lua-stdlib 0.4.1

omniLua's Lua standard library — internal crate; depend on `omnilua`.
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
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
//! Standard mathematical library — `math.*`.
//!
//! The PRNG is xoshiro256** operating on four 64-bit words (the single 64-bit
//! code path; there is no 32-bit fallback to keep).
//!
//! The deprecated `LUA_COMPAT_MATHLIB` roster (`cosh`, `sinh`, `tanh`, `pow`,
//! `log10`, `atan2`) ships in the default 5.1/5.2/5.3/5.4 builds and is dropped
//! in 5.5; `frexp`/`ldexp` survive into 5.5. `atan2` is an alias of `math_atan`.
//! Which functions exist per version is the registration logic in
//! [`luaopen_math`].

use crate::state_stub::{LuaState, LuaStateStubExt as _};
use lua_types::{LuaError, LuaType, LuaValue};

// ── Constants ──────────────────────────────────────────────────────────────

/// `math.pi`. Higher precision than `std::f64::consts::PI` (which is rounded to
/// the nearest `f64`); both round-trip to the same `f64` bit pattern.
const PI: f64 = 3.141592653589793238462643383279502884_f64;

// ── Type aliases for library registration ─────────────────────────────────

/// A Lua C-style function: takes the Lua state, returns count of pushed values.
type LuaCFunction = fn(&mut LuaState) -> Result<usize, LuaError>;

// ── PRNG state ────────────────────────────────────────────────────────────

/// State for the xoshiro256** PRNG.
///
/// In C this is stored as raw `lua_newuserdatauv` memory and accessed by
/// casting the userdata pointer. Until typed-userdata closure upvalues land
/// in Phase B, we keep the PRNG state in a thread-local cell so that
/// `math.random` and `math.randomseed` are callable from Lua. This collapses
/// per-lua_State PRNG isolation to per-thread, which is sufficient for the
/// 5.4 test corpus.
struct RanState {
    s: [u64; 4],
}

thread_local! {
    static RAN_STATE: std::cell::RefCell<RanState> =
        std::cell::RefCell::new(RanState { s: [0xff, 0xff, 0xff, 0xff] });
}

/// The xoshiro256** generator: the bit-exact, load-bearing PRNG core.
///
/// Every function here is pinned byte-for-byte by the behavioral net (the
/// `multiversion_oracle` PRNG-sequence tests on 5.4/5.5). The internals must NOT
/// be reordered or "simplified" — any change to the arithmetic diverges from the
/// reference stream. Grouping them in one private module makes that contract
/// explicit; the callers reach them as `xoshiro::*`.
mod xoshiro {
    /// Number of binary digits in the `f64` mantissa (`DBL_MANT_DIG`).
    const FIGS: u32 = 53;

    /// Bits to discard from a 64-bit random word before float conversion (`= 11`):
    /// the word is reduced to its top [`FIGS`] significant bits.
    const SHIFT64_FIG: u32 = 64 - FIGS;

    /// Advance the xoshiro256** state by one step and return the next raw 64-bit
    /// pseudo-random value.
    pub(super) fn next_rand(s: &mut [u64; 4]) -> u64 {
        let s0 = s[0];
        let s1 = s[1];
        let s2 = s[2] ^ s0;
        let s3 = s[3] ^ s1;
        let res = s1.wrapping_mul(5).rotate_left(7).wrapping_mul(9);
        s[0] = s0 ^ s3;
        s[1] = s1 ^ s2;
        s[2] = s2 ^ (s1 << 17);
        s[3] = s3.rotate_left(45);
        res
    }

    /// Convert a raw 64-bit PRNG output to a float in [0.0, 1.0).
    ///
    /// Takes the top FIGS=53 bits, interprets them as a signed integer, scales
    /// by `scaleFIG = 0.5 / 2^52`, then corrects the two's-complement sign.
    pub(super) fn rand_to_float(x: u64) -> f64 {
        let sx = (x >> SHIFT64_FIG) as i64;
        let scale_fig: f64 = 0.5 / ((1u64 << (FIGS - 1)) as f64);
        let mut res = (sx as f64) * scale_fig;
        if sx < 0 {
            res += 1.0;
        }
        debug_assert!(0.0 <= res && res < 1.0);
        res
    }

    /// Initialise the four PRNG words from two seed values.
    ///
    /// `s[1]` is forced to `0xff` so the state is never all-zero (xoshiro's
    /// fixed point), and sixteen draws are discarded to spread the seed bits before
    /// the first observable output. Takes the word array (not the [`LuaState`]) so
    /// the caller can push the seed values without a borrow conflict against the
    /// `RanState` upvalue.
    pub(super) fn set_seed_words(s: &mut [u64; 4], n1: u64, n2: u64) {
        s[0] = n1;
        s[1] = 0xff;
        s[2] = n2;
        s[3] = 0;
        for _ in 0..16 {
            next_rand(s);
        }
    }

    /// Project `ran` uniformly into `[0, n]` by rejection sampling.
    ///
    /// When `n + 1` is a power of two the low bits are already uniform, so a single
    /// mask suffices. Otherwise `lim` is built as the smallest `2^b - 1` not smaller
    /// than `n` (the bit-smear), and draws outside `[0, n]` are rejected and
    /// redrawn. The `>> 32` smear step is unconditional here because `u64` always
    /// has 64 bits; the C source guards it behind an `#if` on the integer width.
    /// Takes the word array (not `&mut RanState`) to avoid nested borrows at the
    /// call sites.
    pub(super) fn project(mut ran: u64, n: u64, s: &mut [u64; 4]) -> u64 {
        if (n & n.wrapping_add(1)) == 0 {
            return ran & n;
        }
        let mut lim = n;
        lim |= lim >> 1;
        lim |= lim >> 2;
        lim |= lim >> 4;
        lim |= lim >> 8;
        lim |= lim >> 16;
        lim |= lim >> 32;
        debug_assert!((lim & lim.wrapping_add(1)) == 0);
        debug_assert!(lim >= n);
        debug_assert!((lim >> 1) < n);
        loop {
            ran &= lim;
            if ran <= n {
                break;
            }
            ran = next_rand(s);
        }
        ran
    }
}

// ── Helpers ───────────────────────────────────────────────────────────────

/// Whether the argument at stack index `n` carries the integer subtype.
///
/// Several `math.*` functions (`abs`, `floor`, `ceil`, `fmod`, `modf`, `type`)
/// branch on whether an argument is an `Int` versus a `Float`, preserving the
/// integer subtype on the integer path. This names that recurring predicate so
/// the branch reads as intent rather than as a raw `matches!` on the stack slot.
fn arg_is_int(state: &mut LuaState, n: i32) -> bool {
    matches!(state.value_at(n), LuaValue::Int(_))
}

/// The smallest `f64` whose value equals an `i64` exactly: `i64::MIN` as a
/// float (`-2^63`, exactly representable).
const I64_MIN_AS_F64: f64 = i64::MIN as f64;

/// One past the largest representable `i64`, as a float: `2^63`. `i64::MAX`
/// (`2^63 - 1`) is not exactly representable in `f64`, so the in-range test
/// uses a strict `<` against this boundary rather than `<= i64::MAX as f64`.
const I64_MAX_PLUS_1_AS_F64: f64 = -(i64::MIN as f64);

/// Push `d` as an `Int` when it has an integer value that fits exactly in an
/// `i64`, otherwise push it as a `Float`.
///
/// `floor`/`ceil`/`modf` use this to keep their result an integer subtype when
/// it round-trips through `i64` without loss, falling back to a float for
/// magnitudes outside the `i64` range. The bound check is half-open
/// (`I64_MIN_AS_F64 <= d < I64_MAX_PLUS_1_AS_F64`) because `i64::MAX` itself is
/// not exactly representable as an `f64`.
fn push_int_or_float(state: &mut LuaState, d: f64) {
    if d >= I64_MIN_AS_F64 && d < I64_MAX_PLUS_1_AS_F64 {
        state.push(LuaValue::Int(d as i64));
    } else {
        state.push(LuaValue::Float(d));
    }
}

/// Leave the first argument as the sole result (an integer floor/ceil/modf is
/// already its own integer part — return it unchanged, preserving subtype).
///
/// This goes through the public, frame-relative [`lua_vm::api::set_top`] so the
/// truncation is relative to the call frame; the inherent `LuaState::set_top`
/// would treat its argument as an absolute `StackIdx` and discard the argument.
fn keep_first_arg(state: &mut LuaState) -> Result<(), LuaError> {
    lua_vm::api::set_top(state, 1)
}

// ── Basic math functions ──────────────────────────────────────────────────

/// `math.abs(x)` — absolute value, preserving integer type when possible.
///
fn math_abs(state: &mut LuaState) -> Result<usize, LuaError> {
    if arg_is_int(state, 1) {
        let n = state.to_integer(1).unwrap_or(0);
        let n = if n < 0 {
            (0u64.wrapping_sub(n as u64)) as i64
        } else {
            n
        };
        state.push(LuaValue::Int(n));
    } else {
        let x = state.check_number(1)?;
        state.push(LuaValue::Float(x.abs()));
    }
    Ok(1)
}

/// `math.sin(x)` — sine (radians).
///
fn math_sin(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.sin()));
    Ok(1)
}

/// `math.cos(x)` — cosine (radians).
///
fn math_cos(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.cos()));
    Ok(1)
}

/// `math.tan(x)` — tangent (radians).
///
fn math_tan(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.tan()));
    Ok(1)
}

/// `math.asin(x)` — arc-sine, result in radians.
///
fn math_asin(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.asin()));
    Ok(1)
}

/// `math.acos(x)` — arc-cosine, result in radians.
///
fn math_acos(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.acos()));
    Ok(1)
}

/// `math.atan(y [, x])` — arc-tangent of y/x (defaults x=1), result in
/// radians. Subsumes C's `atan2` when x is provided.
///
fn math_atan(state: &mut LuaState) -> Result<usize, LuaError> {
    let y = state.check_number(1)?;
    let x = state.opt_number(2, 1.0)?;
    state.push(LuaValue::Float(y.atan2(x)));
    Ok(1)
}

/// `math.cosh(x)` — hyperbolic cosine. Deprecated `LUA_COMPAT_MATHLIB`
/// function, registered only under the 5.3 backend.
///
fn math_cosh(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.cosh()));
    Ok(1)
}

/// `math.sinh(x)` — hyperbolic sine. Deprecated `LUA_COMPAT_MATHLIB`
/// function, registered only under the 5.3 backend.
///
fn math_sinh(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.sinh()));
    Ok(1)
}

/// `math.tanh(x)` — hyperbolic tangent. Deprecated `LUA_COMPAT_MATHLIB`
/// function, registered only under the 5.3 backend.
///
fn math_tanh(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.tanh()));
    Ok(1)
}

/// `math.pow(x, y)` — x raised to the power y, always returning a float.
/// Deprecated `LUA_COMPAT_MATHLIB` function, registered only under the 5.3
/// backend. Mirrors C `pow(luaL_checknumber, luaL_checknumber)`.
///
fn math_pow(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    let y = state.check_number(2)?;
    state.push(LuaValue::Float(x.powf(y)));
    Ok(1)
}

/// `math.log10(x)` — base-10 logarithm. Deprecated `LUA_COMPAT_MATHLIB`
/// function, registered only under the 5.3 backend.
///
fn math_log10(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.log10()));
    Ok(1)
}

/// `math.ldexp(x, e)` — `x * 2^e`. Deprecated `LUA_COMPAT_MATHLIB` function,
/// registered only under the 5.3 backend. The exponent is an integer argument
/// truncated to C `int` range, matching `ldexp(x, (int)luaL_checkinteger)`.
///
fn math_ldexp(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    let e = state.check_integer(2)? as i32;
    state.push(LuaValue::Float(ldexp(x, e)));
    Ok(1)
}

/// Pure `ldexp`: returns `x * 2^exp` with C `ldexp` semantics.
///
/// A naive `x * 2f64.powi(exp)` underflows (or overflows) the intermediate
/// `2^exp` for large-magnitude exponents, losing subnormal results such as
/// `ldexp(1.0, -1074) == 5e-324`. The scaling is therefore applied in bounded
/// steps so no intermediate factor under/overflows: each step multiplies by a
/// power of two whose magnitude stays inside the normal `f64` range.
fn ldexp(x: f64, exp: i32) -> f64 {
    if x == 0.0 || !x.is_finite() {
        return x;
    }
    let mut result = x;
    let mut e = exp;
    // 2^1023 is the largest power of two representable as a normal f64; chunk
    // the exponent so each `from_bits` factor is always finite and nonzero.
    while e > 1023 {
        result *= f64::from_bits(0x7feu64 << 52); // 2^1023
        e -= 1023;
    }
    while e < -1022 {
        result *= f64::from_bits(0x001u64 << 52); // 2^-1022 (smallest normal)
        e += 1022;
    }
    result * f64::from_bits(((e + 1023) as u64) << 52)
}

/// `math.frexp(x)` — split x into a normalized mantissa and an exponent such
/// that `x == mantissa * 2^exponent` with `0.5 <= |mantissa| < 1`. Returns the
/// float mantissa followed by the **integer** exponent, matching C
/// `frexp` + `lua_pushinteger`. Deprecated `LUA_COMPAT_MATHLIB` function,
/// registered only under the 5.3 backend.
///
/// Rust std has no `frexp`; this replicates C `frexp` via `f64` bit
/// manipulation, including the `frexp(0.0) == (0.0, 0)` special case (and the
/// matching `-0.0`, infinity, and NaN cases, which C leaves unchanged with a
/// zero exponent).
fn math_frexp(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    let (mantissa, exponent) = frexp(x);
    state.push(LuaValue::Float(mantissa));
    state.push(LuaValue::Int(exponent as i64));
    Ok(2)
}

/// Pure `frexp`: returns `(mantissa, exponent)` with `x == mantissa * 2^exp`.
///
/// Replicates C `frexp` semantics for f64. Zero, infinity, and NaN are
/// returned unchanged with a zero exponent.
fn frexp(x: f64) -> (f64, i32) {
    if x == 0.0 || !x.is_finite() {
        return (x, 0);
    }
    let bits = x.to_bits();
    let raw_exp = ((bits >> 52) & 0x7ff) as i32;
    if raw_exp == 0 {
        // Subnormal: scale up by 2^54 to normalize, then correct the exponent.
        let (m, e) = frexp(x * (1u64 << 54) as f64);
        return (m, e - 54);
    }
    // Bias the exponent so the mantissa lands in [0.5, 1): set the stored
    // exponent field to 0x3fe (unbiased -1).
    let exponent = raw_exp - 1022;
    let mantissa_bits = (bits & !(0x7ffu64 << 52)) | (0x3feu64 << 52);
    (f64::from_bits(mantissa_bits), exponent)
}

/// `math.tointeger(x)` — return `x` as an integer, or the fail value when it
/// has no exact integer representation.
///
/// The fail value is `nil`, not `false`: the default 5.3/5.4/5.5 builds expand
/// `luaL_pushfail` to a nil push (only a `LUA_FAILISFALSE` build would push
/// `false`), and the oracle contract pins the nil.
fn math_toint(state: &mut LuaState) -> Result<usize, LuaError> {
    if let Some(n) = state.to_integer_opt(1) {
        state.push(LuaValue::Int(n));
    } else {
        state.check_any(1)?;
        state.push(LuaValue::Nil);
    }
    Ok(1)
}

/// `math.floor(x)` — largest integer ≤ x.
///
fn math_floor(state: &mut LuaState) -> Result<usize, LuaError> {
    if arg_is_int(state, 1) {
        keep_first_arg(state)?;
    } else {
        let d = state.check_number(1)?.floor();
        push_int_or_float(state, d);
    }
    Ok(1)
}

/// `math.ceil(x)` — smallest integer ≥ x.
///
fn math_ceil(state: &mut LuaState) -> Result<usize, LuaError> {
    if arg_is_int(state, 1) {
        keep_first_arg(state)?;
    } else {
        let d = state.check_number(1)?.ceil();
        push_int_or_float(state, d);
    }
    Ok(1)
}

/// `math.fmod(x, y)` — floating-point remainder (same sign as x).
///
fn math_fmod(state: &mut LuaState) -> Result<usize, LuaError> {
    if arg_is_int(state, 1) && arg_is_int(state, 2) {
        let a = state.to_integer(1).unwrap_or(0);
        let d = state.to_integer(2).unwrap_or(0);
        if (d as u64).wrapping_add(1) <= 1 {
            if d == 0 {
                return Err(lua_vm::debug::arg_error_impl(state, 2, b"zero"));
            }
            state.push(LuaValue::Int(0));
        } else {
            state.push(LuaValue::Int(a % d));
        }
    } else {
        let x = state.check_number(1)?;
        let y = state.check_number(2)?;
        state.push(LuaValue::Float(x % y));
    }
    Ok(1)
}

/// `math.modf(x)` — split into integer and fractional parts; returns both.
///
/// The integer part is computed with `ceil`/`floor` + subtraction rather than
/// libc `modf`, avoiding a `double *` out-parameter ABI. An integer argument is
/// its own integer part with a `0.0` fractional part.
fn math_modf(state: &mut LuaState) -> Result<usize, LuaError> {
    if arg_is_int(state, 1) {
        keep_first_arg(state)?;
        state.push(LuaValue::Float(0.0));
    } else {
        let n = state.check_number(1)?;
        let ip = if n < 0.0 { n.ceil() } else { n.floor() };
        push_int_or_float(state, ip);
        let frac = if n == ip { 0.0 } else { n - ip };
        state.push(LuaValue::Float(frac));
    }
    Ok(2)
}

/// `math.sqrt(x)` — square root.
///
fn math_sqrt(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.sqrt()));
    Ok(1)
}

/// `math.ult(m, n)` — unsigned less-than on integers.
///
fn math_ult(state: &mut LuaState) -> Result<usize, LuaError> {
    let a = state.check_integer(1)?;
    let b = state.check_integer(2)?;
    state.push(LuaValue::Bool((a as u64) < (b as u64)));
    Ok(1)
}

/// `math.log(x [, base])` — logarithm; natural if base omitted.
///
fn math_log(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    // Lua 5.1's `math.log` takes a single argument and silently ignores any
    // second; the two-argument base form is a 5.2 addition. Verified against
    // lua5.1.5: `math.log(8,2) == math.log(8) == ln(8)`, and a second arg never
    // errors. See specs/followup/5.1-roster-syntax.md §1.
    if matches!(state.global().lua_version, lua_types::LuaVersion::V51) {
        state.push(LuaValue::Float(x.ln()));
        return Ok(1);
    }
    let res = if matches!(state.type_at(2), LuaType::None | LuaType::Nil) {
        x.ln()
    } else {
        let base = state.check_number(2)?;
        if base == 2.0 {
            x.log2()
        } else if base == 10.0 {
            x.log10()
        } else {
            x.ln() / base.ln()
        }
    };
    state.push(LuaValue::Float(res));
    Ok(1)
}

/// `math.exp(x)` — e raised to the power x.
///
fn math_exp(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x.exp()));
    Ok(1)
}

/// `math.deg(x)` — convert radians to degrees.
///
fn math_deg(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x * (180.0 / PI)));
    Ok(1)
}

/// `math.rad(x)` — convert degrees to radians.
///
fn math_rad(state: &mut LuaState) -> Result<usize, LuaError> {
    let x = state.check_number(1)?;
    state.push(LuaValue::Float(x * (PI / 180.0)));
    Ok(1)
}

/// Whether `math.max`/`math.min` use the float-only `luaL_checknumber` path.
///
/// 5.1 and 5.2 coerce every argument with `luaL_checknumber`, comparing
/// `lua_Number` doubles and returning the coerced number — so a non-number
/// argument raises `number expected, got <type>` and a number-shaped string
/// argument is accepted and returned as a number. 5.3+ rewrote both to use
/// `lua_compare(..., LUA_OPLT)`, which compares the values directly (so strings
/// compare lexicographically and are returned unchanged). 5.4 is the
/// unchangeable baseline, so the float-only path is gated to V51/V52 only.
fn max_min_is_float_only(state: &LuaState) -> bool {
    matches!(
        state.global().lua_version,
        lua_types::LuaVersion::V51 | lua_types::LuaVersion::V52
    )
}

/// `math.min(x, ...)` — minimum of all arguments.
///
/// On 5.1/5.2 every argument is coerced via `luaL_checknumber` and the smallest
/// number is returned (see [`max_min_is_float_only`]); on 5.3+ the smallest
/// argument by Lua `<` comparison is returned unchanged.
fn math_min(state: &mut LuaState) -> Result<usize, LuaError> {
    let n = state.get_top();
    if n < 1 {
        return Err(lua_vm::debug::arg_error_impl(state, 1, b"value expected"));
    }
    if max_min_is_float_only(state) {
        let mut dmin = state.check_number(1)?;
        for i in 2..=n {
            let d = state.check_number(i)?;
            if d < dmin {
                dmin = d;
            }
        }
        state.push(LuaValue::Float(dmin));
        return Ok(1);
    }
    let mut imin: i32 = 1;
    for i in 2..=n {
        if state.compare_lt(i, imin)? {
            imin = i;
        }
    }
    state.push_value(imin)?;
    Ok(1)
}

/// `math.max(x, ...)` — maximum of all arguments.
///
/// On 5.1/5.2 every argument is coerced via `luaL_checknumber` and the largest
/// number is returned (see [`max_min_is_float_only`]); on 5.3+ the largest
/// argument by Lua `<` comparison is returned unchanged.
fn math_max(state: &mut LuaState) -> Result<usize, LuaError> {
    let n = state.get_top();
    if n < 1 {
        return Err(lua_vm::debug::arg_error_impl(state, 1, b"value expected"));
    }
    if max_min_is_float_only(state) {
        let mut dmax = state.check_number(1)?;
        for i in 2..=n {
            let d = state.check_number(i)?;
            if d > dmax {
                dmax = d;
            }
        }
        state.push(LuaValue::Float(dmax));
        return Ok(1);
    }
    let mut imax: i32 = 1;
    for i in 2..=n {
        if state.compare_lt(imax, i)? {
            imax = i;
        }
    }
    state.push_value(imax)?;
    Ok(1)
}

/// `math.type(x)` — return `"integer"`, `"float"`, or the fail value for
/// non-numbers.
///
/// The non-number fail value is `nil`, not `false`, for the same reason as
/// [`math_toint`]: `luaL_pushfail` pushes nil in the default builds and the
/// oracle contract pins the nil.
fn math_type(state: &mut LuaState) -> Result<usize, LuaError> {
    if matches!(state.type_at(1), LuaType::Number) {
        if arg_is_int(state, 1) {
            state.push_string(b"integer")?;
        } else {
            state.push_string(b"float")?;
        }
    } else {
        state.check_any(1)?;
        state.push(LuaValue::Nil);
    }
    Ok(1)
}

// ── PRNG-backed Lua functions ─────────────────────────────────────────────

/// `math.random([m [, n]])` — pseudo-random number generation.
///
/// With no arguments: float in `[0, 1)`. With one argument `n`: integer in
/// `[1, n]` (or a full-range `u64` when `n == 0`, on 5.4/5.5). With two
/// arguments `m, n`: integer in `[m, n]`. The PRNG word is advanced first, then
/// the arguments are read via a separate borrow.
///
/// TODO: the PRNG state is a thread-local rather than per-`lua_State` typed
/// userdata in closure upvalue 1; migrating it to `GcRef<RefCell<RanState>>`
/// (so each state has its own generator) is deferred to the upvalue-userdata
/// work, not changed here.
fn math_random(state: &mut LuaState) -> Result<usize, LuaError> {
    let rv = advance_prng(state)?;
    let n_args = state.get_top();

    if n_args == 0 {
        state.push(LuaValue::Float(xoshiro::rand_to_float(rv)));
        return Ok(1);
    }

    let version = state.global().lua_version;
    let is_v53 = version == lua_types::LuaVersion::V53;
    // 5.1/5.2 are float-only and use the C `rand()` contract: there is no
    // `random(0)` full-range special case (that is a 5.4/5.5 addition), the
    // empty-interval error for `random(m, n)` reports argument index 2 (the
    // upper bound), and integer-valued results are pushed as `Float` to honour
    // the never-construct-`Int` invariant under `FloatOnly`. See
    // specs/followup/5.1-numbers-prng.md §"Impl seams".
    let float_only = version.number_model() == lua_types::NumberModel::FloatOnly;

    let (low, up, empty_arg) = match n_args {
        1 => {
            let up = state.check_integer(1)?;
            // 5.4/5.5 `random(0)` returns a full-range integer; 5.1/5.2/5.3 have
            // no such special case — it is `[1, 0]`, an empty interval.
            if up == 0 && !is_v53 && !float_only {
                state.push(LuaValue::Int(rv as i64));
                return Ok(1);
            }
            (1i64, up, 1)
        }
        2 => {
            let low = state.check_integer(1)?;
            let up = state.check_integer(2)?;
            // 5.1's `luaL_checkint(L, 2)` for the upper bound means its
            // empty-interval `luaL_argerror` reports argument #2; the modern
            // bodies report #1.
            let empty_arg = if float_only { 2 } else { 1 };
            (low, up, empty_arg)
        }
        _ => {
            return Err(LuaError::runtime(format_args!("wrong number of arguments")));
        }
    };

    if low > up {
        return Err(lua_vm::debug::arg_error_impl(
            state,
            empty_arg,
            b"interval is empty",
        ));
    }

    // 5.3 `math_random` rejects intervals whose width overflows a signed integer
    // (`low >= 0 || up <= LUA_MAXINTEGER + low`). 5.4/5.5 use the `project`
    // bit-mask algorithm, which handles the full range without erroring.
    if is_v53 && !(low >= 0 || up <= i64::MAX.wrapping_add(low)) {
        return Err(lua_vm::debug::arg_error_impl(
            state,
            1,
            b"interval too large",
        ));
    }

    let range = (up as u64).wrapping_sub(low as u64);
    let p = project_from_upvalue(state, rv, range)?;
    let result = (p as u64).wrapping_add(low as u64) as i64;
    if float_only {
        state.push(LuaValue::Float(result as f64));
    } else {
        state.push(LuaValue::Int(result));
    }
    Ok(1)
}

/// `math.randomseed([x [, y]])` — seed the PRNG.
///
/// The return shape and auto-seed behavior are version-gated and load-bearing:
/// - 5.1/5.2 (float-only) REQUIRE the seed argument (a missing one raises
///   "number expected, got no value"), take a single seed word, and return NO
///   values.
/// - 5.3+ auto-seed from host entropy when the argument is absent and return the
///   two seed words.
///
/// See `specs/followup/5.1-numbers-prng.md`.
fn math_randomseed(state: &mut LuaState) -> Result<usize, LuaError> {
    let float_only = state.global().lua_version.number_model() == lua_types::NumberModel::FloatOnly;

    if matches!(state.type_at(1), LuaType::None) {
        if float_only {
            let n1 = state.check_integer(1)? as u64;
            apply_set_seed_quiet(state, n1, 0);
            return Ok(0);
        }
        apply_random_seed(state)?;
    } else {
        let n1 = state.check_integer(1)? as u64;
        if float_only {
            apply_set_seed_quiet(state, n1, 0);
            return Ok(0);
        }
        let n2 = state.opt_integer(2, 0)? as u64;
        apply_set_seed(state, n1, n2)?;
    }
    Ok(2)
}

/// Advance the PRNG in the thread-local [`RAN_STATE`] and return the raw 64-bit
/// output.
///
/// The thread-local is the only difference from the C source, which draws from
/// userdata in closure upvalue 1; the [`next_rand`] algorithm is identical (see
/// [`math_random`] for the deferred per-state migration).
fn advance_prng(_state: &mut LuaState) -> Result<u64, LuaError> {
    Ok(RAN_STATE.with(|r| xoshiro::next_rand(&mut r.borrow_mut().s)))
}

/// Project a raw draw into `[0, n]` using the thread-local PRNG for any
/// rejection redraws (see [`advance_prng`] for the thread-local rationale).
fn project_from_upvalue(_state: &mut LuaState, ran: u64, n: u64) -> Result<u64, LuaError> {
    Ok(RAN_STATE.with(|r| xoshiro::project(ran, n, &mut r.borrow_mut().s)))
}

/// Seed the PRNG from the host entropy hook (the 5.3+ auto-seed path).
///
/// The second seed word is derived deterministically from the entropy value;
/// the C source additionally mixes address entropy, which is deferred until a
/// richer host entropy API exists.
fn apply_random_seed(state: &mut LuaState) -> Result<(), LuaError> {
    let entropy = state.global().entropy_hook.map(|hook| hook()).unwrap_or(0);
    let seed1 = entropy;
    let seed2: u64 = entropy.rotate_left(17) ^ 0x9e37_79b9_7f4a_7c15;
    apply_set_seed(state, seed1, seed2)
}

/// Apply explicit seeds to the thread-local PRNG and push them onto the stack
/// (the 5.3+ `randomseed` return shape — the two seed words).
fn apply_set_seed(state: &mut LuaState, n1: u64, n2: u64) -> Result<(), LuaError> {
    RAN_STATE.with(|r| xoshiro::set_seed_words(&mut r.borrow_mut().s, n1, n2));
    state.push(LuaValue::Int(n1 as i64));
    state.push(LuaValue::Int(n2 as i64));
    Ok(())
}

/// Seed the PRNG without pushing the seed words onto the stack.
///
/// 5.1/5.2 `math.randomseed` returns no values, so its seeding path must not
/// push (unlike the modern [`apply_set_seed`], which returns the two words).
fn apply_set_seed_quiet(_state: &mut LuaState, n1: u64, n2: u64) {
    RAN_STATE.with(|r| xoshiro::set_seed_words(&mut r.borrow_mut().s, n1, n2));
}

/// Register `math.random` and `math.randomseed` on the math library table at
/// stack top, after seeding the PRNG.
///
/// C binds the PRNG state as upvalue 1 of both closures; with the PRNG in a
/// thread-local ([`RAN_STATE`]) these are registered as plain (non-closure)
/// entries instead.
fn set_rand_func(state: &mut LuaState) -> Result<(), LuaError> {
    apply_random_seed(state)?;
    state.pop_n(2);

    state.push_c_function(math_random)?;
    state.set_field(-2, b"random")?;
    state.push_c_function(math_randomseed)?;
    state.set_field(-2, b"randomseed")?;
    Ok(())
}

// ── Library registration table ────────────────────────────────────────────

static MATHLIB_FUNCS: &[(&[u8], LuaCFunction)] = &[
    (b"abs", math_abs),
    (b"acos", math_acos),
    (b"asin", math_asin),
    (b"atan", math_atan),
    (b"ceil", math_ceil),
    (b"cos", math_cos),
    (b"deg", math_deg),
    (b"exp", math_exp),
    (b"tointeger", math_toint),
    (b"floor", math_floor),
    (b"fmod", math_fmod),
    (b"ult", math_ult),
    (b"log", math_log),
    (b"max", math_max),
    (b"min", math_min),
    (b"modf", math_modf),
    (b"rad", math_rad),
    (b"sin", math_sin),
    (b"sqrt", math_sqrt),
    (b"tan", math_tan),
    (b"type", math_type),
    // `frexp`/`ldexp` survive into 5.5, unlike the rest of the compat roster:
    // they are registered unconditionally on 5.3/5.4/5.5. Verified against all
    // three reference binaries (`type(math.frexp)`/`type(math.ldexp)` ==
    // "function" on 5.3.6, 5.4.7, 5.5.0).
    (b"frexp", math_frexp),
    (b"ldexp", math_ldexp),
];

// ── Module entry point ────────────────────────────────────────────────────

/// Open the `math` library: create the table, register the version-agnostic
/// functions and the per-version roster delta, populate the constants, and wire
/// up the PRNG functions.
pub fn luaopen_math(state: &mut LuaState) -> Result<usize, LuaError> {
    state.new_lib(MATHLIB_FUNCS)?;

    // Per-version roster delta: the `LUA_COMPAT_MATHLIB`-gated functions
    // (`atan2` as an alias of `math_atan`, plus cosh/sinh/tanh/pow/log10) ship
    // in the default lua5.3.6 build (`LUA_COMPAT_MATHLIB` on) AND the default
    // lua5.4.7 build (its `LUA_COMPAT_5_3` umbrella turns `LUA_COMPAT_MATHLIB`
    // on), but were dropped in lua5.5.0 (macro commented out). Verified by
    // probing all three reference binaries directly. `frexp`/`ldexp` are NOT in
    // this set — they survive into 5.5 and live in the agnostic roster above.
    // `new_lib` leaves the new table on the stack top, so we register into it
    // directly. See `specs/followup/5.3-math.md` (whose 5.4/5.5-absence claim
    // is corrected here against the binaries, the binding oracle).
    // The `LUA_COMPAT_MATHLIB` deprecated roster also ships in the default
    // lua5.2.4 build (verified against the reference binary: `type(math.atan2)`
    // etc. == "function" on 5.2.4). 5.5 drops them.
    if matches!(
        state.global().lua_version,
        lua_types::LuaVersion::V51
            | lua_types::LuaVersion::V52
            | lua_types::LuaVersion::V53
            | lua_types::LuaVersion::V54
    ) {
        const COMPAT_MATH_FUNCS: &[(&[u8], LuaCFunction)] = &[
            (b"atan2", math_atan),
            (b"cosh", math_cosh),
            (b"sinh", math_sinh),
            (b"tanh", math_tanh),
            (b"pow", math_pow),
            (b"log10", math_log10),
        ];
        state.set_funcs_with_upvalues(COMPAT_MATH_FUNCS, 0)?;
    }

    // Lua 5.1 carries `math.mod`, a compat alias of `fmod` predating the rename
    // (`math.mod(7,3) == 1`). It was removed in 5.2. Verified against
    // lua5.1.5: `type(math.mod)` == "function". See
    // specs/followup/5.1-roster-syntax.md §1.
    if matches!(state.global().lua_version, lua_types::LuaVersion::V51) {
        state.push_c_function(math_fmod)?;
        state.set_field(-2, b"mod")?;
    }

    state.push(LuaValue::Float(PI));
    state.set_field(-2, b"pi")?;

    state.push(LuaValue::Float(f64::INFINITY));
    state.set_field(-2, b"huge")?;

    // LUA_MAXINTEGER = i64::MAX (lua_Integer is int64_t in default config).
    state.push(LuaValue::Int(i64::MAX));
    state.set_field(-2, b"maxinteger")?;

    state.push(LuaValue::Int(i64::MIN));
    state.set_field(-2, b"mininteger")?;

    // Lua 5.1/5.2 are float-only: the integer-subtype helpers (`math.type`,
    // `math.tointeger`, `math.ult`) and the integer bounds
    // (`math.maxinteger`/`mininteger`) are 5.3 additions and are absent there.
    // Verified against lua5.2.4: each is `nil`.
    if matches!(
        state.global().lua_version,
        lua_types::LuaVersion::V51 | lua_types::LuaVersion::V52
    ) {
        for field in [
            &b"type"[..],
            &b"tointeger"[..],
            &b"ult"[..],
            &b"maxinteger"[..],
            &b"mininteger"[..],
        ] {
            state.push(LuaValue::Nil);
            state.set_field(-2, field)?;
        }
    }

    set_rand_func(state)?;

    Ok(1)
}

// ──────────────────────────────────────────────────────────────────────────
// PORT STATUS
//   target_crate:  lua-stdlib
//   unsafe_blocks: 0
//   deferred:      one TODO — the PRNG state is a thread-local rather than
//                  per-lua_State typed userdata in a closure upvalue; see
//                  math_random. The PRNG/ldexp/frexp/version-gate behavior is
//                  pinned by the behavioral net (multiversion_oracle PRNG
//                  sequence + subnormal tests, the lua-stdlib FloatOnly test,
//                  math.lua, and check.sh 5.1-5.5). See GRADUATED.md "math".
//   version-gated: math.max/math.min use luaL_checknumber on 5.1/5.2 (reject
//                  non-numbers, return a coerced number) vs lua_compare on 5.3+.
//                  Known residual: the 5.1 arg-error function name is '?' in the
//                  reference but qualified ('math.max') here — same lua-vm
//                  arg_error_impl 5.1 name-resolution gap noted in base.rs.
// ──────────────────────────────────────────────────────────────────────────