Unexpected float value when casted from f32 to f64

[comphead]

I tried this code:

    println!("value_64: {:?}", 128.2_f32 as f64);

I expected to see this happen: value_64: 128.2

Instead, this happened: value_64: 128.1999969482422

Meta

rustc --version --verbose:

Tested on 1.78 Stable, but can see this also on 1.80 nightly. Not sure about earlier versions

Backtrace

<backtrace>

[saethlin]

This is not a bug. You can see the same behavior with slightly different rounding from C:

#include <stdio.h>
int main() {
    float f = 128.2;
    printf("%f\n", (double)f);
    return 0;
}

will print

128.199997

The lossy step is the conversion of 128.2 from base 10 to base 2. The base-2 representation has the equivalent of a repeating decimal.

[the8472]

Not a bug

    println!("value_32: {:.20?} bits: {:b}", 128.2_f32, 128.2_f32.to_bits());
    println!("value_64: {:.20?} bits: {:b}", 128.2_f32 as f64, (128.2_f32 as f64).to_bits());

prints

value_32: 128.19999694824218750000 bits: 1000011000000000011001100110011
value_64: 128.19999694824218750000 bits: 100000001100000000001100110011001100000000000000000000000000000

This is just the printing algorithm selecting a shorter number of the many reals that map to to that float bit pattern

[the8472]

If only we had hex and binary formatting for floats.

[saethlin]

Make it happen! You have the power 💪