Float parser that sacrifices nothing.

• Accurate. It finds the "best" approximation to the input just like the standard library, strconv. Fuzzing tests in addition to standard library tests and parse-number-fxx-test-data are actively done.

• Compatible. Basically, it is an improvement on ParseFloat in the standard library, and the usage is exactly the same.

• Fast. Faster than the standard library on benchmarks with normally distributed floats and bitwise uniform random float inputs. For more information, benchmark it yourself or see below.

go get github.com/sugawarayuuta/refloat

gantt
title norm = normally distributed, bits = bitwise uniform (ns/op - lower is better)
todayMarker off
dateFormat  X
axisFormat %s
section bits 64bit 
strconv: 0,106
refloat: 0,91
section norm 64bit 
strconv: 0,92
refloat: 0,65
section bits 32bit 
strconv: 0,79
refloat: 0,80
section norm 32bit 
strconv: 0,65
refloat: 0,49

There are articles written in English and Japanese.

The icon above is from gopher-stickers by Ueda Takuya.

Awesome talk and inspirational algorithm by Daniel Lemire.

Sollya is used for precomputing the polynomial for approximations.