+contacts you have allowed to use it, matching upstream.
+
+> [!IMPORTANT]
+> **This is a pre-release.** The Mesh America configurator keeps offering 1.17.5c; flash this one by
+> hand. ESP32-S3 light sleep was bench-validated on a XIAO ESP32-S3; on the Heltec boards that
+> enable it by default it is build-tested only, so reports from Heltec repeaters are especially
+> welcome (`get pm`, see below).
 
 > [!NOTE]
 > A normal upgrade keeps your identity, settings, contacts and phone pairing. Preferences are
@@ -41,37 +47,58 @@ Three related bugs were fixed along the way:
 - A repeater's GPS duty cycle, deliberately set to 300 seconds, was being silently rewritten back to
   48 hours on every boot. That migration path now only runs against legacy prefs files.
 
-## ESP32-S3 repeaters can light-sleep again
+## ESP32-S3 repeaters now actually light-sleep
 
-Light sleep on S3 repeaters has never actually worked, for two independent reasons, both now fixed:
+Repeater builds for the Heltec WiFi LoRa 32 V3, V4 and V4.3 and the Wireless Tracker and Wireless
+Tracker V2 have shipped with light sleep enabled, but it never worked, for two independent reasons,
+both now fixed:
 
 - Every S3 repeater carries WiFi OTA, and the WiFi driver started WiFi at boot — which holds a modem
   power-management lock for as long as WiFi is up. Measured on a XIAO: zero sleep entries with WiFi
   compiled in. WiFi now stops at boot on power-management builds, and starts again on `start ota` or
   a station connect.
-- The radio interrupt line (DIO1) was never a real wake source. Since 1.17.1, boards without a
-  hardware RTC pad on DIO1 (XIAO GPIO39, Station G2 GPIO48) couldn't wake from sleep on an incoming
-  packet at all. DIO1 and the user button are now armed as proper level-triggered GPIO wakes, and the
-  interrupt a sleep swallowed is re-raised afterwards so the radio and button handlers still run.
+- The radio interrupt line (DIO1) could not wake the chip. The wake mechanism used (ESP32 "EXT1")
+  allows only one trigger polarity for all its pins; since 1.17.1 the active-low user button was
+  armed on it as well, and that silently disarmed the active-high DIO1. DIO1 and the user button are
+  now per-pin level wakes that work on any pin, and the interrupt a sleep swallowed is re-raised
+  afterwards so the radio and button handlers still run. Boards with DIO1 on a non-RTC pin (XIAO
+  ESP32-S3, Station G2) can now light-sleep too; they are not switched on by default yet.
 
-Verified on a XIAO ESP32-S3 repeater against a T1000-E on the same channel: asleep 96.6% of uptime,
-packet-for-packet parity with the awake baseline.
+The pins the radio depends on (chip select, reset, RF switch, FEM enable and supply) are now held at
+their level through each sleep: light sleep otherwise lets every unheld pin float while the radio
+keeps receiving.
+
+Verified on a XIAO ESP32-S3 repeater on a live mesh against a T1000-E on the same channel: asleep
+96.6% of uptime, 119 packets heard vs 119, and transmit counters correct.
 
 > [!NOTE]
-> **Nothing to change on your side.** `powersaving` (already an upstream command) gates this and
-> defaults to on for light-sleep-capable builds; a previously-stored "off" is respected. Companion
-> light sleep is still off by default — it isn't ready to ship yet.
+> **Nothing to change on your side.** `powersaving` (already an upstream command, stored but ignored
+> until now) gates light sleep. It defaults to on for light-sleep builds, and the "off" older firmware
+> stored without acting on it is read as on once. `powersaving off` keeps the node awake and is kept
+> across reboots. `get pm` shows how much of the time the node sleeps and what wakes it.
+>
+> **USB console:** for 10 minutes after boot, and again after every press of the user button, the node
+> stays awake and the USB console works normally. After that, characters typed at a sleeping node are
+> lost, and on boards whose console is the chip's own USB port (V4, V4.3, Wireless Tracker V2) the USB
+> device disconnects and reconnects as the chip sleeps and wakes. Press the button, or send
+> `powersaving off` over LoRa, to get the console back. Remote admin over LoRa is unaffected.
+>
+> Companion light sleep is still off by default — it isn't ready to ship yet.
 
 ## Less flash wear from contact updates and admin logins
 
 The companion was marking its contact list dirty on every advert from a node it *declined* to
-auto-add (full table, hop limit, wrong type) — forcing a full rewrite of the contact store roughly
-every 5 seconds on a busy mesh, even though nothing was actually added. It now only dirties on a real
+auto-add (full table, hop limit, wrong type) — each such advert forced a full rewrite of the contact
+store 5 seconds later, even though nothing was actually added. It now only dirties on a real
 addition or a changed name, type or position; a plain re-advert just extends the node's liveness
 deadline (now 1 hour instead of 10 minutes) without touching flash.
 
 Repeaters and room servers likewise stopped rewriting their access-control list on every admin login
-— it's now saved only when the role or secret actually changed.
+— it's now saved only when the role or secret actually changed — and a preferences save that would
+write exactly what is already stored is skipped.
+
+Pending writes are now flushed before every clean reboot and power-off (CLI, UI, bootloader and OTA
+paths), not only before the companion app's reboot command.
 
 ## Remote CLI now executes on the device (protocol v14)
 
@@ -84,8 +111,14 @@ the command types without executing anything sent over the air.
 
 - Servers now honour `gps_enabled` at boot — it was previously lost across a reboot. `zc.gps_set`
   upgrades older stored prefs to "on" once, so this doesn't silently turn GPS off for anyone.
-- GPS can be turned off on boards with no power-control line, by putting the UART to sleep instead.
-- The sleep lock is held only while a fix is being acquired, not indefinitely.
+- GPS can be turned off on boards with no power-control line, by sending the module its sleep
+  command over the UART instead.
+- On nRF boards the GPS UART now also stops listening whenever the GPS is off or in standby on boards
+  without a power-control line (RAK4631, RAK3401 1W, GAT562 30S, R1 Neo, XIAO nRF54L15), as it already
+  did on the others. Tested with 900 suspend/resume cycles into a streaming GPS.
+- XIAO nRF54L15: an optional GPS can be wired to the header serial pins (D6 = TX, D7 = RX, 9600 baud
+  NMEA) and works with the release firmware.
+- On ESP32, light sleep is held off only while a fix is being acquired, not indefinitely.
 - The observer role now parks its GPS instead of leaving it running unconfigured.
 
 ## Battery and power reporting
@@ -95,6 +128,8 @@ the command types without executing anything sent over the air.
 - Every companion, including headless ones with no display, now has low-battery auto-shutdown with a
   `v`-contact alert beforehand, and an `autoshutdown` CLI command to control it. Below 2000 mV counts
   as "no battery installed" rather than critically low.
+- New upstream commands: `poweroff`/`shutdown` (on nRF, wake with the user button; elsewhere, a power
+  cycle) and `get pwrmgt.support`, `.source`, `.bootreason`, `.bootmv`.
 - `get pwrmgt.bootreason` no longer reports "Unknown" after a flash: ESP32 reset reasons that Zephyr's
   `hwinfo` driver maps to 0 (USB-Serial-JTAG reset, power glitch, CPU lockup, eFuse) are now labelled
   correctly.
@@ -121,6 +156,11 @@ partial ZephCore-specific decode.
 
 ## Also in this release
 
+- **XIAO ESP32-S3 (Wio-SX1262 kit): user button fixed.** GPIO21 is both the kit's button and the
+  XIAO's onboard LED, and the heartbeat LED kept driving the button pin. The heartbeat LED is gone on
+  this board; the button now has a pull-up and works.
+- **Station G2: user button polarity fixed** (it was read inverted, as active-high; upstream reads it
+  active-low).
 - **Radio drivers were unified.** SX126x, SX127x, LR1110 and LR2021 now share one `LoRaRadio` adapter
   over per-family operation tables instead of four separate classes. The LR1110 and LR2021 SDKs now
   compile from Semtech's LoRa Basics Modem source instead of a vendored fork.

zephcore/CMakeLists.txt

@@ -33,7 +33,7 @@ list(APPEND BOARD_ROOT ${CMAKE_CURRENT_SOURCE_DIR})
 # Service advertises (CONFIG_BT_DIS_FW_REV_STR, generated from it below), what the
 # GitHub release is tagged/named, and what the Mesh America catalog uses as its
 # version key. The release workflow reads this line directly — keep its form.
-set(ZEPHCORE_FIRMWARE_VERSION "1.17.5c-zephcore")
+set(ZEPHCORE_FIRMWARE_VERSION "1.17.6a-zephcore")
 
 # ========== Helpers ==========
 # Append FILE to the overlay list once, keeping its FIRST position. Position is