Custom floats precision per field?

[rycee]

It is currently possible to set a float precision for a point, but perhaps it would be nice to be able to specify the precision per field?

Something like

point.addField("foo", PrecisionedFloat{0.1234, 2})

This could then be serialized in the std::visit using something like

[&convert](PrecisionedFloat v) { convert << std::setprecision(v.precision) << std::fixed << v.value; }

[offa]

Are there any docs regarding this on the Influxdb side?

[rycee]

I think this is entirely an influxdb-cxx side feature. It is already possible to set a desired precision for float fields sent to InfluxDB, this seems to have been introduced in 317eb29.

The current solution is a little limited, however, in that the precision is set for the whole Point structure, making it difficult to choose different precisions for different float/double fields. My suggestion is to introduce a new simple data structure that simply stores a pair, the double value and the desired precision. I called this PrecisionedFloat in my original post for lack of a better name.

For example, if you want to write a point containing air temperature with a single decimal and air pressure with three decimals then it would be possible to do something like this:

double temp = GetTempFromSensor();
double press = GetPressureFromSensor();
influxdb::Point{"met"}
  .addField("temp", influxdb::PrecisionedFloat{temp, 1})
  .addField("press", influxdb::PrecisionedFloat{press, 3});

No change in existing behavior would be necessary, that is, if you call addField with a double or float you get the default precision as before.

[offa]

The current solution is a little limited, however, in that the precision is set for the whole Point structure, making it difficult to choose different precisions for different float/double fields.

I'm not sure if it was a good idea to at this feature actually. As far as I know there's a precision only for timestamps, but not for doubles. So no matter what precision is used on client side, eventually it ends as a 64 bit floating type in the DB.

Adding precisions to types, not supported on the DB side, somewhat adds business logic to the client.

Using your example you maybe even could write:

double temp = GetTempFromSensor();
double press = GetPressureFromSensor();
influxdb::Point{"met"}
  .addField("temp", PrecisionedFloat{temp, 1}) // where PrecisionedFloat converts to double
  .addField("press", toPrecision(press, 3)); // or as a function

❓ Regardless, I would be interested to know whether floatsPrecision is really useful and how widespread its use is.