How long can the cables be in practice ?
what is your experience with that?
A customer used a couple of meters of CAT5E cable and did not reported any problem. I think the overall length is something dependent to a lot of factors like noisy environment and so on. Just imagine that the communication does not use differential signals but TTL level signals... so I expect to have no more than "some meters".
Reducing the bus speed will help to reduce the communication errors could be inferred by cable length and/or noise.
The averaged formats, as there are 8 measurements, how many digits precission is there?
Internally are counted as float. But the result is exported with 1 precision digit. Having more than one digit seems not significant because sensors could not assure so high precision.
does it return an int or a float
We use three bytes to store the result in a packet transfer. It's a proprietary format used to optimize the communication between the shield and Arduino.
The provided library is able to decode to int (for raw datas), float (for averaged datas) and plain ASCII text.
Is it possible to get a " lowpass " value of the measurements?
Yes, This could be done at the Arduino side reading the full raw measures (provided as number between 0 to 1024) instead of the averaged ones. The shield buffer contains at maximum 8 samples for each channel (temperature, pressure, humidity) a user could read bypassing any averaging performed by the shield.
Reading the raw values allows users to calculate whatever he need. The sample buffer is a "circular" buffer. User could read any of the samples or could ask for "the latest one".
Just a quick note: the averaged values are "temperature compensated" by the shield following the sensor datasheet specifications. Temperature compensation is not applied to raw samples. For this reason you can have some differences between straight average calculation from raw values and the results provided by the shield if you don't compensate your values.