math/big: ParseFloat loses precision with trailing zeros after decimal

[josharian]

package main

import (
    "fmt"
    "math/big"
)

func main() {
    threes := [...]string{"3", "03", "3.", "3.0", "3.00", "3.000"}

    for _, three := range threes {
        f, b, err := big.ParseFloat(three, 10, 350, big.ToNearestEven)
        fmt.Println(three, f, b, err)
        fmt.Println(f.Format('b', 350))
    }
}

yields

3 3 10 EOF
0p-348
03 3 10 EOF
0p-348
3. 3 10 EOF
0p-348
3.0 3 10 EOF
1720123961992553633708115671476565205597423741876210842803191629540192157066363606052513914832594264915968p-348
3.00 3 10 EOF
1720123961992553633708115671476565205597423741876210842803191629540192157066363606052513914832594264915968p-348
3.000 3 10 EOF
1720123961992553633708115671476565205597423741876210842803191629540192157066363606052513914832594264915968p-348

Something similar is true of gc's constants; see my comments in #7740.

[griesemer]

This issue is unrelated to #7740.

This is simply a bug that crept in after a relaxation of the internal invariant that the mantissa is always at least as long as the precision. The conversion routine was making the assumption that this was still the case.

https://go-review.googlesource.com/5640