Revision control
Copy as Markdown
Other Tools
//! The purpose of these tests is to cover corner cases of iterators
//! and adaptors.
//!
//! In particular we test the tedious size_hint and exact size correctness.
use quickcheck as qc;
use std::default::Default;
use std::num::Wrapping;
use std::ops::Range;
use std::cmp::{max, min, Ordering};
use std::collections::{HashMap, HashSet};
use itertools::Itertools;
use itertools::{
multizip,
EitherOrBoth,
iproduct,
izip,
};
use itertools::free::{
cloned,
enumerate,
multipeek,
peek_nth,
put_back,
put_back_n,
rciter,
zip,
zip_eq,
};
use rand::Rng;
use rand::seq::SliceRandom;
use quickcheck::TestResult;
/// Trait for size hint modifier types
trait HintKind: Copy + Send + qc::Arbitrary {
fn loosen_bounds(&self, org_hint: (usize, Option<usize>)) -> (usize, Option<usize>);
}
/// Exact size hint variant that leaves hints unchanged
#[derive(Clone, Copy, Debug)]
struct Exact {}
impl HintKind for Exact {
fn loosen_bounds(&self, org_hint: (usize, Option<usize>)) -> (usize, Option<usize>) {
org_hint
}
}
impl qc::Arbitrary for Exact {
fn arbitrary<G: qc::Gen>(_: &mut G) -> Self {
Exact {}
}
}
/// Inexact size hint variant to simulate imprecise (but valid) size hints
///
/// Will always decrease the lower bound and increase the upper bound
/// of the size hint by set amounts.
#[derive(Clone, Copy, Debug)]
struct Inexact {
underestimate: usize,
overestimate: usize,
}
impl HintKind for Inexact {
fn loosen_bounds(&self, org_hint: (usize, Option<usize>)) -> (usize, Option<usize>) {
let (org_lower, org_upper) = org_hint;
(org_lower.saturating_sub(self.underestimate),
org_upper.and_then(move |x| x.checked_add(self.overestimate)))
}
}
impl qc::Arbitrary for Inexact {
fn arbitrary<G: qc::Gen>(g: &mut G) -> Self {
let ue_value = usize::arbitrary(g);
let oe_value = usize::arbitrary(g);
// Compensate for quickcheck using extreme values too rarely
let ue_choices = &[0, ue_value, usize::max_value()];
let oe_choices = &[0, oe_value, usize::max_value()];
Inexact {
underestimate: *ue_choices.choose(g).unwrap(),
overestimate: *oe_choices.choose(g).unwrap(),
}
}
fn shrink(&self) -> Box<dyn Iterator<Item=Self>> {
let underestimate_value = self.underestimate;
let overestimate_value = self.overestimate;
Box::new(
underestimate_value.shrink().flat_map(move |ue_value|
overestimate_value.shrink().map(move |oe_value|
Inexact {
underestimate: ue_value,
overestimate: oe_value,
}
)
)
)
}
}
/// Our base iterator that we can impl Arbitrary for
///
/// By default we'll return inexact bounds estimates for size_hint
/// to make tests harder to pass.
///
/// NOTE: Iter is tricky and is not fused, to help catch bugs.
/// At the end it will return None once, then return Some(0),
/// then return None again.
#[derive(Clone, Debug)]
struct Iter<T, SK: HintKind = Inexact> {
iterator: Range<T>,
// fuse/done flag
fuse_flag: i32,
hint_kind: SK,
}
impl<T, HK> Iter<T, HK> where HK: HintKind
{
fn new(it: Range<T>, hint_kind: HK) -> Self {
Iter {
iterator: it,
fuse_flag: 0,
hint_kind,
}
}
}
impl<T, HK> Iterator for Iter<T, HK>
where Range<T>: Iterator,
<Range<T> as Iterator>::Item: Default,
HK: HintKind,
{
type Item = <Range<T> as Iterator>::Item;
fn next(&mut self) -> Option<Self::Item>
{
let elt = self.iterator.next();
if elt.is_none() {
self.fuse_flag += 1;
// check fuse flag
if self.fuse_flag == 2 {
return Some(Default::default())
}
}
elt
}
fn size_hint(&self) -> (usize, Option<usize>)
{
let org_hint = self.iterator.size_hint();
self.hint_kind.loosen_bounds(org_hint)
}
}
impl<T, HK> DoubleEndedIterator for Iter<T, HK>
where Range<T>: DoubleEndedIterator,
<Range<T> as Iterator>::Item: Default,
HK: HintKind
{
fn next_back(&mut self) -> Option<Self::Item> { self.iterator.next_back() }
}
impl<T> ExactSizeIterator for Iter<T, Exact> where Range<T>: ExactSizeIterator,
<Range<T> as Iterator>::Item: Default,
{ }
impl<T, HK> qc::Arbitrary for Iter<T, HK>
where T: qc::Arbitrary,
HK: HintKind,
{
fn arbitrary<G: qc::Gen>(g: &mut G) -> Self
{
Iter::new(T::arbitrary(g)..T::arbitrary(g), HK::arbitrary(g))
}
fn shrink(&self) -> Box<dyn Iterator<Item=Iter<T, HK>>>
{
let r = self.iterator.clone();
let hint_kind = self.hint_kind;
Box::new(
r.start.shrink().flat_map(move |a|
r.end.shrink().map(move |b|
Iter::new(a.clone()..b, hint_kind)
)
)
)
}
}
/// A meta-iterator which yields `Iter<i32>`s whose start/endpoints are
/// increased or decreased linearly on each iteration.
#[derive(Clone, Debug)]
struct ShiftRange<HK = Inexact> {
range_start: i32,
range_end: i32,
start_step: i32,
end_step: i32,
iter_count: u32,
hint_kind: HK,
}
impl<HK> Iterator for ShiftRange<HK> where HK: HintKind {
type Item = Iter<i32, HK>;
fn next(&mut self) -> Option<Self::Item> {
if self.iter_count == 0 {
return None;
}
let iter = Iter::new(self.range_start..self.range_end, self.hint_kind);
self.range_start += self.start_step;
self.range_end += self.end_step;
self.iter_count -= 1;
Some(iter)
}
}
impl ExactSizeIterator for ShiftRange<Exact> { }
impl<HK> qc::Arbitrary for ShiftRange<HK>
where HK: HintKind
{
fn arbitrary<G: qc::Gen>(g: &mut G) -> Self {
const MAX_STARTING_RANGE_DIFF: i32 = 32;
const MAX_STEP_MODULO: i32 = 8;
const MAX_ITER_COUNT: u32 = 3;
let range_start = qc::Arbitrary::arbitrary(g);
let range_end = range_start + g.gen_range(0, MAX_STARTING_RANGE_DIFF + 1);
let start_step = g.gen_range(-MAX_STEP_MODULO, MAX_STEP_MODULO + 1);
let end_step = g.gen_range(-MAX_STEP_MODULO, MAX_STEP_MODULO + 1);
let iter_count = g.gen_range(0, MAX_ITER_COUNT + 1);
let hint_kind = qc::Arbitrary::arbitrary(g);
ShiftRange {
range_start,
range_end,
start_step,
end_step,
iter_count,
hint_kind,
}
}
}
fn correct_count<I, F>(get_it: F) -> bool
where
I: Iterator,
F: Fn() -> I
{
let mut counts = vec![get_it().count()];
'outer: loop {
let mut it = get_it();
for _ in 0..(counts.len() - 1) {
#[allow(clippy::manual_assert)]
if it.next().is_none() {
panic!("Iterator shouldn't be finished, may not be deterministic");
}
}
if it.next().is_none() {
break 'outer;
}
counts.push(it.count());
}
let total_actual_count = counts.len() - 1;
for (i, returned_count) in counts.into_iter().enumerate() {
let actual_count = total_actual_count - i;
if actual_count != returned_count {
println!("Total iterations: {} True count: {} returned count: {}", i, actual_count, returned_count);
return false;
}
}
true
}
fn correct_size_hint<I: Iterator>(mut it: I) -> bool {
// record size hint at each iteration
let initial_hint = it.size_hint();
let mut hints = Vec::with_capacity(initial_hint.0 + 1);
hints.push(initial_hint);
while let Some(_) = it.next() {
hints.push(it.size_hint())
}
let mut true_count = hints.len(); // start off +1 too much
// check all the size hints
for &(low, hi) in &hints {
true_count -= 1;
if low > true_count ||
(hi.is_some() && hi.unwrap() < true_count)
{
println!("True size: {:?}, size hint: {:?}", true_count, (low, hi));
//println!("All hints: {:?}", hints);
return false
}
}
true
}
fn exact_size<I: ExactSizeIterator>(mut it: I) -> bool {
// check every iteration
let (mut low, mut hi) = it.size_hint();
if Some(low) != hi { return false; }
while let Some(_) = it.next() {
let (xlow, xhi) = it.size_hint();
if low != xlow + 1 { return false; }
low = xlow;
hi = xhi;
if Some(low) != hi { return false; }
}
let (low, hi) = it.size_hint();
low == 0 && hi == Some(0)
}
// Exact size for this case, without ExactSizeIterator
fn exact_size_for_this<I: Iterator>(mut it: I) -> bool {
// check every iteration
let (mut low, mut hi) = it.size_hint();
if Some(low) != hi { return false; }
while let Some(_) = it.next() {
let (xlow, xhi) = it.size_hint();
if low != xlow + 1 { return false; }
low = xlow;
hi = xhi;
if Some(low) != hi { return false; }
}
let (low, hi) = it.size_hint();
low == 0 && hi == Some(0)
}
/*
* NOTE: Range<i8> is broken!
* (all signed ranges are)
#[quickcheck]
fn size_range_i8(a: Iter<i8>) -> bool {
exact_size(a)
}
#[quickcheck]
fn size_range_i16(a: Iter<i16>) -> bool {
exact_size(a)
}
#[quickcheck]
fn size_range_u8(a: Iter<u8>) -> bool {
exact_size(a)
}
*/
macro_rules! quickcheck {
// accept several property function definitions
// The property functions can use pattern matching and `mut` as usual
// in the function arguments, but the functions can not be generic.
{$($(#$attr:tt)* fn $fn_name:ident($($arg:tt)*) -> $ret:ty { $($code:tt)* })*} => (
$(
#[test]
$(#$attr)*
fn $fn_name() {
fn prop($($arg)*) -> $ret {
$($code)*
}
::quickcheck::quickcheck(quickcheck!(@fn prop [] $($arg)*));
}
)*
);
// parse argument list (with patterns allowed) into prop as fn(_, _) -> _
(@fn $f:ident [$($t:tt)*]) => {
$f as fn($($t),*) -> _
};
(@fn $f:ident [$($p:tt)*] : $($tail:tt)*) => {
quickcheck!(@fn $f [$($p)* _] $($tail)*)
};
(@fn $f:ident [$($p:tt)*] $t:tt $($tail:tt)*) => {
quickcheck!(@fn $f [$($p)*] $($tail)*)
};
}
quickcheck! {
fn size_product(a: Iter<u16>, b: Iter<u16>) -> bool {
correct_size_hint(a.cartesian_product(b))
}
fn size_product3(a: Iter<u16>, b: Iter<u16>, c: Iter<u16>) -> bool {
correct_size_hint(iproduct!(a, b, c))
}
fn correct_cartesian_product3(a: Iter<u16>, b: Iter<u16>, c: Iter<u16>,
take_manual: usize) -> ()
{
// test correctness of iproduct through regular iteration (take)
// and through fold.
let ac = a.clone();
let br = &b.clone();
let cr = &c.clone();
let answer: Vec<_> = ac.flat_map(move |ea| br.clone().flat_map(move |eb| cr.clone().map(move |ec| (ea, eb, ec)))).collect();
let mut product_iter = iproduct!(a, b, c);
let mut actual = Vec::new();
actual.extend((&mut product_iter).take(take_manual));
if actual.len() == take_manual {
product_iter.fold((), |(), elt| actual.push(elt));
}
assert_eq!(answer, actual);
}
fn size_multi_product(a: ShiftRange) -> bool {
correct_size_hint(a.multi_cartesian_product())
}
fn correct_multi_product3(a: ShiftRange, take_manual: usize) -> () {
// Fix no. of iterators at 3
let a = ShiftRange { iter_count: 3, ..a };
// test correctness of MultiProduct through regular iteration (take)
// and through fold.
let mut iters = a.clone();
let i0 = iters.next().unwrap();
let i1r = &iters.next().unwrap();
let i2r = &iters.next().unwrap();
let answer: Vec<_> = i0.flat_map(move |ei0| i1r.clone().flat_map(move |ei1| i2r.clone().map(move |ei2| vec![ei0, ei1, ei2]))).collect();
let mut multi_product = a.clone().multi_cartesian_product();
let mut actual = Vec::new();
actual.extend((&mut multi_product).take(take_manual));
if actual.len() == take_manual {
multi_product.fold((), |(), elt| actual.push(elt));
}
assert_eq!(answer, actual);
assert_eq!(answer.into_iter().last(), a.multi_cartesian_product().last());
}
#[allow(deprecated)]
fn size_step(a: Iter<i16, Exact>, s: usize) -> bool {
let mut s = s;
if s == 0 {
s += 1; // never zero
}
let filt = a.clone().dedup();
correct_size_hint(filt.step(s)) &&
exact_size(a.step(s))
}
#[allow(deprecated)]
fn equal_step(a: Iter<i16>, s: usize) -> bool {
let mut s = s;
if s == 0 {
s += 1; // never zero
}
let mut i = 0;
itertools::equal(a.clone().step(s), a.filter(|_| {
let keep = i % s == 0;
i += 1;
keep
}))
}
#[allow(deprecated)]
fn equal_step_vec(a: Vec<i16>, s: usize) -> bool {
let mut s = s;
if s == 0 {
s += 1; // never zero
}
let mut i = 0;
itertools::equal(a.iter().step(s), a.iter().filter(|_| {
let keep = i % s == 0;
i += 1;
keep
}))
}
fn size_multipeek(a: Iter<u16, Exact>, s: u8) -> bool {
let mut it = multipeek(a);
// peek a few times
for _ in 0..s {
it.peek();
}
exact_size(it)
}
fn size_peek_nth(a: Iter<u16, Exact>, s: u8) -> bool {
let mut it = peek_nth(a);
// peek a few times
for n in 0..s {
it.peek_nth(n as usize);
}
exact_size(it)
}
fn equal_merge(mut a: Vec<i16>, mut b: Vec<i16>) -> bool {
a.sort();
b.sort();
let mut merged = a.clone();
merged.extend(b.iter().cloned());
merged.sort();
itertools::equal(&merged, a.iter().merge(&b))
}
fn size_merge(a: Iter<u16>, b: Iter<u16>) -> bool {
correct_size_hint(a.merge(b))
}
fn size_zip(a: Iter<i16, Exact>, b: Iter<i16, Exact>, c: Iter<i16, Exact>) -> bool {
let filt = a.clone().dedup();
correct_size_hint(multizip((filt, b.clone(), c.clone()))) &&
exact_size(multizip((a, b, c)))
}
fn size_zip_rc(a: Iter<i16>, b: Iter<i16>) -> bool {
let rc = rciter(a);
correct_size_hint(multizip((&rc, &rc, b)))
}
fn size_zip_macro(a: Iter<i16, Exact>, b: Iter<i16, Exact>, c: Iter<i16, Exact>) -> bool {
let filt = a.clone().dedup();
correct_size_hint(izip!(filt, b.clone(), c.clone())) &&
exact_size(izip!(a, b, c))
}
fn equal_kmerge(mut a: Vec<i16>, mut b: Vec<i16>, mut c: Vec<i16>) -> bool {
use itertools::free::kmerge;
a.sort();
b.sort();
c.sort();
let mut merged = a.clone();
merged.extend(b.iter().cloned());
merged.extend(c.iter().cloned());
merged.sort();
itertools::equal(merged.into_iter(), kmerge(vec![a, b, c]))
}
// Any number of input iterators
fn equal_kmerge_2(mut inputs: Vec<Vec<i16>>) -> bool {
use itertools::free::kmerge;
// sort the inputs
for input in &mut inputs {
input.sort();
}
let mut merged = inputs.concat();
merged.sort();
itertools::equal(merged.into_iter(), kmerge(inputs))
}
// Any number of input iterators
fn equal_kmerge_by_ge(mut inputs: Vec<Vec<i16>>) -> bool {
// sort the inputs
for input in &mut inputs {
input.sort();
input.reverse();
}
let mut merged = inputs.concat();
merged.sort();
merged.reverse();
itertools::equal(merged.into_iter(),
inputs.into_iter().kmerge_by(|x, y| x >= y))
}
// Any number of input iterators
fn equal_kmerge_by_lt(mut inputs: Vec<Vec<i16>>) -> bool {
// sort the inputs
for input in &mut inputs {
input.sort();
}
let mut merged = inputs.concat();
merged.sort();
itertools::equal(merged.into_iter(),
inputs.into_iter().kmerge_by(|x, y| x < y))
}
// Any number of input iterators
fn equal_kmerge_by_le(mut inputs: Vec<Vec<i16>>) -> bool {
// sort the inputs
for input in &mut inputs {
input.sort();
}
let mut merged = inputs.concat();
merged.sort();
itertools::equal(merged.into_iter(),
inputs.into_iter().kmerge_by(|x, y| x <= y))
}
fn size_kmerge(a: Iter<i16>, b: Iter<i16>, c: Iter<i16>) -> bool {
use itertools::free::kmerge;
correct_size_hint(kmerge(vec![a, b, c]))
}
fn equal_zip_eq(a: Vec<i32>, b: Vec<i32>) -> bool {
let len = std::cmp::min(a.len(), b.len());
let a = &a[..len];
let b = &b[..len];
itertools::equal(zip_eq(a, b), zip(a, b))
}
fn size_zip_longest(a: Iter<i16, Exact>, b: Iter<i16, Exact>) -> bool {
let filt = a.clone().dedup();
let filt2 = b.clone().dedup();
correct_size_hint(filt.zip_longest(b.clone())) &&
correct_size_hint(a.clone().zip_longest(filt2)) &&
exact_size(a.zip_longest(b))
}
fn size_2_zip_longest(a: Iter<i16>, b: Iter<i16>) -> bool {
let it = a.clone().zip_longest(b.clone());
let jt = a.clone().zip_longest(b.clone());
itertools::equal(a,
it.filter_map(|elt| match elt {
EitherOrBoth::Both(x, _) => Some(x),
EitherOrBoth::Left(x) => Some(x),
_ => None,
}
))
&&
itertools::equal(b,
jt.filter_map(|elt| match elt {
EitherOrBoth::Both(_, y) => Some(y),
EitherOrBoth::Right(y) => Some(y),
_ => None,
}
))
}
fn size_interleave(a: Iter<i16>, b: Iter<i16>) -> bool {
correct_size_hint(a.interleave(b))
}
fn exact_interleave(a: Iter<i16, Exact>, b: Iter<i16, Exact>) -> bool {
exact_size_for_this(a.interleave(b))
}
fn size_interleave_shortest(a: Iter<i16>, b: Iter<i16>) -> bool {
correct_size_hint(a.interleave_shortest(b))
}
fn exact_interleave_shortest(a: Vec<()>, b: Vec<()>) -> bool {
exact_size_for_this(a.iter().interleave_shortest(&b))
}
fn size_intersperse(a: Iter<i16>, x: i16) -> bool {
correct_size_hint(a.intersperse(x))
}
fn equal_intersperse(a: Vec<i32>, x: i32) -> bool {
let mut inter = false;
let mut i = 0;
for elt in a.iter().cloned().intersperse(x) {
if inter {
if elt != x { return false }
} else {
if elt != a[i] { return false }
i += 1;
}
inter = !inter;
}
true
}
fn equal_combinations_2(a: Vec<u8>) -> bool {
let mut v = Vec::new();
for (i, x) in enumerate(&a) {
for y in &a[i + 1..] {
v.push((x, y));
}
}
itertools::equal(a.iter().tuple_combinations::<(_, _)>(), v)
}
fn collect_tuple_matches_size(a: Iter<i16>) -> bool {
let size = a.clone().count();
a.collect_tuple::<(_, _, _)>().is_some() == (size == 3)
}
fn correct_permutations(vals: HashSet<i32>, k: usize) -> () {
// Test permutations only on iterators of distinct integers, to prevent
// false positives.
const MAX_N: usize = 5;
let n = min(vals.len(), MAX_N);
let vals: HashSet<i32> = vals.into_iter().take(n).collect();
let perms = vals.iter().permutations(k);
let mut actual = HashSet::new();
for perm in perms {
assert_eq!(perm.len(), k);
let all_items_valid = perm.iter().all(|p| vals.contains(p));
assert!(all_items_valid, "perm contains value not from input: {:?}", perm);
// Check that all perm items are distinct
let distinct_len = {
let perm_set: HashSet<_> = perm.iter().collect();
perm_set.len()
};
assert_eq!(perm.len(), distinct_len);
// Check that the perm is new
assert!(actual.insert(perm.clone()), "perm already encountered: {:?}", perm);
}
}
fn permutations_lexic_order(a: usize, b: usize) -> () {
let a = a % 6;
let b = b % 6;
let n = max(a, b);
let k = min (a, b);
let expected_first: Vec<usize> = (0..k).collect();
let expected_last: Vec<usize> = ((n - k)..n).rev().collect();
let mut perms = (0..n).permutations(k);
let mut curr_perm = match perms.next() {
Some(p) => p,
None => { return; }
};
assert_eq!(expected_first, curr_perm);
for next_perm in perms {
assert!(
next_perm > curr_perm,
"next perm isn't greater-than current; next_perm={:?} curr_perm={:?} n={}",
next_perm, curr_perm, n
);
curr_perm = next_perm;
}
assert_eq!(expected_last, curr_perm);
}
fn permutations_count(n: usize, k: usize) -> bool {
let n = n % 6;
correct_count(|| (0..n).permutations(k))
}
fn permutations_size(a: Iter<i32>, k: usize) -> bool {
correct_size_hint(a.take(5).permutations(k))
}
fn permutations_k0_yields_once(n: usize) -> () {
let k = 0;
let expected: Vec<Vec<usize>> = vec![vec![]];
let actual = (0..n).permutations(k).collect_vec();
assert_eq!(expected, actual);
}
}
quickcheck! {
fn dedup_via_coalesce(a: Vec<i32>) -> bool {
let mut b = a.clone();
b.dedup();
itertools::equal(
&b,
a
.iter()
.coalesce(|x, y| {
if x==y {
Ok(x)
} else {
Err((x, y))
}
})
.fold(vec![], |mut v, n| {
v.push(n);
v
})
)
}
}
quickcheck! {
fn equal_dedup(a: Vec<i32>) -> bool {
let mut b = a.clone();
b.dedup();
itertools::equal(&b, a.iter().dedup())
}
}
quickcheck! {
fn equal_dedup_by(a: Vec<(i32, i32)>) -> bool {
let mut b = a.clone();
b.dedup_by(|x, y| x.0==y.0);
itertools::equal(&b, a.iter().dedup_by(|x, y| x.0==y.0))
}
}
quickcheck! {
fn size_dedup(a: Vec<i32>) -> bool {
correct_size_hint(a.iter().dedup())
}
}
quickcheck! {
fn size_dedup_by(a: Vec<(i32, i32)>) -> bool {
correct_size_hint(a.iter().dedup_by(|x, y| x.0==y.0))
}
}
quickcheck! {
fn exact_repeatn((n, x): (usize, i32)) -> bool {
let it = itertools::repeat_n(x, n);
exact_size(it)
}
}
quickcheck! {
fn size_put_back(a: Vec<u8>, x: Option<u8>) -> bool {
let mut it = put_back(a.into_iter());
match x {
Some(t) => it.put_back(t),
None => {}
}
correct_size_hint(it)
}
}
quickcheck! {
fn size_put_backn(a: Vec<u8>, b: Vec<u8>) -> bool {
let mut it = put_back_n(a.into_iter());
for elt in b {
it.put_back(elt)
}
correct_size_hint(it)
}
}
quickcheck! {
fn size_tee(a: Vec<u8>) -> bool {
let (mut t1, mut t2) = a.iter().tee();
t1.next();
t1.next();
t2.next();
exact_size(t1) && exact_size(t2)
}
}
quickcheck! {
fn size_tee_2(a: Vec<u8>) -> bool {
let (mut t1, mut t2) = a.iter().dedup().tee();
t1.next();
t1.next();
t2.next();
correct_size_hint(t1) && correct_size_hint(t2)
}
}
quickcheck! {
fn size_take_while_ref(a: Vec<u8>, stop: u8) -> bool {
correct_size_hint(a.iter().take_while_ref(|x| **x != stop))
}
}
quickcheck! {
fn equal_partition(a: Vec<i32>) -> bool {
let mut a = a;
let mut ap = a.clone();
let split_index = itertools::partition(&mut ap, |x| *x >= 0);
let parted = (0..split_index).all(|i| ap[i] >= 0) &&
(split_index..a.len()).all(|i| ap[i] < 0);
a.sort();
ap.sort();
parted && (a == ap)
}
}
quickcheck! {
fn size_combinations(it: Iter<i16>) -> bool {
correct_size_hint(it.tuple_combinations::<(_, _)>())
}
}
quickcheck! {
fn equal_combinations(it: Iter<i16>) -> bool {
let values = it.clone().collect_vec();
let mut cmb = it.tuple_combinations();
for i in 0..values.len() {
for j in i+1..values.len() {
let pair = (values[i], values[j]);
if pair != cmb.next().unwrap() {
return false;
}
}
}
cmb.next() == None
}
}
quickcheck! {
fn size_pad_tail(it: Iter<i8>, pad: u8) -> bool {
correct_size_hint(it.clone().pad_using(pad as usize, |_| 0)) &&
correct_size_hint(it.dropping(1).rev().pad_using(pad as usize, |_| 0))
}
}
quickcheck! {
fn size_pad_tail2(it: Iter<i8, Exact>, pad: u8) -> bool {
exact_size(it.pad_using(pad as usize, |_| 0))
}
}
quickcheck! {
fn size_powerset(it: Iter<u8, Exact>) -> bool {
// Powerset cardinality gets large very quickly, limit input to keep test fast.
correct_size_hint(it.take(12).powerset())
}
}
quickcheck! {
fn size_duplicates(it: Iter<i8>) -> bool {
correct_size_hint(it.duplicates())
}
}
quickcheck! {
fn size_unique(it: Iter<i8>) -> bool {
correct_size_hint(it.unique())
}
fn count_unique(it: Vec<i8>, take_first: u8) -> () {
let answer = {
let mut v = it.clone();
v.sort(); v.dedup();
v.len()
};
let mut iter = cloned(&it).unique();
let first_count = (&mut iter).take(take_first as usize).count();
let rest_count = iter.count();
assert_eq!(answer, first_count + rest_count);
}
}
quickcheck! {
fn fuzz_group_by_lazy_1(it: Iter<u8>) -> bool {
let jt = it.clone();
let groups = it.group_by(|k| *k);
itertools::equal(jt, groups.into_iter().flat_map(|(_, x)| x))
}
}
quickcheck! {
fn fuzz_group_by_lazy_2(data: Vec<u8>) -> bool {
let groups = data.iter().group_by(|k| *k / 10);
let res = itertools::equal(data.iter(), groups.into_iter().flat_map(|(_, x)| x));
res
}
}
quickcheck! {
fn fuzz_group_by_lazy_3(data: Vec<u8>) -> bool {
let grouper = data.iter().group_by(|k| *k / 10);
let groups = grouper.into_iter().collect_vec();
let res = itertools::equal(data.iter(), groups.into_iter().flat_map(|(_, x)| x));
res
}
}
quickcheck! {
fn fuzz_group_by_lazy_duo(data: Vec<u8>, order: Vec<(bool, bool)>) -> bool {
let grouper = data.iter().group_by(|k| *k / 3);
let mut groups1 = grouper.into_iter();
let mut groups2 = grouper.into_iter();
let mut elts = Vec::<&u8>::new();
let mut old_groups = Vec::new();
let tup1 = |(_, b)| b;
for &(ord, consume_now) in &order {
let iter = &mut [&mut groups1, &mut groups2][ord as usize];
match iter.next() {
Some((_, gr)) => if consume_now {
for og in old_groups.drain(..) {
elts.extend(og);
}
elts.extend(gr);
} else {
old_groups.push(gr);
},
None => break,
}
}
for og in old_groups.drain(..) {
elts.extend(og);
}
for gr in groups1.map(&tup1) { elts.extend(gr); }
for gr in groups2.map(&tup1) { elts.extend(gr); }
itertools::assert_equal(&data, elts);
true
}
}
quickcheck! {
fn equal_chunks_lazy(a: Vec<u8>, size: u8) -> bool {
let mut size = size;
if size == 0 {
size += 1;
}
let chunks = a.iter().chunks(size as usize);
let it = a.chunks(size as usize);
for (a, b) in chunks.into_iter().zip(it) {
if !itertools::equal(a, b) {
return false;
}
}
true
}
}
quickcheck! {
fn equal_tuple_windows_1(a: Vec<u8>) -> bool {
let x = a.windows(1).map(|s| (&s[0], ));
let y = a.iter().tuple_windows::<(_,)>();
itertools::equal(x, y)
}
fn equal_tuple_windows_2(a: Vec<u8>) -> bool {
let x = a.windows(2).map(|s| (&s[0], &s[1]));
let y = a.iter().tuple_windows::<(_, _)>();
itertools::equal(x, y)
}
fn equal_tuple_windows_3(a: Vec<u8>) -> bool {
let x = a.windows(3).map(|s| (&s[0], &s[1], &s[2]));
let y = a.iter().tuple_windows::<(_, _, _)>();
itertools::equal(x, y)
}
fn equal_tuple_windows_4(a: Vec<u8>) -> bool {
let x = a.windows(4).map(|s| (&s[0], &s[1], &s[2], &s[3]));
let y = a.iter().tuple_windows::<(_, _, _, _)>();
itertools::equal(x, y)
}
fn equal_tuples_1(a: Vec<u8>) -> bool {
let x = a.chunks(1).map(|s| (&s[0], ));
let y = a.iter().tuples::<(_,)>();
itertools::equal(x, y)
}
fn equal_tuples_2(a: Vec<u8>) -> bool {
let x = a.chunks(2).filter(|s| s.len() == 2).map(|s| (&s[0], &s[1]));
let y = a.iter().tuples::<(_, _)>();
itertools::equal(x, y)
}
fn equal_tuples_3(a: Vec<u8>) -> bool {
let x = a.chunks(3).filter(|s| s.len() == 3).map(|s| (&s[0], &s[1], &s[2]));
let y = a.iter().tuples::<(_, _, _)>();
itertools::equal(x, y)
}
fn equal_tuples_4(a: Vec<u8>) -> bool {
let x = a.chunks(4).filter(|s| s.len() == 4).map(|s| (&s[0], &s[1], &s[2], &s[3]));
let y = a.iter().tuples::<(_, _, _, _)>();
itertools::equal(x, y)
}
fn exact_tuple_buffer(a: Vec<u8>) -> bool {
let mut iter = a.iter().tuples::<(_, _, _, _)>();
(&mut iter).last();
let buffer = iter.into_buffer();
assert_eq!(buffer.len(), a.len() % 4);
exact_size(buffer)
}
}
// with_position
quickcheck! {
fn with_position_exact_size_1(a: Vec<u8>) -> bool {
exact_size_for_this(a.iter().with_position())
}
fn with_position_exact_size_2(a: Iter<u8, Exact>) -> bool {
exact_size_for_this(a.with_position())
}
}
quickcheck! {
fn correct_group_map_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let count = a.len();
let lookup = a.into_iter().map(|i| (i % modulo, i)).into_group_map();
assert_eq!(lookup.values().flat_map(|vals| vals.iter()).count(), count);
for (&key, vals) in lookup.iter() {
assert!(vals.iter().all(|&val| val % modulo == key));
}
}
}
/// A peculiar type: Equality compares both tuple items, but ordering only the
/// first item. This is so we can check the stability property easily.
#[derive(Clone, Debug, PartialEq, Eq)]
struct Val(u32, u32);
impl PartialOrd<Val> for Val {
fn partial_cmp(&self, other: &Val) -> Option<Ordering> {
self.0.partial_cmp(&other.0)
}
}
impl Ord for Val {
fn cmp(&self, other: &Val) -> Ordering {
self.0.cmp(&other.0)
}
}
impl qc::Arbitrary for Val {
fn arbitrary<G: qc::Gen>(g: &mut G) -> Self {
let (x, y) = <(u32, u32)>::arbitrary(g);
Val(x, y)
}
fn shrink(&self) -> Box<dyn Iterator<Item = Self>> {
Box::new((self.0, self.1).shrink().map(|(x, y)| Val(x, y)))
}
}
quickcheck! {
fn minmax(a: Vec<Val>) -> bool {
use itertools::MinMaxResult;
let minmax = a.iter().minmax();
let expected = match a.len() {
0 => MinMaxResult::NoElements,
1 => MinMaxResult::OneElement(&a[0]),
_ => MinMaxResult::MinMax(a.iter().min().unwrap(),
a.iter().max().unwrap()),
};
minmax == expected
}
}
quickcheck! {
fn minmax_f64(a: Vec<f64>) -> TestResult {
use itertools::MinMaxResult;
if a.iter().any(|x| x.is_nan()) {
return TestResult::discard();
}
let min = cloned(&a).fold1(f64::min);
let max = cloned(&a).fold1(f64::max);
let minmax = cloned(&a).minmax();
let expected = match a.len() {
0 => MinMaxResult::NoElements,
1 => MinMaxResult::OneElement(min.unwrap()),
_ => MinMaxResult::MinMax(min.unwrap(), max.unwrap()),
};
TestResult::from_bool(minmax == expected)
}
}
quickcheck! {
#[allow(deprecated)]
fn tree_fold1_f64(mut a: Vec<f64>) -> TestResult {
fn collapse_adjacent<F>(x: Vec<f64>, mut f: F) -> Vec<f64>
where F: FnMut(f64, f64) -> f64
{
let mut out = Vec::new();
for i in (0..x.len()).step(2) {
if i == x.len()-1 {
out.push(x[i])
} else {
out.push(f(x[i], x[i+1]));
}
}
out
}
if a.iter().any(|x| x.is_nan()) {
return TestResult::discard();
}
let actual = a.iter().cloned().tree_fold1(f64::atan2);
while a.len() > 1 {
a = collapse_adjacent(a, f64::atan2);
}
let expected = a.pop();
TestResult::from_bool(actual == expected)
}
}
quickcheck! {
fn exactly_one_i32(a: Vec<i32>) -> TestResult {
let ret = a.iter().cloned().exactly_one();
match a.len() {
1 => TestResult::from_bool(ret.unwrap() == a[0]),
_ => TestResult::from_bool(ret.unwrap_err().eq(a.iter().cloned())),
}
}
}
quickcheck! {
fn at_most_one_i32(a: Vec<i32>) -> TestResult {
let ret = a.iter().cloned().at_most_one();
match a.len() {
0 => TestResult::from_bool(ret.unwrap() == None),
1 => TestResult::from_bool(ret.unwrap() == Some(a[0])),
_ => TestResult::from_bool(ret.unwrap_err().eq(a.iter().cloned())),
}
}
}
quickcheck! {
fn consistent_grouping_map_with_by(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup_grouping_map = a.iter().copied().map(|i| (i % modulo, i)).into_grouping_map().collect::<Vec<_>>();
let lookup_grouping_map_by = a.iter().copied().into_grouping_map_by(|i| i % modulo).collect::<Vec<_>>();
assert_eq!(lookup_grouping_map, lookup_grouping_map_by);
}
fn correct_grouping_map_by_aggregate_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo < 2 { 2 } else { modulo } as u64; // Avoid `% 0`
let lookup = a.iter()
.map(|&b| b as u64) // Avoid overflows
.into_grouping_map_by(|i| i % modulo)
.aggregate(|acc, &key, val| {
assert!(val % modulo == key);
if val % (modulo - 1) == 0 {
None
} else {
Some(acc.unwrap_or(0) + val)
}
});
let group_map_lookup = a.iter()
.map(|&b| b as u64)
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.filter_map(|(key, vals)| {
vals.into_iter().fold(None, |acc, val| {
if val % (modulo - 1) == 0 {
None
} else {
Some(acc.unwrap_or(0) + val)
}
}).map(|new_val| (key, new_val))
})
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for m in 0..modulo {
assert_eq!(
lookup.get(&m).copied(),
a.iter()
.map(|&b| b as u64)
.filter(|&val| val % modulo == m)
.fold(None, |acc, val| {
if val % (modulo - 1) == 0 {
None
} else {
Some(acc.unwrap_or(0) + val)
}
})
);
}
}
fn correct_grouping_map_by_fold_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo } as u64; // Avoid `% 0`
let lookup = a.iter().map(|&b| b as u64) // Avoid overflows
.into_grouping_map_by(|i| i % modulo)
.fold(0u64, |acc, &key, val| {
assert!(val % modulo == key);
acc + val
});
let group_map_lookup = a.iter()
.map(|&b| b as u64)
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().sum()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &sum) in lookup.iter() {
assert_eq!(sum, a.iter().map(|&b| b as u64).filter(|&val| val % modulo == key).sum::<u64>());
}
}
fn correct_grouping_map_by_fold_first_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo } as u64; // Avoid `% 0`
let lookup = a.iter().map(|&b| b as u64) // Avoid overflows
.into_grouping_map_by(|i| i % modulo)
.fold_first(|acc, &key, val| {
assert!(val % modulo == key);
acc + val
});
// TODO: Swap `fold1` with stdlib's `fold_first` when it's stabilized
let group_map_lookup = a.iter()
.map(|&b| b as u64)
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().fold1(|acc, val| acc + val).unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &sum) in lookup.iter() {
assert_eq!(sum, a.iter().map(|&b| b as u64).filter(|&val| val % modulo == key).sum::<u64>());
}
}
fn correct_grouping_map_by_collect_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup_grouping_map = a.iter().copied().into_grouping_map_by(|i| i % modulo).collect::<Vec<_>>();
let lookup_group_map = a.iter().copied().map(|i| (i % modulo, i)).into_group_map();
assert_eq!(lookup_grouping_map, lookup_group_map);
}
fn correct_grouping_map_by_max_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).max();
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().max().unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &max) in lookup.iter() {
assert_eq!(Some(max), a.iter().copied().filter(|&val| val % modulo == key).max());
}
}
fn correct_grouping_map_by_max_by_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).max_by(|_, v1, v2| v1.cmp(v2));
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().max_by(|v1, v2| v1.cmp(v2)).unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &max) in lookup.iter() {
assert_eq!(Some(max), a.iter().copied().filter(|&val| val % modulo == key).max_by(|v1, v2| v1.cmp(v2)));
}
}
fn correct_grouping_map_by_max_by_key_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).max_by_key(|_, &val| val);
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().max_by_key(|&val| val).unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &max) in lookup.iter() {
assert_eq!(Some(max), a.iter().copied().filter(|&val| val % modulo == key).max_by_key(|&val| val));
}
}
fn correct_grouping_map_by_min_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).min();
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().min().unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &min) in lookup.iter() {
assert_eq!(Some(min), a.iter().copied().filter(|&val| val % modulo == key).min());
}
}
fn correct_grouping_map_by_min_by_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).min_by(|_, v1, v2| v1.cmp(v2));
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().min_by(|v1, v2| v1.cmp(v2)).unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &min) in lookup.iter() {
assert_eq!(Some(min), a.iter().copied().filter(|&val| val % modulo == key).min_by(|v1, v2| v1.cmp(v2)));
}
}
fn correct_grouping_map_by_min_by_key_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).min_by_key(|_, &val| val);
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().min_by_key(|&val| val).unwrap()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &min) in lookup.iter() {
assert_eq!(Some(min), a.iter().copied().filter(|&val| val % modulo == key).min_by_key(|&val| val));
}
}
fn correct_grouping_map_by_minmax_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).minmax();
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().minmax()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &minmax) in lookup.iter() {
assert_eq!(minmax, a.iter().copied().filter(|&val| val % modulo == key).minmax());
}
}
fn correct_grouping_map_by_minmax_by_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).minmax_by(|_, v1, v2| v1.cmp(v2));
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().minmax_by(|v1, v2| v1.cmp(v2))))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &minmax) in lookup.iter() {
assert_eq!(minmax, a.iter().copied().filter(|&val| val % modulo == key).minmax_by(|v1, v2| v1.cmp(v2)));
}
}
fn correct_grouping_map_by_minmax_by_key_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo }; // Avoid `% 0`
let lookup = a.iter().copied().into_grouping_map_by(|i| i % modulo).minmax_by_key(|_, &val| val);
let group_map_lookup = a.iter().copied()
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().minmax_by_key(|&val| val)))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &minmax) in lookup.iter() {
assert_eq!(minmax, a.iter().copied().filter(|&val| val % modulo == key).minmax_by_key(|&val| val));
}
}
fn correct_grouping_map_by_sum_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = if modulo == 0 { 1 } else { modulo } as u64; // Avoid `% 0`
let lookup = a.iter().map(|&b| b as u64) // Avoid overflows
.into_grouping_map_by(|i| i % modulo)
.sum();
let group_map_lookup = a.iter().map(|&b| b as u64)
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().sum()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &sum) in lookup.iter() {
assert_eq!(sum, a.iter().map(|&b| b as u64).filter(|&val| val % modulo == key).sum::<u64>());
}
}
fn correct_grouping_map_by_product_modulo_key(a: Vec<u8>, modulo: u8) -> () {
let modulo = Wrapping(if modulo == 0 { 1 } else { modulo } as u64); // Avoid `% 0`
let lookup = a.iter().map(|&b| Wrapping(b as u64)) // Avoid overflows
.into_grouping_map_by(|i| i % modulo)
.product();
let group_map_lookup = a.iter().map(|&b| Wrapping(b as u64))
.map(|i| (i % modulo, i))
.into_group_map()
.into_iter()
.map(|(key, vals)| (key, vals.into_iter().product::<Wrapping<u64>>()))
.collect::<HashMap<_,_>>();
assert_eq!(lookup, group_map_lookup);
for (&key, &prod) in lookup.iter() {
assert_eq!(
prod,
a.iter()
.map(|&b| Wrapping(b as u64))
.filter(|&val| val % modulo == key)
.product::<Wrapping<u64>>()
);
}
}
// This should check that if multiple elements are equally minimum or maximum
// then `max`, `min` and `minmax` pick the first minimum and the last maximum.
// This is to be consistent with `std::iter::max` and `std::iter::min`.
fn correct_grouping_map_by_min_max_minmax_order_modulo_key() -> () {
use itertools::MinMaxResult;
let lookup = (0..=10)
.into_grouping_map_by(|_| 0)
.max_by(|_, _, _| Ordering::Equal);
assert_eq!(lookup[&0], 10);
let lookup = (0..=10)
.into_grouping_map_by(|_| 0)
.min_by(|_, _, _| Ordering::Equal);
assert_eq!(lookup[&0], 0);
let lookup = (0..=10)
.into_grouping_map_by(|_| 0)
.minmax_by(|_, _, _| Ordering::Equal);
assert_eq!(lookup[&0], MinMaxResult::MinMax(0, 10));
}
}
quickcheck! {
fn counts(nums: Vec<isize>) -> TestResult {
let counts = nums.iter().counts();
for (&item, &count) in counts.iter() {
#[allow(clippy::absurd_extreme_comparisons)]
if count <= 0 {
return TestResult::failed();
}
if count != nums.iter().filter(|&x| x == item).count() {
return TestResult::failed();
}
}
for item in nums.iter() {
if !counts.contains_key(item) {
return TestResult::failed();
}
}
TestResult::passed()
}
}
quickcheck! {
fn test_double_ended_zip_2(a: Vec<u8>, b: Vec<u8>) -> TestResult {
let mut x =
multizip((a.clone().into_iter(), b.clone().into_iter()))
.collect_vec();
x.reverse();
let y =
multizip((a.into_iter(), b.into_iter()))
.rfold(Vec::new(), |mut vec, e| { vec.push(e); vec });
TestResult::from_bool(itertools::equal(x, y))
}
fn test_double_ended_zip_3(a: Vec<u8>, b: Vec<u8>, c: Vec<u8>) -> TestResult {
let mut x =
multizip((a.clone().into_iter(), b.clone().into_iter(), c.clone().into_iter()))
.collect_vec();
x.reverse();
let y =
multizip((a.into_iter(), b.into_iter(), c.into_iter()))
.rfold(Vec::new(), |mut vec, e| { vec.push(e); vec });
TestResult::from_bool(itertools::equal(x, y))
}
}
fn is_fused<I: Iterator>(mut it: I) -> bool
{
for _ in it.by_ref() {}
for _ in 0..10{
if it.next().is_some(){
return false;
}
}
true
}
quickcheck! {
fn fused_combination(a: Iter<i16>) -> bool
{
is_fused(a.clone().combinations(1)) &&
is_fused(a.combinations(3))
}
fn fused_combination_with_replacement(a: Iter<i16>) -> bool
{
is_fused(a.clone().combinations_with_replacement(1)) &&
is_fused(a.combinations_with_replacement(3))
}
fn fused_tuple_combination(a: Iter<i16>) -> bool
{
is_fused(a.clone().fuse().tuple_combinations::<(_,)>()) &&
is_fused(a.fuse().tuple_combinations::<(_,_,_)>())
}
fn fused_unique(a: Iter<i16>) -> bool
{
is_fused(a.fuse().unique())
}
fn fused_unique_by(a: Iter<i16>) -> bool
{
is_fused(a.fuse().unique_by(|x| x % 100))
}
fn fused_interleave_shortest(a: Iter<i16>, b: Iter<i16>) -> bool
{
!is_fused(a.clone().interleave_shortest(b.clone())) &&
is_fused(a.fuse().interleave_shortest(b.fuse()))
}
fn fused_product(a: Iter<i16>, b: Iter<i16>) -> bool
{
is_fused(a.fuse().cartesian_product(b.fuse()))
}
fn fused_merge(a: Iter<i16>, b: Iter<i16>) -> bool
{
is_fused(a.fuse().merge(b.fuse()))
}
fn fused_filter_ok(a: Iter<i16>) -> bool
{
is_fused(a.map(|x| if x % 2 == 0 {Ok(x)} else {Err(x)} )
.filter_ok(|x| x % 3 == 0)
.fuse())
}
fn fused_filter_map_ok(a: Iter<i16>) -> bool
{
is_fused(a.map(|x| if x % 2 == 0 {Ok(x)} else {Err(x)} )
.filter_map_ok(|x| if x % 3 == 0 {Some(x / 3)} else {None})
.fuse())
}
fn fused_positions(a: Iter<i16>) -> bool
{
!is_fused(a.clone().positions(|x|x%2==0)) &&
is_fused(a.fuse().positions(|x|x%2==0))
}
fn fused_update(a: Iter<i16>) -> bool
{
!is_fused(a.clone().update(|x|*x+=1)) &&
is_fused(a.fuse().update(|x|*x+=1))
}
fn fused_tuple_windows(a: Iter<i16>) -> bool
{
is_fused(a.fuse().tuple_windows::<(_,_)>())
}
fn fused_pad_using(a: Iter<i16>) -> bool
{
is_fused(a.fuse().pad_using(100,|_|0))
}
}
quickcheck! {
fn min_set_contains_min(a: Vec<(usize, char)>) -> bool {
let result_set = a.iter().min_set();
if let Some(result_element) = a.iter().min() {
result_set.contains(&result_element)
} else {
result_set.is_empty()
}
}
fn min_set_by_contains_min(a: Vec<(usize, char)>) -> bool {
let compare = |x: &&(usize, char), y: &&(usize, char)| x.1.cmp(&y.1);
let result_set = a.iter().min_set_by(compare);
if let Some(result_element) = a.iter().min_by(compare) {
result_set.contains(&result_element)
} else {
result_set.is_empty()
}
}
fn min_set_by_key_contains_min(a: Vec<(usize, char)>) -> bool {
let key = |x: &&(usize, char)| x.1;
let result_set = a.iter().min_set_by_key(&key);
if let Some(result_element) = a.iter().min_by_key(&key) {
result_set.contains(&result_element)
} else {
result_set.is_empty()
}
}
fn max_set_contains_max(a: Vec<(usize, char)>) -> bool {
let result_set = a.iter().max_set();
if let Some(result_element) = a.iter().max() {
result_set.contains(&result_element)
} else {
result_set.is_empty()
}
}
fn max_set_by_contains_max(a: Vec<(usize, char)>) -> bool {
let compare = |x: &&(usize, char), y: &&(usize, char)| x.1.cmp(&y.1);
let result_set = a.iter().max_set_by(compare);
if let Some(result_element) = a.iter().max_by(compare) {
result_set.contains(&result_element)
} else {
result_set.is_empty()
}
}
fn max_set_by_key_contains_max(a: Vec<(usize, char)>) -> bool {
let key = |x: &&(usize, char)| x.1;
let result_set = a.iter().max_set_by_key(&key);
if let Some(result_element) = a.iter().max_by_key(&key) {
result_set.contains(&result_element)
} else {
result_set.is_empty()
}
}
}