RLE Column returns Infinity for large floating point numbers about nemo HOT 2 CLOSED

3.4028235e38 looks (at least) suspiciously close to f32::MAX (due to rounding on parsing the float value, it might actually even be equal to it). We might want to discuss whether we want to allow non-integer increments at all, since there might also be issues with non-exactly representable values such as 0.1, which, over the run of a full column, might accumulate quite some inaccuracy.

from nemo.

Some checks for this are done during construction of the RLE column, i.e. it is checked that the original values can be constructed again using the increment stored in the RLE. If this is not possible then a new RLE "block" is added.
Inaccuracies may still occur in the get function since the increment is just multiplied by the target index here. This is also what causes the problem above.

After our last discussion I'm also wondering if we need a RLE for floating point numbers at all or if we just need to support u64 anyway (see #38). I will discuss this offline before trying to fix this bug.

from nemo.