Devices

espos_devices is whole devices rather than parts: a class that owns its nodes and wires them in its constructor, so a firmware says what the thing is rather than how it is plumbed.

#include "espos_devices.hpp"

espos::flow::Graph g;

espos::devices::TankLevel fresh(g, {
    .id = "fresh", .gpio = 4,
    .path = "tanks.freshWater.0.currentLevel",
    .empty_ohms = 190.0f, .full_ohms = 3.0f,
});
fresh.start();

The same tank wired by hand is four nodes and five edges, and the wiring is the part a firmware gets wrong: an ADC emitting volts into a transform expecting ohms gives a tank that reads plausibly and is wrong all season.

This is the shape SensESP asked for in #264 and never had.

Nothing is hidden

Every node stays reachable — analog(), level(), output() — because a device class that cannot be taken apart is a worse deal than the wiring it replaced. Add a filter, retitle a parameter, wire a second consumer:

auto& avg = g.make<Ema<float>>("smooth", 0.2f);
fresh.ohms() >> avg >> fresh.level();     // insert a filter mid-chain
fresh.level() >> my_alarm;                 // a second consumer

What each one publishes

Device	Publishes	The part that is easy to get wrong
TankLevel	tanks.*.currentLevel (ratio 0–1)	volts → ohms → ratio is three conversions, not one
EngineRpm	propulsion.*.revolutions	revolutions per SECOND, not RPM
EngineHours	propulsion.*.runTime	seconds, and it does not persist on its own
SmartSwitch	electrical.switches.*.state	the device must publish before the server will route a PUT to it
BilgeSwitch	the same, plus a notification	a float switch bobs; without debounce the alarm cries wolf
OneWireTemperature	environment.*.temperature (kelvin)	pass the ROM address, never an index

Three things worth reading twice

Signal K wants SI units. revolutions is revolutions per second, not RPM; temperatures are kelvin; tank levels are a ratio, not a percentage. Publishing 2400 where the server expects 40 looks right on a gauge that happens to scale it and wrong everywhere else. Each device does its conversion once, here, rather than leaving a magic constant in every firmware.

A switch has to publish before it can be switched. signalk-server routes a PUT by the (path, $source) pairs it has seen a device publish, so a switch that only listens is one the server answers 405 for — the request never reaches the device. SmartSwitch publishes its initial state in its constructor for exactly this reason.

EngineHours does not survive a reboot by itself. RunHours has no config persistence: it counts from zero at every start. Persist total_s() and restore it with set_total_s() at boot, or accept that it resets. Said plainly because an hour meter that silently resets is worse than none — it reads plausibly.

1-Wire is opt-in

OneWireTemperature exists only when CONFIG_ESPOS_SENSORS_ONEWIRE is on, and that needs two lines in your manifest, because the component manager resolves dependencies before Kconfig exists and every firmware would otherwise download a driver it has no sensor for:

dependencies:
  espressif/ds18b20: "^0.3.1"

The bus is yours, not the device's: one wire carries many sensors and each conversion is a bus-wide operation, so a device that owned its bus would give you one GPIO per thermometer.

espos::sensors::OneWireBus<4> bus(4);
espos::devices::OneWireTemperature engine(g, bus, {
    .id = "engrm", .addr = 0x28FF1234567890AB,
    .path = "environment.inside.engineRoom.temperature",
});
bus.start(5000);          // one cycle drives every sensor on the wire

What is not here yet

BatteryMonitor (INA219), Environment (BME280 + dew point), StatusLed and SystemButton need drivers or registry components that do not exist in this tree yet. They are the same shape when they arrive; nothing above changes.