Float64 conversion

[dcullender-cb]

When converting to float64 using Float64(), the value can differ from the input string while shopspring/decimal seems to produce the correct result. for example

Appreciate that the value cannot fit into a float64 as per ericOk but curious why the shopspring value matches?

[ericlagergren]

This is because shopspring converts their decimal to a big.Rat, then converts the Rat to a float64. My library converts as-is using float64.

The upside to shopspring’s method is that it won’t round worse than the decimal’s current precision during the conversion. The downside is that converting to a rational number is expensive: you end up with huge numerators and denominators. Plus, you have to rescale the decimal to create the rational number.

The upside to my method is it’s fast and allocation-free (for small floats, that is). The downside is you occasionally have rounding issues.

There might be a better way of performing the conversion (more bit twiddling instead of float division/multiplication), but I haven’t looked too closely at it. I’m not entirely sure why people are using the Float64 method—it seems to defeat the purpose of using a decimal library.