This is a C# implementation of Homie Convention 4.0.0.
Homie Convention targets MQTT transport and adds an application-level protocol to standartize ways of how devices talk to each other, since MQTT doesn't really have any such facilities. Also, Homie emphasizes discoverability, so, for example, if a new device joins the network, any controller could immediatelly figure out its properties and capabilities.
Apparently, you need an MQTT broker first. Mosquitto is probably the most popular free one. It run on anything, but running it on NAS in a Docker container is probably the most convenient option.
Then, check out the TestApp
. It has multiple simulated devices, like AirConditioner
, which not only have some Homie properties implemented, but they also simulate real-world counterparts, so you could, for example, set the target temperature and see how it actually slowly changes. You can add these devices to OpenHAB
, HomeAssistant
or other home automation platforms and start clicking button right away.
Of course, if developing new devices, you need some better debugging tools. To watch raw MQTT traffic, check out MQTT explorer. To quickly check the controls you created, try HoDD (although it only works if downloaded and run from the PC directly). If you have Docker installed, use BHD
YAHI is fully-Homie compliant, however, it is a slightly more strict subset.
YAHI defines three distinct types of properties: States
(non-settable, retained), Commands
(settable, non-retained) and Parameters
(settable and retained). Homie allows a fourth type (non-settable, non-retained), which I found redundant and very weakly defined, so it has not made it into YAHI. But rest assured, you don't need it at all.
States have read-only access, and is something that user cannot change directly. It is usually a measured or calculated value, like, for example, an ambient temperature. Reading a state multiple times may result in different values.
Commands have write-only access, and, as the name suggests, sending a command invokes some action. For example, "turn on" would be a command. It is possible to send the same command multiple times, but it is up for the device whether to execute it or not (like, one can't really turn on a device when it is already on).
Parameters have read/write access, and are used in calculations or for defining behaviour. For example, "target temperature" of an A/C unit. Important implementation thing about parameters is that they can only be changed by user. Device cannot alter parameter value by itself, it has to be either directly set by a user, or via indirect command like "reset to factory defaults", which is still invoked manually. Rreading a parameter multiple time should always return last set value.
In short: duration
, integer
, percent
and boolean
data types aren't supported, or are supported partially, by design.
percent
is a very confusing data type. It requires $unit
field to be set to "%", and the payload can be either an integer
or a float
. That itself creates enough confusion and parsing problems. However, one may also define percent payload using casual integer
or float
, so we get three ways to define percentage payload:
$datatype=percent
,$unit="%"
$datatype=float
,$unit="%"
$datatype=integer
,$unit="%"
So, support for percent
data type was dropped, because the same can be achieved using casual number data types.
Support for boolean
was dropped because it is just a subset of enum
data type with possible values being true
and false
. Also, usage of boolean
type feels very unnatural. Homie IOT Convention is quite a humanly one, but humans do not speak in true
and false
. If there's a question "is the light switch turned on?", nobody answers "false"; natural answers would be "no", "nope", "it is not" and similar.
This is also problematic for frontends and GUIs. Systems like OpenHAB are generic ones; boolean
data type had two predefined states negating each other, so that begs to render a simple visual switch. But switch positions must have labels of some sort, and boolean
data type forces to use "true" and "false", resulting in a very unnatural visual representation. However, one may hardly find such "true/false" switch in reality. It is always "on/off", "yes/no", maybe even "1/0", but not "true/false".
So for these reasons, boolean
had to go, and is only partially supported on client side. Even then, it is silently converted to a enum
property.
This brings us to integer
and float
data types. These, just as boolean
, comes from computer science and is meant for computers, not humans. Humans do not use integers of floats, they use decimal numbers. Having to choose between an integer
or float
again present confusion to Homie implementers and consumers: so which one to pick? It is not always obvious. Temperatures usually benefit from having a decimal place or two, as change of 1°C is pretty significant even for humans (although there are plenty of sensors that return temperature in whole numbers, too). How about relative humidity? It is a positive number, with meaningful values below 100. Does it need decimal places, or is it fine to go with whole numbers? 1% change in relative humidity certainly doesn't matter much for humans, but are there any use cases needing to measure humidity more precisely? What if a sensor only returns whole numbers (which is pretty common), and implementer creates a integer
Homie property, but later sensor is upgraded to more precise one? Precision is then lost because switching to float
data type would result in serious breaking changes?
It is actually hard to think of any real world application for integer
data type that is not imposed by computer science. Besides, Homie Iot defines float
data type as double precision, and doubles can represent exact integers from −2⁵³ to 2⁵³, which is more than enough for the application that Homie is targeting.
Due to these considerations, support for integer
was dropped and any host number property created by YAHI will have data type set to float
. integer
is still partially supported on client side by silently converting it to a float
property.
As for duration
, it is a complicated data type. For starters, Homie asks for ISO 8601, which has a lot of different representations, but also kind of limits to "PTHHMMSS" format, already creating confusion. Parsing this type is really difficult, and ensuring compatibility between different producers and consumers is even more so.
Then, "PTHHMMSS" is ambiguous by itself, as there are multiple ways to define the same duration. For example, "PT2H", "PT120M", "PT1H59.5M30S are equal values, representing a time span of 2 hours. That by itself is a great source of confusion.
Given that Homie 4.0.0 defines duration only as hours+minutes+seconds, it is arguably much better to just go with a floating point number, "2.0" for example, which is very easy to perceive, parse, consume and compare.
So for those reasons, and also because duration
data type is pretty rare (I haven't encountered it in any of my 30+ devices that I own), YAHI does not support duration
neither on host, nor on client side. It might support client side in the future, though.
So you've got a new device and want to make it talk to Homie broker. Here are some best practices of how to do that.
Anything that is measured becomes a state, no exceptions. For example: temperature, humidity, illuminance, wind speed, power consumption, etc.