coordinode-lsm-tree 7.4.2

use super::*;
use crate::ValueType;

fn new_map() -> SkipMap {
    SkipMap::new(crate::comparator::default_comparator())
}

/// The word-at-a-time comparison orders every pair of keys as `<[u8]>::cmp`
/// does: every length up to past the longest key compared in words, a
/// difference at each position on either side of a word boundary, a key that
/// is a prefix of the other, a short key against a long one, and
/// pseudo-random pairs on both sides of the length limit. Guards against a
/// word compare that reads the bytes in the wrong order, lets the length
/// decide too early, or picks the path by the wrong key's length.
#[test]
fn the_word_comparison_orders_keys_as_the_byte_comparison() {
    let mut pairs: Vec<(Vec<u8>, Vec<u8>)> = Vec::new();
    for len in 0..=WORD_COMPARE_MAX + 12 {
        let base: Vec<u8> = (0..len)
            .map(|i| u8::try_from((i * 37 + 11) % 256).expect("below 256"))
            .collect();
        pairs.push((base.clone(), base.clone()));
        for at in 0..len {
            for delta in [1u8, 0x80, 0xFF] {
                let mut other = base.clone();
                other[at] = other[at].wrapping_add(delta);
                pairs.push((base.clone(), other.clone()));
                pairs.push((other, base.clone()));
            }
        }
        for shorter in 0..len {
            pairs.push((base[..shorter].to_vec(), base.clone()));
            pairs.push((base.clone(), base[..shorter].to_vec()));
        }
    }
    let mut state = 0x9E37_79B9_7F4A_7C15_u64;
    let mut next = || {
        state ^= state << 13;
        state ^= state >> 7;
        state ^= state << 17;
        state
    };
    for round in 0..10_000 {
        let mut below = |n: u64| next() % n;
        let (la, lb) = (below(160), below(160));
        // A small alphabet makes long common prefixes likely; every other
        // pair also shares a prefix longer than the word-compare limit.
        let mut a: Vec<u8> = (0..la)
            .map(|_| u8::try_from(below(3)).expect("below 3"))
            .collect();
        let mut b: Vec<u8> = (0..lb)
            .map(|_| u8::try_from(below(3)).expect("below 3"))
            .collect();
        if round % 2 == 0 {
            let shared = [7u8; WORD_COMPARE_MAX + 4];
            a = [&shared[..], &a].concat();
            b = [&shared[..], &b].concat();
        }
        pairs.push((a, b));
    }
    for (a, b) in pairs {
        assert_eq!(lexicographic(&a, &b), a.cmp(&b), "{a:?} vs {b:?}");
    }
}

fn make_key(user_key: &[u8], seqno: SeqNo) -> InternalKey {
    InternalKey::new(user_key.to_vec(), seqno, ValueType::Value)
}

fn make_value(data: &[u8]) -> UserValue {
    UserValue::from(data)
}

use crate::runtime_config::ChecksumAlgorithm;

/// Low-32 logical digest of an entry under `algo` (the value stored at
/// insert under `KvChecksumComputePoint::AtInsert`).
fn digest4(key: &InternalKey, val: &UserValue, algo: ChecksumAlgorithm) -> u32 {
    let item = InternalValue::new(key.clone(), val.clone());
    #[expect(
        clippy::cast_possible_truncation,
        clippy::expect_used,
        reason = "4-byte algo fits u32; test helper"
    )]
    let d = crate::table::block::kv_checksum::kv_digest(&item, algo).expect("xxh3 always available")
        as u32;
    d
}

#[test]
fn verify_kv_digests_passes_when_uncorrupted() {
    // Every node carries the digest computed over its own bytes, so a
    // recompute at flush matches and verify succeeds.
    let algo = ChecksumAlgorithm::Xxh3Low32;
    let map = new_map();
    for i in 0..8u8 {
        let key = make_key(&[b'k', i], u64::from(i) + 1);
        let val = make_value(&[b'v', i, i, i]);
        map.insert_with_kv_digest(&key, val.clone(), Some((digest4(&key, &val, algo), algo)));
    }
    assert!(map.verify_kv_digests().is_ok());
}

#[test]
#[expect(clippy::expect_used, reason = "test asserts the error via expect_err")]
fn verify_kv_digests_detects_residence_corruption() {
    // Simulate a RAM bit-flip during residence: store the correct digest at
    // insert, then corrupt the entry's key bytes in the arena. The recompute
    // over the corrupted bytes diverges from the stored digest, so verify
    // reports a MemtableKvChecksumMismatch.
    let algo = ChecksumAlgorithm::Xxh3Low32;
    let map = new_map();
    let key = make_key(b"victim", 42);
    let val = make_value(b"payload");
    map.insert_with_kv_digest(&key, val.clone(), Some((digest4(&key, &val, algo), algo)));

    // Flip a bit in the (only) node's user-key bytes, simulating residence
    // corruption.
    map.test_flip_first_key_byte();

    let err = map
        .verify_kv_digests()
        .expect_err("corruption must be detected");
    assert!(
        matches!(err, crate::Error::MemtableKvChecksumMismatch { .. }),
        "expected MemtableKvChecksumMismatch, got {err:?}"
    );
}

#[test]
fn verify_kv_digests_skips_nodes_without_a_digest() {
    // Mixed-variant memtable (Off -> AtInsert toggle): nodes inserted
    // without a digest are never verified, even if their bytes are
    // corrupted. Only digest-bearing nodes are checked.
    let algo = ChecksumAlgorithm::Xxh3Low32;
    let map = new_map();

    // No-digest node (pre-toggle): corrupt it later; must be skipped.
    let k0 = make_key(b"aaa", 1);
    let v0 = make_value(b"no-digest");
    map.insert_with_kv_digest(&k0, v0, None);

    // Digest-bearing node (post-toggle): left intact.
    let k1 = make_key(b"bbb", 2);
    let v1 = make_value(b"has-digest");
    map.insert_with_kv_digest(&k1, v1.clone(), Some((digest4(&k1, &v1, algo), algo)));

    // Corrupt the FIRST node's key (the no-digest "aaa" entry sorts first).
    map.test_flip_first_key_byte();

    // The corrupted node has no digest, so verify skips it and passes.
    assert!(
        map.verify_kv_digests().is_ok(),
        "nodes without a stored digest must not be verified"
    );
}

#[test]
fn verify_kv_digests_no_digests_is_ok() {
    // A memtable with only plain inserts has nothing to verify.
    let map = new_map();
    for i in 0..4u8 {
        let key = make_key(&[b'x', i], u64::from(i) + 1);
        map.insert(&key, make_value(b"plain"));
    }
    assert!(map.verify_kv_digests().is_ok());
}

#[test]
#[expect(clippy::expect_used, reason = "test helper computes a known digest")]
fn verify_kv_digests_rejects_non_4_byte_algorithm_tag() {
    // AtInsert only ever stores a 4-byte algorithm. A present-digest node
    // whose algorithm bits decode to a non-4-byte algorithm (e.g. Xxh3_64
    // after a single-bit flip from Xxh3Low32's tag 1 to tag 0) is metadata
    // corruption. Store the Xxh3_64 digest truncated to u32 under the
    // 8-byte Xxh3_64 tag: the worst case where comparing the digest alone
    // would pass. Verification must reject it on the algorithm tag instead.
    let map = new_map();
    let key = make_key(b"k", 1);
    let val = make_value(b"v");
    let item = InternalValue::new(key.clone(), val.clone());
    #[expect(clippy::cast_possible_truncation, reason = "low 32 bits on purpose")]
    let d64 = crate::table::block::kv_checksum::kv_digest(&item, ChecksumAlgorithm::Xxh3_64)
        .expect("xxh3 always available") as u32;
    map.insert_with_kv_digest(&key, val, Some((d64, ChecksumAlgorithm::Xxh3_64)));

    assert!(
        map.verify_kv_digests().is_err(),
        "a present digest under a non-4-byte algorithm must be rejected, not verified"
    );
}

#[test]
fn verify_kv_digests_fails_closed_on_corrupt_value_type() {
    // A digest-bearing node whose value_type bits were corrupted to an
    // invalid discriminant must surface a typed error from the residence
    // verifier, not panic (the verifier reconstructs the key, which decodes
    // value_type). Fail closed.
    let algo = ChecksumAlgorithm::Xxh3Low32;
    let map = new_map();
    let key = make_key(b"k", 1);
    let val = make_value(b"v");
    map.insert_with_kv_digest(&key, val.clone(), Some((digest4(&key, &val, algo), algo)));

    map.test_corrupt_first_node_value_type();

    assert!(
        map.verify_kv_digests().is_err(),
        "corrupt value_type on a digest-bearing node must fail closed, not panic"
    );
}

#[test]
fn insert_and_get_single() {
    let map = new_map();
    let key = make_key(b"hello", 1);
    let val = make_value(b"world");
    map.insert(&key, val);

    assert_eq!(map.len(), 1);
    assert!(!map.is_empty());

    let mut iter = map.iter();
    let entry = iter.next().expect("one entry");
    assert_eq!(&*entry.key().user_key, b"hello");
    assert_eq!(entry.key().seqno, 1);
    assert_eq!(&*entry.value(), b"world");
    assert!(iter.next().is_none());
}

#[test]
fn custom_comparator_orders_and_reverse_iterates() {
    // A non-lexicographic comparator (orders user keys in REVERSE byte
    // order) exercises the trait-dispatch (`else`) branch of compare_key
    // (via insert / find_splice) and compare_nodes (via reverse iteration /
    // find_predecessor) — the paths the default lexicographic fast path
    // skips.
    #[derive(Debug)]
    struct ReverseComparator;
    impl crate::comparator::UserComparator for ReverseComparator {
        fn name(&self) -> &'static str {
            "reverse-test"
        }
        fn compare(&self, a: &[u8], b: &[u8]) -> core::cmp::Ordering {
            b.cmp(a)
        }
        fn is_lexicographic(&self) -> bool {
            false
        }
    }

    let map = SkipMap::new(alloc::sync::Arc::new(ReverseComparator));
    assert!(
        !map.is_lexicographic,
        "a custom comparator must take the trait-dispatch path"
    );

    map.insert(&make_key(b"aaa", 1), make_value(b"a"));
    map.insert(&make_key(b"ccc", 1), make_value(b"c"));
    map.insert(&make_key(b"bbb", 1), make_value(b"b"));

    // Forward iteration follows the custom (reverse) order.
    let fwd: Vec<Vec<u8>> = map.iter().map(|e| e.key().user_key.to_vec()).collect();
    assert_eq!(
        fwd,
        vec![b"ccc".to_vec(), b"bbb".to_vec(), b"aaa".to_vec()],
        "custom comparator orders entries in reverse byte order"
    );

    // Reverse iteration (drives compare_nodes via find_predecessor) mirrors it.
    let rev: Vec<Vec<u8>> = map
        .iter()
        .rev()
        .map(|e| e.key().user_key.to_vec())
        .collect();
    assert_eq!(
        rev,
        vec![b"aaa".to_vec(), b"bbb".to_vec(), b"ccc".to_vec()],
        "reverse iteration mirrors the custom order"
    );
}

#[test]
fn ordering_user_key_asc_seqno_desc() {
    let map = new_map();

    // Same user_key, different seqnos → should iterate highest seqno first.
    map.insert(&make_key(b"abc", 1), make_value(b"v1"));
    map.insert(&make_key(b"abc", 3), make_value(b"v3"));
    map.insert(&make_key(b"abc", 2), make_value(b"v2"));

    let seqnos: Vec<SeqNo> = map.iter().map(|e| e.key().seqno).collect();
    assert_eq!(seqnos, vec![3, 2, 1]);

    // Different user_keys → ascending.
    map.insert(&make_key(b"zzz", 10), make_value(b"z"));
    map.insert(&make_key(b"aaa", 10), make_value(b"a"));

    let keys: Vec<Vec<u8>> = map.iter().map(|e| e.key().user_key.to_vec()).collect();
    assert_eq!(
        keys,
        vec![
            b"aaa".to_vec(),
            b"abc".to_vec(),
            b"abc".to_vec(),
            b"abc".to_vec(),
            b"zzz".to_vec(),
        ]
    );
}

#[test]
fn range_lower_bound() {
    let map = new_map();
    for i in 0u8..10 {
        let key = vec![b'a' + i];
        map.insert(&make_key(&key, 0), make_value(&[i]));
    }

    // Range from 'e' onwards → e, f, g, h, i, j
    let bound = make_key(b"e", crate::MAX_SEQNO);
    let keys: Vec<u8> = map.range(bound..).map(|e| e.key().user_key[0]).collect();
    assert_eq!(keys, vec![b'e', b'f', b'g', b'h', b'i', b'j']);
}

#[test]
fn range_bounded() {
    let map = new_map();
    for i in 0u8..10 {
        let key = vec![b'a' + i];
        map.insert(&make_key(&key, 0), make_value(&[i]));
    }

    let lo = make_key(b"c", crate::MAX_SEQNO);
    let hi = make_key(b"f", 0);
    let keys: Vec<u8> = map.range(lo..=hi).map(|e| e.key().user_key[0]).collect();
    assert_eq!(keys, vec![b'c', b'd', b'e', b'f']);
}

#[test]
fn double_ended_iter() {
    let map = new_map();
    for i in 0u8..5 {
        let key = vec![b'a' + i];
        map.insert(&make_key(&key, 0), make_value(&[i]));
    }

    let mut iter = map.iter();
    assert_eq!(iter.next().unwrap().key().user_key[0], b'a');
    assert_eq!(iter.next_back().unwrap().key().user_key[0], b'e');
    assert_eq!(iter.next().unwrap().key().user_key[0], b'b');
    assert_eq!(iter.next_back().unwrap().key().user_key[0], b'd');
    assert_eq!(iter.next().unwrap().key().user_key[0], b'c');
    assert!(iter.next().is_none());
    assert!(iter.next_back().is_none());
}

#[test]
fn double_ended_range() {
    let map = new_map();
    for i in 0u8..10 {
        let key = vec![b'a' + i];
        map.insert(&make_key(&key, 0), make_value(&[i]));
    }

    let lo = make_key(b"c", crate::MAX_SEQNO);
    let hi = make_key(b"g", 0);
    let rev: Vec<u8> = map
        .range(lo..=hi)
        .rev()
        .map(|e| e.key().user_key[0])
        .collect();
    assert_eq!(rev, vec![b'g', b'f', b'e', b'd', b'c']);
}

/// Backward steps over a list far larger than one back batch hand out every
/// node once, in reverse order, for the whole list and for a bounded range.
#[test]
fn backward_steps_over_many_batches_reverse_the_forward_order() {
    let map = new_map();
    for i in 0u32..5_000 {
        map.insert(
            &make_key(&i.to_be_bytes(), u64::from(i % 3)),
            make_value(b"v"),
        );
    }

    let forward: Vec<_> = map.iter().map(|e| e.key()).collect();
    let mut backward: Vec<_> = map.iter().rev().map(|e| e.key()).collect();
    backward.reverse();
    assert_eq!(backward, forward);

    let lo = make_key(&700u32.to_be_bytes(), crate::MAX_SEQNO);
    let hi = make_key(&4_321u32.to_be_bytes(), 0);
    let forward: Vec<_> = map
        .range(lo.clone()..=hi.clone())
        .map(|e| e.key())
        .collect();
    let mut backward: Vec<_> = map.range(lo..=hi).rev().map(|e| e.key()).collect();
    backward.reverse();
    assert_eq!(backward, forward);
    assert_eq!(forward.len(), 4_321 - 700 + 1);
}

/// Both ends taken in turn meet once over a long list: nothing repeated,
/// nothing skipped, wherever the meeting point falls in a back batch.
#[test]
fn alternating_ends_over_many_batches_meet_once() {
    let map = new_map();
    for i in 0u32..3_001 {
        map.insert(&make_key(&i.to_be_bytes(), 0), make_value(b"v"));
    }
    for front_steps in [0usize, 1, 7, 100, 1_499, 1_500, 2_999] {
        let mut iter = map.iter();
        let mut seen = Vec::new();
        for _ in 0..front_steps {
            seen.push(iter.next().unwrap().key());
        }
        while let Some(e) = iter.next_back() {
            seen.push(e.key());
            let Some(e) = iter.next() else { break };
            seen.push(e.key());
        }
        seen.sort();
        let all: Vec<_> = map.iter().map(|e| e.key()).collect();
        assert_eq!(seen, all, "front_steps={front_steps}");
    }
}

/// A node linked between nodes the back end has already batched, while the
/// front is still before them, must not make either end hand a node out
/// twice: the ends meet by key order, not by meeting the same node.
#[test]
fn an_insert_inside_a_back_batch_never_repeats_a_key() {
    for (range, alternate) in [(false, true), (true, true), (false, false), (true, false)] {
        let map = new_map();
        for k in [b"a", b"c", b"e", b"g"] {
            map.insert(&make_key(k, 0), make_value(b"v"));
        }
        let lo = make_key(b"a", crate::MAX_SEQNO);
        let hi = make_key(b"z", 0);
        let mut iter: Box<dyn DoubleEndedIterator<Item = Entry<'_>>> = if range {
            Box::new(map.range(lo..=hi))
        } else {
            Box::new(map.iter())
        };

        let mut seen = vec![iter.next().unwrap().key()];
        seen.push(iter.next_back().unwrap().key());
        map.insert(&make_key(b"d", 0), make_value(b"v"));
        // Front takes "c", back takes "e" and pops "c" as its next node,
        // which the front has already handed out.
        seen.push(iter.next().unwrap().key());
        seen.push(iter.next_back().unwrap().key());
        if alternate {
            while let Some(e) = iter.next() {
                seen.push(e.key());
                let Some(e) = iter.next_back() else { break };
                seen.push(e.key());
            }
        } else {
            // The front alone first: it reaches the inserted "d", then "e",
            // which the back has already handed out.
            seen.extend(iter.by_ref().map(|e| e.key()));
            seen.extend(iter.rev().map(|e| e.key()));
        }

        let mut keys: Vec<_> = seen.iter().map(|k| k.user_key.to_vec()).collect();
        keys.sort();
        let len = keys.len();
        keys.dedup();
        assert_eq!(keys.len(), len, "a key came out twice (range={range})");
        for k in [b"a", b"c", b"e", b"g"] {
            assert!(keys.contains(&k.to_vec()), "{k:?} missing (range={range})");
        }
    }
}

#[test]
fn empty_value() {
    let map = new_map();
    map.insert(&make_key(b"k", 0), make_value(b""));
    let entry = map.iter().next().unwrap();
    assert!(entry.value().is_empty());
}

#[test]
fn concurrent_inserts() {
    use std::sync::Arc;

    let map = Arc::new(new_map());
    let n_threads = 8;
    let n_per_thread = 1000;

    let handles: Vec<_> = (0..n_threads)
        .map(|t| {
            let map = Arc::clone(&map);
            std::thread::spawn(move || {
                for i in 0..n_per_thread {
                    let key = format!("t{t:02}_k{i:05}");
                    map.insert(&make_key(key.as_bytes(), i as u64), make_value(b"v"));
                }
            })
        })
        .collect();

    for h in handles {
        h.join().expect("thread panicked");
    }

    assert_eq!(map.len(), n_threads * n_per_thread);

    // Verify sorted order.
    let entries: Vec<_> = map.iter().collect();
    for pair in entries.windows(2) {
        let a = pair[0].key();
        let b = pair[1].key();
        assert!(a <= b, "out of order: {a:?} > {b:?}");
    }
}

#[test]
fn mvcc_point_lookup_via_range() {
    let map = new_map();

    // Insert 3 versions of "key" at seqnos 1, 2, 3.
    map.insert(&make_key(b"key", 1), make_value(b"v1"));
    map.insert(&make_key(b"key", 2), make_value(b"v2"));
    map.insert(&make_key(b"key", 3), make_value(b"v3"));

    // Memtable MVCC read at read_seqno=3 (visible: seqno <= 2).
    // The memtable uses lower_bound = InternalKey("key", read_seqno - 1).
    // With InternalKey ordering (user_key ASC, seqno DESC), range(("key", 2)..)
    // yields entries starting from seqno=2 downward.
    let lower = InternalKey::new(b"key".to_vec(), 2, ValueType::Value);
    let mut iter = map.range(lower..);
    let entry = iter
        .next()
        .filter(|e| &*e.key().user_key == b"key")
        .expect("should find key");
    assert_eq!(entry.key().seqno, 2);
    assert_eq!(&*entry.value(), b"v2");

    // At read_seqno=2, lower_bound = ("key", 1), yields seqno=1.
    let lower2 = InternalKey::new(b"key".to_vec(), 1, ValueType::Value);
    let entry2 = map
        .range(lower2..)
        .next()
        .filter(|e| &*e.key().user_key == b"key")
        .expect("should find key");
    assert_eq!(entry2.key().seqno, 1);
    assert_eq!(&*entry2.value(), b"v1");

    // At read_seqno=crate::MAX_SEQNO, lower_bound = ("key", MAX-1), yields seqno=3 (latest).
    let lower3 = InternalKey::new(b"key".to_vec(), crate::MAX_SEQNO - 1, ValueType::Value);
    let entry3 = map
        .range(lower3..)
        .next()
        .filter(|e| &*e.key().user_key == b"key")
        .expect("should find key");
    assert_eq!(entry3.key().seqno, 3);
    assert_eq!(&*entry3.value(), b"v3");
}

#[test]
fn empty_iter_next_back() {
    let map = new_map();
    let mut iter = map.iter();
    assert!(iter.next().is_none());
    assert!(iter.next_back().is_none());
}

#[test]
fn empty_range_next_back() {
    let map = new_map();
    let lo = make_key(b"a", crate::MAX_SEQNO);
    let hi = make_key(b"z", 0);
    let mut range = map.range(lo..=hi);
    assert!(range.next().is_none());
    assert!(range.next_back().is_none());
}

#[test]
fn range_excluded_end_next_back() {
    let map = new_map();
    for i in 0u8..5 {
        map.insert(&make_key(&[b'a' + i], 0), make_value(&[i]));
    }

    // Excluded end "d" → range is [a, d) = a, b, c
    let lo = make_key(b"a", crate::MAX_SEQNO);
    let hi = make_key(b"d", crate::MAX_SEQNO);
    let rev: Vec<u8> = map
        .range(lo..hi)
        .rev()
        .map(|e| e.key().user_key[0])
        .collect();
    assert_eq!(rev, vec![b'c', b'b', b'a']);
}

#[test]
fn seek_le_all_greater_returns_none() {
    let map = new_map();
    map.insert(&make_key(b"m", 0), make_value(b"v"));

    // All keys > "a", so seek_le("a") returns UNSET → next_back = None
    let hi = make_key(b"a", 0);
    let mut range = map.range(..=hi);
    assert!(range.next_back().is_none());
}

#[test]
fn next_back_on_first_element() {
    let map = new_map();
    map.insert(&make_key(b"only", 0), make_value(b"v"));

    let mut iter = map.iter();
    // next_back on single-element list
    let entry = iter.next_back().expect("one entry");
    assert_eq!(&*entry.key().user_key, b"only");
    assert!(iter.next().is_none());
    assert!(iter.next_back().is_none());
}

/// Regression test for SIGBUS on aarch64: concurrent inserts + reads
/// caused misaligned AtomicU32 access when a skiplist next-pointer
/// contained a key-data offset (align=1) instead of a node offset
/// (align=4).
///
/// The test stresses concurrent insert + iteration to surface the
/// race.  Prior to the fix, this would SIGBUS on Apple Silicon.
#[test]
fn concurrent_insert_and_iter_no_sigbus() {
    use std::sync::{Arc, Barrier};

    let map = Arc::new(new_map());
    let barrier = Arc::new(Barrier::new(9)); // 8 writers + 1 reader

    // 8 writer threads
    let writers: Vec<_> = (0..8)
        .map(|t| {
            let map = Arc::clone(&map);
            let barrier = Arc::clone(&barrier);
            std::thread::spawn(move || {
                barrier.wait();
                for i in 0..500 {
                    let key = format!("t{t:02}_k{i:04}");
                    map.insert(&make_key(key.as_bytes(), i as u64), make_value(b"val"));
                }
            })
        })
        .collect();

    // 1 reader thread doing concurrent iteration
    let reader = {
        let map = Arc::clone(&map);
        let barrier = Arc::clone(&barrier);
        std::thread::spawn(move || {
            barrier.wait();
            let mut count = 0u64;
            for _ in 0..100 {
                // Iterate the skiplist while writers are active.
                // This exercises tower_atomic / next_at on every node.
                for entry in map.iter() {
                    let _ = entry.key();
                    let _ = entry.value();
                    count += 1;
                }
            }
            count
        })
    };

    for w in writers {
        w.join().expect("writer panicked");
    }
    let reads = reader.join().expect("reader panicked");

    // Sanity: all entries inserted
    assert_eq!(map.len(), 4000);
    // Reader count may be 0 if writers finished before reader iterated.
    // The key assertion is that no SIGBUS/panic occurred during iteration.
    let _ = reads;
}

/// A reconstructed key is a zero-copy view into the arena block; the view's
/// reference must keep the block's bytes alive even after the whole map (and
/// its arena) is dropped. Uses a key longer than the inline threshold so the
/// view actually shares the block, and a second map to give a use-after-free
/// a chance to scribble over recycled memory before the assertion.
#[test]
fn key_view_after_map_drop_remains_valid() {
    let long_key = vec![0xA5u8; 64];
    let map = new_map();
    map.insert(&make_key(&long_key, 1), make_value(b"payload"));

    #[expect(
        clippy::expect_used,
        reason = "test inserts one entry before reading it"
    )]
    let key = map.iter().next().expect("one entry").key();
    drop(map);

    // Recycle allocator pages with a fresh map full of different bytes.
    let other = new_map();
    for i in 0..100u64 {
        other.insert(&make_key(&[0x5Au8; 64], i), make_value(&[0xFFu8; 128]));
    }

    assert_eq!(&*key.user_key, long_key.as_slice());
}

/// The sequential-insert fast path (tail hint) must never change WHERE a
/// node lands: appends take it, everything else (a key below the tail, a
/// newer version of the tail's own key, which sorts BEFORE the older one)
/// must fall back to the full splice search. Interleaves all three and
/// checks the final order is the exact internal-key order with nothing
/// lost.
#[test]
fn tail_hint_keeps_internal_key_order_under_mixed_inserts() {
    let map = new_map();
    let mut expected: Vec<(Vec<u8>, SeqNo)> = Vec::new();

    for i in 0..2_000u64 {
        // Append: strictly increasing key (the hint path for height-1 nodes).
        let key = format!("k{i:06}").into_bytes();
        map.insert(&make_key(&key, 1), make_value(b"v"));
        expected.push((key.clone(), 1));

        if i % 7 == 3 {
            // Newer version of the tail key: same user key, higher seqno.
            // Sorts BEFORE the version just appended, so it is NOT a tail append.
            map.insert(&make_key(&key, 2), make_value(b"v2"));
            expected.push((key.clone(), 2));
        }
        if i % 11 == 5 {
            // A key below the current tail (out of order): full search path.
            let below = format!("k{:06}", i / 2).into_bytes();
            map.insert(&make_key(&below, 3), make_value(b"v3"));
            expected.push((below, 3));
        }
    }

    // Internal-key order: user key ascending, seqno DESCENDING within a key.
    expected.sort_by(|a, b| a.0.cmp(&b.0).then_with(|| b.1.cmp(&a.1)));

    let got: Vec<(Vec<u8>, SeqNo)> = map
        .iter()
        .map(|e| {
            let k = e.key();
            (k.user_key.to_vec(), k.seqno)
        })
        .collect();
    assert_eq!(got.len(), expected.len(), "an insert was lost");
    assert_eq!(
        got, expected,
        "iteration order must be the internal-key order"
    );
    assert_eq!(map.len(), expected.len());
}

/// Arms the seek hook to insert `(user_key, seqno)` into the window between a
/// seek's level-0 comparison and its return, the way a concurrent writer can.
fn insert_during_next_seek(map: &SkipMap, user_key: &'static [u8], seqno: SeqNo) {
    map.set_seek_hook(alloc::boxed::Box::new(move |map: &SkipMap| {
        map.insert(&make_key(user_key, seqno), make_value(b"concurrent"));
    }));
}

/// An insert whose predecessor gains another node between the splice search
/// and the link re-searches and retries, and the retry is counted; both keys
/// end up in order. Guards against a counter that misses the contended path,
/// which would report a skewed writer load as free of contention.
#[cfg(feature = "metrics")]
#[test]
fn an_insert_raced_at_its_splice_counts_a_cas_retry() {
    let map = new_map();
    map.insert(&make_key(b"a", 1), make_value(b"a"));
    map.insert(&make_key(b"c", 1), make_value(b"c"));
    assert_eq!(map.cas_retries(), 0, "no contention yet");
    map.set_link_hook(alloc::boxed::Box::new(|map: &SkipMap| {
        map.insert(&make_key(b"b", 1), make_value(b"concurrent"));
    }));
    map.insert(&make_key(b"bb", 1), make_value(b"bb"));
    assert!(map.cas_retries() >= 1, "the raced link was retried");
    let keys: Vec<Vec<u8>> = map.iter().map(|e| e.key().user_key.to_vec()).collect();
    assert_eq!(keys, [&b"a"[..], b"b", b"bb", b"c"]);
}

/// Inserts no other writer races leave the retry count at zero, whatever the
/// order: the sequential tail append and the full splice search alike.
#[cfg(feature = "metrics")]
#[test]
fn uncontended_inserts_count_no_cas_retry() {
    let map = new_map();
    for i in 0..64u8 {
        map.insert(&make_key(&[i], 1), make_value(b"v"));
    }
    for i in (0..64u8).rev() {
        map.insert(&make_key(&[i], 2), make_value(b"v"));
    }
    assert_eq!(map.cas_retries(), 0);
}

/// `(user key, seqno)` of every entry a range yields.
fn yielded(range: Range<'_>) -> Vec<(Vec<u8>, SeqNo)> {
    range
        .map(|e| {
            let k = e.key();
            (k.user_key.to_vec(), k.seqno)
        })
        .collect()
}

/// A key linked below the lower bound after the seek compared its successor
/// must not be yielded: the seek is the range's only lower-bound check.
#[test]
fn a_range_never_yields_a_key_inserted_below_its_bound_during_the_seek() {
    let map = new_map();
    map.insert(&make_key(b"a", 1), make_value(b"a"));
    map.insert(&make_key(b"d", 1), make_value(b"d"));
    insert_during_next_seek(&map, b"b", 1);
    let got = yielded(map.range(make_key(b"c", SeqNo::MAX)..));
    assert_eq!(got, vec![(b"d".to_vec(), 1)]);
}

/// The same when the compared successor was the end of the list: a key
/// appended below the bound afterwards is not yielded either.
#[test]
fn a_range_past_the_last_key_never_yields_a_key_appended_below_its_bound() {
    let map = new_map();
    map.insert(&make_key(b"a", 1), make_value(b"a"));
    insert_during_next_seek(&map, b"b", 1);
    let got = yielded(map.range(make_key(b"c", SeqNo::MAX)..));
    assert!(got.is_empty(), "yielded {got:?}");
}

/// An exclusive lower bound excludes a key equal to it, even one linked
/// after the seek compared the successor.
#[test]
fn an_exclusive_bound_never_yields_an_equal_key_inserted_during_the_seek() {
    let map = new_map();
    map.insert(&make_key(b"a", 1), make_value(b"a"));
    map.insert(&make_key(b"d", 1), make_value(b"d"));
    insert_during_next_seek(&map, b"c", 1);
    let got = yielded(map.range((Bound::Excluded(make_key(b"c", 1)), Bound::Unbounded)));
    assert_eq!(got, vec![(b"d".to_vec(), 1)]);
}

/// Another key linked before the target during the seek must not hide the
/// target: `point_get` finds the version it compared.
#[test]
fn point_get_finds_the_key_when_another_is_inserted_before_it_during_the_seek() {
    let map = new_map();
    map.insert(&make_key(b"a", 1), make_value(b"a"));
    map.insert(&make_key(b"k", 3), make_value(b"k3"));
    insert_during_next_seek(&map, b"b", 1);
    let got = map.point_get(b"k", 5).map(|v| v.key.seqno);
    assert_eq!(got, Some(3));
}

/// A newer version of the key sorts before the visible one; linked during the
/// seek it must stay invisible to a read below its seqno, including one right
/// at `max_seqno + 1`.
#[test]
fn point_get_never_returns_a_version_above_its_snapshot_inserted_during_the_seek() {
    let got: Vec<(SeqNo, Option<SeqNo>)> = [9, 6]
        .into_iter()
        .map(|newer| {
            let map = new_map();
            map.insert(&make_key(b"a", 1), make_value(b"a"));
            map.insert(&make_key(b"k", 3), make_value(b"k3"));
            insert_during_next_seek(&map, b"k", newer);
            (newer, map.point_get(b"k", 5).map(|v| v.key.seqno))
        })
        .collect();
    assert_eq!(got, vec![(9, Some(3)), (6, Some(3))]);
}