decimal-scaled 0.2.1

Const-generic base-10 fixed-point decimals (D9/D18/D38/D76/D153/D307) with integer-only transcendentals correctly rounded to within 0.5 ULP — exact at the type's last representable place. Deterministic across every platform; no_std-friendly.
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
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
//! Mathematical constants and float-compatibility constants for every
//! decimal width.
//!
//! # Constants provided
//!
//! The [`DecimalConsts`] trait exposes `pi`, `tau`, `half_pi`,
//! `quarter_pi`, `golden`, and `e` as methods on every width. The
//! native-tier (`D38` and narrower) impls live here; the wide tier
//! (`D76` / `D153` / `D307`) impls live in `consts_wide.rs`.
//!
//! Two inherent associated constants, `EPSILON` and `MIN_POSITIVE`, are
//! provided as analogues to `f64::EPSILON` and `f64::MIN_POSITIVE` so
//! that generic code parameterised over numeric types continues to
//! compile when `T` is any of the decimal widths.
//!
//! # Precision strategy
//!
//! Constants are derived from raw integer references — no `f64`
//! anywhere. Each tier has its own reference at the tier's maximum
//! storage precision:
//!
//! | Tier           | Storage of reference | `SCALE_REF` (= reference digits) | Source file       |
//! |----------------|---------------------|----------------------------------|-------------------|
//! | D9 / D18 / D38 | `i128`              | 37                               | this file         |
//! | D76            | `Int256`            | 75                               | `consts_wide.rs`  |
//! | D153           | `Int512`            | 153                              | `consts_wide.rs`  |
//! | D307           | `Int1024`           | 307                              | `consts_wide.rs`  |
//!
//! The rescale from `SCALE_REF` to the caller's `SCALE` uses integer
//! division with the crate-default [`RoundingMode`] (half-to-even by
//! default; overridable via the `rounding-*` Cargo features). Going
//! through `f64` would cap precision at ~15–17 decimal digits; the
//! raw-integer path preserves the full per-tier reference width.
//!
//! At `SCALE ≤ SCALE_REF` (every supported scale on D9 / D18 / D76 /
//! D153 / D307, and every D38 scale up to 37) the result is within
//! **0.5 ULP** of the canonical decimal expansion. The single
//! exception is `D38<38>`: the largest D38 reference fits 37 fractional
//! digits in `i128` (`tau ≈ 6.28×10³⁷` is below `i128::MAX ≈
//! 1.7×10³⁸`), so `SCALE = 38` (the D38 maximum) is rescaled
//! upward — multiplying the 37-digit reference by 10 — which appends
//! a placeholder zero rather than adding precision. The error there
//! is bounded at ≈ 5 ULP for `pi` / `tau` / `e` / `golden`; `half_pi`
//! and `quarter_pi` (smaller in magnitude) remain inside 0.5 ULP.
//!
//! [`RoundingMode`]: crate::rounding::RoundingMode
//!
//! # Sources
//!
//! Each raw constant is the half-to-even rounding of the canonical
//! decimal expansion to the tier's `SCALE_REF` fractional digits. ISO
//! 80000-2 (pi, tau, pi/2, pi/4), OEIS A001113 (e), OEIS A001622
//! (golden ratio).

use crate::core_type::D38;

/// Reference scale for the high-precision raw constants below.
///
/// Every constant fits in `i128` at this scale; the largest
/// (tau ≈ 6.28×10³⁷) is below `i128::MAX ≈ 1.7×10³⁸`. Caller scales
/// above this value rescale up by `10^(SCALE - SCALE_REF)`, which
/// appends placeholder zeros without adding precision — `SCALE = 38`
/// loses up to ≈ 5 ULP this way. Caller scales at or below
/// `SCALE_REF` rescale down via the crate-default [`RoundingMode`],
/// preserving the 0.5 ULP contract.
///
/// # Precision
///
/// N/A: constant value, no arithmetic performed.
///
/// [`RoundingMode`]: crate::rounding::RoundingMode
const SCALE_REF: u32 = 37;

// Raw i128 constants at SCALE_REF = 37, materialised at build time
// by `build.rs` (the same hand-rolled multi-precision generator that
// emits the wide-tier constants). Sources: ISO 80000-2 (pi, tau,
// pi/2, pi/4), OEIS A001113 (e), OEIS A001622 (golden ratio).
//
// The build-time string -> i128 parse is `const fn` (Rust 1.83+).

include!(concat!(env!("OUT_DIR"), "/wide_consts.rs"));

const PI_RAW_S37: i128 = match i128::from_str_radix(PI_D38_S37, 10) {
    Ok(v) => v,
    Err(_) => panic!("consts: PI_D38_S37 not parseable"),
};
const TAU_RAW_S37: i128 = match i128::from_str_radix(TAU_D38_S37, 10) {
    Ok(v) => v,
    Err(_) => panic!("consts: TAU_D38_S37 not parseable"),
};
const HALF_PI_RAW_S37: i128 = match i128::from_str_radix(HALF_PI_D38_S37, 10) {
    Ok(v) => v,
    Err(_) => panic!("consts: HALF_PI_D38_S37 not parseable"),
};
const QUARTER_PI_RAW_S37: i128 = match i128::from_str_radix(QUARTER_PI_D38_S37, 10) {
    Ok(v) => v,
    Err(_) => panic!("consts: QUARTER_PI_D38_S37 not parseable"),
};
const E_RAW_S37: i128 = match i128::from_str_radix(E_D38_S37, 10) {
    Ok(v) => v,
    Err(_) => panic!("consts: E_D38_S37 not parseable"),
};
const GOLDEN_RAW_S37: i128 = match i128::from_str_radix(GOLDEN_D38_S37, 10) {
    Ok(v) => v,
    Err(_) => panic!("consts: GOLDEN_D38_S37 not parseable"),
};

// Rescaling from SCALE_REF to the caller's SCALE is delegated to
// `D38::rescale` (which uses round-half-to-even by default; see
// `src/rescale.rs`). The constants below construct a `D38<SCALE_REF>`
// from the raw integer literal and then rescale to the caller's
// `D38<SCALE>`.

/// Well-known mathematical constants available on every decimal width
/// (`D9` / `D18` / `D38` / `D76` / `D153` / `D307`).
///
/// Import this trait to call `D38s12::pi()`, `D76::<35>::e()`, etc.
///
/// All returned values are computed from a raw integer reference at
/// each tier's maximum storage precision (37 digits for D9/D18/D38, 75
/// for D76, 153 for D153, 307 for D307) without passing through `f64`.
/// The result is within 0.5 ULP of the canonical decimal expansion at
/// the target `SCALE` for every supported scale, with one exception:
/// `D38<38>` (the D38 maximum) rescales the 37-digit reference upward
/// by 10, appending a placeholder zero rather than adding precision;
/// the error there is bounded at ≈ 5 ULP for the larger-magnitude
/// constants. See the module-level docs for the per-tier table.
pub trait DecimalConsts: Sized {
    /// Pi (~3.14159265...). One half-turn in radians.
    ///
    /// Source: ISO 80000-2 / OEIS A000796. Rescaled per-tier (see the
    /// module-level table) to the caller's `SCALE` via the crate-default
    /// rounding mode.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    fn pi() -> Self;

    /// Tau (~6.28318530...). One full turn in radians.
    ///
    /// Defined as `2 * pi`. Rescaled per-tier (see the module-level table) to the caller's `SCALE` via the crate-default rounding mode.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    fn tau() -> Self;

    /// Half-pi (~1.57079632...). One quarter-turn in radians.
    ///
    /// Defined as `pi / 2`. Rescaled per-tier (see the module-level table) to the caller's `SCALE` via the crate-default rounding mode.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    fn half_pi() -> Self;

    /// Quarter-pi (~0.78539816...). One eighth-turn in radians.
    ///
    /// Defined as `pi / 4`. Rescaled per-tier (see the module-level table) to the caller's `SCALE` via the crate-default rounding mode.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    fn quarter_pi() -> Self;

    /// The golden ratio (~1.61803398...). Dimensionless.
    ///
    /// Defined as `(1 + sqrt(5)) / 2`. Source: OEIS A001622. Rescaled
    /// per-tier (see the module-level table) to the caller's `SCALE`
    /// via the crate-default rounding mode.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    fn golden() -> Self;

    /// Euler's number (~2.71828182...). Dimensionless.
    ///
    /// Source: OEIS A001113. Rescaled per-tier (see the module-level table) to the caller's `SCALE` via the crate-default rounding mode.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    fn e() -> Self;
}

// Public-to-crate helpers that return each constant's rescaled bits at
// the caller's target SCALE. Used by the `decl_decimal_consts!` macro
// to provide DecimalConsts for narrower widths (D9, D18) without
// duplicating the rescale logic.

pub(crate) fn pi_at_target<const TARGET: u32>() -> i128 {
    D38::<SCALE_REF>::from_bits(PI_RAW_S37).rescale::<TARGET>().to_bits()
}
pub(crate) fn tau_at_target<const TARGET: u32>() -> i128 {
    D38::<SCALE_REF>::from_bits(TAU_RAW_S37).rescale::<TARGET>().to_bits()
}
pub(crate) fn half_pi_at_target<const TARGET: u32>() -> i128 {
    D38::<SCALE_REF>::from_bits(HALF_PI_RAW_S37).rescale::<TARGET>().to_bits()
}
pub(crate) fn quarter_pi_at_target<const TARGET: u32>() -> i128 {
    D38::<SCALE_REF>::from_bits(QUARTER_PI_RAW_S37).rescale::<TARGET>().to_bits()
}
pub(crate) fn golden_at_target<const TARGET: u32>() -> i128 {
    D38::<SCALE_REF>::from_bits(GOLDEN_RAW_S37).rescale::<TARGET>().to_bits()
}
pub(crate) fn e_at_target<const TARGET: u32>() -> i128 {
    D38::<SCALE_REF>::from_bits(E_RAW_S37).rescale::<TARGET>().to_bits()
}

// The `DecimalConsts` impl for `D38<SCALE>` is emitted by the
// `decl_decimal_consts!` macro — the same macro D9 / D18 / D76+ use.
// It expands to `Self(pi_at_target::<SCALE>())` etc., which is
// identical to the previous hand-coded
// `D38::<SCALE_REF>::from_bits(PI_RAW_S37).rescale::<SCALE>()` because
// `pi_at_target` is defined as exactly that, then `.to_bits()`.
crate::macros::consts::decl_decimal_consts!(D38, i128);

// Inherent associated constants: EPSILON / MIN_POSITIVE.
//
// These mirror `f64::EPSILON` and `f64::MIN_POSITIVE` so that generic
// numeric code that calls `T::EPSILON` or `T::MIN_POSITIVE` compiles
// when `T = D38<SCALE>`. For D38 both equal `D38(1)` -- the smallest
// representable positive value (1 LSB = 10^-SCALE). There are no subnormals.

impl<const SCALE: u32> D38<SCALE> {
    /// Smallest representable positive value: 1 LSB = `10^-SCALE`.
    ///
    /// Provided as an analogue to `f64::EPSILON` for generic numeric code.
    /// Note that this differs from the f64 definition ("difference between
    /// 1.0 and the next-larger f64"): for `D38` the LSB is uniform across
    /// the entire representable range.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    pub const EPSILON: Self = Self(1);

    /// Smallest positive value (equal to [`Self::EPSILON`]).
    ///
    /// Provided as an analogue to `f64::MIN_POSITIVE` for generic numeric
    /// code. Unlike `f64`, `D38` has no subnormals, so `MIN_POSITIVE`
    /// and `EPSILON` are the same value.
    ///
    /// # Precision
    ///
    /// N/A: constant value, no arithmetic performed.
    pub const MIN_POSITIVE: Self = Self(1);
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::core_type::D38s12;

    // Bit-exact assertions at SCALE = 12.
    //
    // At SCALE = 12 each constant is the 37-digit raw integer divided by
    // 10^23, rounded half-to-even.

    /// pi at SCALE=12: raw / 10^23.
    /// Truncated 13 digits: 3_141_592_653_589.
    /// 14th digit is 7 (from position 14 of the raw) -> round up.
    /// Expected: 3_141_592_653_590.
    #[test]
    fn pi_is_bit_exact_at_scale_12() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        assert_eq!(D38s12::pi().to_bits(), 3_141_592_653_590_i128);
    }

    /// tau at SCALE=12: raw / 10^23.
    /// Truncated 13 digits: 6_283_185_307_179.
    /// 14th digit is 5 -> round up. Expected: 6_283_185_307_180.
    #[test]
    fn tau_is_bit_exact_at_scale_12() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        assert_eq!(D38s12::tau().to_bits(), 6_283_185_307_180_i128);
    }

    /// half_pi at SCALE=12: raw / 10^23.
    /// Truncated 13 digits: 1_570_796_326_794.
    /// 14th digit is 8 -> round up. Expected: 1_570_796_326_795.
    #[test]
    fn half_pi_is_bit_exact_at_scale_12() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        assert_eq!(D38s12::half_pi().to_bits(), 1_570_796_326_795_i128);
    }

    /// quarter_pi at SCALE=12: raw / 10^23.
    /// Truncated 12 digits: 785_398_163_397.
    /// 13th digit is 4 -> no round-up. Expected: 785_398_163_397.
    #[test]
    fn quarter_pi_is_bit_exact_at_scale_12() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        assert_eq!(D38s12::quarter_pi().to_bits(), 785_398_163_397_i128);
    }

    /// e at SCALE=12: raw / 10^23.
    /// Truncated 13 digits: 2_718_281_828_459.
    /// 14th digit is 0 -> no round-up. Expected: 2_718_281_828_459.
    #[test]
    fn e_is_bit_exact_at_scale_12() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        assert_eq!(D38s12::e().to_bits(), 2_718_281_828_459_i128);
    }

    /// golden at SCALE=12: raw / 10^23.
    /// Truncated 13 digits: 1_618_033_988_749.
    /// 14th digit is 8 -> round up. Expected: 1_618_033_988_750.
    #[test]
    fn golden_is_bit_exact_at_scale_12() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        assert_eq!(D38s12::golden().to_bits(), 1_618_033_988_750_i128);
    }

    // Closeness checks against core::f64::consts.
    // These verify that the correct reference digits were selected; the
    // bit-exact tests above are the primary acceptance criteria.

    /// pi() converted to f64 is within 1e-11 of `core::f64::consts::PI`.
    /// At SCALE=12, 1 LSB = 1e-12, so 1e-11 covers rescale rounding plus
    /// the f64 conversion step.
    #[test]
    fn pi_close_to_f64_pi() {
        let diff = (D38s12::pi().to_f64() - core::f64::consts::PI).abs();
        assert!(diff < 1e-11, "pi diverges from f64 PI by {diff}");
    }

    #[test]
    fn tau_close_to_f64_tau() {
        let diff = (D38s12::tau().to_f64() - core::f64::consts::TAU).abs();
        assert!(diff < 1e-11, "tau diverges from f64 TAU by {diff}");
    }

    #[test]
    fn half_pi_close_to_f64_frac_pi_2() {
        let diff =
            (D38s12::half_pi().to_f64() - core::f64::consts::FRAC_PI_2).abs();
        assert!(diff < 1e-11, "half_pi diverges from f64 FRAC_PI_2 by {diff}");
    }

    #[test]
    fn quarter_pi_close_to_f64_frac_pi_4() {
        let diff =
            (D38s12::quarter_pi().to_f64() - core::f64::consts::FRAC_PI_4).abs();
        assert!(
            diff < 1e-11,
            "quarter_pi diverges from f64 FRAC_PI_4 by {diff}"
        );
    }

    #[test]
    fn e_close_to_f64_e() {
        let diff = (D38s12::e().to_f64() - core::f64::consts::E).abs();
        assert!(diff < 1e-11, "e diverges from f64 E by {diff}");
    }

    /// golden() converted to f64 is within 1e-11 of the closed form
    /// `(1 + sqrt(5)) / 2`. Requires std for `f64::sqrt`.
    #[cfg(feature = "std")]
    #[test]
    fn golden_close_to_closed_form() {
        let expected = (1.0_f64 + 5.0_f64.sqrt()) / 2.0;
        let diff = (D38s12::golden().to_f64() - expected).abs();
        assert!(diff < 1e-11, "golden diverges from closed-form by {diff}");
    }

    // EPSILON / MIN_POSITIVE

    #[test]
    fn epsilon_is_one_ulp() {
        assert_eq!(D38s12::EPSILON.to_bits(), 1_i128);
        assert!(D38s12::EPSILON > D38s12::ZERO);
    }

    #[test]
    fn min_positive_is_one_ulp() {
        assert_eq!(D38s12::MIN_POSITIVE.to_bits(), 1_i128);
        assert_eq!(D38s12::MIN_POSITIVE, D38s12::EPSILON);
    }

    /// At SCALE = 6 the LSB is 10^-6; EPSILON is still raw 1.
    #[test]
    fn epsilon_at_scale_6_is_one_ulp() {
        type D6 = D38<6>;
        assert_eq!(D6::EPSILON.to_bits(), 1_i128);
        assert_eq!(D6::MIN_POSITIVE.to_bits(), 1_i128);
    }

    // Cross-scale exercises

    /// At SCALE = 6, pi() should equal 3.141593 (rounded half-to-even from
    /// 3.1415926535...). Expected raw bits: 3_141_593.
    #[test]
    fn pi_at_scale_6_is_bit_exact() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        type D6 = D38<6>;
        assert_eq!(D6::pi().to_bits(), 3_141_593_i128);
    }

    /// At SCALE = 0, pi() rounds to 3 (first fractional digit is 1, no
    /// round-up).
    #[test]
    fn pi_at_scale_0_is_three() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        type D0 = D38<0>;
        assert_eq!(D0::pi().to_bits(), 3_i128);
    }

    /// At SCALE = SCALE_REF (37), pi() returns exactly the raw constant.
    #[test]
    fn pi_at_scale_ref_is_raw_constant() {
        type D37 = D38<37>;
        assert_eq!(D37::pi().to_bits(), PI_RAW_S37);
    }

    /// At SCALE = SCALE_REF + 1 (38), pi() multiplies by 10, appending
    /// one trailing zero digit. PI_RAW_S37 * 10 ≈ 3.14×10³⁸ which is
    /// larger than i128::MAX ≈ 1.7×10³⁸, so this case overflows
    /// `D38<38>` storage at compile time — exercising the upper end
    /// of the rescale-up path is left to the SCALE = 37 case above.
    #[test]
    fn pi_at_scale_37_is_raw_constant() {
        type D37 = D38<37>;
        assert_eq!(D37::pi().to_bits(), PI_RAW_S37);
    }

    /// Negative-side rounding: negating pi gives the expected raw bits.
    #[test]
    fn neg_pi_round_trip() {
        if !crate::rounding::DEFAULT_IS_HALF_TO_EVEN { return; }
        let pi = D38s12::pi();
        let neg_pi = -pi;
        assert_eq!(neg_pi.to_bits(), -3_141_592_653_590_i128);
    }

    // (`rescale_from_ref` boundary tests removed: the rounding logic now
    // lives in `D38::rescale` / `src/rounding.rs::apply_rounding` and is
    // covered by the tests in those modules.)
}