1 | //! The purpose of these tests is to cover corner cases of iterators |
2 | //! and adaptors. |
3 | //! |
4 | //! In particular we test the tedious size_hint and exact size correctness. |
5 | |
6 | use quickcheck as qc; |
7 | use std::default::Default; |
8 | use std::num::Wrapping; |
9 | use std::ops::Range; |
10 | use std::cmp::{max, min, Ordering}; |
11 | use std::collections::{HashMap, HashSet}; |
12 | use itertools::Itertools; |
13 | use itertools::{ |
14 | multizip, |
15 | EitherOrBoth, |
16 | iproduct, |
17 | izip, |
18 | }; |
19 | use itertools::free::{ |
20 | cloned, |
21 | enumerate, |
22 | multipeek, |
23 | peek_nth, |
24 | put_back, |
25 | put_back_n, |
26 | rciter, |
27 | zip, |
28 | zip_eq, |
29 | }; |
30 | |
31 | use rand::Rng; |
32 | use rand::seq::SliceRandom; |
33 | use quickcheck::TestResult; |
34 | |
35 | /// Trait for size hint modifier types |
36 | trait HintKind: Copy + Send + qc::Arbitrary { |
37 | fn loosen_bounds(&self, org_hint: (usize, Option<usize>)) -> (usize, Option<usize>); |
38 | } |
39 | |
40 | /// Exact size hint variant that leaves hints unchanged |
41 | #[derive(Clone, Copy, Debug)] |
42 | struct Exact {} |
43 | |
44 | impl HintKind for Exact { |
45 | fn loosen_bounds(&self, org_hint: (usize, Option<usize>)) -> (usize, Option<usize>) { |
46 | org_hint |
47 | } |
48 | } |
49 | |
50 | impl qc::Arbitrary for Exact { |
51 | fn arbitrary<G: qc::Gen>(_: &mut G) -> Self { |
52 | Exact {} |
53 | } |
54 | } |
55 | |
56 | /// Inexact size hint variant to simulate imprecise (but valid) size hints |
57 | /// |
58 | /// Will always decrease the lower bound and increase the upper bound |
59 | /// of the size hint by set amounts. |
60 | #[derive(Clone, Copy, Debug)] |
61 | struct Inexact { |
62 | underestimate: usize, |
63 | overestimate: usize, |
64 | } |
65 | |
66 | impl HintKind for Inexact { |
67 | fn loosen_bounds(&self, org_hint: (usize, Option<usize>)) -> (usize, Option<usize>) { |
68 | let (org_lower, org_upper) = org_hint; |
69 | (org_lower.saturating_sub(self.underestimate), |
70 | org_upper.and_then(move |x| x.checked_add(self.overestimate))) |
71 | } |
72 | } |
73 | |
74 | impl qc::Arbitrary for Inexact { |
75 | fn arbitrary<G: qc::Gen>(g: &mut G) -> Self { |
76 | let ue_value = usize::arbitrary(g); |
77 | let oe_value = usize::arbitrary(g); |
78 | // Compensate for quickcheck using extreme values too rarely |
79 | let ue_choices = &[0, ue_value, usize::max_value()]; |
80 | let oe_choices = &[0, oe_value, usize::max_value()]; |
81 | Inexact { |
82 | underestimate: *ue_choices.choose(g).unwrap(), |
83 | overestimate: *oe_choices.choose(g).unwrap(), |
84 | } |
85 | } |
86 | |
87 | fn shrink(&self) -> Box<dyn Iterator<Item=Self>> { |
88 | let underestimate_value = self.underestimate; |
89 | let overestimate_value = self.overestimate; |
90 | Box::new( |
91 | underestimate_value.shrink().flat_map(move |ue_value| |
92 | overestimate_value.shrink().map(move |oe_value| |
93 | Inexact { |
94 | underestimate: ue_value, |
95 | overestimate: oe_value, |
96 | } |
97 | ) |
98 | ) |
99 | ) |
100 | } |
101 | } |
102 | |
103 | /// Our base iterator that we can impl Arbitrary for |
104 | /// |
105 | /// By default we'll return inexact bounds estimates for size_hint |
106 | /// to make tests harder to pass. |
107 | /// |
108 | /// NOTE: Iter is tricky and is not fused, to help catch bugs. |
109 | /// At the end it will return None once, then return Some(0), |
110 | /// then return None again. |
111 | #[derive(Clone, Debug)] |
112 | struct Iter<T, SK: HintKind = Inexact> { |
113 | iterator: Range<T>, |
114 | // fuse/done flag |
115 | fuse_flag: i32, |
116 | hint_kind: SK, |
117 | } |
118 | |
119 | impl<T, HK> Iter<T, HK> where HK: HintKind |
120 | { |
121 | fn new(it: Range<T>, hint_kind: HK) -> Self { |
122 | Iter { |
123 | iterator: it, |
124 | fuse_flag: 0, |
125 | hint_kind, |
126 | } |
127 | } |
128 | } |
129 | |
130 | impl<T, HK> Iterator for Iter<T, HK> |
131 | where Range<T>: Iterator, |
132 | <Range<T> as Iterator>::Item: Default, |
133 | HK: HintKind, |
134 | { |
135 | type Item = <Range<T> as Iterator>::Item; |
136 | |
137 | fn next(&mut self) -> Option<Self::Item> |
138 | { |
139 | let elt = self.iterator.next(); |
140 | if elt.is_none() { |
141 | self.fuse_flag += 1; |
142 | // check fuse flag |
143 | if self.fuse_flag == 2 { |
144 | return Some(Default::default()) |
145 | } |
146 | } |
147 | elt |
148 | } |
149 | |
150 | fn size_hint(&self) -> (usize, Option<usize>) |
151 | { |
152 | let org_hint = self.iterator.size_hint(); |
153 | self.hint_kind.loosen_bounds(org_hint) |
154 | } |
155 | } |
156 | |
157 | impl<T, HK> DoubleEndedIterator for Iter<T, HK> |
158 | where Range<T>: DoubleEndedIterator, |
159 | <Range<T> as Iterator>::Item: Default, |
160 | HK: HintKind |
161 | { |
162 | fn next_back(&mut self) -> Option<Self::Item> { self.iterator.next_back() } |
163 | } |
164 | |
165 | impl<T> ExactSizeIterator for Iter<T, Exact> where Range<T>: ExactSizeIterator, |
166 | <Range<T> as Iterator>::Item: Default, |
167 | { } |
168 | |
169 | impl<T, HK> qc::Arbitrary for Iter<T, HK> |
170 | where T: qc::Arbitrary, |
171 | HK: HintKind, |
172 | { |
173 | fn arbitrary<G: qc::Gen>(g: &mut G) -> Self |
174 | { |
175 | Iter::new(T::arbitrary(g)..T::arbitrary(g), HK::arbitrary(g)) |
176 | } |
177 | |
178 | fn shrink(&self) -> Box<dyn Iterator<Item=Iter<T, HK>>> |
179 | { |
180 | let r = self.iterator.clone(); |
181 | let hint_kind = self.hint_kind; |
182 | Box::new( |
183 | r.start.shrink().flat_map(move |a| |
184 | r.end.shrink().map(move |b| |
185 | Iter::new(a.clone()..b, hint_kind) |
186 | ) |
187 | ) |
188 | ) |
189 | } |
190 | } |
191 | |
192 | /// A meta-iterator which yields `Iter<i32>`s whose start/endpoints are |
193 | /// increased or decreased linearly on each iteration. |
194 | #[derive(Clone, Debug)] |
195 | struct ShiftRange<HK = Inexact> { |
196 | range_start: i32, |
197 | range_end: i32, |
198 | start_step: i32, |
199 | end_step: i32, |
200 | iter_count: u32, |
201 | hint_kind: HK, |
202 | } |
203 | |
204 | impl<HK> Iterator for ShiftRange<HK> where HK: HintKind { |
205 | type Item = Iter<i32, HK>; |
206 | |
207 | fn next(&mut self) -> Option<Self::Item> { |
208 | if self.iter_count == 0 { |
209 | return None; |
210 | } |
211 | |
212 | let iter = Iter::new(self.range_start..self.range_end, self.hint_kind); |
213 | |
214 | self.range_start += self.start_step; |
215 | self.range_end += self.end_step; |
216 | self.iter_count -= 1; |
217 | |
218 | Some(iter) |
219 | } |
220 | } |
221 | |
222 | impl ExactSizeIterator for ShiftRange<Exact> { } |
223 | |
224 | impl<HK> qc::Arbitrary for ShiftRange<HK> |
225 | where HK: HintKind |
226 | { |
227 | fn arbitrary<G: qc::Gen>(g: &mut G) -> Self { |
228 | const MAX_STARTING_RANGE_DIFF: i32 = 32; |
229 | const MAX_STEP_MODULO: i32 = 8; |
230 | const MAX_ITER_COUNT: u32 = 3; |
231 | |
232 | let range_start = qc::Arbitrary::arbitrary(g); |
233 | let range_end = range_start + g.gen_range(0, MAX_STARTING_RANGE_DIFF + 1); |
234 | let start_step = g.gen_range(-MAX_STEP_MODULO, MAX_STEP_MODULO + 1); |
235 | let end_step = g.gen_range(-MAX_STEP_MODULO, MAX_STEP_MODULO + 1); |
236 | let iter_count = g.gen_range(0, MAX_ITER_COUNT + 1); |
237 | let hint_kind = qc::Arbitrary::arbitrary(g); |
238 | |
239 | ShiftRange { |
240 | range_start, |
241 | range_end, |
242 | start_step, |
243 | end_step, |
244 | iter_count, |
245 | hint_kind, |
246 | } |
247 | } |
248 | } |
249 | |
250 | fn correct_count<I, F>(get_it: F) -> bool |
251 | where |
252 | I: Iterator, |
253 | F: Fn() -> I |
254 | { |
255 | let mut counts = vec![get_it().count()]; |
256 | |
257 | 'outer: loop { |
258 | let mut it = get_it(); |
259 | |
260 | for _ in 0..(counts.len() - 1) { |
261 | #[allow (clippy::manual_assert)] |
262 | if it.next().is_none() { |
263 | panic!("Iterator shouldn't be finished, may not be deterministic" ); |
264 | } |
265 | } |
266 | |
267 | if it.next().is_none() { |
268 | break 'outer; |
269 | } |
270 | |
271 | counts.push(it.count()); |
272 | } |
273 | |
274 | let total_actual_count = counts.len() - 1; |
275 | |
276 | for (i, returned_count) in counts.into_iter().enumerate() { |
277 | let actual_count = total_actual_count - i; |
278 | if actual_count != returned_count { |
279 | println!("Total iterations: {} True count: {} returned count: {}" , i, actual_count, returned_count); |
280 | |
281 | return false; |
282 | } |
283 | } |
284 | |
285 | true |
286 | } |
287 | |
288 | fn correct_size_hint<I: Iterator>(mut it: I) -> bool { |
289 | // record size hint at each iteration |
290 | let initial_hint = it.size_hint(); |
291 | let mut hints = Vec::with_capacity(initial_hint.0 + 1); |
292 | hints.push(initial_hint); |
293 | while let Some(_) = it.next() { |
294 | hints.push(it.size_hint()) |
295 | } |
296 | |
297 | let mut true_count = hints.len(); // start off +1 too much |
298 | |
299 | // check all the size hints |
300 | for &(low, hi) in &hints { |
301 | true_count -= 1; |
302 | if low > true_count || |
303 | (hi.is_some() && hi.unwrap() < true_count) |
304 | { |
305 | println!("True size: {:?}, size hint: {:?}" , true_count, (low, hi)); |
306 | //println!("All hints: {:?}", hints); |
307 | return false |
308 | } |
309 | } |
310 | true |
311 | } |
312 | |
313 | fn exact_size<I: ExactSizeIterator>(mut it: I) -> bool { |
314 | // check every iteration |
315 | let (mut low, mut hi) = it.size_hint(); |
316 | if Some(low) != hi { return false; } |
317 | while let Some(_) = it.next() { |
318 | let (xlow, xhi) = it.size_hint(); |
319 | if low != xlow + 1 { return false; } |
320 | low = xlow; |
321 | hi = xhi; |
322 | if Some(low) != hi { return false; } |
323 | } |
324 | let (low, hi) = it.size_hint(); |
325 | low == 0 && hi == Some(0) |
326 | } |
327 | |
328 | // Exact size for this case, without ExactSizeIterator |
329 | fn exact_size_for_this<I: Iterator>(mut it: I) -> bool { |
330 | // check every iteration |
331 | let (mut low, mut hi) = it.size_hint(); |
332 | if Some(low) != hi { return false; } |
333 | while let Some(_) = it.next() { |
334 | let (xlow, xhi) = it.size_hint(); |
335 | if low != xlow + 1 { return false; } |
336 | low = xlow; |
337 | hi = xhi; |
338 | if Some(low) != hi { return false; } |
339 | } |
340 | let (low, hi) = it.size_hint(); |
341 | low == 0 && hi == Some(0) |
342 | } |
343 | |
344 | /* |
345 | * NOTE: Range<i8> is broken! |
346 | * (all signed ranges are) |
347 | #[quickcheck] |
348 | fn size_range_i8(a: Iter<i8>) -> bool { |
349 | exact_size(a) |
350 | } |
351 | |
352 | #[quickcheck] |
353 | fn size_range_i16(a: Iter<i16>) -> bool { |
354 | exact_size(a) |
355 | } |
356 | |
357 | #[quickcheck] |
358 | fn size_range_u8(a: Iter<u8>) -> bool { |
359 | exact_size(a) |
360 | } |
361 | */ |
362 | |
363 | macro_rules! quickcheck { |
364 | // accept several property function definitions |
365 | // The property functions can use pattern matching and `mut` as usual |
366 | // in the function arguments, but the functions can not be generic. |
367 | {$($(#$attr:tt)* fn $fn_name:ident($($arg:tt)*) -> $ret:ty { $($code:tt)* })*} => ( |
368 | $( |
369 | #[test] |
370 | $(#$attr)* |
371 | fn $fn_name() { |
372 | fn prop($($arg)*) -> $ret { |
373 | $($code)* |
374 | } |
375 | ::quickcheck::quickcheck(quickcheck!(@fn prop [] $($arg)*)); |
376 | } |
377 | )* |
378 | ); |
379 | // parse argument list (with patterns allowed) into prop as fn(_, _) -> _ |
380 | (@fn $f:ident [$($t:tt)*]) => { |
381 | $f as fn($($t),*) -> _ |
382 | }; |
383 | (@fn $f:ident [$($p:tt)*] : $($tail:tt)*) => { |
384 | quickcheck!(@fn $f [$($p)* _] $($tail)*) |
385 | }; |
386 | (@fn $f:ident [$($p:tt)*] $t:tt $($tail:tt)*) => { |
387 | quickcheck!(@fn $f [$($p)*] $($tail)*) |
388 | }; |
389 | } |
390 | |
391 | quickcheck! { |
392 | |
393 | fn size_product(a: Iter<u16>, b: Iter<u16>) -> bool { |
394 | correct_size_hint(a.cartesian_product(b)) |
395 | } |
396 | fn size_product3(a: Iter<u16>, b: Iter<u16>, c: Iter<u16>) -> bool { |
397 | correct_size_hint(iproduct!(a, b, c)) |
398 | } |
399 | |
400 | fn correct_cartesian_product3(a: Iter<u16>, b: Iter<u16>, c: Iter<u16>, |
401 | take_manual: usize) -> () |
402 | { |
403 | // test correctness of iproduct through regular iteration (take) |
404 | // and through fold. |
405 | let ac = a.clone(); |
406 | let br = &b.clone(); |
407 | let cr = &c.clone(); |
408 | let answer: Vec<_> = ac.flat_map(move |ea| br.clone().flat_map(move |eb| cr.clone().map(move |ec| (ea, eb, ec)))).collect(); |
409 | let mut product_iter = iproduct!(a, b, c); |
410 | let mut actual = Vec::new(); |
411 | |
412 | actual.extend((&mut product_iter).take(take_manual)); |
413 | if actual.len() == take_manual { |
414 | product_iter.fold((), |(), elt| actual.push(elt)); |
415 | } |
416 | assert_eq!(answer, actual); |
417 | } |
418 | |
419 | fn size_multi_product(a: ShiftRange) -> bool { |
420 | correct_size_hint(a.multi_cartesian_product()) |
421 | } |
422 | fn correct_multi_product3(a: ShiftRange, take_manual: usize) -> () { |
423 | // Fix no. of iterators at 3 |
424 | let a = ShiftRange { iter_count: 3, ..a }; |
425 | |
426 | // test correctness of MultiProduct through regular iteration (take) |
427 | // and through fold. |
428 | let mut iters = a.clone(); |
429 | let i0 = iters.next().unwrap(); |
430 | let i1r = &iters.next().unwrap(); |
431 | let i2r = &iters.next().unwrap(); |
432 | let answer: Vec<_> = i0.flat_map(move |ei0| i1r.clone().flat_map(move |ei1| i2r.clone().map(move |ei2| vec![ei0, ei1, ei2]))).collect(); |
433 | let mut multi_product = a.clone().multi_cartesian_product(); |
434 | let mut actual = Vec::new(); |
435 | |
436 | actual.extend((&mut multi_product).take(take_manual)); |
437 | if actual.len() == take_manual { |
438 | multi_product.fold((), |(), elt| actual.push(elt)); |
439 | } |
440 | assert_eq!(answer, actual); |
441 | |
442 | assert_eq!(answer.into_iter().last(), a.multi_cartesian_product().last()); |
443 | } |
444 | |
445 | #[allow (deprecated)] |
446 | fn size_step(a: Iter<i16, Exact>, s: usize) -> bool { |
447 | let mut s = s; |
448 | if s == 0 { |
449 | s += 1; // never zero |
450 | } |
451 | let filt = a.clone().dedup(); |
452 | correct_size_hint(filt.step(s)) && |
453 | exact_size(a.step(s)) |
454 | } |
455 | |
456 | #[allow (deprecated)] |
457 | fn equal_step(a: Iter<i16>, s: usize) -> bool { |
458 | let mut s = s; |
459 | if s == 0 { |
460 | s += 1; // never zero |
461 | } |
462 | let mut i = 0; |
463 | itertools::equal(a.clone().step(s), a.filter(|_| { |
464 | let keep = i % s == 0; |
465 | i += 1; |
466 | keep |
467 | })) |
468 | } |
469 | |
470 | #[allow (deprecated)] |
471 | fn equal_step_vec(a: Vec<i16>, s: usize) -> bool { |
472 | let mut s = s; |
473 | if s == 0 { |
474 | s += 1; // never zero |
475 | } |
476 | let mut i = 0; |
477 | itertools::equal(a.iter().step(s), a.iter().filter(|_| { |
478 | let keep = i % s == 0; |
479 | i += 1; |
480 | keep |
481 | })) |
482 | } |
483 | |
484 | fn size_multipeek(a: Iter<u16, Exact>, s: u8) -> bool { |
485 | let mut it = multipeek(a); |
486 | // peek a few times |
487 | for _ in 0..s { |
488 | it.peek(); |
489 | } |
490 | exact_size(it) |
491 | } |
492 | |
493 | fn size_peek_nth(a: Iter<u16, Exact>, s: u8) -> bool { |
494 | let mut it = peek_nth(a); |
495 | // peek a few times |
496 | for n in 0..s { |
497 | it.peek_nth(n as usize); |
498 | } |
499 | exact_size(it) |
500 | } |
501 | |
502 | fn equal_merge(mut a: Vec<i16>, mut b: Vec<i16>) -> bool { |
503 | a.sort(); |
504 | b.sort(); |
505 | let mut merged = a.clone(); |
506 | merged.extend(b.iter().cloned()); |
507 | merged.sort(); |
508 | itertools::equal(&merged, a.iter().merge(&b)) |
509 | } |
510 | fn size_merge(a: Iter<u16>, b: Iter<u16>) -> bool { |
511 | correct_size_hint(a.merge(b)) |
512 | } |
513 | fn size_zip(a: Iter<i16, Exact>, b: Iter<i16, Exact>, c: Iter<i16, Exact>) -> bool { |
514 | let filt = a.clone().dedup(); |
515 | correct_size_hint(multizip((filt, b.clone(), c.clone()))) && |
516 | exact_size(multizip((a, b, c))) |
517 | } |
518 | fn size_zip_rc(a: Iter<i16>, b: Iter<i16>) -> bool { |
519 | let rc = rciter(a); |
520 | correct_size_hint(multizip((&rc, &rc, b))) |
521 | } |
522 | |
523 | fn size_zip_macro(a: Iter<i16, Exact>, b: Iter<i16, Exact>, c: Iter<i16, Exact>) -> bool { |
524 | let filt = a.clone().dedup(); |
525 | correct_size_hint(izip!(filt, b.clone(), c.clone())) && |
526 | exact_size(izip!(a, b, c)) |
527 | } |
528 | fn equal_kmerge(mut a: Vec<i16>, mut b: Vec<i16>, mut c: Vec<i16>) -> bool { |
529 | use itertools::free::kmerge; |
530 | a.sort(); |
531 | b.sort(); |
532 | c.sort(); |
533 | let mut merged = a.clone(); |
534 | merged.extend(b.iter().cloned()); |
535 | merged.extend(c.iter().cloned()); |
536 | merged.sort(); |
537 | itertools::equal(merged.into_iter(), kmerge(vec![a, b, c])) |
538 | } |
539 | |
540 | // Any number of input iterators |
541 | fn equal_kmerge_2(mut inputs: Vec<Vec<i16>>) -> bool { |
542 | use itertools::free::kmerge; |
543 | // sort the inputs |
544 | for input in &mut inputs { |
545 | input.sort(); |
546 | } |
547 | let mut merged = inputs.concat(); |
548 | merged.sort(); |
549 | itertools::equal(merged.into_iter(), kmerge(inputs)) |
550 | } |
551 | |
552 | // Any number of input iterators |
553 | fn equal_kmerge_by_ge(mut inputs: Vec<Vec<i16>>) -> bool { |
554 | // sort the inputs |
555 | for input in &mut inputs { |
556 | input.sort(); |
557 | input.reverse(); |
558 | } |
559 | let mut merged = inputs.concat(); |
560 | merged.sort(); |
561 | merged.reverse(); |
562 | itertools::equal(merged.into_iter(), |
563 | inputs.into_iter().kmerge_by(|x, y| x >= y)) |
564 | } |
565 | |
566 | // Any number of input iterators |
567 | fn equal_kmerge_by_lt(mut inputs: Vec<Vec<i16>>) -> bool { |
568 | // sort the inputs |
569 | for input in &mut inputs { |
570 | input.sort(); |
571 | } |
572 | let mut merged = inputs.concat(); |
573 | merged.sort(); |
574 | itertools::equal(merged.into_iter(), |
575 | inputs.into_iter().kmerge_by(|x, y| x < y)) |
576 | } |
577 | |
578 | // Any number of input iterators |
579 | fn equal_kmerge_by_le(mut inputs: Vec<Vec<i16>>) -> bool { |
580 | // sort the inputs |
581 | for input in &mut inputs { |
582 | input.sort(); |
583 | } |
584 | let mut merged = inputs.concat(); |
585 | merged.sort(); |
586 | itertools::equal(merged.into_iter(), |
587 | inputs.into_iter().kmerge_by(|x, y| x <= y)) |
588 | } |
589 | fn size_kmerge(a: Iter<i16>, b: Iter<i16>, c: Iter<i16>) -> bool { |
590 | use itertools::free::kmerge; |
591 | correct_size_hint(kmerge(vec![a, b, c])) |
592 | } |
593 | fn equal_zip_eq(a: Vec<i32>, b: Vec<i32>) -> bool { |
594 | let len = std::cmp::min(a.len(), b.len()); |
595 | let a = &a[..len]; |
596 | let b = &b[..len]; |
597 | itertools::equal(zip_eq(a, b), zip(a, b)) |
598 | } |
599 | fn size_zip_longest(a: Iter<i16, Exact>, b: Iter<i16, Exact>) -> bool { |
600 | let filt = a.clone().dedup(); |
601 | let filt2 = b.clone().dedup(); |
602 | correct_size_hint(filt.zip_longest(b.clone())) && |
603 | correct_size_hint(a.clone().zip_longest(filt2)) && |
604 | exact_size(a.zip_longest(b)) |
605 | } |
606 | fn size_2_zip_longest(a: Iter<i16>, b: Iter<i16>) -> bool { |
607 | let it = a.clone().zip_longest(b.clone()); |
608 | let jt = a.clone().zip_longest(b.clone()); |
609 | itertools::equal(a, |
610 | it.filter_map(|elt| match elt { |
611 | EitherOrBoth::Both(x, _) => Some(x), |
612 | EitherOrBoth::Left(x) => Some(x), |
613 | _ => None, |
614 | } |
615 | )) |
616 | && |
617 | itertools::equal(b, |
618 | jt.filter_map(|elt| match elt { |
619 | EitherOrBoth::Both(_, y) => Some(y), |
620 | EitherOrBoth::Right(y) => Some(y), |
621 | _ => None, |
622 | } |
623 | )) |
624 | } |
625 | fn size_interleave(a: Iter<i16>, b: Iter<i16>) -> bool { |
626 | correct_size_hint(a.interleave(b)) |
627 | } |
628 | fn exact_interleave(a: Iter<i16, Exact>, b: Iter<i16, Exact>) -> bool { |
629 | exact_size_for_this(a.interleave(b)) |
630 | } |
631 | fn size_interleave_shortest(a: Iter<i16>, b: Iter<i16>) -> bool { |
632 | correct_size_hint(a.interleave_shortest(b)) |
633 | } |
634 | fn exact_interleave_shortest(a: Vec<()>, b: Vec<()>) -> bool { |
635 | exact_size_for_this(a.iter().interleave_shortest(&b)) |
636 | } |
637 | fn size_intersperse(a: Iter<i16>, x: i16) -> bool { |
638 | correct_size_hint(a.intersperse(x)) |
639 | } |
640 | fn equal_intersperse(a: Vec<i32>, x: i32) -> bool { |
641 | let mut inter = false; |
642 | let mut i = 0; |
643 | for elt in a.iter().cloned().intersperse(x) { |
644 | if inter { |
645 | if elt != x { return false } |
646 | } else { |
647 | if elt != a[i] { return false } |
648 | i += 1; |
649 | } |
650 | inter = !inter; |
651 | } |
652 | true |
653 | } |
654 | |
655 | fn equal_combinations_2(a: Vec<u8>) -> bool { |
656 | let mut v = Vec::new(); |
657 | for (i, x) in enumerate(&a) { |
658 | for y in &a[i + 1..] { |
659 | v.push((x, y)); |
660 | } |
661 | } |
662 | itertools::equal(a.iter().tuple_combinations::<(_, _)>(), v) |
663 | } |
664 | |
665 | fn collect_tuple_matches_size(a: Iter<i16>) -> bool { |
666 | let size = a.clone().count(); |
667 | a.collect_tuple::<(_, _, _)>().is_some() == (size == 3) |
668 | } |
669 | |
670 | fn correct_permutations(vals: HashSet<i32>, k: usize) -> () { |
671 | // Test permutations only on iterators of distinct integers, to prevent |
672 | // false positives. |
673 | |
674 | const MAX_N: usize = 5; |
675 | |
676 | let n = min(vals.len(), MAX_N); |
677 | let vals: HashSet<i32> = vals.into_iter().take(n).collect(); |
678 | |
679 | let perms = vals.iter().permutations(k); |
680 | |
681 | let mut actual = HashSet::new(); |
682 | |
683 | for perm in perms { |
684 | assert_eq!(perm.len(), k); |
685 | |
686 | let all_items_valid = perm.iter().all(|p| vals.contains(p)); |
687 | assert!(all_items_valid, "perm contains value not from input: {:?}" , perm); |
688 | |
689 | // Check that all perm items are distinct |
690 | let distinct_len = { |
691 | let perm_set: HashSet<_> = perm.iter().collect(); |
692 | perm_set.len() |
693 | }; |
694 | assert_eq!(perm.len(), distinct_len); |
695 | |
696 | // Check that the perm is new |
697 | assert!(actual.insert(perm.clone()), "perm already encountered: {:?}" , perm); |
698 | } |
699 | } |
700 | |
701 | fn permutations_lexic_order(a: usize, b: usize) -> () { |
702 | let a = a % 6; |
703 | let b = b % 6; |
704 | |
705 | let n = max(a, b); |
706 | let k = min (a, b); |
707 | |
708 | let expected_first: Vec<usize> = (0..k).collect(); |
709 | let expected_last: Vec<usize> = ((n - k)..n).rev().collect(); |
710 | |
711 | let mut perms = (0..n).permutations(k); |
712 | |
713 | let mut curr_perm = match perms.next() { |
714 | Some(p) => p, |
715 | None => { return; } |
716 | }; |
717 | |
718 | assert_eq!(expected_first, curr_perm); |
719 | |
720 | for next_perm in perms { |
721 | assert!( |
722 | next_perm > curr_perm, |
723 | "next perm isn't greater-than current; next_perm={:?} curr_perm={:?} n={}" , |
724 | next_perm, curr_perm, n |
725 | ); |
726 | |
727 | curr_perm = next_perm; |
728 | } |
729 | |
730 | assert_eq!(expected_last, curr_perm); |
731 | |
732 | } |
733 | |
734 | fn permutations_count(n: usize, k: usize) -> bool { |
735 | let n = n % 6; |
736 | |
737 | correct_count(|| (0..n).permutations(k)) |
738 | } |
739 | |
740 | fn permutations_size(a: Iter<i32>, k: usize) -> bool { |
741 | correct_size_hint(a.take(5).permutations(k)) |
742 | } |
743 | |
744 | fn permutations_k0_yields_once(n: usize) -> () { |
745 | let k = 0; |
746 | let expected: Vec<Vec<usize>> = vec![vec![]]; |
747 | let actual = (0..n).permutations(k).collect_vec(); |
748 | |
749 | assert_eq!(expected, actual); |
750 | } |
751 | } |
752 | |
753 | quickcheck! { |
754 | fn dedup_via_coalesce(a: Vec<i32>) -> bool { |
755 | let mut b = a.clone(); |
756 | b.dedup(); |
757 | itertools::equal( |
758 | &b, |
759 | a |
760 | .iter() |
761 | .coalesce(|x, y| { |
762 | if x==y { |
763 | Ok(x) |
764 | } else { |
765 | Err((x, y)) |
766 | } |
767 | }) |
768 | .fold(vec![], |mut v, n| { |
769 | v.push(n); |
770 | v |
771 | }) |
772 | ) |
773 | } |
774 | } |
775 | |
776 | quickcheck! { |
777 | fn equal_dedup(a: Vec<i32>) -> bool { |
778 | let mut b = a.clone(); |
779 | b.dedup(); |
780 | itertools::equal(&b, a.iter().dedup()) |
781 | } |
782 | } |
783 | |
784 | quickcheck! { |
785 | fn equal_dedup_by(a: Vec<(i32, i32)>) -> bool { |
786 | let mut b = a.clone(); |
787 | b.dedup_by(|x, y| x.0==y.0); |
788 | itertools::equal(&b, a.iter().dedup_by(|x, y| x.0==y.0)) |
789 | } |
790 | } |
791 | |
792 | quickcheck! { |
793 | fn size_dedup(a: Vec<i32>) -> bool { |
794 | correct_size_hint(a.iter().dedup()) |
795 | } |
796 | } |
797 | |
798 | quickcheck! { |
799 | fn size_dedup_by(a: Vec<(i32, i32)>) -> bool { |
800 | correct_size_hint(a.iter().dedup_by(|x, y| x.0==y.0)) |
801 | } |
802 | } |
803 | |
804 | quickcheck! { |
805 | fn exact_repeatn((n, x): (usize, i32)) -> bool { |
806 | let it = itertools::repeat_n(x, n); |
807 | exact_size(it) |
808 | } |
809 | } |
810 | |
811 | quickcheck! { |
812 | fn size_put_back(a: Vec<u8>, x: Option<u8>) -> bool { |
813 | let mut it = put_back(a.into_iter()); |
814 | match x { |
815 | Some(t) => it.put_back(t), |
816 | None => {} |
817 | } |
818 | correct_size_hint(it) |
819 | } |
820 | } |
821 | |
822 | quickcheck! { |
823 | fn size_put_backn(a: Vec<u8>, b: Vec<u8>) -> bool { |
824 | let mut it = put_back_n(a.into_iter()); |
825 | for elt in b { |
826 | it.put_back(elt) |
827 | } |
828 | correct_size_hint(it) |
829 | } |
830 | } |
831 | |
832 | quickcheck! { |
833 | fn size_tee(a: Vec<u8>) -> bool { |
834 | let (mut t1, mut t2) = a.iter().tee(); |
835 | t1.next(); |
836 | t1.next(); |
837 | t2.next(); |
838 | exact_size(t1) && exact_size(t2) |
839 | } |
840 | } |
841 | |
842 | quickcheck! { |
843 | fn size_tee_2(a: Vec<u8>) -> bool { |
844 | let (mut t1, mut t2) = a.iter().dedup().tee(); |
845 | t1.next(); |
846 | t1.next(); |
847 | t2.next(); |
848 | correct_size_hint(t1) && correct_size_hint(t2) |
849 | } |
850 | } |
851 | |
852 | quickcheck! { |
853 | fn size_take_while_ref(a: Vec<u8>, stop: u8) -> bool { |
854 | correct_size_hint(a.iter().take_while_ref(|x| **x != stop)) |
855 | } |
856 | } |
857 | |
858 | quickcheck! { |
859 | fn equal_partition(a: Vec<i32>) -> bool { |
860 | let mut a = a; |
861 | let mut ap = a.clone(); |
862 | let split_index = itertools::partition(&mut ap, |x| *x >= 0); |
863 | let parted = (0..split_index).all(|i| ap[i] >= 0) && |
864 | (split_index..a.len()).all(|i| ap[i] < 0); |
865 | |
866 | a.sort(); |
867 | ap.sort(); |
868 | parted && (a == ap) |
869 | } |
870 | } |
871 | |
872 | quickcheck! { |
873 | fn size_combinations(it: Iter<i16>) -> bool { |
874 | correct_size_hint(it.tuple_combinations::<(_, _)>()) |
875 | } |
876 | } |
877 | |
878 | quickcheck! { |
879 | fn equal_combinations(it: Iter<i16>) -> bool { |
880 | let values = it.clone().collect_vec(); |
881 | let mut cmb = it.tuple_combinations(); |
882 | for i in 0..values.len() { |
883 | for j in i+1..values.len() { |
884 | let pair = (values[i], values[j]); |
885 | if pair != cmb.next().unwrap() { |
886 | return false; |
887 | } |
888 | } |
889 | } |
890 | cmb.next() == None |
891 | } |
892 | } |
893 | |
894 | quickcheck! { |
895 | fn size_pad_tail(it: Iter<i8>, pad: u8) -> bool { |
896 | correct_size_hint(it.clone().pad_using(pad as usize, |_| 0)) && |
897 | correct_size_hint(it.dropping(1).rev().pad_using(pad as usize, |_| 0)) |
898 | } |
899 | } |
900 | |
901 | quickcheck! { |
902 | fn size_pad_tail2(it: Iter<i8, Exact>, pad: u8) -> bool { |
903 | exact_size(it.pad_using(pad as usize, |_| 0)) |
904 | } |
905 | } |
906 | |
907 | quickcheck! { |
908 | fn size_powerset(it: Iter<u8, Exact>) -> bool { |
909 | // Powerset cardinality gets large very quickly, limit input to keep test fast. |
910 | correct_size_hint(it.take(12).powerset()) |
911 | } |
912 | } |
913 | |
914 | quickcheck! { |
915 | fn size_duplicates(it: Iter<i8>) -> bool { |
916 | correct_size_hint(it.duplicates()) |
917 | } |
918 | } |
919 | |
920 | quickcheck! { |
921 | fn size_unique(it: Iter<i8>) -> bool { |
922 | correct_size_hint(it.unique()) |
923 | } |
924 | |
925 | fn count_unique(it: Vec<i8>, take_first: u8) -> () { |
926 | let answer = { |
927 | let mut v = it.clone(); |
928 | v.sort(); v.dedup(); |
929 | v.len() |
930 | }; |
931 | let mut iter = cloned(&it).unique(); |
932 | let first_count = (&mut iter).take(take_first as usize).count(); |
933 | let rest_count = iter.count(); |
934 | assert_eq!(answer, first_count + rest_count); |
935 | } |
936 | } |
937 | |
938 | quickcheck! { |
939 | fn fuzz_group_by_lazy_1(it: Iter<u8>) -> bool { |
940 | let jt = it.clone(); |
941 | let groups = it.group_by(|k| *k); |
942 | itertools::equal(jt, groups.into_iter().flat_map(|(_, x)| x)) |
943 | } |
944 | } |
945 | |
946 | quickcheck! { |
947 | fn fuzz_group_by_lazy_2(data: Vec<u8>) -> bool { |
948 | let groups = data.iter().group_by(|k| *k / 10); |
949 | let res = itertools::equal(data.iter(), groups.into_iter().flat_map(|(_, x)| x)); |
950 | res |
951 | } |
952 | } |
953 | |
954 | quickcheck! { |
955 | fn fuzz_group_by_lazy_3(data: Vec<u8>) -> bool { |
956 | let grouper = data.iter().group_by(|k| *k / 10); |
957 | let groups = grouper.into_iter().collect_vec(); |
958 | let res = itertools::equal(data.iter(), groups.into_iter().flat_map(|(_, x)| x)); |
959 | res |
960 | } |
961 | } |
962 | |
963 | quickcheck! { |
964 | fn fuzz_group_by_lazy_duo(data: Vec<u8>, order: Vec<(bool, bool)>) -> bool { |
965 | let grouper = data.iter().group_by(|k| *k / 3); |
966 | let mut groups1 = grouper.into_iter(); |
967 | let mut groups2 = grouper.into_iter(); |
968 | let mut elts = Vec::<&u8>::new(); |
969 | let mut old_groups = Vec::new(); |
970 | |
971 | let tup1 = |(_, b)| b; |
972 | for &(ord, consume_now) in &order { |
973 | let iter = &mut [&mut groups1, &mut groups2][ord as usize]; |
974 | match iter.next() { |
975 | Some((_, gr)) => if consume_now { |
976 | for og in old_groups.drain(..) { |
977 | elts.extend(og); |
978 | } |
979 | elts.extend(gr); |
980 | } else { |
981 | old_groups.push(gr); |
982 | }, |
983 | None => break, |
984 | } |
985 | } |
986 | for og in old_groups.drain(..) { |
987 | elts.extend(og); |
988 | } |
989 | for gr in groups1.map(&tup1) { elts.extend(gr); } |
990 | for gr in groups2.map(&tup1) { elts.extend(gr); } |
991 | itertools::assert_equal(&data, elts); |
992 | true |
993 | } |
994 | } |
995 | |
996 | quickcheck! { |
997 | fn equal_chunks_lazy(a: Vec<u8>, size: u8) -> bool { |
998 | let mut size = size; |
999 | if size == 0 { |
1000 | size += 1; |
1001 | } |
1002 | let chunks = a.iter().chunks(size as usize); |
1003 | let it = a.chunks(size as usize); |
1004 | for (a, b) in chunks.into_iter().zip(it) { |
1005 | if !itertools::equal(a, b) { |
1006 | return false; |
1007 | } |
1008 | } |
1009 | true |
1010 | } |
1011 | } |
1012 | |
1013 | quickcheck! { |
1014 | fn equal_tuple_windows_1(a: Vec<u8>) -> bool { |
1015 | let x = a.windows(1).map(|s| (&s[0], )); |
1016 | let y = a.iter().tuple_windows::<(_,)>(); |
1017 | itertools::equal(x, y) |
1018 | } |
1019 | |
1020 | fn equal_tuple_windows_2(a: Vec<u8>) -> bool { |
1021 | let x = a.windows(2).map(|s| (&s[0], &s[1])); |
1022 | let y = a.iter().tuple_windows::<(_, _)>(); |
1023 | itertools::equal(x, y) |
1024 | } |
1025 | |
1026 | fn equal_tuple_windows_3(a: Vec<u8>) -> bool { |
1027 | let x = a.windows(3).map(|s| (&s[0], &s[1], &s[2])); |
1028 | let y = a.iter().tuple_windows::<(_, _, _)>(); |
1029 | itertools::equal(x, y) |
1030 | } |
1031 | |
1032 | fn equal_tuple_windows_4(a: Vec<u8>) -> bool { |
1033 | let x = a.windows(4).map(|s| (&s[0], &s[1], &s[2], &s[3])); |
1034 | let y = a.iter().tuple_windows::<(_, _, _, _)>(); |
1035 | itertools::equal(x, y) |
1036 | } |
1037 | |
1038 | fn equal_tuples_1(a: Vec<u8>) -> bool { |
1039 | let x = a.chunks(1).map(|s| (&s[0], )); |
1040 | let y = a.iter().tuples::<(_,)>(); |
1041 | itertools::equal(x, y) |
1042 | } |
1043 | |
1044 | fn equal_tuples_2(a: Vec<u8>) -> bool { |
1045 | let x = a.chunks(2).filter(|s| s.len() == 2).map(|s| (&s[0], &s[1])); |
1046 | let y = a.iter().tuples::<(_, _)>(); |
1047 | itertools::equal(x, y) |
1048 | } |
1049 | |
1050 | fn equal_tuples_3(a: Vec<u8>) -> bool { |
1051 | let x = a.chunks(3).filter(|s| s.len() == 3).map(|s| (&s[0], &s[1], &s[2])); |
1052 | let y = a.iter().tuples::<(_, _, _)>(); |
1053 | itertools::equal(x, y) |
1054 | } |
1055 | |
1056 | fn equal_tuples_4(a: Vec<u8>) -> bool { |
1057 | let x = a.chunks(4).filter(|s| s.len() == 4).map(|s| (&s[0], &s[1], &s[2], &s[3])); |
1058 | let y = a.iter().tuples::<(_, _, _, _)>(); |
1059 | itertools::equal(x, y) |
1060 | } |
1061 | |
1062 | fn exact_tuple_buffer(a: Vec<u8>) -> bool { |
1063 | let mut iter = a.iter().tuples::<(_, _, _, _)>(); |
1064 | (&mut iter).last(); |
1065 | let buffer = iter.into_buffer(); |
1066 | assert_eq!(buffer.len(), a.len() % 4); |
1067 | exact_size(buffer) |
1068 | } |
1069 | } |
1070 | |
1071 | // with_position |
1072 | quickcheck! { |
1073 | fn with_position_exact_size_1(a: Vec<u8>) -> bool { |
1074 | exact_size_for_this(a.iter().with_position()) |
1075 | } |
1076 | fn with_position_exact_size_2(a: Iter<u8, Exact>) -> bool { |
1077 | exact_size_for_this(a.with_position()) |
1078 | } |
1079 | } |
1080 | |
1081 | quickcheck! { |
1082 | fn correct_group_map_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1083 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1084 | let count = a.len(); |
1085 | let lookup = a.into_iter().map(|i| (i % modulo, i)).into_group_map(); |
1086 | |
1087 | assert_eq!(lookup.values().flat_map(|vals| vals.iter()).count(), count); |
1088 | |
1089 | for (&key, vals) in lookup.iter() { |
1090 | assert!(vals.iter().all(|&val| val % modulo == key)); |
1091 | } |
1092 | } |
1093 | } |
1094 | |
1095 | /// A peculiar type: Equality compares both tuple items, but ordering only the |
1096 | /// first item. This is so we can check the stability property easily. |
1097 | #[derive(Clone, Debug, PartialEq, Eq)] |
1098 | struct Val(u32, u32); |
1099 | |
1100 | impl PartialOrd<Val> for Val { |
1101 | fn partial_cmp(&self, other: &Val) -> Option<Ordering> { |
1102 | self.0.partial_cmp(&other.0) |
1103 | } |
1104 | } |
1105 | |
1106 | impl Ord for Val { |
1107 | fn cmp(&self, other: &Val) -> Ordering { |
1108 | self.0.cmp(&other.0) |
1109 | } |
1110 | } |
1111 | |
1112 | impl qc::Arbitrary for Val { |
1113 | fn arbitrary<G: qc::Gen>(g: &mut G) -> Self { |
1114 | let (x, y) = <(u32, u32)>::arbitrary(g); |
1115 | Val(x, y) |
1116 | } |
1117 | fn shrink(&self) -> Box<dyn Iterator<Item = Self>> { |
1118 | Box::new((self.0, self.1).shrink().map(|(x, y)| Val(x, y))) |
1119 | } |
1120 | } |
1121 | |
1122 | quickcheck! { |
1123 | fn minmax(a: Vec<Val>) -> bool { |
1124 | use itertools::MinMaxResult; |
1125 | |
1126 | |
1127 | let minmax = a.iter().minmax(); |
1128 | let expected = match a.len() { |
1129 | 0 => MinMaxResult::NoElements, |
1130 | 1 => MinMaxResult::OneElement(&a[0]), |
1131 | _ => MinMaxResult::MinMax(a.iter().min().unwrap(), |
1132 | a.iter().max().unwrap()), |
1133 | }; |
1134 | minmax == expected |
1135 | } |
1136 | } |
1137 | |
1138 | quickcheck! { |
1139 | fn minmax_f64(a: Vec<f64>) -> TestResult { |
1140 | use itertools::MinMaxResult; |
1141 | |
1142 | if a.iter().any(|x| x.is_nan()) { |
1143 | return TestResult::discard(); |
1144 | } |
1145 | |
1146 | let min = cloned(&a).fold1(f64::min); |
1147 | let max = cloned(&a).fold1(f64::max); |
1148 | |
1149 | let minmax = cloned(&a).minmax(); |
1150 | let expected = match a.len() { |
1151 | 0 => MinMaxResult::NoElements, |
1152 | 1 => MinMaxResult::OneElement(min.unwrap()), |
1153 | _ => MinMaxResult::MinMax(min.unwrap(), max.unwrap()), |
1154 | }; |
1155 | TestResult::from_bool(minmax == expected) |
1156 | } |
1157 | } |
1158 | |
1159 | quickcheck! { |
1160 | #[allow (deprecated)] |
1161 | fn tree_fold1_f64(mut a: Vec<f64>) -> TestResult { |
1162 | fn collapse_adjacent<F>(x: Vec<f64>, mut f: F) -> Vec<f64> |
1163 | where F: FnMut(f64, f64) -> f64 |
1164 | { |
1165 | let mut out = Vec::new(); |
1166 | for i in (0..x.len()).step(2) { |
1167 | if i == x.len()-1 { |
1168 | out.push(x[i]) |
1169 | } else { |
1170 | out.push(f(x[i], x[i+1])); |
1171 | } |
1172 | } |
1173 | out |
1174 | } |
1175 | |
1176 | if a.iter().any(|x| x.is_nan()) { |
1177 | return TestResult::discard(); |
1178 | } |
1179 | |
1180 | let actual = a.iter().cloned().tree_fold1(f64::atan2); |
1181 | |
1182 | while a.len() > 1 { |
1183 | a = collapse_adjacent(a, f64::atan2); |
1184 | } |
1185 | let expected = a.pop(); |
1186 | |
1187 | TestResult::from_bool(actual == expected) |
1188 | } |
1189 | } |
1190 | |
1191 | quickcheck! { |
1192 | fn exactly_one_i32(a: Vec<i32>) -> TestResult { |
1193 | let ret = a.iter().cloned().exactly_one(); |
1194 | match a.len() { |
1195 | 1 => TestResult::from_bool(ret.unwrap() == a[0]), |
1196 | _ => TestResult::from_bool(ret.unwrap_err().eq(a.iter().cloned())), |
1197 | } |
1198 | } |
1199 | } |
1200 | |
1201 | quickcheck! { |
1202 | fn at_most_one_i32(a: Vec<i32>) -> TestResult { |
1203 | let ret = a.iter().cloned().at_most_one(); |
1204 | match a.len() { |
1205 | 0 => TestResult::from_bool(ret.unwrap() == None), |
1206 | 1 => TestResult::from_bool(ret.unwrap() == Some(a[0])), |
1207 | _ => TestResult::from_bool(ret.unwrap_err().eq(a.iter().cloned())), |
1208 | } |
1209 | } |
1210 | } |
1211 | |
1212 | quickcheck! { |
1213 | fn consistent_grouping_map_with_by(a: Vec<u8>, modulo: u8) -> () { |
1214 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1215 | |
1216 | let lookup_grouping_map = a.iter().copied().map(|i| (i % modulo, i)).into_grouping_map().collect::<Vec<_>>(); |
1217 | let lookup_grouping_map_by = a.iter().copied().into_grouping_map_by(|i| i % modulo).collect::<Vec<_>>(); |
1218 | |
1219 | assert_eq!(lookup_grouping_map, lookup_grouping_map_by); |
1220 | } |
1221 | |
1222 | fn correct_grouping_map_by_aggregate_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1223 | let modulo = if modulo < 2 { 2 } else { modulo } as u64; // Avoid `% 0` |
1224 | let lookup = a.iter() |
1225 | .map(|&b| b as u64) // Avoid overflows |
1226 | .into_grouping_map_by(|i| i % modulo) |
1227 | .aggregate(|acc, &key, val| { |
1228 | assert!(val % modulo == key); |
1229 | if val % (modulo - 1) == 0 { |
1230 | None |
1231 | } else { |
1232 | Some(acc.unwrap_or(0) + val) |
1233 | } |
1234 | }); |
1235 | |
1236 | let group_map_lookup = a.iter() |
1237 | .map(|&b| b as u64) |
1238 | .map(|i| (i % modulo, i)) |
1239 | .into_group_map() |
1240 | .into_iter() |
1241 | .filter_map(|(key, vals)| { |
1242 | vals.into_iter().fold(None, |acc, val| { |
1243 | if val % (modulo - 1) == 0 { |
1244 | None |
1245 | } else { |
1246 | Some(acc.unwrap_or(0) + val) |
1247 | } |
1248 | }).map(|new_val| (key, new_val)) |
1249 | }) |
1250 | .collect::<HashMap<_,_>>(); |
1251 | assert_eq!(lookup, group_map_lookup); |
1252 | |
1253 | for m in 0..modulo { |
1254 | assert_eq!( |
1255 | lookup.get(&m).copied(), |
1256 | a.iter() |
1257 | .map(|&b| b as u64) |
1258 | .filter(|&val| val % modulo == m) |
1259 | .fold(None, |acc, val| { |
1260 | if val % (modulo - 1) == 0 { |
1261 | None |
1262 | } else { |
1263 | Some(acc.unwrap_or(0) + val) |
1264 | } |
1265 | }) |
1266 | ); |
1267 | } |
1268 | } |
1269 | |
1270 | fn correct_grouping_map_by_fold_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1271 | let modulo = if modulo == 0 { 1 } else { modulo } as u64; // Avoid `% 0` |
1272 | let lookup = a.iter().map(|&b| b as u64) // Avoid overflows |
1273 | .into_grouping_map_by(|i| i % modulo) |
1274 | .fold(0u64, |acc, &key, val| { |
1275 | assert!(val % modulo == key); |
1276 | acc + val |
1277 | }); |
1278 | |
1279 | let group_map_lookup = a.iter() |
1280 | .map(|&b| b as u64) |
1281 | .map(|i| (i % modulo, i)) |
1282 | .into_group_map() |
1283 | .into_iter() |
1284 | .map(|(key, vals)| (key, vals.into_iter().sum())) |
1285 | .collect::<HashMap<_,_>>(); |
1286 | assert_eq!(lookup, group_map_lookup); |
1287 | |
1288 | for (&key, &sum) in lookup.iter() { |
1289 | assert_eq!(sum, a.iter().map(|&b| b as u64).filter(|&val| val % modulo == key).sum::<u64>()); |
1290 | } |
1291 | } |
1292 | |
1293 | fn correct_grouping_map_by_fold_first_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1294 | let modulo = if modulo == 0 { 1 } else { modulo } as u64; // Avoid `% 0` |
1295 | let lookup = a.iter().map(|&b| b as u64) // Avoid overflows |
1296 | .into_grouping_map_by(|i| i % modulo) |
1297 | .fold_first(|acc, &key, val| { |
1298 | assert!(val % modulo == key); |
1299 | acc + val |
1300 | }); |
1301 | |
1302 | // TODO: Swap `fold1` with stdlib's `fold_first` when it's stabilized |
1303 | let group_map_lookup = a.iter() |
1304 | .map(|&b| b as u64) |
1305 | .map(|i| (i % modulo, i)) |
1306 | .into_group_map() |
1307 | .into_iter() |
1308 | .map(|(key, vals)| (key, vals.into_iter().fold1(|acc, val| acc + val).unwrap())) |
1309 | .collect::<HashMap<_,_>>(); |
1310 | assert_eq!(lookup, group_map_lookup); |
1311 | |
1312 | for (&key, &sum) in lookup.iter() { |
1313 | assert_eq!(sum, a.iter().map(|&b| b as u64).filter(|&val| val % modulo == key).sum::<u64>()); |
1314 | } |
1315 | } |
1316 | |
1317 | fn correct_grouping_map_by_collect_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1318 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1319 | let lookup_grouping_map = a.iter().copied().into_grouping_map_by(|i| i % modulo).collect::<Vec<_>>(); |
1320 | let lookup_group_map = a.iter().copied().map(|i| (i % modulo, i)).into_group_map(); |
1321 | |
1322 | assert_eq!(lookup_grouping_map, lookup_group_map); |
1323 | } |
1324 | |
1325 | fn correct_grouping_map_by_max_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1326 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1327 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).max(); |
1328 | |
1329 | let group_map_lookup = a.iter().copied() |
1330 | .map(|i| (i % modulo, i)) |
1331 | .into_group_map() |
1332 | .into_iter() |
1333 | .map(|(key, vals)| (key, vals.into_iter().max().unwrap())) |
1334 | .collect::<HashMap<_,_>>(); |
1335 | assert_eq!(lookup, group_map_lookup); |
1336 | |
1337 | for (&key, &max) in lookup.iter() { |
1338 | assert_eq!(Some(max), a.iter().copied().filter(|&val| val % modulo == key).max()); |
1339 | } |
1340 | } |
1341 | |
1342 | fn correct_grouping_map_by_max_by_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1343 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1344 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).max_by(|_, v1, v2| v1.cmp(v2)); |
1345 | |
1346 | let group_map_lookup = a.iter().copied() |
1347 | .map(|i| (i % modulo, i)) |
1348 | .into_group_map() |
1349 | .into_iter() |
1350 | .map(|(key, vals)| (key, vals.into_iter().max_by(|v1, v2| v1.cmp(v2)).unwrap())) |
1351 | .collect::<HashMap<_,_>>(); |
1352 | assert_eq!(lookup, group_map_lookup); |
1353 | |
1354 | for (&key, &max) in lookup.iter() { |
1355 | assert_eq!(Some(max), a.iter().copied().filter(|&val| val % modulo == key).max_by(|v1, v2| v1.cmp(v2))); |
1356 | } |
1357 | } |
1358 | |
1359 | fn correct_grouping_map_by_max_by_key_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1360 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1361 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).max_by_key(|_, &val| val); |
1362 | |
1363 | let group_map_lookup = a.iter().copied() |
1364 | .map(|i| (i % modulo, i)) |
1365 | .into_group_map() |
1366 | .into_iter() |
1367 | .map(|(key, vals)| (key, vals.into_iter().max_by_key(|&val| val).unwrap())) |
1368 | .collect::<HashMap<_,_>>(); |
1369 | assert_eq!(lookup, group_map_lookup); |
1370 | |
1371 | for (&key, &max) in lookup.iter() { |
1372 | assert_eq!(Some(max), a.iter().copied().filter(|&val| val % modulo == key).max_by_key(|&val| val)); |
1373 | } |
1374 | } |
1375 | |
1376 | fn correct_grouping_map_by_min_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1377 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1378 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).min(); |
1379 | |
1380 | let group_map_lookup = a.iter().copied() |
1381 | .map(|i| (i % modulo, i)) |
1382 | .into_group_map() |
1383 | .into_iter() |
1384 | .map(|(key, vals)| (key, vals.into_iter().min().unwrap())) |
1385 | .collect::<HashMap<_,_>>(); |
1386 | assert_eq!(lookup, group_map_lookup); |
1387 | |
1388 | for (&key, &min) in lookup.iter() { |
1389 | assert_eq!(Some(min), a.iter().copied().filter(|&val| val % modulo == key).min()); |
1390 | } |
1391 | } |
1392 | |
1393 | fn correct_grouping_map_by_min_by_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1394 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1395 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).min_by(|_, v1, v2| v1.cmp(v2)); |
1396 | |
1397 | let group_map_lookup = a.iter().copied() |
1398 | .map(|i| (i % modulo, i)) |
1399 | .into_group_map() |
1400 | .into_iter() |
1401 | .map(|(key, vals)| (key, vals.into_iter().min_by(|v1, v2| v1.cmp(v2)).unwrap())) |
1402 | .collect::<HashMap<_,_>>(); |
1403 | assert_eq!(lookup, group_map_lookup); |
1404 | |
1405 | for (&key, &min) in lookup.iter() { |
1406 | assert_eq!(Some(min), a.iter().copied().filter(|&val| val % modulo == key).min_by(|v1, v2| v1.cmp(v2))); |
1407 | } |
1408 | } |
1409 | |
1410 | fn correct_grouping_map_by_min_by_key_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1411 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1412 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).min_by_key(|_, &val| val); |
1413 | |
1414 | let group_map_lookup = a.iter().copied() |
1415 | .map(|i| (i % modulo, i)) |
1416 | .into_group_map() |
1417 | .into_iter() |
1418 | .map(|(key, vals)| (key, vals.into_iter().min_by_key(|&val| val).unwrap())) |
1419 | .collect::<HashMap<_,_>>(); |
1420 | assert_eq!(lookup, group_map_lookup); |
1421 | |
1422 | for (&key, &min) in lookup.iter() { |
1423 | assert_eq!(Some(min), a.iter().copied().filter(|&val| val % modulo == key).min_by_key(|&val| val)); |
1424 | } |
1425 | } |
1426 | |
1427 | fn correct_grouping_map_by_minmax_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1428 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1429 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).minmax(); |
1430 | |
1431 | let group_map_lookup = a.iter().copied() |
1432 | .map(|i| (i % modulo, i)) |
1433 | .into_group_map() |
1434 | .into_iter() |
1435 | .map(|(key, vals)| (key, vals.into_iter().minmax())) |
1436 | .collect::<HashMap<_,_>>(); |
1437 | assert_eq!(lookup, group_map_lookup); |
1438 | |
1439 | for (&key, &minmax) in lookup.iter() { |
1440 | assert_eq!(minmax, a.iter().copied().filter(|&val| val % modulo == key).minmax()); |
1441 | } |
1442 | } |
1443 | |
1444 | fn correct_grouping_map_by_minmax_by_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1445 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1446 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).minmax_by(|_, v1, v2| v1.cmp(v2)); |
1447 | |
1448 | let group_map_lookup = a.iter().copied() |
1449 | .map(|i| (i % modulo, i)) |
1450 | .into_group_map() |
1451 | .into_iter() |
1452 | .map(|(key, vals)| (key, vals.into_iter().minmax_by(|v1, v2| v1.cmp(v2)))) |
1453 | .collect::<HashMap<_,_>>(); |
1454 | assert_eq!(lookup, group_map_lookup); |
1455 | |
1456 | for (&key, &minmax) in lookup.iter() { |
1457 | assert_eq!(minmax, a.iter().copied().filter(|&val| val % modulo == key).minmax_by(|v1, v2| v1.cmp(v2))); |
1458 | } |
1459 | } |
1460 | |
1461 | fn correct_grouping_map_by_minmax_by_key_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1462 | let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0` |
1463 | let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).minmax_by_key(|_, &val| val); |
1464 | |
1465 | let group_map_lookup = a.iter().copied() |
1466 | .map(|i| (i % modulo, i)) |
1467 | .into_group_map() |
1468 | .into_iter() |
1469 | .map(|(key, vals)| (key, vals.into_iter().minmax_by_key(|&val| val))) |
1470 | .collect::<HashMap<_,_>>(); |
1471 | assert_eq!(lookup, group_map_lookup); |
1472 | |
1473 | for (&key, &minmax) in lookup.iter() { |
1474 | assert_eq!(minmax, a.iter().copied().filter(|&val| val % modulo == key).minmax_by_key(|&val| val)); |
1475 | } |
1476 | } |
1477 | |
1478 | fn correct_grouping_map_by_sum_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1479 | let modulo = if modulo == 0 { 1 } else { modulo } as u64; // Avoid `% 0` |
1480 | let lookup = a.iter().map(|&b| b as u64) // Avoid overflows |
1481 | .into_grouping_map_by(|i| i % modulo) |
1482 | .sum(); |
1483 | |
1484 | let group_map_lookup = a.iter().map(|&b| b as u64) |
1485 | .map(|i| (i % modulo, i)) |
1486 | .into_group_map() |
1487 | .into_iter() |
1488 | .map(|(key, vals)| (key, vals.into_iter().sum())) |
1489 | .collect::<HashMap<_,_>>(); |
1490 | assert_eq!(lookup, group_map_lookup); |
1491 | |
1492 | for (&key, &sum) in lookup.iter() { |
1493 | assert_eq!(sum, a.iter().map(|&b| b as u64).filter(|&val| val % modulo == key).sum::<u64>()); |
1494 | } |
1495 | } |
1496 | |
1497 | fn correct_grouping_map_by_product_modulo_key(a: Vec<u8>, modulo: u8) -> () { |
1498 | let modulo = Wrapping(if modulo == 0 { 1 } else { modulo } as u64); // Avoid `% 0` |
1499 | let lookup = a.iter().map(|&b| Wrapping(b as u64)) // Avoid overflows |
1500 | .into_grouping_map_by(|i| i % modulo) |
1501 | .product(); |
1502 | |
1503 | let group_map_lookup = a.iter().map(|&b| Wrapping(b as u64)) |
1504 | .map(|i| (i % modulo, i)) |
1505 | .into_group_map() |
1506 | .into_iter() |
1507 | .map(|(key, vals)| (key, vals.into_iter().product::<Wrapping<u64>>())) |
1508 | .collect::<HashMap<_,_>>(); |
1509 | assert_eq!(lookup, group_map_lookup); |
1510 | |
1511 | for (&key, &prod) in lookup.iter() { |
1512 | assert_eq!( |
1513 | prod, |
1514 | a.iter() |
1515 | .map(|&b| Wrapping(b as u64)) |
1516 | .filter(|&val| val % modulo == key) |
1517 | .product::<Wrapping<u64>>() |
1518 | ); |
1519 | } |
1520 | } |
1521 | |
1522 | // This should check that if multiple elements are equally minimum or maximum |
1523 | // then `max`, `min` and `minmax` pick the first minimum and the last maximum. |
1524 | // This is to be consistent with `std::iter::max` and `std::iter::min`. |
1525 | fn correct_grouping_map_by_min_max_minmax_order_modulo_key() -> () { |
1526 | use itertools::MinMaxResult; |
1527 | |
1528 | let lookup = (0..=10) |
1529 | .into_grouping_map_by(|_| 0) |
1530 | .max_by(|_, _, _| Ordering::Equal); |
1531 | |
1532 | assert_eq!(lookup[&0], 10); |
1533 | |
1534 | let lookup = (0..=10) |
1535 | .into_grouping_map_by(|_| 0) |
1536 | .min_by(|_, _, _| Ordering::Equal); |
1537 | |
1538 | assert_eq!(lookup[&0], 0); |
1539 | |
1540 | let lookup = (0..=10) |
1541 | .into_grouping_map_by(|_| 0) |
1542 | .minmax_by(|_, _, _| Ordering::Equal); |
1543 | |
1544 | assert_eq!(lookup[&0], MinMaxResult::MinMax(0, 10)); |
1545 | } |
1546 | } |
1547 | |
1548 | quickcheck! { |
1549 | fn counts(nums: Vec<isize>) -> TestResult { |
1550 | let counts = nums.iter().counts(); |
1551 | for (&item, &count) in counts.iter() { |
1552 | #[allow (clippy::absurd_extreme_comparisons)] |
1553 | if count <= 0 { |
1554 | return TestResult::failed(); |
1555 | } |
1556 | if count != nums.iter().filter(|&x| x == item).count() { |
1557 | return TestResult::failed(); |
1558 | } |
1559 | } |
1560 | for item in nums.iter() { |
1561 | if !counts.contains_key(item) { |
1562 | return TestResult::failed(); |
1563 | } |
1564 | } |
1565 | TestResult::passed() |
1566 | } |
1567 | } |
1568 | |
1569 | quickcheck! { |
1570 | fn test_double_ended_zip_2(a: Vec<u8>, b: Vec<u8>) -> TestResult { |
1571 | let mut x = |
1572 | multizip((a.clone().into_iter(), b.clone().into_iter())) |
1573 | .collect_vec(); |
1574 | x.reverse(); |
1575 | |
1576 | let y = |
1577 | multizip((a.into_iter(), b.into_iter())) |
1578 | .rfold(Vec::new(), |mut vec, e| { vec.push(e); vec }); |
1579 | |
1580 | TestResult::from_bool(itertools::equal(x, y)) |
1581 | } |
1582 | |
1583 | fn test_double_ended_zip_3(a: Vec<u8>, b: Vec<u8>, c: Vec<u8>) -> TestResult { |
1584 | let mut x = |
1585 | multizip((a.clone().into_iter(), b.clone().into_iter(), c.clone().into_iter())) |
1586 | .collect_vec(); |
1587 | x.reverse(); |
1588 | |
1589 | let y = |
1590 | multizip((a.into_iter(), b.into_iter(), c.into_iter())) |
1591 | .rfold(Vec::new(), |mut vec, e| { vec.push(e); vec }); |
1592 | |
1593 | TestResult::from_bool(itertools::equal(x, y)) |
1594 | } |
1595 | } |
1596 | |
1597 | |
1598 | fn is_fused<I: Iterator>(mut it: I) -> bool |
1599 | { |
1600 | for _ in it.by_ref() {} |
1601 | for _ in 0..10{ |
1602 | if it.next().is_some(){ |
1603 | return false; |
1604 | } |
1605 | } |
1606 | true |
1607 | } |
1608 | |
1609 | quickcheck! { |
1610 | fn fused_combination(a: Iter<i16>) -> bool |
1611 | { |
1612 | is_fused(a.clone().combinations(1)) && |
1613 | is_fused(a.combinations(3)) |
1614 | } |
1615 | |
1616 | fn fused_combination_with_replacement(a: Iter<i16>) -> bool |
1617 | { |
1618 | is_fused(a.clone().combinations_with_replacement(1)) && |
1619 | is_fused(a.combinations_with_replacement(3)) |
1620 | } |
1621 | |
1622 | fn fused_tuple_combination(a: Iter<i16>) -> bool |
1623 | { |
1624 | is_fused(a.clone().fuse().tuple_combinations::<(_,)>()) && |
1625 | is_fused(a.fuse().tuple_combinations::<(_,_,_)>()) |
1626 | } |
1627 | |
1628 | fn fused_unique(a: Iter<i16>) -> bool |
1629 | { |
1630 | is_fused(a.fuse().unique()) |
1631 | } |
1632 | |
1633 | fn fused_unique_by(a: Iter<i16>) -> bool |
1634 | { |
1635 | is_fused(a.fuse().unique_by(|x| x % 100)) |
1636 | } |
1637 | |
1638 | fn fused_interleave_shortest(a: Iter<i16>, b: Iter<i16>) -> bool |
1639 | { |
1640 | !is_fused(a.clone().interleave_shortest(b.clone())) && |
1641 | is_fused(a.fuse().interleave_shortest(b.fuse())) |
1642 | } |
1643 | |
1644 | fn fused_product(a: Iter<i16>, b: Iter<i16>) -> bool |
1645 | { |
1646 | is_fused(a.fuse().cartesian_product(b.fuse())) |
1647 | } |
1648 | |
1649 | fn fused_merge(a: Iter<i16>, b: Iter<i16>) -> bool |
1650 | { |
1651 | is_fused(a.fuse().merge(b.fuse())) |
1652 | } |
1653 | |
1654 | fn fused_filter_ok(a: Iter<i16>) -> bool |
1655 | { |
1656 | is_fused(a.map(|x| if x % 2 == 0 {Ok(x)} else {Err(x)} ) |
1657 | .filter_ok(|x| x % 3 == 0) |
1658 | .fuse()) |
1659 | } |
1660 | |
1661 | fn fused_filter_map_ok(a: Iter<i16>) -> bool |
1662 | { |
1663 | is_fused(a.map(|x| if x % 2 == 0 {Ok(x)} else {Err(x)} ) |
1664 | .filter_map_ok(|x| if x % 3 == 0 {Some(x / 3)} else {None}) |
1665 | .fuse()) |
1666 | } |
1667 | |
1668 | fn fused_positions(a: Iter<i16>) -> bool |
1669 | { |
1670 | !is_fused(a.clone().positions(|x|x%2==0)) && |
1671 | is_fused(a.fuse().positions(|x|x%2==0)) |
1672 | } |
1673 | |
1674 | fn fused_update(a: Iter<i16>) -> bool |
1675 | { |
1676 | !is_fused(a.clone().update(|x|*x+=1)) && |
1677 | is_fused(a.fuse().update(|x|*x+=1)) |
1678 | } |
1679 | |
1680 | fn fused_tuple_windows(a: Iter<i16>) -> bool |
1681 | { |
1682 | is_fused(a.fuse().tuple_windows::<(_,_)>()) |
1683 | } |
1684 | |
1685 | fn fused_pad_using(a: Iter<i16>) -> bool |
1686 | { |
1687 | is_fused(a.fuse().pad_using(100,|_|0)) |
1688 | } |
1689 | } |
1690 | |
1691 | quickcheck! { |
1692 | fn min_set_contains_min(a: Vec<(usize, char)>) -> bool { |
1693 | let result_set = a.iter().min_set(); |
1694 | if let Some(result_element) = a.iter().min() { |
1695 | result_set.contains(&result_element) |
1696 | } else { |
1697 | result_set.is_empty() |
1698 | } |
1699 | } |
1700 | |
1701 | fn min_set_by_contains_min(a: Vec<(usize, char)>) -> bool { |
1702 | let compare = |x: &&(usize, char), y: &&(usize, char)| x.1.cmp(&y.1); |
1703 | let result_set = a.iter().min_set_by(compare); |
1704 | if let Some(result_element) = a.iter().min_by(compare) { |
1705 | result_set.contains(&result_element) |
1706 | } else { |
1707 | result_set.is_empty() |
1708 | } |
1709 | } |
1710 | |
1711 | fn min_set_by_key_contains_min(a: Vec<(usize, char)>) -> bool { |
1712 | let key = |x: &&(usize, char)| x.1; |
1713 | let result_set = a.iter().min_set_by_key(&key); |
1714 | if let Some(result_element) = a.iter().min_by_key(&key) { |
1715 | result_set.contains(&result_element) |
1716 | } else { |
1717 | result_set.is_empty() |
1718 | } |
1719 | } |
1720 | |
1721 | fn max_set_contains_max(a: Vec<(usize, char)>) -> bool { |
1722 | let result_set = a.iter().max_set(); |
1723 | if let Some(result_element) = a.iter().max() { |
1724 | result_set.contains(&result_element) |
1725 | } else { |
1726 | result_set.is_empty() |
1727 | } |
1728 | } |
1729 | |
1730 | fn max_set_by_contains_max(a: Vec<(usize, char)>) -> bool { |
1731 | let compare = |x: &&(usize, char), y: &&(usize, char)| x.1.cmp(&y.1); |
1732 | let result_set = a.iter().max_set_by(compare); |
1733 | if let Some(result_element) = a.iter().max_by(compare) { |
1734 | result_set.contains(&result_element) |
1735 | } else { |
1736 | result_set.is_empty() |
1737 | } |
1738 | } |
1739 | |
1740 | fn max_set_by_key_contains_max(a: Vec<(usize, char)>) -> bool { |
1741 | let key = |x: &&(usize, char)| x.1; |
1742 | let result_set = a.iter().max_set_by_key(&key); |
1743 | if let Some(result_element) = a.iter().max_by_key(&key) { |
1744 | result_set.contains(&result_element) |
1745 | } else { |
1746 | result_set.is_empty() |
1747 | } |
1748 | } |
1749 | } |
1750 | |