flexfloat 1.0.1

A Rust library for arbitrary-precision floating-point arithmetic with growable exponents and fixed-size fractions, extending IEEE 754 double-precision range.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148

//! Deterministic tests for FlexFloat operations at the boundaries of f64 range.
//!
//! These tests cover values where f64 itself overflows to ±∞ or underflows
//! to ±0; FlexFloat must remain finite and non-zero with a grown exponent
//! field, while preserving the sign.

use flexfloat::FlexFloat;
use flexfloat::prelude::*;

const EXTREME_VALUES: &[f64] = &[
    f64::MAX,
    f64::MIN_POSITIVE,
    f64::EPSILON,
    1e300,
    1e-300,
    1e308,
    5e-324,
    -f64::MAX,
    -f64::MIN_POSITIVE,
    -1e300,
    -1e-300,
    1.0,
    -1.0,
    0.0,
    -0.0,
];

fn classify_and_check_finite(name: &str, ff: &FlexFloat, f: f64) {
    if f.is_nan() {
        assert!(ff.is_nan(), "{name}: expected NaN, got {ff:?}");
        return;
    }
    if f.is_infinite() {
        assert!(
            !ff.is_infinite() && !ff.is_nan(),
            "{name}: f64 was inf but FlexFloat must stay finite: {ff:?}",
        );
        assert!(
            ff.exponent().len() > 11,
            "{name}: f64 overflow but FlexFloat exponent did not grow: {ff:?}",
        );
        return;
    }
    if f == 0.0 && !ff.is_zero() {
        assert!(
            ff.exponent().len() > 11 || ff.is_subnormal(),
            "{name}: underflow but result is neither grown nor subnormal: {ff:?}",
        );
        return;
    }
    // Check relative error entirely in FlexFloat arithmetic — no to_f64 conversion.
    let expected = FlexFloat::from(f);
    if ff != &expected {
        // rel = |ff - expected| / max(|expected|, MIN_POSITIVE)
        let diff = (ff.clone() - expected.clone()).abs();
        let denom = expected.abs().max(FlexFloat::from(1e-300_f64));
        let rel = diff / denom;
        assert!(
            rel < FlexFloat::from(1e-10_f64),
            "{name}: round-trip mismatch: ff={ff:?}, expected f={f:?}",
        );
    }
}

#[test]
fn add_at_f64_boundaries() {
    for &a in EXTREME_VALUES {
        for &b in EXTREME_VALUES {
            let ff_a = FlexFloat::from(a);
            let ff_b = FlexFloat::from(b);
            let result = ff_a + ff_b;
            let expected = a + b;
            classify_and_check_finite(&format!("{a:?} + {b:?}"), &result, expected);
        }
    }
}

#[test]
fn sub_at_f64_boundaries() {
    for &a in EXTREME_VALUES {
        for &b in EXTREME_VALUES {
            let ff_a = FlexFloat::from(a);
            let ff_b = FlexFloat::from(b);
            let result = ff_a - ff_b;
            let expected = a - b;
            classify_and_check_finite(&format!("{a:?} - {b:?}"), &result, expected);
        }
    }
}

#[test]
fn mul_at_f64_boundaries() {
    for &a in EXTREME_VALUES {
        for &b in EXTREME_VALUES {
            let ff_a = FlexFloat::from(a);
            let ff_b = FlexFloat::from(b);
            let result = ff_a * ff_b;
            let expected = a * b;
            classify_and_check_finite(&format!("{a:?} * {b:?}"), &result, expected);
        }
    }
}

#[test]
fn div_at_f64_boundaries() {
    for &a in EXTREME_VALUES {
        for &b in EXTREME_VALUES {
            if b == 0.0 {
                continue;
            }
            let ff_a = FlexFloat::from(a);
            let ff_b = FlexFloat::from(b);
            let result = ff_a / ff_b;
            let expected = a / b;
            classify_and_check_finite(&format!("{a:?} / {b:?}"), &result, expected);
        }
    }
}

#[test]
fn arithmetic_crossing_f64_overflow() {
    let max = FlexFloat::from(f64::MAX);
    let two = FlexFloat::from(2.0);
    let huge = max.clone() * two;
    assert!(
        !huge.is_infinite(),
        "MAX * 2 should stay finite, got {huge:?}"
    );
    assert!(
        huge.exponent().len() > 11,
        "MAX * 2 must grow exponent, got {huge:?}"
    );
}

#[test]
fn arithmetic_crossing_f64_underflow() {
    let tiny = FlexFloat::from(f64::MIN_POSITIVE);
    let half = FlexFloat::from(0.5);
    let smaller = tiny * half;
    assert!(
        !smaller.is_zero(),
        "MIN_POSITIVE * 0.5 should stay nonzero, got {smaller:?}"
    );
    assert!(
        smaller.exponent().len() > 11 || smaller.is_subnormal(),
        "MIN_POSITIVE * 0.5 must stay representable, got {smaller:?}",
    );
}