paupino / rust-decimal

Decimal number implementation written in pure Rust suitable for financial and fixed-precision calculations.
https://docs.rs/rust_decimal/
MIT License
972 stars 177 forks source link

Scaling string representation differs from dividing by 10^x #627

Open Zarathustra2 opened 9 months ago

Zarathustra2 commented 9 months ago

Hi there,

#[test]
fn scaling_issue() {
    let num = 3850;
    let dec = Decimal::from(num) / Decimal::ONE_THOUSAND;

    let mut dec_scale = Decimal::from(num);
    dec_scale.set_scale(3).unwrap();

    // Passes
    assert_eq!(dec, dec_scale);
    // Fails
    assert_eq!(dec.to_string(), dec_scale.to_string());
    // thread 'scaling_issue' panicked at 'assertion failed: `(left == right)`
    //  left: `"3.85"`,
    // right: `"3.850"`', tests/decimal_tests.rs:4818:5
    // note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

    // dec 3.85 flags 131072 hi 0 lo 385 mid 0
    // dec_scale 3.850 flags 196608 hi 0 lo 3850 mid 0
}

Is this intended behaviour? If so is there a way to make scaling behave the same way as if I would divided by any 10^x number so that when I convert it to String I don't have any trailing zeros?

Best regards, Dario

Tony-Samuels commented 9 months ago

You want the normalize[1] or normalize_assign[2] functions, to strip the extra 0.

[1] https://docs.rs/rust_decimal/latest/rust_decimal/struct.Decimal.html#method.normalize [2] https://docs.rs/rust_decimal/latest/rust_decimal/struct.Decimal.html#method.normalize_assign

Zarathustra2 commented 9 months ago

Ah awesome!

Should the docs of set_scale maybe be adjusted so that future users know about it?

paupino commented 9 months ago

Thanks @Tony-Samuels for your reply - exactly right.

@Zarathustra2 - I think there is always an opportunity to make the documentation clearer. In this case, would it help to describe what set_scale is doing under the hood? i.e. we maintain the same mantissa, but just change where the decimal point is - hence retaining precision without rounding. e.g. in this example 3850 is having the decimal point moved to the left 3 places.

Zarathustra2 commented 9 months ago

@paupino Yes some clearer docs would be nice. Maybe we can also add a code example directly to the docs as example. I personally think a coding example with comments is more clear than plain text.

    let num = 3850;
    let dec = Decimal::from(num) / Decimal::ONE_THOUSAND;

    let mut dec_scale = Decimal::from(num);
    dec_scale.set_scale(3).unwrap();

    // Passes since same number
    assert_eq!(dec, dec_scale);
    assert_eq!(dec.to_string(), "3.85".to_string());
    // set_scale just moves the decimal point and keeps original mantisse. 
    // this effects the string representation but both dec & dec_scale represent
    // the same number
    assert_eq!(dec_scale.to_string(), "3.850".to_string());
    assert_eq!(dec_scale, dec);
    assert_ne!(dec_scale.to_string(), dec.to_string());
    // Use normalize to normalize the mantisse
    assert_eq!(dec_scale.normalize().to_string(), "3.85".to_string());