use itertools::Itertools; use ordermap::{OrderMap, OrderSet}; use quickcheck::Arbitrary; use quickcheck::Gen; use quickcheck::QuickCheck; use quickcheck::TestResult; use fnv::FnvHasher; use std::hash::{BuildHasher, BuildHasherDefault}; type FnvBuilder = BuildHasherDefault; type OrderMapFnv = OrderMap; use std::cmp::min; use std::collections::HashMap; use std::collections::HashSet; use std::fmt::Debug; use std::hash::Hash; use std::ops::Bound; use std::ops::Deref; use ordermap::map::Entry; use std::collections::hash_map::Entry as StdEntry; fn set<'a, T: 'a, I>(iter: I) -> HashSet where I: IntoIterator, T: Copy + Hash + Eq, { iter.into_iter().copied().collect() } fn ordermap<'a, T: 'a, I>(iter: I) -> OrderMap where I: IntoIterator, T: Copy + Hash + Eq, { OrderMap::from_iter(iter.into_iter().copied().map(|k| (k, ()))) } // Helper macro to allow us to use smaller quickcheck limits under miri. macro_rules! quickcheck_limit { (@as_items $($i:item)*) => ($($i)*); { $( $(#[$m:meta])* fn $fn_name:ident($($arg_name:ident : $arg_ty:ty),*) -> $ret:ty { $($code:tt)* } )* } => ( quickcheck::quickcheck! { @as_items $( #[test] $(#[$m])* fn $fn_name() { fn prop($($arg_name: $arg_ty),*) -> $ret { $($code)* } let mut quickcheck = QuickCheck::new(); if cfg!(miri) { quickcheck = quickcheck .gen(Gen::new(10)) .tests(10) .max_tests(100); } quickcheck.quickcheck(prop as fn($($arg_ty),*) -> $ret); } )* } ) } quickcheck_limit! { fn contains(insert: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } insert.iter().all(|&key| map.get(&key).is_some()) } fn contains_not(insert: Vec, not: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } let nots = &set(¬) - &set(&insert); nots.iter().all(|&key| map.get(&key).is_none()) } fn insert_remove(insert: Vec, remove: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } for &key in &remove { map.swap_remove(&key); } let elements = &set(&insert) - &set(&remove); map.len() == elements.len() && map.iter().count() == elements.len() && elements.iter().all(|k| map.get(k).is_some()) } fn insertion_order(insert: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } itertools::assert_equal(insert.iter().unique(), map.keys()); true } fn insert_sorted(insert: Vec<(u32, u32)>) -> bool { let mut hmap = HashMap::new(); let mut map = OrderMap::new(); let mut map2 = OrderMap::new(); for &(key, value) in &insert { hmap.insert(key, value); map.insert_sorted(key, value); match map2.entry(key) { Entry::Occupied(e) => *e.into_mut() = value, Entry::Vacant(e) => { e.insert_sorted(value); } } } itertools::assert_equal(hmap.iter().sorted(), &map); itertools::assert_equal(&map, &map2); true } fn pop(insert: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } let mut pops = Vec::new(); while let Some((key, _v)) = map.pop() { pops.push(key); } pops.reverse(); itertools::assert_equal(insert.iter().unique(), &pops); true } fn with_cap(template: Vec<()>) -> bool { let cap = template.len(); let map: OrderMap = OrderMap::with_capacity(cap); println!("wish: {}, got: {} (diff: {})", cap, map.capacity(), map.capacity() as isize - cap as isize); map.capacity() >= cap } fn drain_full(insert: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } let mut clone = map.clone(); let drained = clone.drain(..); for (key, _) in drained { map.swap_remove(&key); } map.is_empty() } fn drain_bounds(insert: Vec, range: (Bound, Bound)) -> TestResult { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } // First see if `Vec::drain` is happy with this range. let result = std::panic::catch_unwind(|| { let mut keys: Vec = map.keys().copied().collect(); keys.drain(range); keys }); if let Ok(keys) = result { map.drain(range); // Check that our `drain` matches the same key order. assert!(map.keys().eq(&keys)); // Check that hash lookups all work too. assert!(keys.iter().all(|key| map.contains_key(key))); TestResult::passed() } else { // If `Vec::drain` panicked, so should we. TestResult::must_fail(move || { map.drain(range); }) } } // aka shift_remove fn remove(insert: Vec, remove: Vec) -> bool { let mut map = OrderMap::new(); for &key in &insert { map.insert(key, ()); } for &key in &remove { map.remove(&key); } let elements = &set(&insert) - &set(&remove); // Check that order is preserved after removals let mut iter = map.keys(); for &key in insert.iter().unique() { if elements.contains(&key) { assert_eq!(Some(&key), iter.next()); } } map.len() == elements.len() && map.iter().count() == elements.len() && elements.iter().all(|k| map.get(k).is_some()) } fn indexing(insert: Vec) -> bool { let mut map: OrderMap<_, _> = insert.into_iter().map(|x| (x, x)).collect(); let set: OrderSet<_> = map.keys().copied().collect(); assert_eq!(map.len(), set.len()); for (i, &key) in set.iter().enumerate() { assert_eq!(map.get_index(i), Some((&key, &key))); assert_eq!(set.get_index(i), Some(&key)); assert_eq!(map[i], key); assert_eq!(set[i], key); *map.get_index_mut(i).unwrap().1 >>= 1; map[i] <<= 1; } set.iter().enumerate().all(|(i, &key)| { let value = key & !1; map[&key] == value && map[i] == value }) } // Use `u8` test indices so quickcheck is less likely to go out of bounds. fn set_swap_indices(vec: Vec, a: u8, b: u8) -> TestResult { let mut set = OrderSet::::from_iter(vec); let a = usize::from(a); let b = usize::from(b); if a >= set.len() || b >= set.len() { return TestResult::discard(); } let mut vec = Vec::from_iter(set.iter().cloned()); vec.swap(a, b); set.swap_indices(a, b); // Check both iteration order and hash lookups assert!(set.iter().eq(vec.iter())); assert!(vec.iter().enumerate().all(|(i, x)| { set.get_index_of(x) == Some(i) })); TestResult::passed() } fn map_swap_indices(vec: Vec, from: u8, to: u8) -> TestResult { test_map_swap_indices(vec, from, to, OrderMap::swap_indices) } fn occupied_entry_swap_indices(vec: Vec, from: u8, to: u8) -> TestResult { test_map_swap_indices(vec, from, to, |map, from, to| { let key = map.keys()[from]; match map.entry(key) { Entry::Occupied(entry) => entry.swap_indices(to), _ => unreachable!(), } }) } fn indexed_entry_swap_indices(vec: Vec, from: u8, to: u8) -> TestResult { test_map_swap_indices(vec, from, to, |map, from, to| { map.get_index_entry(from).unwrap().swap_indices(to); }) } fn raw_occupied_entry_swap_indices(vec: Vec, from: u8, to: u8) -> TestResult { use ordermap::map::raw_entry_v1::{RawEntryApiV1, RawEntryMut}; test_map_swap_indices(vec, from, to, |map, from, to| { let key = map.keys()[from]; match map.raw_entry_mut_v1().from_key(&key) { RawEntryMut::Occupied(entry) => entry.swap_indices(to), _ => unreachable!(), } }) } // Use `u8` test indices so quickcheck is less likely to go out of bounds. fn set_move_index(vec: Vec, from: u8, to: u8) -> TestResult { let mut set = OrderSet::::from_iter(vec); let from = usize::from(from); let to = usize::from(to); if from >= set.len() || to >= set.len() { return TestResult::discard(); } let mut vec = Vec::from_iter(set.iter().cloned()); let x = vec.remove(from); vec.insert(to, x); set.move_index(from, to); // Check both iteration order and hash lookups assert!(set.iter().eq(vec.iter())); assert!(vec.iter().enumerate().all(|(i, x)| { set.get_index_of(x) == Some(i) })); TestResult::passed() } fn map_move_index(vec: Vec, from: u8, to: u8) -> TestResult { test_map_move_index(vec, from, to, OrderMap::move_index) } fn occupied_entry_move_index(vec: Vec, from: u8, to: u8) -> TestResult { test_map_move_index(vec, from, to, |map, from, to| { let key = map.keys()[from]; match map.entry(key) { Entry::Occupied(entry) => entry.move_index(to), _ => unreachable!(), } }) } fn indexed_entry_move_index(vec: Vec, from: u8, to: u8) -> TestResult { test_map_move_index(vec, from, to, |map, from, to| { map.get_index_entry(from).unwrap().move_index(to); }) } fn raw_occupied_entry_move_index(vec: Vec, from: u8, to: u8) -> TestResult { use ordermap::map::raw_entry_v1::{RawEntryApiV1, RawEntryMut}; test_map_move_index(vec, from, to, |map, from, to| { let key = map.keys()[from]; match map.raw_entry_mut_v1().from_key(&key) { RawEntryMut::Occupied(entry) => entry.move_index(to), _ => unreachable!(), } }) } fn occupied_entry_shift_insert(vec: Vec, i: u8) -> TestResult { test_map_shift_insert(vec, i, |map, i, key| { match map.entry(key) { Entry::Vacant(entry) => entry.shift_insert(i, ()), _ => unreachable!(), }; }) } fn raw_occupied_entry_shift_insert(vec: Vec, i: u8) -> TestResult { use ordermap::map::raw_entry_v1::{RawEntryApiV1, RawEntryMut}; test_map_shift_insert(vec, i, |map, i, key| { match map.raw_entry_mut_v1().from_key(&key) { RawEntryMut::Vacant(entry) => entry.shift_insert(i, key, ()), _ => unreachable!(), }; }) } } fn test_map_swap_indices(vec: Vec, a: u8, b: u8, swap_indices: F) -> TestResult where F: FnOnce(&mut OrderMap, usize, usize), { let mut map = OrderMap::::from_iter(vec.into_iter().map(|k| (k, ()))); let a = usize::from(a); let b = usize::from(b); if a >= map.len() || b >= map.len() { return TestResult::discard(); } let mut vec = Vec::from_iter(map.keys().copied()); vec.swap(a, b); swap_indices(&mut map, a, b); // Check both iteration order and hash lookups assert!(map.keys().eq(vec.iter())); assert!(vec .iter() .enumerate() .all(|(i, x)| { map.get_index_of(x) == Some(i) })); TestResult::passed() } fn test_map_move_index(vec: Vec, from: u8, to: u8, move_index: F) -> TestResult where F: FnOnce(&mut OrderMap, usize, usize), { let mut map = OrderMap::::from_iter(vec.into_iter().map(|k| (k, ()))); let from = usize::from(from); let to = usize::from(to); if from >= map.len() || to >= map.len() { return TestResult::discard(); } let mut vec = Vec::from_iter(map.keys().copied()); let x = vec.remove(from); vec.insert(to, x); move_index(&mut map, from, to); // Check both iteration order and hash lookups assert!(map.keys().eq(vec.iter())); assert!(vec .iter() .enumerate() .all(|(i, x)| { map.get_index_of(x) == Some(i) })); TestResult::passed() } fn test_map_shift_insert(vec: Vec, i: u8, shift_insert: F) -> TestResult where F: FnOnce(&mut OrderMap, usize, u8), { let mut map = OrderMap::::from_iter(vec.into_iter().map(|k| (k, ()))); let i = usize::from(i); if i >= map.len() { return TestResult::discard(); } let mut vec = Vec::from_iter(map.keys().copied()); let x = vec.pop().unwrap(); vec.insert(i, x); let (last, ()) = map.pop().unwrap(); assert_eq!(x, last); map.shrink_to_fit(); // so we might have to grow and rehash the table shift_insert(&mut map, i, last); // Check both iteration order and hash lookups assert!(map.keys().eq(vec.iter())); assert!(vec .iter() .enumerate() .all(|(i, x)| { map.get_index_of(x) == Some(i) })); TestResult::passed() } use crate::Op::*; #[derive(Copy, Clone, Debug)] enum Op { Add(K, V), Remove(K), AddEntry(K, V), RemoveEntry(K), } impl Arbitrary for Op where K: Arbitrary, V: Arbitrary, { fn arbitrary(g: &mut Gen) -> Self { match u32::arbitrary(g) % 4 { 0 => Add(K::arbitrary(g), V::arbitrary(g)), 1 => AddEntry(K::arbitrary(g), V::arbitrary(g)), 2 => Remove(K::arbitrary(g)), _ => RemoveEntry(K::arbitrary(g)), } } } fn do_ops(ops: &[Op], a: &mut OrderMap, b: &mut HashMap) where K: Hash + Eq + Clone, V: Clone, S: BuildHasher, { for op in ops { match *op { Add(ref k, ref v) => { a.insert(k.clone(), v.clone()); b.insert(k.clone(), v.clone()); } AddEntry(ref k, ref v) => { a.entry(k.clone()).or_insert_with(|| v.clone()); b.entry(k.clone()).or_insert_with(|| v.clone()); } Remove(ref k) => { a.swap_remove(k); b.remove(k); } RemoveEntry(ref k) => { if let Entry::Occupied(ent) = a.entry(k.clone()) { ent.swap_remove_entry(); } if let StdEntry::Occupied(ent) = b.entry(k.clone()) { ent.remove_entry(); } } } //println!("{:?}", a); } } fn assert_maps_equivalent(a: &OrderMap, b: &HashMap) -> bool where K: Hash + Eq + Debug, V: Eq + Debug, { assert_eq!(a.len(), b.len()); assert_eq!(a.iter().next().is_some(), b.iter().next().is_some()); for key in a.keys() { assert!(b.contains_key(key), "b does not contain {:?}", key); } for key in b.keys() { assert!(a.get(key).is_some(), "a does not contain {:?}", key); } for key in a.keys() { assert_eq!(a[key], b[key]); } true } quickcheck_limit! { fn operations_i8(ops: Large>>) -> bool { let mut map = OrderMap::new(); let mut reference = HashMap::new(); do_ops(&ops, &mut map, &mut reference); assert_maps_equivalent(&map, &reference) } fn operations_string(ops: Vec>) -> bool { let mut map = OrderMap::new(); let mut reference = HashMap::new(); do_ops(&ops, &mut map, &mut reference); assert_maps_equivalent(&map, &reference) } fn keys_values(ops: Large>>) -> bool { let mut map = OrderMap::new(); let mut reference = HashMap::new(); do_ops(&ops, &mut map, &mut reference); let mut visit = OrderMap::new(); for (k, v) in map.keys().zip(map.values()) { assert_eq!(&map[k], v); assert!(!visit.contains_key(k)); visit.insert(*k, *v); } assert_eq!(visit.len(), reference.len()); true } fn keys_values_mut(ops: Large>>) -> bool { let mut map = OrderMap::new(); let mut reference = HashMap::new(); do_ops(&ops, &mut map, &mut reference); let mut visit = OrderMap::new(); let keys = Vec::from_iter(map.keys().copied()); for (k, v) in keys.iter().zip(map.values_mut()) { assert_eq!(&reference[k], v); assert!(!visit.contains_key(k)); visit.insert(*k, *v); } assert_eq!(visit.len(), reference.len()); true } fn equality(ops1: Vec>, removes: Vec) -> bool { let mut map = OrderMap::new(); let mut reference = HashMap::new(); do_ops(&ops1, &mut map, &mut reference); let mut ops2 = ops1.clone(); for &r in &removes { if !ops2.is_empty() { let i = r % ops2.len(); ops2.remove(i); } } let mut map2 = OrderMapFnv::default(); let mut reference2 = HashMap::new(); do_ops(&ops2, &mut map2, &mut reference2); map.sort_keys(); map2.sort_keys(); assert_eq!(map == map2, reference == reference2); true } fn retain_ordered(keys: Large>, remove: Large>) -> () { let mut map = ordermap(keys.iter()); let initial_map = map.clone(); // deduplicated in-order input let remove_map = ordermap(remove.iter()); let keys_s = set(keys.iter()); let remove_s = set(remove.iter()); let answer = &keys_s - &remove_s; map.retain(|k, _| !remove_map.contains_key(k)); // check the values assert_eq!(map.len(), answer.len()); for key in &answer { assert!(map.contains_key(key)); } // check the order itertools::assert_equal(map.keys(), initial_map.keys().filter(|&k| !remove_map.contains_key(k))); } fn sort_1(keyvals: Large>) -> () { let mut map: OrderMap<_, _> = OrderMap::from_iter(keyvals.to_vec()); let mut answer = keyvals.0; answer.sort_by_key(|t| t.0); // reverse dedup: Because OrderMap::from_iter keeps the last value for // identical keys answer.reverse(); answer.dedup_by_key(|t| t.0); answer.reverse(); map.sort_by(|k1, _, k2, _| Ord::cmp(k1, k2)); // check it contains all the values it should for &(key, val) in &answer { assert_eq!(map[&key], val); } // check the order let mapv = Vec::from_iter(map); assert_eq!(answer, mapv); } fn sort_2(keyvals: Large>) -> () { let mut map: OrderMap<_, _> = OrderMap::from_iter(keyvals.to_vec()); map.sort_by(|_, v1, _, v2| Ord::cmp(v1, v2)); assert_sorted_by_key(map, |t| t.1); } fn sort_3(keyvals: Large>) -> () { let mut map: OrderMap<_, _> = OrderMap::from_iter(keyvals.to_vec()); map.sort_by_cached_key(|&k, _| std::cmp::Reverse(k)); assert_sorted_by_key(map, |t| std::cmp::Reverse(t.0)); } fn reverse(keyvals: Large>) -> () { let mut map: OrderMap<_, _> = OrderMap::from_iter(keyvals.to_vec()); fn generate_answer(input: &Vec<(i8, i8)>) -> Vec<(i8, i8)> { // to mimic what `OrderMap::from_iter` does: // need to get (A) the unique keys in forward order, and (B) the // last value of each of those keys. // create (A): an iterable that yields the unique keys in ltr order let mut seen_keys = HashSet::new(); let unique_keys_forward = input.iter().filter_map(move |(k, _)| { if seen_keys.contains(k) { None } else { seen_keys.insert(*k); Some(*k) } }); // create (B): a mapping of keys to the last value seen for that key // this is the same as reversing the input and taking the first // value seen for that key! let mut last_val_per_key = HashMap::new(); for &(k, v) in input.iter().rev() { if !last_val_per_key.contains_key(&k) { last_val_per_key.insert(k, v); } } // iterate over the keys in (A) in order, and match each one with // the corresponding last value from (B) let mut ans: Vec<_> = unique_keys_forward .map(|k| (k, *last_val_per_key.get(&k).unwrap())) .collect(); // finally, since this test is testing `.reverse()`, reverse the // answer in-place ans.reverse(); ans } let answer = generate_answer(&keyvals.0); // perform the work map.reverse(); // check it contains all the values it should for &(key, val) in &answer { assert_eq!(map[&key], val); } // check the order let mapv = Vec::from_iter(map); assert_eq!(answer, mapv); } } fn assert_sorted_by_key(iterable: I, key: Key) where I: IntoIterator, I::Item: Ord + Clone + Debug, Key: Fn(&I::Item) -> X, X: Ord, { let input = Vec::from_iter(iterable); let mut sorted = input.clone(); sorted.sort_by_key(key); assert_eq!(input, sorted); } #[derive(Clone, Debug, Hash, PartialEq, Eq)] struct Alpha(String); impl Deref for Alpha { type Target = String; fn deref(&self) -> &String { &self.0 } } const ALPHABET: &[u8] = b"abcdefghijklmnopqrstuvwxyz"; impl Arbitrary for Alpha { fn arbitrary(g: &mut Gen) -> Self { let len = usize::arbitrary(g) % g.size(); let len = min(len, 16); Alpha( (0..len) .map(|_| ALPHABET[usize::arbitrary(g) % ALPHABET.len()] as char) .collect(), ) } fn shrink(&self) -> Box> { Box::new((**self).shrink().map(Alpha)) } } /// quickcheck Arbitrary adaptor -- make a larger vec #[derive(Clone, Debug)] struct Large(T); impl Deref for Large { type Target = T; fn deref(&self) -> &T { &self.0 } } impl Arbitrary for Large> where T: Arbitrary, { fn arbitrary(g: &mut Gen) -> Self { let len = usize::arbitrary(g) % (g.size() * 10); Large((0..len).map(|_| T::arbitrary(g)).collect()) } fn shrink(&self) -> Box> { Box::new((**self).shrink().map(Large)) } }