excelerate 0.14.0

Read, write and recalculate spreadsheets: xlsx, xls, ods, csv, html, and a 518-function formula engine
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
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
//! Statistics.

use super::{Arg, aggregate_numbers, cells, selected};
use crate::error::CellError;
use crate::formula::value::Value;

/// `AVERAGE(number1, ...)`
pub fn average(args: &[Arg]) -> Value {
    match aggregate_numbers(args) {
        // An average of nothing has no answer; Excel says so with `#DIV/0!`.
        Ok(ns) if ns.is_empty() => Value::Error(CellError::Div0),
        #[expect(
            clippy::cast_precision_loss,
            reason = "a count that large cannot be reached"
        )]
        Ok(ns) => Value::Number(ns.iter().sum::<f64>() / ns.len() as f64),
        Err(e) => Value::Error(e),
    }
}

/// `MIN(number1, ...)`
pub fn min(args: &[Arg]) -> Value {
    extreme(args, f64::min)
}

/// `MAX(number1, ...)`
pub fn max(args: &[Arg]) -> Value {
    extreme(args, f64::max)
}

/// Shared body of `MIN` and `MAX`, which both answer 0 when nothing numeric
/// was passed.
fn extreme(args: &[Arg], pick: fn(f64, f64) -> f64) -> Value {
    match aggregate_numbers(args) {
        Ok(ns) => Value::Number(ns.into_iter().reduce(pick).unwrap_or(0.0)),
        Err(e) => Value::Error(e),
    }
}

/// `MEDIAN(number1, ...)`
pub fn median(args: &[Arg]) -> Value {
    match aggregate_numbers(args) {
        Ok(ns) if ns.is_empty() => Value::Error(CellError::Num),
        Ok(mut ns) => {
            ns.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
            let mid = ns.len() / 2;
            Value::Number(if ns.len() % 2 == 0 {
                f64::midpoint(ns[mid - 1], ns[mid])
            } else {
                ns[mid]
            })
        }
        Err(e) => Value::Error(e),
    }
}

/// `COUNT(value1, ...)` - how many of them are numbers.
pub fn count(args: &[Arg]) -> Value {
    let mut n = 0usize;
    for arg in args {
        let mut flat = Vec::new();
        arg.value.flatten(&mut flat);
        for v in flat {
            let counts = match v {
                Value::Number(_) => true,
                Value::Blank | Value::Error(_) | Value::Array(_) | Value::Lambda(_) => false,
                // Written into the formula, text and booleans count if they
                // convert; read out of a cell they never do.
                Value::Bool(_) | Value::Text(_) => !arg.reference && v.number().is_ok(),
            };
            n += usize::from(counts);
        }
    }
    count_value(n)
}

/// `COUNTA(value1, ...)` - how many are not empty.
pub fn counta(args: &[Arg]) -> Value {
    let mut n = 0usize;
    for arg in args {
        let mut flat = Vec::new();
        arg.value.flatten(&mut flat);
        n += flat.iter().filter(|v| !matches!(v, Value::Blank)).count();
    }
    count_value(n)
}

/// `COUNTBLANK(range)` - how many are empty.
///
/// The count only covers the part of the sheet that holds anything: a whole
/// column is clipped to its used rows before it reaches here, so
/// `COUNTBLANK(A:A)` answers about the used range rather than about a million
/// cells.
pub fn countblank(args: &[Arg]) -> Value {
    let mut n = 0usize;
    for arg in args {
        let mut flat = Vec::new();
        arg.value.flatten(&mut flat);
        n += flat
            .iter()
            .filter(|v| matches!(v, Value::Blank) || matches!(v, Value::Text(t) if t.is_empty()))
            .count();
    }
    count_value(n)
}

#[expect(
    clippy::cast_precision_loss,
    reason = "a sheet cannot hold enough cells to lose a digit here"
)]
fn count_value(n: usize) -> Value {
    Value::Number(n as f64)
}

/// `COUNTIF(range, criterion)`
pub fn countif(args: &[Arg]) -> Value {
    let [range, criterion] = args else {
        return Value::Error(CellError::Value);
    };
    conditional(&[range, criterion], None, |kept, _| count_value(kept.len()))
}

/// `COUNTIFS(range1, criterion1, ...)`
pub fn countifs(args: &[Arg]) -> Value {
    if args.is_empty() || !args.len().is_multiple_of(2) {
        return Value::Error(CellError::Value);
    }
    let pairs: Vec<&Arg> = args.iter().collect();
    conditional(&pairs, None, |kept, _| count_value(kept.len()))
}

/// `SUMIF(range, criterion, [sum_range])`
pub fn sumif(args: &[Arg]) -> Value {
    one_condition(args, |ns| {
        Value::Number(ns.iter().fold(0.0, |acc, n| acc + n))
    })
}

/// `SUMIFS(sum_range, range1, criterion1, ...)`
pub fn sumifs(args: &[Arg]) -> Value {
    many_conditions(args, |ns| {
        Value::Number(ns.iter().fold(0.0, |acc, n| acc + n))
    })
}

/// `AVERAGEIF(range, criterion, [average_range])`
pub fn averageif(args: &[Arg]) -> Value {
    one_condition(args, mean)
}

/// `AVERAGEIFS(average_range, range1, criterion1, ...)`
pub fn averageifs(args: &[Arg]) -> Value {
    many_conditions(args, mean)
}

/// `MEDIANIF(range, criteria, [median_range])` - the median of the values a
/// criterion keeps.
///
/// Undocumented by Microsoft and a stub (`Functions::DUMMY`), so
/// there is no oracle for it; it is spelled here the way `AVERAGEIF` is,
/// because that is the function it mirrors.
pub fn medianif(args: &[Arg]) -> Value {
    one_condition(args, middle)
}

/// The median of the numbers that were kept, or `#NUM!` when none were -
/// which is what `MEDIAN` itself answers for nothing.
fn middle(ns: &[f64]) -> Value {
    if ns.is_empty() {
        return Value::Error(CellError::Num);
    }
    let mut ns = ns.to_vec();
    ns.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
    let mid = ns.len() / 2;
    Value::Number(if ns.len().is_multiple_of(2) {
        f64::midpoint(ns[mid - 1], ns[mid])
    } else {
        ns[mid]
    })
}

/// The average of the numbers that were kept, or `#DIV/0!` when none were.
fn mean(ns: &[f64]) -> Value {
    if ns.is_empty() {
        return Value::Error(CellError::Div0);
    }
    #[expect(
        clippy::cast_precision_loss,
        reason = "a count that large cannot be reached"
    )]
    let count = ns.len() as f64;
    Value::Number(ns.iter().sum::<f64>() / count)
}

/// Shared body of `SUMIF` and `AVERAGEIF`, whose value range is optional and
/// comes last.
fn one_condition(args: &[Arg], body: fn(&[f64]) -> Value) -> Value {
    let (range, criterion, values) = match args {
        [r, c] => (r, c, r),
        [r, c, v] => (r, c, v),
        _ => return Value::Error(CellError::Value),
    };
    conditional(&[range, criterion], Some(values), |kept, values| {
        body(&numbers_at(&kept, values))
    })
}

/// Shared body of `SUMIFS` and `AVERAGEIFS`, whose value range comes first.
fn many_conditions(args: &[Arg], body: fn(&[f64]) -> Value) -> Value {
    let Some((values, rest)) = args.split_first() else {
        return Value::Error(CellError::Value);
    };
    if rest.is_empty() || !rest.len().is_multiple_of(2) {
        return Value::Error(CellError::Value);
    }
    let pairs: Vec<&Arg> = rest.iter().collect();
    conditional(&pairs, Some(values), |kept, values| {
        body(&numbers_at(&kept, values))
    })
}

/// Applies the criteria and hands the surviving positions to a body.
///
/// The value range, when there is one, is walked in step with the criteria
/// ranges: what matters is the position, not the value itself.
fn conditional(
    pairs: &[&Arg],
    values: Option<&Arg>,
    body: impl Fn(Vec<usize>, &[&Value]) -> Value,
) -> Value {
    let Some(first) = pairs.first() else {
        return Value::Error(CellError::Value);
    };
    // A range that is itself an error - `INDIRECT` of a name that does not
    // exist - is no range to count in. An error inside a range is a cell like
    // any other and only fails the criteria.
    let ranges = pairs.iter().step_by(2).copied().chain(values);
    for range in ranges {
        if let Value::Error(e) = range.value {
            return Value::Error(e);
        }
    }
    let len = cells(first).len();
    let keep = match selected(pairs, len) {
        Ok(keep) => keep,
        Err(e) => return Value::Error(e),
    };
    let positions: Vec<usize> = keep
        .iter()
        .enumerate()
        .filter_map(|(i, on)| on.then_some(i))
        .collect();
    let values = values.map(cells).unwrap_or_default();
    body(positions, &values)
}

/// The numbers at the given positions of a range, skipping anything else.
fn numbers_at(positions: &[usize], values: &[&Value]) -> Vec<f64> {
    positions
        .iter()
        .filter_map(|i| match values.get(*i) {
            Some(Value::Number(n)) => Some(*n),
            _ => None,
        })
        .collect()
}

/// `LARGE(array, k)` - the k-th largest number.
pub fn large(args: &[Arg]) -> Value {
    nth(args, true)
}

/// `SMALL(array, k)` - the k-th smallest.
pub fn small(args: &[Arg]) -> Value {
    nth(args, false)
}

/// Shared body of `LARGE` and `SMALL`.
fn nth(args: &[Arg], from_top: bool) -> Value {
    let [array, k] = args else {
        return Value::Error(CellError::Value);
    };
    let (Ok(mut ns), Ok(k)) = (aggregate_numbers(std::slice::from_ref(array)), k.number()) else {
        return super::read_error(&[k.number()]);
    };
    if ns.is_empty() || k < 1.0 {
        return Value::Error(CellError::Num);
    }
    ns.sort_by(|a, b| a.partial_cmp(b).unwrap_or(std::cmp::Ordering::Equal));
    #[expect(
        clippy::cast_possible_truncation,
        clippy::cast_sign_loss,
        reason = "checked at least 1; a larger k is caught as #NUM! below"
    )]
    let index = k.trunc() as usize - 1;
    let index = if from_top {
        match ns.len().checked_sub(index + 1) {
            Some(i) => i,
            None => return Value::Error(CellError::Num),
        }
    } else {
        index
    };
    ns.get(index)
        .map_or(Value::Error(CellError::Num), |n| Value::Number(*n))
}

/// `RANK(number, range, [order])` - 0 or omitted ranks downwards.
pub fn rank(args: &[Arg]) -> Value {
    let (number, range, ascending) = match args {
        [n, r] => (n.number(), r, Ok(0.0)),
        [n, r, o] => (n.number(), r, o.number()),
        _ => return Value::Error(CellError::Value),
    };
    let (Ok(number), Ok(order)) = (number, ascending) else {
        return super::read_error(&[number, ascending]);
    };
    let Ok(ns) = aggregate_numbers(std::slice::from_ref(range)) else {
        return Value::Error(CellError::Value);
    };
    if !ns.iter().any(|n| (n - number).abs() < f64::EPSILON) {
        return Value::Error(CellError::Na);
    }
    let ahead = ns
        .iter()
        .filter(|n| {
            if order == 0.0 {
                **n > number
            } else {
                **n < number
            }
        })
        .count();
    count_value(ahead + 1)
}

/// `STDEV(number1, ...)` - over a sample.
pub fn stdev(args: &[Arg]) -> Value {
    spread(args, true, true)
}

/// `STDEVP(number1, ...)` - over the whole population.
pub fn stdevp(args: &[Arg]) -> Value {
    spread(args, false, true)
}

/// `VAR(number1, ...)` - over a sample.
pub fn var(args: &[Arg]) -> Value {
    spread(args, true, false)
}

/// `VARP(number1, ...)` - over the whole population.
pub fn varp(args: &[Arg]) -> Value {
    spread(args, false, false)
}

/// Shared body of the spread functions: a sample divides by one less than the
/// count, a population by the count itself.
fn spread(args: &[Arg], sample: bool, root: bool) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    let wanted = if sample { 2 } else { 1 };
    if ns.len() < wanted {
        return Value::Error(CellError::Div0);
    }
    #[expect(
        clippy::cast_precision_loss,
        reason = "a count that large cannot be reached"
    )]
    let count = ns.len() as f64;
    let mean = ns.iter().sum::<f64>() / count;
    let sum_squares: f64 = ns.iter().map(|n| (n - mean) * (n - mean)).sum();
    let divisor = if sample { count - 1.0 } else { count };
    let variance = sum_squares / divisor;
    Value::Number(if root { variance.sqrt() } else { variance })
}

/// `AVEDEV(number1, ...)` - the mean distance from the mean.
pub fn avedev(args: &[Arg]) -> Value {
    around_mean(args, |ns, mean, count| {
        ns.iter().map(|n| (n - mean).abs()).sum::<f64>() / count
    })
}

/// `DEVSQ(number1, ...)` - the sum of the squared distances from the mean.
pub fn devsq(args: &[Arg]) -> Value {
    around_mean(args, |ns, mean, _| {
        ns.iter().map(|n| (n - mean) * (n - mean)).sum()
    })
}

/// Shared body of the two that measure spread around the mean directly.
fn around_mean(args: &[Arg], body: impl Fn(&[f64], f64, f64) -> f64) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    if ns.is_empty() {
        return Value::Error(CellError::Num);
    }
    let count = count_of(&ns);
    let mean = ns.iter().sum::<f64>() / count;
    Value::Number(body(&ns, mean, count))
}

/// `GEOMEAN(number1, ...)` - the n-th root of the product, which needs every
/// number to be positive.
pub fn geomean(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    if ns.is_empty() || ns.iter().any(|n| *n <= 0.0) {
        return Value::Error(CellError::Num);
    }
    // Summed as logarithms: the product of a long column overflows long before
    // its geometric mean does.
    let logs: f64 = ns.iter().map(|n| n.ln()).sum();
    Value::Number((logs / count_of(&ns)).exp())
}

/// `HARMEAN(number1, ...)` - the reciprocal of the mean of the reciprocals.
pub fn harmean(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    if ns.is_empty() || ns.iter().any(|n| *n <= 0.0) {
        return Value::Error(CellError::Num);
    }
    let sum: f64 = ns.iter().map(|n| 1.0 / n).sum();
    Value::Number(count_of(&ns) / sum)
}

/// `SKEW(number1, ...)` - how lopsided the distribution is, from a sample.
pub fn skew(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    let count = count_of(&ns);
    if ns.len() < 3 {
        return Value::Error(CellError::Div0);
    }
    let mean = ns.iter().sum::<f64>() / count;
    let deviation = sample_deviation(&ns, mean, count);
    if deviation == 0.0 {
        return Value::Error(CellError::Div0);
    }
    let sum: f64 = ns.iter().map(|n| ((n - mean) / deviation).powi(3)).sum();
    Value::Number(sum * count / ((count - 1.0) * (count - 2.0)))
}

/// `SKEW.P(number1, ...)` - the same for a whole population.
pub fn skewp(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    if ns.is_empty() {
        return Value::Error(CellError::Div0);
    }
    let count = count_of(&ns);
    let mean = ns.iter().sum::<f64>() / count;
    let variance = ns.iter().map(|n| (n - mean) * (n - mean)).sum::<f64>() / count;
    if variance == 0.0 {
        return Value::Error(CellError::Div0);
    }
    let deviation = variance.sqrt();
    let sum: f64 = ns.iter().map(|n| ((n - mean) / deviation).powi(3)).sum();
    Value::Number(sum / count)
}

/// `KURT(number1, ...)` - how heavy the tails are, against a normal curve.
pub fn kurt(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    if ns.len() < 4 {
        return Value::Error(CellError::Div0);
    }
    let count = count_of(&ns);
    let mean = ns.iter().sum::<f64>() / count;
    let deviation = sample_deviation(&ns, mean, count);
    if deviation == 0.0 {
        return Value::Error(CellError::Div0);
    }
    let sum: f64 = ns.iter().map(|n| ((n - mean) / deviation).powi(4)).sum();
    let scale = count * (count + 1.0) / ((count - 1.0) * (count - 2.0) * (count - 3.0));
    let correction = 3.0 * (count - 1.0).powi(2) / ((count - 2.0) * (count - 3.0));
    Value::Number(sum * scale - correction)
}

/// The sample standard deviation, which the shape statistics divide by.
fn sample_deviation(ns: &[f64], mean: f64, count: f64) -> f64 {
    (ns.iter().map(|n| (n - mean) * (n - mean)).sum::<f64>() / (count - 1.0)).sqrt()
}

/// `MODE(number1, ...)`, and `MODE.SNGL`, which is the same function.
///
/// The most common number; where several tie, the one that appears first.
pub fn mode(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    let mut best: Option<(f64, usize)> = None;
    for candidate in &ns {
        // The same number, exactly: `MODE` counts repeats, not near misses.
        let seen = ns
            .iter()
            .filter(|n| n.total_cmp(candidate) == core::cmp::Ordering::Equal)
            .count();
        // Strictly greater, so a tie is settled by whichever came first.
        if seen > 1 && best.is_none_or(|(_, most)| seen > most) {
            best = Some((*candidate, seen));
        }
    }
    // Excel has no mode for a list where nothing repeats.
    best.map_or(Value::Error(CellError::Na), |(n, _)| Value::Number(n))
}

/// `TRIMMEAN(array, percent)` - the mean after the extremes are dropped.
pub fn trimmean(args: &[Arg]) -> Value {
    let Some((percent, rest)) = args.split_last() else {
        return Value::Error(CellError::Value);
    };
    let Ok(percent) = percent.number() else {
        return Value::Error(CellError::Value);
    };
    if !(0.0..=1.0).contains(&percent) {
        return Value::Error(CellError::Num);
    }
    let mut ns = match aggregate_numbers(rest) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    if ns.is_empty() {
        return Value::Error(CellError::Num);
    }
    ns.sort_by(f64::total_cmp);
    // The count to drop is rounded down and then split between the two ends,
    // so the same number goes from each.
    #[expect(
        clippy::cast_possible_truncation,
        clippy::cast_sign_loss,
        reason = "a fraction of a count, floored, is a small non-negative number"
    )]
    let discard = (count_of(&ns) * percent / 2.0).floor() as usize;
    if discard * 2 >= ns.len() {
        return Value::Error(CellError::Num);
    }
    mean(&ns[discard..ns.len() - discard])
}

/// `STANDARDIZE(value, mean, deviation)` - how many deviations from the mean.
pub fn standardize(args: &[Arg]) -> Value {
    let [x, mean, deviation] = args else {
        return Value::Error(CellError::Value);
    };
    let (Ok(x), Ok(mean), Ok(deviation)) = (x.number(), mean.number(), deviation.number()) else {
        return Value::Error(CellError::Value);
    };
    if deviation <= 0.0 {
        return Value::Error(CellError::Num);
    }
    Value::Number((x - mean) / deviation)
}

/// `PERMUTATIONA(n, k)` - arrangements of k from n with repeats allowed.
pub fn permutationa(args: &[Arg]) -> Value {
    let [n, k] = args else {
        return Value::Error(CellError::Value);
    };
    let (Ok(n), Ok(k)) = (n.number(), k.number()) else {
        return Value::Error(CellError::Value);
    };
    let (n, k) = (n.trunc(), k.trunc());
    if n < 0.0 || k < 0.0 {
        return Value::Error(CellError::Num);
    }
    Value::Number(n.powf(k))
}

/// `PERCENTILE(array, k)`, and `PERCENTILE.INC`, which is the same function.
///
/// `k` runs from 0 to 1 over the sorted values, and a `k` landing between two
/// of them is interpolated.
pub fn percentile(args: &[Arg]) -> Value {
    let Some((k, rest)) = args.split_last() else {
        return Value::Error(CellError::Value);
    };
    let Ok(k) = k.number() else {
        return Value::Error(CellError::Value);
    };
    if !(0.0..=1.0).contains(&k) {
        return Value::Error(CellError::Num);
    }
    let Some(sorted) = sorted_numbers(rest) else {
        return Value::Error(CellError::Num);
    };
    Value::Number(at_position(&sorted, k * (count_of(&sorted) - 1.0)))
}

/// `PERCENTILE.EXC(array, k)` - the exclusive form, where `k` may not reach
/// either end: with n values it runs strictly between `1/(n+1)` and `n/(n+1)`.
pub fn percentile_exc(args: &[Arg]) -> Value {
    let Some((k, rest)) = args.split_last() else {
        return Value::Error(CellError::Value);
    };
    let Ok(k) = k.number() else {
        return Value::Error(CellError::Value);
    };
    let Some(sorted) = sorted_numbers(rest) else {
        return Value::Error(CellError::Num);
    };
    let count = count_of(&sorted);
    let position = k * (count + 1.0) - 1.0;
    if k <= 0.0 || k >= 1.0 || position < 0.0 || position > count - 1.0 {
        return Value::Error(CellError::Num);
    }
    Value::Number(at_position(&sorted, position))
}

/// `QUARTILE(array, quarter)`, and `QUARTILE.INC`: the percentile at a
/// quarter of the way along.
pub fn quartile(args: &[Arg]) -> Value {
    quarter_of(args, percentile)
}

/// `QUARTILE.EXC(array, quarter)`
pub fn quartile_exc(args: &[Arg]) -> Value {
    quarter_of(args, percentile_exc)
}

/// Shared body of the two quartile functions: a quarter is a percentile.
fn quarter_of(args: &[Arg], body: fn(&[Arg]) -> Value) -> Value {
    let Some((quarter, rest)) = args.split_last() else {
        return Value::Error(CellError::Value);
    };
    let Ok(quarter) = quarter.number() else {
        return Value::Error(CellError::Value);
    };
    let k = quarter.floor() / 4.0;
    if !(0.0..=1.0).contains(&k) {
        return Value::Error(CellError::Num);
    }
    let mut with_k: Vec<Arg> = rest.to_vec();
    with_k.push(Arg {
        value: Value::Number(k),
        reference: false,
    });
    body(&with_k)
}

/// `PERCENTRANK(array, value, [digits])`, and `PERCENTRANK.INC`.
pub fn percentrank(args: &[Arg]) -> Value {
    rank_within(args, false)
}

/// `PERCENTRANK.EXC(array, value, [digits])`
pub fn percentrank_exc(args: &[Arg]) -> Value {
    rank_within(args, true)
}

/// Shared body of the two percent-rank functions.
///
/// The inclusive form divides by `n - 1`, so the smallest value ranks 0 and
/// the largest 1; the exclusive form divides by `n + 1`, leaving room at both
/// ends. Both truncate rather than round, to three digits unless told.
fn rank_within(args: &[Arg], exclusive: bool) -> Value {
    let (array, value, digits) = match args {
        [a, v] => (a, v, Ok(3.0)),
        [a, v, d] => (a, v, d.number()),
        _ => return Value::Error(CellError::Value),
    };
    let (Ok(value), Ok(digits)) = (value.number(), digits) else {
        return Value::Error(CellError::Value);
    };
    if digits < 1.0 {
        return Value::Error(CellError::Num);
    }
    let Some(sorted) = sorted_numbers(core::slice::from_ref(array)) else {
        return Value::Error(CellError::Num);
    };
    let count = count_of(&sorted);
    if value < sorted[0] || value > sorted[sorted.len() - 1] {
        return Value::Error(CellError::Na);
    }
    // Where the value sits among the sorted ones, interpolated when it falls
    // between two of them.
    let exact = sorted
        .iter()
        .position(|n| n.total_cmp(&value) == core::cmp::Ordering::Equal);
    let position = if let Some(at) = exact {
        count_of(&sorted[..at])
    } else {
        let above = sorted.iter().position(|n| *n > value).unwrap_or(0);
        let below = above - 1;
        count_of(&sorted[..below]) + (value - sorted[below]) / (sorted[above] - sorted[below])
    };
    let rank = if exclusive {
        (position + 1.0) / (count + 1.0)
    } else {
        position / (count - 1.0)
    };
    #[expect(
        clippy::cast_possible_truncation,
        reason = "a digit count Excel caps well below what i32 holds"
    )]
    let scale = 10f64.powi(digits.trunc() as i32);
    Value::Number((rank * scale).trunc() / scale)
}

/// `CORREL(y, x)`, and `PEARSON`, which is the same coefficient.
pub fn correl(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let (sxx, syy, sxy) = moments(xs, ys);
        let denominator = (sxx * syy).sqrt();
        (denominator != 0.0).then(|| sxy / denominator)
    })
}

/// `RSQ(y, x)` - the square of the correlation.
pub fn rsq(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let (sxx, syy, sxy) = moments(xs, ys);
        let denominator = sxx * syy;
        (denominator != 0.0).then(|| sxy * sxy / denominator)
    })
}

/// `COVAR(y, x)`, and `COVARIANCE.P`: the covariance of a whole population.
pub fn covar(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let (_, _, sxy) = moments(xs, ys);
        Some(sxy / count_of(xs))
    })
}

/// `COVARIANCE.S(y, x)` - the same from a sample, so divided by one less.
pub fn covariance_s(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let (_, _, sxy) = moments(xs, ys);
        let count = count_of(xs);
        (count > 1.0).then(|| sxy / (count - 1.0))
    })
}

/// `SLOPE(y, x)` - the gradient of the line of best fit.
pub fn slope(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let (sxx, _, sxy) = moments(xs, ys);
        (sxx != 0.0).then(|| sxy / sxx)
    })
}

/// `INTERCEPT(y, x)` - where that line crosses the axis.
pub fn intercept(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let (sxx, _, sxy) = moments(xs, ys);
        if sxx == 0.0 {
            return None;
        }
        let count = count_of(xs);
        let (mx, my) = (
            xs.iter().sum::<f64>() / count,
            ys.iter().sum::<f64>() / count,
        );
        Some(my - (sxy / sxx) * mx)
    })
}

/// `STEYX(y, x)` - the standard error of the predicted y.
pub fn steyx(args: &[Arg]) -> Value {
    paired_stat(args, |xs, ys| {
        let count = count_of(xs);
        if count < 3.0 {
            return None;
        }
        let (sxx, syy, sxy) = moments(xs, ys);
        if sxx == 0.0 {
            return None;
        }
        Some(((syy - sxy * sxy / sxx) / (count - 2.0)).sqrt())
    })
}

/// `FORECAST(x, y_values, x_values)`, and `FORECAST.LINEAR`: the line of best
/// fit read off at a given x.
pub fn forecast(args: &[Arg]) -> Value {
    let [x, ys, xs] = args else {
        return Value::Error(CellError::Value);
    };
    let Ok(x) = x.number() else {
        return Value::Error(CellError::Value);
    };
    paired_stat(&[ys.clone(), xs.clone()], |xs, ys| {
        let (sxx, _, sxy) = moments(xs, ys);
        if sxx == 0.0 {
            return None;
        }
        let count = count_of(xs);
        let (mx, my) = (
            xs.iter().sum::<f64>() / count,
            ys.iter().sum::<f64>() / count,
        );
        let gradient = sxy / sxx;
        Some(my - gradient * mx + gradient * x)
    })
}

/// `FISHER(x)` - the transformation that makes a correlation normal.
pub fn fisher(args: &[Arg]) -> Value {
    super::one(args, |x| {
        if x <= -1.0 || x >= 1.0 {
            return Value::Error(CellError::Num);
        }
        Value::Number(0.5 * ((1.0 + x) / (1.0 - x)).ln())
    })
}

/// `FISHERINV(y)` - its inverse.
pub fn fisherinv(args: &[Arg]) -> Value {
    super::one(args, |y| {
        let e = (2.0 * y).exp();
        Value::Number((e - 1.0) / (e + 1.0))
    })
}

/// `PHI(x)` - the height of the standard normal curve at x.
pub fn phi(args: &[Arg]) -> Value {
    super::one(args, |x| {
        Value::Number((-x * x / 2.0).exp() / (2.0 * core::f64::consts::PI).sqrt())
    })
}

/// The three sums the regression statistics are all built from.
fn moments(xs: &[f64], ys: &[f64]) -> (f64, f64, f64) {
    let count = count_of(xs);
    let (mx, my) = (
        xs.iter().sum::<f64>() / count,
        ys.iter().sum::<f64>() / count,
    );
    let (mut sxx, mut syy, mut sxy) = (0.0, 0.0, 0.0);
    for (x, y) in xs.iter().zip(ys) {
        sxx += (x - mx) * (x - mx);
        syy += (y - my) * (y - my);
        sxy += (x - mx) * (y - my);
    }
    (sxx, syy, sxy)
}

/// Applies a body to two ranges walked in step.
///
/// A pair where either side is not a number is dropped, which is what keeps a
/// header row or a gap from skewing the line.
fn paired_stat(args: &[Arg], body: impl Fn(&[f64], &[f64]) -> Option<f64>) -> Value {
    let [ys, xs] = args else {
        return Value::Error(CellError::Value);
    };
    if let Some(e) = super::first_error(args) {
        return Value::Error(e);
    }
    let (ys, xs) = (cells(ys), cells(xs));
    if ys.len() != xs.len() {
        return Value::Error(CellError::Na);
    }
    let (mut x_kept, mut y_kept) = (Vec::new(), Vec::new());
    for (y, x) in ys.iter().zip(&xs) {
        if let (Value::Number(y), Value::Number(x)) = (y.scalar(), x.scalar()) {
            y_kept.push(*y);
            x_kept.push(*x);
        }
    }
    if x_kept.is_empty() {
        return Value::Error(CellError::Div0);
    }
    body(&x_kept, &y_kept).map_or(Value::Error(CellError::Div0), Value::Number)
}

/// The numbers of the arguments, sorted, or nothing when there are none.
fn sorted_numbers(args: &[Arg]) -> Option<Vec<f64>> {
    let mut ns = aggregate_numbers(args).ok()?;
    if ns.is_empty() {
        return None;
    }
    ns.sort_by(f64::total_cmp);
    Some(ns)
}

/// The value at a fractional position along a sorted list.
fn at_position(sorted: &[f64], position: f64) -> f64 {
    let floor = position.floor();
    #[expect(
        clippy::cast_possible_truncation,
        clippy::cast_sign_loss,
        reason = "a position inside a list this crate could build"
    )]
    let base = floor as usize;
    // Landing exactly on a value needs no special case: the interpolation
    // below multiplies the step by zero, and the last value has no next one
    // to step towards.
    let next = sorted.get(base + 1).copied().unwrap_or(sorted[base]);
    sorted[base] + (next - sorted[base]) * (position - floor)
}

/// A count as a float, for the arithmetic every one of these does.
#[expect(
    clippy::cast_precision_loss,
    reason = "a count that large cannot be reached"
)]
fn count_of<T>(items: &[T]) -> f64 {
    items.len() as f64
}

/// `AVERAGEA(value1, ...)` - the mean counting text as zero.
pub fn averagea(args: &[Arg]) -> Value {
    with_text_as_zero(args, mean)
}

/// `MAXA(value1, ...)`
pub fn maxa(args: &[Arg]) -> Value {
    with_text_as_zero(args, |ns| extreme_of(ns, f64::max))
}

/// `MINA(value1, ...)`
pub fn mina(args: &[Arg]) -> Value {
    with_text_as_zero(args, |ns| extreme_of(ns, f64::min))
}

/// `STDEVA(value1, ...)`
pub fn stdeva(args: &[Arg]) -> Value {
    with_text_as_zero(args, |ns| spread_of(ns, true, true))
}

/// `STDEVPA(value1, ...)`
pub fn stdevpa(args: &[Arg]) -> Value {
    with_text_as_zero(args, |ns| spread_of(ns, false, true))
}

/// `VARA(value1, ...)`
pub fn vara(args: &[Arg]) -> Value {
    with_text_as_zero(args, |ns| spread_of(ns, true, false))
}

/// `VARPA(value1, ...)`
pub fn varpa(args: &[Arg]) -> Value {
    with_text_as_zero(args, |ns| spread_of(ns, false, false))
}

/// The shared body of the `A` forms.
///
/// Where `AVERAGE` skips the text and the booleans it finds in cells, these
/// count them: text is zero, `TRUE` is one and `FALSE` is zero. An empty cell
/// is still skipped by both.
fn with_text_as_zero(args: &[Arg], body: impl Fn(&[f64]) -> Value) -> Value {
    let mut ns = Vec::new();
    for arg in args {
        // Text read out of cells or an array counts as zero; text written as
        // the argument itself has to be a number, as `AVERAGEA(1,"x")` is
        // `#VALUE!` in Excel.
        let written = !arg.reference && !matches!(arg.value, Value::Array(_));
        let mut flat = Vec::new();
        arg.value.flatten(&mut flat);
        for v in flat {
            match v {
                Value::Error(e) => return Value::Error(*e),
                Value::Number(n) => ns.push(*n),
                Value::Bool(b) => ns.push(f64::from(u8::from(*b))),
                Value::Text(_) if written => match v.number() {
                    Ok(n) => ns.push(n),
                    Err(e) => return Value::Error(e),
                },
                Value::Text(_) => ns.push(0.0),
                Value::Blank | Value::Array(_) | Value::Lambda(_) => {}
            }
        }
    }
    body(&ns)
}

/// The largest or smallest of a list already reduced to numbers.
fn extreme_of(ns: &[f64], pick: fn(f64, f64) -> f64) -> Value {
    if ns.is_empty() {
        return Value::Number(0.0);
    }
    Value::Number(ns.iter().copied().fold(ns[0], pick))
}

/// The variance or standard deviation of a list already reduced to numbers.
fn spread_of(ns: &[f64], sample: bool, root: bool) -> Value {
    if ns.len() < usize::from(sample) + 1 {
        return Value::Error(CellError::Div0);
    }
    let count = count_of(ns);
    let mean = ns.iter().sum::<f64>() / count;
    let squares: f64 = ns.iter().map(|n| (n - mean) * (n - mean)).sum();
    let variance = squares / if sample { count - 1.0 } else { count };
    Value::Number(if root { variance.sqrt() } else { variance })
}

/// `MAXIFS(values, range, criterion, ...)`
pub fn maxifs(args: &[Arg]) -> Value {
    extreme_ifs(args, f64::max)
}

/// `MINIFS(values, range, criterion, ...)`
pub fn minifs(args: &[Arg]) -> Value {
    extreme_ifs(args, f64::min)
}

/// Shared body of `MAXIFS` and `MINIFS`, which take their value range first
/// the way `SUMIFS` does.
fn extreme_ifs(args: &[Arg], pick: fn(f64, f64) -> f64) -> Value {
    let Some((values, rest)) = args.split_first() else {
        return Value::Error(CellError::Value);
    };
    if rest.is_empty() || !rest.len().is_multiple_of(2) {
        return Value::Error(CellError::Value);
    }
    let pairs: Vec<&Arg> = rest.iter().collect();
    conditional(&pairs, Some(values), |kept, values| {
        let ns = numbers_at(&kept, values);
        // Excel answers zero when nothing matched, rather than an error.
        extreme_of(&ns, pick)
    })
}

/// `RANK.AVG(number, ref, [order])` - like `RANK`, but ties share the average
/// of the places they take up rather than all taking the first of them.
pub fn rank_avg(args: &[Arg]) -> Value {
    let (needle, list, ascending) = match args {
        [n, r] => (n.number(), r, Ok(0.0)),
        [n, r, o] => (n.number(), r, o.number()),
        _ => return Value::Error(CellError::Value),
    };
    let (Ok(needle), Ok(ascending)) = (needle, ascending) else {
        return Value::Error(CellError::Value);
    };
    let ns: Vec<f64> = cells(list)
        .into_iter()
        .filter_map(|v| match v.scalar() {
            Value::Number(n) => Some(*n),
            _ => None,
        })
        .collect();
    let ahead = ns
        .iter()
        .filter(|n| {
            if ascending == 0.0 {
                **n > needle
            } else {
                **n < needle
            }
        })
        .count();
    let tied = ns
        .iter()
        .filter(|n| n.total_cmp(&needle) == core::cmp::Ordering::Equal)
        .count();
    if tied == 0 {
        return Value::Error(CellError::Na);
    }
    // The places the tie takes up run from `ahead + 1` upwards; their average
    // is the first plus half of what is left.
    let first = count_of(&ns[..ahead]) + 1.0;
    Value::Number(first + (count_of(&ns[..tied]) - 1.0) / 2.0)
}

/// `MODE.MULT(number1, ...)` - every value tied for most common, as a column.
pub fn mode_mult(args: &[Arg]) -> Value {
    let ns = match aggregate_numbers(args) {
        Ok(ns) => ns,
        Err(e) => return Value::Error(e),
    };
    let count_same = |candidate: &f64| {
        ns.iter()
            .filter(|n| n.total_cmp(candidate) == core::cmp::Ordering::Equal)
            .count()
    };
    let most = ns.iter().map(count_same).max().unwrap_or(0);
    if most < 2 {
        return Value::Error(CellError::Na);
    }
    let mut out: Vec<Vec<Value>> = Vec::new();
    for (i, n) in ns.iter().enumerate() {
        // Each tied value once, at the place it first appeared.
        let first = ns[..i]
            .iter()
            .all(|earlier| earlier.total_cmp(n) != core::cmp::Ordering::Equal);
        if first && count_same(n) == most {
            out.push(vec![Value::Number(*n)]);
        }
    }
    Value::array(out)
}

/// `FREQUENCY(data, bins)` - how many of the data fall in each bin, as a
/// column one longer than the bins: the last row is everything above them.
pub fn frequency(args: &[Arg]) -> Value {
    let [data, bins] = args else {
        return Value::Error(CellError::Value);
    };
    if let Some(e) = super::first_error(args) {
        return Value::Error(e);
    }
    let numbers = |arg: &Arg| -> Vec<f64> {
        cells(arg)
            .into_iter()
            .filter_map(|v| match v.scalar() {
                Value::Number(n) => Some(*n),
                _ => None,
            })
            .collect()
    };
    let data = numbers(data);
    let mut edges = numbers(bins);
    edges.sort_by(f64::total_cmp);
    let mut counts = vec![0.0; edges.len() + 1];
    for value in data {
        // The first bin whose edge the value does not exceed takes it.
        let bin = edges.iter().position(|edge| value <= *edge);
        counts[bin.unwrap_or(edges.len())] += 1.0;
    }
    Value::array(counts.into_iter().map(|n| vec![Value::Number(n)]).collect())
}

/// `PROB(values, probabilities, low, [high])` - the chance of landing in a
/// range, given the chance of each value.
pub fn prob(args: &[Arg]) -> Value {
    if let Some(e) = super::first_error(args) {
        return Value::Error(e);
    }
    let (values, chances, low, high) = match args {
        [v, c, l] => (v, c, l.number(), None),
        [v, c, l, h] => (v, c, l.number(), Some(h.number())),
        _ => return Value::Error(CellError::Value),
    };
    let Ok(low) = low else {
        return Value::Error(CellError::Value);
    };
    let high = match high {
        Some(Ok(h)) => h,
        Some(Err(_)) => return Value::Error(CellError::Value),
        // With one bound the range is that single value.
        None => low,
    };
    let (values, chances) = (cells(values), cells(chances));
    if values.len() != chances.len() {
        return Value::Error(CellError::Na);
    }
    let mut total = 0.0;
    let mut all = 0.0;
    for (value, chance) in values.iter().zip(&chances) {
        let (Value::Number(value), Value::Number(chance)) = (value.scalar(), chance.scalar())
        else {
            continue;
        };
        if *chance < 0.0 || *chance > 1.0 {
            return Value::Error(CellError::Num);
        }
        all += chance;
        if (low..=high).contains(value) {
            total += chance;
        }
    }
    // Excel insists the chances add up to one before it answers.
    if (all - 1.0).abs() > 1.0e-9 {
        return Value::Error(CellError::Num);
    }
    Value::Number(total)
}