Selection guide

Picking a schema

Start with inference. Fate.log("pose", myObject) builds a ReflectedClassSchema over the public properties, and for most robot data that is all you need.

Use CustomStructSchema when you want fewer bytes or different ones. You can drop fields you won't plot, round a double, or convert units as you log. You give it the component names, their schemas, and a function that pulls the values out. Pass it to log(name, value, schema), or register it so plain log finds it.

TranslatedSchema is the smaller version of the same idea. It converts one value to another type and encodes that, so a Rotation2d can go into the log as a plain double.

What costs what

Primitive schemas are the cheap case. Their sizes are fixed, so nothing has to be measured before writing. An Int takes 4 bytes, a Long or Double 8, a Boolean 1, and a String a 4-byte length prefix plus its UTF-8 bytes.

ReflectedClassSchema looks up its property accessors once per class and caches them. It still reads each property through reflection on every log. A CustomStructSchema reads them directly.

objSize and encodeObject both run on every log, so a CustomStructSchema encoder is called twice per logged value. Keep it to field reads and arithmetic. At 50 Hz, anything slower adds up.

Register custom schemas in init() rather than mid-match, so the reflection happens before the match starts.

Registering schemas

FateSchema.registerSchema overwrites any previous entry for the same class and gives no error. A second registration changes the schema for every path first logged after it. Register each schema once, early in initialization.

Use AS_TYPE

A class with an AS_TYPE constant gets that name recorded in the log next to its fields. FateWeaver doesn't require it. It gives viewers and your own analysis scripts something to key on besides the path.