lance_table/

rowids.rs

// SPDX-License-Identifier: Apache-2.0
// SPDX-FileCopyrightText: Copyright The Lance Authors
//! Indices for mapping row ids to their corresponding addresses.
//!
//! Each fragment in a table has a [RowIdSequence] that contains the row ids
//! in the order they appear in the fragment. The [RowIdIndex] aggregates these
//! sequences and maps row ids to their corresponding addresses across the
//! whole dataset.
//!
//! [RowIdSequence]s are serialized individually and stored in the fragment
//! metadata. Use [read_row_ids] and [write_row_ids] to read and write these
//! sequences. The on-disk format is designed to align well with the in-memory
//! representation, to avoid unnecessary deserialization.
use std::ops::Range;
// TODO: replace this with Arrow BooleanBuffer.

// These are all internal data structures, and are private.
mod bitmap;
mod encoded_array;
mod index;
pub mod segment;
mod serde;
pub mod version;

use deepsize::DeepSizeOf;
// These are the public API.
pub use index::FragmentRowIdIndex;
pub use index::RowIdIndex;
use lance_core::{
    Error, Result,
    utils::mask::{RowAddrMask, RowAddrTreeMap},
};
use lance_io::ReadBatchParams;
pub use serde::{read_row_ids, write_row_ids};

use crate::utils::LanceIteratorExtension;
use lance_core::utils::mask::RowSetOps;
use segment::U64Segment;
use tracing::instrument;

/// A sequence of row ids.
///
/// Row ids are u64s that:
///
/// 1. Are **unique** within a table (except for tombstones)
/// 2. Are *often* but not always sorted and/or contiguous.
///
/// This sequence of row ids is optimized to be compact when the row ids are
/// contiguous and sorted. However, it does not require that the row ids are
/// contiguous or sorted.
///
/// We can make optimizations that assume uniqueness.
#[derive(Debug, Clone, DeepSizeOf, PartialEq, Eq, Default)]
pub struct RowIdSequence(Vec<U64Segment>);

impl std::fmt::Display for RowIdSequence {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        let mut iter = self.iter();
        let mut first_10 = Vec::new();
        let mut last_10 = Vec::new();
        for row_id in iter.by_ref() {
            first_10.push(row_id);
            if first_10.len() > 10 {
                break;
            }
        }

        while let Some(row_id) = iter.next_back() {
            last_10.push(row_id);
            if last_10.len() > 10 {
                break;
            }
        }
        last_10.reverse();

        let theres_more = iter.next().is_some();

        write!(f, "[")?;
        for row_id in first_10 {
            write!(f, "{}", row_id)?;
        }
        if theres_more {
            write!(f, ", ...")?;
        }
        for row_id in last_10 {
            write!(f, ", {}", row_id)?;
        }
        write!(f, "]")
    }
}

impl From<Range<u64>> for RowIdSequence {
    fn from(range: Range<u64>) -> Self {
        Self(vec![U64Segment::Range(range)])
    }
}

impl From<&[u64]> for RowIdSequence {
    fn from(row_ids: &[u64]) -> Self {
        Self(vec![U64Segment::from_slice(row_ids)])
    }
}

impl RowIdSequence {
    pub fn new() -> Self {
        Self::default()
    }

    pub fn iter(&self) -> impl DoubleEndedIterator<Item = u64> + '_ {
        self.0.iter().flat_map(|segment| segment.iter())
    }

    pub fn len(&self) -> u64 {
        self.0.iter().map(|segment| segment.len() as u64).sum()
    }

    pub fn is_empty(&self) -> bool {
        self.0.is_empty()
    }

    /// Combines this row id sequence with another row id sequence.
    pub fn extend(&mut self, other: Self) {
        // If the last element of this sequence and the first element of next
        // sequence are ranges, we might be able to combine them into a single
        // range.
        if let (Some(U64Segment::Range(range1)), Some(U64Segment::Range(range2))) =
            (self.0.last(), other.0.first())
            && range1.end == range2.start
        {
            let new_range = U64Segment::Range(range1.start..range2.end);
            self.0.pop();
            self.0.push(new_range);
            self.0.extend(other.0.into_iter().skip(1));
            return;
        }
        // TODO: add other optimizations, such as combining two RangeWithHoles.
        self.0.extend(other.0);
    }

    /// Remove a set of row ids from the sequence.
    pub fn delete(&mut self, row_ids: impl IntoIterator<Item = u64>) {
        // Order the row ids by position in which they appear in the sequence.
        let (row_ids, offsets) = self.find_ids(row_ids);

        let capacity = self.0.capacity();
        let old_segments = std::mem::replace(&mut self.0, Vec::with_capacity(capacity));
        let mut remaining_segments = old_segments.as_slice();

        for (segment_idx, range) in offsets {
            let segments_handled = old_segments.len() - remaining_segments.len();
            let segments_to_add = segment_idx - segments_handled;
            self.0
                .extend_from_slice(&remaining_segments[..segments_to_add]);
            remaining_segments = &remaining_segments[segments_to_add..];

            let segment;
            (segment, remaining_segments) = remaining_segments.split_first().unwrap();

            let segment_ids = &row_ids[range];
            self.0.push(segment.delete(segment_ids));
        }

        // Add the remaining segments.
        self.0.extend_from_slice(remaining_segments);
    }

    /// Delete row ids by position.
    pub fn mask(&mut self, positions: impl IntoIterator<Item = u32>) -> Result<()> {
        let mut local_positions = Vec::new();
        let mut positions_iter = positions.into_iter();
        let mut curr_position = positions_iter.next();
        let mut offset = 0;
        let mut cutoff = 0;

        for segment in &mut self.0 {
            // Make vector of local positions
            cutoff += segment.len() as u32;
            while let Some(position) = curr_position {
                if position >= cutoff {
                    break;
                }
                local_positions.push(position - offset);
                curr_position = positions_iter.next();
            }

            if !local_positions.is_empty() {
                segment.mask(&local_positions);
                local_positions.clear();
            }
            offset = cutoff;
        }

        self.0.retain(|segment| !segment.is_empty());

        Ok(())
    }

    /// Find the row ids in the sequence.
    ///
    /// Returns the row ids sorted by their appearance in the sequence.
    /// Also returns the segment index and the range where that segment's
    /// row id matches are found in the returned row id vector.
    fn find_ids(
        &self,
        row_ids: impl IntoIterator<Item = u64>,
    ) -> (Vec<u64>, Vec<(usize, Range<usize>)>) {
        // Often, the row ids will already be provided in the order they appear.
        // So the optimal way to search will be to cycle through rather than
        // restarting the search from the beginning each time.
        let mut segment_iter = self.0.iter().enumerate().cycle();

        let mut segment_matches = vec![Vec::new(); self.0.len()];

        row_ids.into_iter().for_each(|row_id| {
            let mut i = 0;
            // If we've cycled through all segments, we know the row id is not in the sequence.
            while i < self.0.len() {
                let (segment_idx, segment) = segment_iter.next().unwrap();
                if segment.range().is_some_and(|range| range.contains(&row_id))
                    && let Some(offset) = segment.position(row_id)
                {
                    segment_matches.get_mut(segment_idx).unwrap().push(offset);
                    // The row id was not found it the segment. It might be in a later segment.
                }
                i += 1;
            }
        });
        for matches in &mut segment_matches {
            matches.sort_unstable();
        }

        let mut offset = 0;
        let segment_ranges = segment_matches
            .iter()
            .enumerate()
            .filter(|(_, matches)| !matches.is_empty())
            .map(|(segment_idx, matches)| {
                let range = offset..offset + matches.len();
                offset += matches.len();
                (segment_idx, range)
            })
            .collect();
        let row_ids = segment_matches
            .into_iter()
            .enumerate()
            .flat_map(|(segment_idx, offset)| {
                offset
                    .into_iter()
                    .map(move |offset| self.0[segment_idx].get(offset).unwrap())
            })
            .collect();

        (row_ids, segment_ranges)
    }

    pub fn slice(&self, offset: usize, len: usize) -> RowIdSeqSlice<'_> {
        if len == 0 {
            return RowIdSeqSlice {
                segments: &[],
                offset_start: 0,
                offset_last: 0,
            };
        }

        // Find the starting position
        let mut offset_start = offset;
        let mut segment_offset = 0;
        for segment in &self.0 {
            let segment_len = segment.len();
            if offset_start < segment_len {
                break;
            }
            offset_start -= segment_len;
            segment_offset += 1;
        }

        // Find the ending position
        let mut offset_last = offset_start + len;
        let mut segment_offset_last = segment_offset;
        for segment in &self.0[segment_offset..] {
            let segment_len = segment.len();
            if offset_last <= segment_len {
                break;
            }
            offset_last -= segment_len;
            segment_offset_last += 1;
        }

        RowIdSeqSlice {
            segments: &self.0[segment_offset..=segment_offset_last],
            offset_start,
            offset_last,
        }
    }

    /// Get the row id at the given index.
    ///
    /// If the index is out of bounds, this will return None.
    pub fn get(&self, index: usize) -> Option<u64> {
        let mut offset = 0;
        for segment in &self.0 {
            let segment_len = segment.len();
            if index < offset + segment_len {
                return segment.get(index - offset);
            }
            offset += segment_len;
        }
        None
    }

    /// Get row ids from the sequence based on the provided _sorted_ offsets
    ///
    /// Any out of bounds offsets will be ignored
    ///
    /// # Panics
    ///
    /// If the input selection is not sorted, this function will panic
    pub fn select<'a>(
        &'a self,
        selection: impl Iterator<Item = usize> + 'a,
    ) -> impl Iterator<Item = u64> + 'a {
        let mut seg_iter = self.0.iter();
        let mut cur_seg = seg_iter.next();
        let mut rows_passed = 0;
        let mut cur_seg_len = cur_seg.map(|seg| seg.len()).unwrap_or(0);
        let mut last_index = 0;
        selection.filter_map(move |index| {
            if index < last_index {
                panic!("Selection is not sorted");
            }
            last_index = index;

            cur_seg?;

            while (index - rows_passed) >= cur_seg_len {
                rows_passed += cur_seg_len;
                cur_seg = seg_iter.next();
                if let Some(cur_seg) = cur_seg {
                    cur_seg_len = cur_seg.len();
                } else {
                    return None;
                }
            }

            Some(cur_seg.unwrap().get(index - rows_passed).unwrap())
        })
    }

    /// Given a mask of row ids, calculate the offset ranges of the row ids that are present
    /// in the sequence.
    ///
    /// For example, given a mask that selects all even ids and a sequence that is
    /// [80..85, 86..90, 14]
    ///
    /// this will return [0, 2, 4, 5, 7, 9]
    /// because the range expands to
    ///
    /// [80, 81, 82, 83, 84, 86, 87, 88, 89, 14] with offsets
    /// [ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9]
    ///
    /// This function is useful when determining which row offsets to read from a fragment given
    /// a mask.
    #[instrument(level = "debug", skip_all)]
    pub fn mask_to_offset_ranges(&self, mask: &RowAddrMask) -> Vec<Range<u64>> {
        let mut offset = 0;
        let mut ranges = Vec::new();
        for segment in &self.0 {
            match segment {
                U64Segment::Range(range) => {
                    let mut ids = RowAddrTreeMap::from(range.clone());
                    ids.mask(mask);
                    ranges.extend(GroupingIterator::new(
                        unsafe { ids.into_addr_iter() }.map(|addr| addr - range.start + offset),
                    ));
                    offset += range.end - range.start;
                }
                U64Segment::RangeWithHoles { range, holes } => {
                    let offset_start = offset;
                    let mut ids = RowAddrTreeMap::from(range.clone());
                    offset += range.end - range.start;
                    for hole in holes.iter() {
                        if ids.remove(hole) {
                            offset -= 1;
                        }
                    }
                    ids.mask(mask);

                    // Sadly we can't just subtract the offset because of the holes
                    let mut sorted_holes = holes.clone().into_iter().collect::<Vec<_>>();
                    sorted_holes.sort_unstable();
                    let mut next_holes_iter = sorted_holes.into_iter().peekable();
                    let mut holes_passed = 0;
                    ranges.extend(GroupingIterator::new(unsafe { ids.into_addr_iter() }.map(
                        |addr| {
                            while let Some(next_hole) = next_holes_iter.peek() {
                                if *next_hole < addr {
                                    next_holes_iter.next();
                                    holes_passed += 1;
                                } else {
                                    break;
                                }
                            }
                            addr - range.start + offset_start - holes_passed
                        },
                    )));
                }
                U64Segment::RangeWithBitmap { range, bitmap } => {
                    let mut ids = RowAddrTreeMap::from(range.clone());
                    let offset_start = offset;
                    offset += range.end - range.start;
                    for (i, val) in range.clone().enumerate() {
                        if !bitmap.get(i) && ids.remove(val) {
                            offset -= 1;
                        }
                    }
                    ids.mask(mask);
                    let mut bitmap_iter = bitmap.iter();
                    let mut bitmap_iter_pos = 0;
                    let mut holes_passed = 0;
                    ranges.extend(GroupingIterator::new(unsafe { ids.into_addr_iter() }.map(
                        |addr| {
                            let offset_no_holes = addr - range.start + offset_start;
                            while bitmap_iter_pos < offset_no_holes {
                                if !bitmap_iter.next().unwrap() {
                                    holes_passed += 1;
                                }
                                bitmap_iter_pos += 1;
                            }
                            offset_no_holes - holes_passed
                        },
                    )));
                }
                U64Segment::SortedArray(array) | U64Segment::Array(array) => {
                    // TODO: Could probably optimize the sorted array case to be O(N) instead of O(N log N)
                    ranges.extend(GroupingIterator::new(array.iter().enumerate().filter_map(
                        |(off, id)| {
                            if mask.selected(id) {
                                Some(off as u64 + offset)
                            } else {
                                None
                            }
                        },
                    )));
                    offset += array.len() as u64;
                }
            }
        }
        ranges
    }
}

/// An iterator that groups row ids into ranges
///
/// For example, given an input iterator of [1, 2, 3, 5, 6, 7, 10, 11, 12]
/// this will return an iterator of [(1..4), (5..8), (10..13)]
struct GroupingIterator<I: Iterator<Item = u64>> {
    iter: I,
    cur_range: Option<Range<u64>>,
}

impl<I: Iterator<Item = u64>> GroupingIterator<I> {
    fn new(iter: I) -> Self {
        Self {
            iter,
            cur_range: None,
        }
    }
}

impl<I: Iterator<Item = u64>> Iterator for GroupingIterator<I> {
    type Item = Range<u64>;

    fn next(&mut self) -> Option<Self::Item> {
        for id in self.iter.by_ref() {
            if let Some(range) = self.cur_range.as_mut() {
                if range.end == id {
                    range.end = id + 1;
                } else {
                    let ret = Some(range.clone());
                    self.cur_range = Some(id..id + 1);
                    return ret;
                }
            } else {
                self.cur_range = Some(id..id + 1);
            }
        }
        self.cur_range.take()
    }
}

impl From<&RowIdSequence> for RowAddrTreeMap {
    fn from(row_ids: &RowIdSequence) -> Self {
        let mut tree_map = Self::new();
        for segment in &row_ids.0 {
            match segment {
                U64Segment::Range(range) => {
                    tree_map.insert_range(range.clone());
                }
                U64Segment::RangeWithBitmap { range, bitmap } => {
                    tree_map.insert_range(range.clone());
                    for (i, val) in range.clone().enumerate() {
                        if !bitmap.get(i) {
                            tree_map.remove(val);
                        }
                    }
                }
                U64Segment::RangeWithHoles { range, holes } => {
                    tree_map.insert_range(range.clone());
                    for hole in holes.iter() {
                        tree_map.remove(hole);
                    }
                }
                U64Segment::SortedArray(array) | U64Segment::Array(array) => {
                    for val in array.iter() {
                        tree_map.insert(val);
                    }
                }
            }
        }
        tree_map
    }
}

#[derive(Debug)]
pub struct RowIdSeqSlice<'a> {
    /// Current slice of the segments we cover
    segments: &'a [U64Segment],
    /// Offset into the first segment to start iterating from
    offset_start: usize,
    /// Offset into the last segment to stop iterating at
    offset_last: usize,
}

impl RowIdSeqSlice<'_> {
    pub fn iter(&self) -> impl Iterator<Item = u64> + '_ {
        let mut known_size = self.segments.iter().map(|segment| segment.len()).sum();
        known_size -= self.offset_start;
        known_size -= self.segments.last().map(|s| s.len()).unwrap_or_default() - self.offset_last;

        let end = self.segments.len().saturating_sub(1);
        self.segments
            .iter()
            .enumerate()
            .flat_map(move |(i, segment)| {
                match i {
                    0 if self.segments.len() == 1 => {
                        let len = self.offset_last - self.offset_start;
                        // TODO: Optimize this so we don't have to use skip
                        // (take is probably fine though.)
                        Box::new(segment.iter().skip(self.offset_start).take(len))
                            as Box<dyn Iterator<Item = u64>>
                    }
                    0 => Box::new(segment.iter().skip(self.offset_start)),
                    i if i == end => Box::new(segment.iter().take(self.offset_last)),
                    _ => Box::new(segment.iter()),
                }
            })
            .exact_size(known_size)
    }
}

/// Re-chunk a sequences of row ids into chunks of a given size.
///
/// The sequences may less than chunk sizes, because the sequences only
/// contains the row ids that we want to keep, they come from the updates records.
/// But the chunk sizes are based on the fragment physical rows(may contain inserted records).
/// So if the sequences are smaller than the chunk sizes, we need to
/// assign the incremental row ids in the further step. This behavior is controlled by the
/// `allow_incomplete` parameter.
///
/// # Errors
///
/// If `allow_incomplete` is false, will return an error if the sum of the chunk sizes
/// is not equal to the total number of row ids in the sequences.
pub fn rechunk_sequences(
    sequences: impl IntoIterator<Item = RowIdSequence>,
    chunk_sizes: impl IntoIterator<Item = u64>,
    allow_incomplete: bool,
) -> Result<Vec<RowIdSequence>> {
    // TODO: return an iterator. (with a good size hint?)
    let chunk_sizes_vec: Vec<u64> = chunk_sizes.into_iter().collect();
    let total_chunks = chunk_sizes_vec.len();
    let mut chunked_sequences = Vec::with_capacity(total_chunks);
    let mut segment_iter = sequences
        .into_iter()
        .flat_map(|sequence| sequence.0.into_iter())
        .peekable();

    let too_few_segments_error = |chunk_index: usize, expected_chunk_size: u64, remaining: u64| {
        Error::invalid_input(format!(
            "Got too few segments for chunk {}. Expected chunk size: {}, remaining needed: {}",
            chunk_index, expected_chunk_size, remaining
        ))
    };

    let too_many_segments_error = |processed_chunks: usize, total_chunk_sizes: usize| {
        Error::invalid_input(format!(
            "Got too many segments for the provided chunk lengths. Processed {} chunks out of {} expected",
            processed_chunks, total_chunk_sizes
        ))
    };

    let mut segment_offset = 0_u64;

    for (chunk_index, chunk_size) in chunk_sizes_vec.iter().enumerate() {
        let chunk_size = *chunk_size;
        let mut sequence = RowIdSequence(Vec::new());
        let mut remaining = chunk_size;

        while remaining > 0 {
            let remaining_in_segment = segment_iter
                .peek()
                .map_or(0, |segment| segment.len() as u64 - segment_offset);

            // Step 1: Handle segment remaining to be empty(also empty seg) - skip and continue
            if remaining_in_segment == 0 {
                if segment_iter.next().is_some() {
                    segment_offset = 0;
                    continue;
                } else {
                    // No more segments available
                    if allow_incomplete {
                        break;
                    } else {
                        return Err(too_few_segments_error(chunk_index, chunk_size, remaining));
                    }
                }
            }

            // Step 2: Handle still remaining segment based on size comparison
            match remaining_in_segment.cmp(&remaining) {
                std::cmp::Ordering::Greater => {
                    // Segment is larger than remaining space - slice it
                    let segment = segment_iter
                        .peek()
                        .ok_or_else(|| too_few_segments_error(chunk_index, chunk_size, remaining))?
                        .slice(segment_offset as usize, remaining as usize);
                    sequence.extend(RowIdSequence(vec![segment]));
                    segment_offset += remaining;
                    remaining = 0;
                }
                std::cmp::Ordering::Equal | std::cmp::Ordering::Less => {
                    // UNIFIED HANDLING: Both equal and less cases subtract from remaining
                    // Equal case: remaining -= remaining_in_segment (remaining becomes 0)
                    // Less case: remaining -= remaining_in_segment (remaining becomes positive)
                    let segment = segment_iter
                        .next()
                        .ok_or_else(|| too_few_segments_error(chunk_index, chunk_size, remaining))?
                        .slice(segment_offset as usize, remaining_in_segment as usize);
                    sequence.extend(RowIdSequence(vec![segment]));
                    segment_offset = 0;
                    remaining -= remaining_in_segment;
                }
            }
        }

        chunked_sequences.push(sequence);
    }

    if segment_iter.peek().is_some() {
        return Err(too_many_segments_error(
            chunked_sequences.len(),
            total_chunks,
        ));
    }

    Ok(chunked_sequences)
}

/// Selects the row ids from a sequence based on the provided offsets.
pub fn select_row_ids<'a>(
    sequence: &'a RowIdSequence,
    offsets: &'a ReadBatchParams,
) -> Result<Vec<u64>> {
    let out_of_bounds_err = |offset: u32| {
        Error::invalid_input(format!(
            "Index out of bounds: {} for sequence of length {}",
            offset,
            sequence.len()
        ))
    };

    match offsets {
        // TODO: Optimize this if indices are sorted, which is a common case.
        ReadBatchParams::Indices(indices) => indices
            .values()
            .iter()
            .map(|index| {
                sequence
                    .get(*index as usize)
                    .ok_or_else(|| out_of_bounds_err(*index))
            })
            .collect(),
        ReadBatchParams::Range(range) => {
            if range.end > sequence.len() as usize {
                return Err(out_of_bounds_err(range.end as u32));
            }
            let sequence = sequence.slice(range.start, range.end - range.start);
            Ok(sequence.iter().collect())
        }
        ReadBatchParams::Ranges(ranges) => {
            let num_rows = ranges
                .iter()
                .map(|r| (r.end - r.start) as usize)
                .sum::<usize>();
            let mut result = Vec::with_capacity(num_rows);
            for range in ranges.as_ref() {
                if range.end > sequence.len() {
                    return Err(out_of_bounds_err(range.end as u32));
                }
                let sequence =
                    sequence.slice(range.start as usize, (range.end - range.start) as usize);
                result.extend(sequence.iter());
            }
            Ok(result)
        }

        ReadBatchParams::RangeFull => Ok(sequence.iter().collect()),
        ReadBatchParams::RangeTo(to) => {
            if to.end > sequence.len() as usize {
                return Err(out_of_bounds_err(to.end as u32));
            }
            let len = to.end;
            let sequence = sequence.slice(0, len);
            Ok(sequence.iter().collect())
        }
        ReadBatchParams::RangeFrom(from) => {
            let sequence = sequence.slice(from.start, sequence.len() as usize - from.start);
            Ok(sequence.iter().collect())
        }
    }
}

#[cfg(test)]
mod test {
    use super::*;

    use pretty_assertions::assert_eq;
    use test::bitmap::Bitmap;

    #[test]
    fn test_row_id_sequence_from_range() {
        let sequence = RowIdSequence::from(0..10);
        assert_eq!(sequence.len(), 10);
        assert_eq!(sequence.is_empty(), false);

        let iter = sequence.iter();
        assert_eq!(iter.collect::<Vec<_>>(), (0..10).collect::<Vec<_>>());
    }

    #[test]
    fn test_row_id_sequence_extend() {
        let mut sequence = RowIdSequence::from(0..10);
        sequence.extend(RowIdSequence::from(10..20));
        assert_eq!(sequence.0, vec![U64Segment::Range(0..20)]);

        let mut sequence = RowIdSequence::from(0..10);
        sequence.extend(RowIdSequence::from(20..30));
        assert_eq!(
            sequence.0,
            vec![U64Segment::Range(0..10), U64Segment::Range(20..30)]
        );
    }

    #[test]
    fn test_row_id_sequence_delete() {
        let mut sequence = RowIdSequence::from(0..10);
        sequence.delete(vec![1, 3, 5, 7, 9]);
        let mut expected_bitmap = Bitmap::new_empty(9);
        for i in [0, 2, 4, 6, 8] {
            expected_bitmap.set(i as usize);
        }
        assert_eq!(
            sequence.0,
            vec![U64Segment::RangeWithBitmap {
                range: 0..9,
                bitmap: expected_bitmap
            },]
        );

        let mut sequence = RowIdSequence::from(0..10);
        sequence.extend(RowIdSequence::from(12..20));
        sequence.delete(vec![0, 9, 10, 11, 12, 13]);
        assert_eq!(
            sequence.0,
            vec![U64Segment::Range(1..9), U64Segment::Range(14..20),]
        );

        let mut sequence = RowIdSequence::from(0..10);
        sequence.delete(vec![0, 1, 2, 3, 4, 5, 6, 7, 8, 9]);
        assert_eq!(sequence.0, vec![U64Segment::Range(0..0)]);
    }

    #[test]
    fn test_row_id_slice() {
        // The type of sequence isn't that relevant to the implementation, so
        // we can just have a single one with all the segment types.
        let sequence = RowIdSequence(vec![
            U64Segment::Range(30..35), // 5
            U64Segment::RangeWithHoles {
                // 8
                range: 50..60,
                holes: vec![53, 54].into(),
            },
            U64Segment::SortedArray(vec![7, 9].into()), // 2
            U64Segment::RangeWithBitmap {
                range: 0..5,
                bitmap: [true, false, true, false, true].as_slice().into(),
            },
            U64Segment::Array(vec![35, 39].into()),
            U64Segment::Range(40..50),
        ]);

        // All possible offsets and lengths
        for offset in 0..sequence.len() as usize {
            for len in 0..sequence.len() as usize {
                if offset + len > sequence.len() as usize {
                    continue;
                }
                let slice = sequence.slice(offset, len);

                let actual = slice.iter().collect::<Vec<_>>();
                let expected = sequence.iter().skip(offset).take(len).collect::<Vec<_>>();
                assert_eq!(
                    actual, expected,
                    "Failed for offset {} and len {}",
                    offset, len
                );

                let (claimed_size, claimed_max) = slice.iter().size_hint();
                assert_eq!(claimed_max, Some(claimed_size)); // Exact size hint
                assert_eq!(claimed_size, actual.len()); // Correct size hint
            }
        }
    }

    #[test]
    fn test_row_id_slice_empty() {
        let sequence = RowIdSequence::from(0..10);
        let slice = sequence.slice(10, 0);
        assert_eq!(slice.iter().collect::<Vec<_>>(), Vec::<u64>::new());
    }

    #[test]
    fn test_row_id_sequence_rechunk() {
        fn assert_rechunked(
            input: Vec<RowIdSequence>,
            chunk_sizes: Vec<u64>,
            expected: Vec<RowIdSequence>,
        ) {
            let chunked = rechunk_sequences(input, chunk_sizes, false).unwrap();
            assert_eq!(chunked, expected);
        }

        // Small pieces to larger ones
        let many_segments = vec![
            RowIdSequence(vec![U64Segment::Range(0..5), U64Segment::Range(35..40)]),
            RowIdSequence::from(10..18),
            RowIdSequence::from(18..28),
            RowIdSequence::from(28..30),
        ];
        let fewer_segments = vec![
            RowIdSequence(vec![U64Segment::Range(0..5), U64Segment::Range(35..40)]),
            RowIdSequence::from(10..30),
        ];
        assert_rechunked(
            many_segments.clone(),
            fewer_segments.iter().map(|seq| seq.len()).collect(),
            fewer_segments.clone(),
        );

        // Large pieces to smaller ones
        assert_rechunked(
            fewer_segments,
            many_segments.iter().map(|seq| seq.len()).collect(),
            many_segments.clone(),
        );

        // Equal pieces
        assert_rechunked(
            many_segments.clone(),
            many_segments.iter().map(|seq| seq.len()).collect(),
            many_segments.clone(),
        );

        // Too few segments -> error
        let result = rechunk_sequences(many_segments.clone(), vec![100], false);
        assert!(result.is_err());

        // Too many segments -> error
        let result = rechunk_sequences(many_segments, vec![5], false);
        assert!(result.is_err());
    }

    #[test]
    fn test_select_row_ids() {
        // All forms of offsets
        let offsets = [
            ReadBatchParams::Indices(vec![1, 3, 9, 5, 7, 6].into()),
            ReadBatchParams::Range(2..8),
            ReadBatchParams::RangeFull,
            ReadBatchParams::RangeTo(..5),
            ReadBatchParams::RangeFrom(5..),
            ReadBatchParams::Ranges(vec![2..3, 5..10].into()),
        ];

        // Sequences with all segment types. These have at least 10 elements,
        // so they are valid for all the above offsets.
        let sequences = [
            RowIdSequence(vec![
                U64Segment::Range(0..5),
                U64Segment::RangeWithHoles {
                    range: 50..60,
                    holes: vec![53, 54].into(),
                },
                U64Segment::SortedArray(vec![7, 9].into()),
            ]),
            RowIdSequence(vec![
                U64Segment::RangeWithBitmap {
                    range: 0..5,
                    bitmap: [true, false, true, false, true].as_slice().into(),
                },
                U64Segment::Array(vec![30, 20, 10].into()),
                U64Segment::Range(40..50),
            ]),
        ];

        for params in offsets {
            for sequence in &sequences {
                let row_ids = select_row_ids(sequence, &params).unwrap();
                let flat_sequence = sequence.iter().collect::<Vec<_>>();

                // Transform params into bounded ones
                let selection: Vec<usize> = match &params {
                    ReadBatchParams::RangeFull => (0..flat_sequence.len()).collect(),
                    ReadBatchParams::RangeTo(to) => (0..to.end).collect(),
                    ReadBatchParams::RangeFrom(from) => (from.start..flat_sequence.len()).collect(),
                    ReadBatchParams::Range(range) => range.clone().collect(),
                    ReadBatchParams::Ranges(ranges) => ranges
                        .iter()
                        .flat_map(|r| r.start as usize..r.end as usize)
                        .collect(),
                    ReadBatchParams::Indices(indices) => {
                        indices.values().iter().map(|i| *i as usize).collect()
                    }
                };

                let expected = selection
                    .into_iter()
                    .map(|i| flat_sequence[i])
                    .collect::<Vec<_>>();
                assert_eq!(
                    row_ids, expected,
                    "Failed for params {:?} on the sequence {:?}",
                    &params, sequence
                );
            }
        }
    }

    #[test]
    fn test_select_row_ids_out_of_bounds() {
        let offsets = [
            ReadBatchParams::Indices(vec![1, 1000, 4].into()),
            ReadBatchParams::Range(2..1000),
            ReadBatchParams::RangeTo(..1000),
        ];

        let sequence = RowIdSequence::from(0..10);

        for params in offsets {
            let result = select_row_ids(&sequence, &params);
            assert!(result.is_err());
            assert!(matches!(result.unwrap_err(), Error::InvalidInput { .. }));
        }
    }

    #[test]
    fn test_row_id_sequence_to_treemap() {
        let sequence = RowIdSequence(vec![
            U64Segment::Range(0..5),
            U64Segment::RangeWithHoles {
                range: 50..60,
                holes: vec![53, 54].into(),
            },
            U64Segment::SortedArray(vec![7, 9].into()),
            U64Segment::RangeWithBitmap {
                range: 10..15,
                bitmap: [true, false, true, false, true].as_slice().into(),
            },
            U64Segment::Array(vec![35, 39].into()),
            U64Segment::Range(40..50),
        ]);

        let tree_map = RowAddrTreeMap::from(&sequence);
        let expected = vec![
            0, 1, 2, 3, 4, 7, 9, 10, 12, 14, 35, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50,
            51, 52, 55, 56, 57, 58, 59,
        ]
        .into_iter()
        .collect::<RowAddrTreeMap>();
        assert_eq!(tree_map, expected);
    }

    #[test]
    fn test_row_addr_mask() {
        // 0, 1, 2, 3, 4
        // 50, 51, 52, 55, 56, 57, 58, 59
        // 7, 9
        // 10, 12, 14
        // 35, 39
        let sequence = RowIdSequence(vec![
            U64Segment::Range(0..5),
            U64Segment::RangeWithHoles {
                range: 50..60,
                holes: vec![53, 54].into(),
            },
            U64Segment::SortedArray(vec![7, 9].into()),
            U64Segment::RangeWithBitmap {
                range: 10..15,
                bitmap: [true, false, true, false, true].as_slice().into(),
            },
            U64Segment::Array(vec![35, 39].into()),
        ]);

        // Masking one in each segment
        let values_to_remove = [4, 55, 7, 12, 39];
        let positions_to_remove = sequence
            .iter()
            .enumerate()
            .filter_map(|(i, val)| {
                if values_to_remove.contains(&val) {
                    Some(i as u32)
                } else {
                    None
                }
            })
            .collect::<Vec<_>>();
        let mut sequence = sequence;
        sequence.mask(positions_to_remove).unwrap();
        let expected = RowIdSequence(vec![
            U64Segment::Range(0..4),
            U64Segment::RangeWithBitmap {
                range: 50..60,
                bitmap: [
                    true, true, true, false, false, false, true, true, true, true,
                ]
                .as_slice()
                .into(),
            },
            U64Segment::Range(9..10),
            U64Segment::RangeWithBitmap {
                range: 10..15,
                bitmap: [true, false, false, false, true].as_slice().into(),
            },
            U64Segment::Array(vec![35].into()),
        ]);
        assert_eq!(sequence, expected);
    }

    #[test]
    fn test_row_addr_mask_everything() {
        let mut sequence = RowIdSequence(vec![
            U64Segment::Range(0..5),
            U64Segment::SortedArray(vec![7, 9].into()),
        ]);
        sequence.mask(0..sequence.len() as u32).unwrap();
        let expected = RowIdSequence(vec![]);
        assert_eq!(sequence, expected);
    }

    #[test]
    fn test_selection() {
        let sequence = RowIdSequence(vec![
            U64Segment::Range(0..5),
            U64Segment::Range(10..15),
            U64Segment::Range(20..25),
        ]);
        let selection = sequence.select(vec![2, 4, 13, 14, 57].into_iter());
        assert_eq!(selection.collect::<Vec<_>>(), vec![2, 4, 23, 24]);
    }

    #[test]
    #[should_panic]
    fn test_selection_unsorted() {
        let sequence = RowIdSequence(vec![
            U64Segment::Range(0..5),
            U64Segment::Range(10..15),
            U64Segment::Range(20..25),
        ]);
        let _ = sequence
            .select(vec![2, 4, 3].into_iter())
            .collect::<Vec<_>>();
    }

    #[test]
    fn test_mask_to_offset_ranges() {
        // Tests with a simple range segment
        let sequence = RowIdSequence(vec![U64Segment::Range(0..10)]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[0, 2, 4, 6, 8]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..1, 2..3, 4..5, 6..7, 8..9]);

        let sequence = RowIdSequence(vec![U64Segment::Range(40..60)]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[54]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![14..15]);

        let sequence = RowIdSequence(vec![U64Segment::Range(40..60)]);
        let mask = RowAddrMask::from_block(RowAddrTreeMap::from_iter(&[54]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..14, 15..20]);

        // Test with a range segment with holes
        // 0, 1, 3, 4, 5, 7, 8, 9
        let sequence = RowIdSequence(vec![U64Segment::RangeWithHoles {
            range: 0..10,
            holes: vec![2, 6].into(),
        }]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[0, 2, 4, 6, 8]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..1, 3..4, 6..7]);

        let sequence = RowIdSequence(vec![U64Segment::RangeWithHoles {
            range: 40..60,
            holes: vec![47, 43].into(),
        }]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[44]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![3..4]);

        let sequence = RowIdSequence(vec![U64Segment::RangeWithHoles {
            range: 40..60,
            holes: vec![47, 43].into(),
        }]);
        let mask = RowAddrMask::from_block(RowAddrTreeMap::from_iter(&[44]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..3, 4..18]);

        // Test with a range segment with bitmap
        // 0, 1, 4, 5, 6, 7
        let sequence = RowIdSequence(vec![U64Segment::RangeWithBitmap {
            range: 0..10,
            bitmap: [
                true, true, false, false, true, true, true, true, false, false,
            ]
            .as_slice()
            .into(),
        }]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[0, 2, 4, 6, 8]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..1, 2..3, 4..5]);

        let sequence = RowIdSequence(vec![U64Segment::RangeWithBitmap {
            range: 40..45,
            bitmap: [true, true, false, false, true].as_slice().into(),
        }]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[44]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![2..3]);

        let sequence = RowIdSequence(vec![U64Segment::RangeWithBitmap {
            range: 40..45,
            bitmap: [true, true, false, false, true].as_slice().into(),
        }]);
        let mask = RowAddrMask::from_block(RowAddrTreeMap::from_iter(&[44]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..2]);

        // Test with a sorted array segment
        let sequence = RowIdSequence(vec![U64Segment::SortedArray(vec![0, 2, 4, 6, 8].into())]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[0, 6, 8]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..1, 3..5]);

        let sequence = RowIdSequence(vec![U64Segment::Array(vec![8, 2, 6, 0, 4].into())]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[0, 6, 8]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..1, 2..4]);

        // Test with multiple segments
        // 0, 1, 2, 3, 4, 100, 101, 102, 104, 44, 46, 78
        // *, -, *, -, -, ***, ---, ---, ***, --, **, --
        // 0, 1, 2, 3, 4,   5,   6,   7,   8,  9, 10, 11
        let sequence = RowIdSequence(vec![
            U64Segment::Range(0..5),
            U64Segment::RangeWithHoles {
                range: 100..105,
                holes: vec![103].into(),
            },
            U64Segment::SortedArray(vec![44, 46, 78].into()),
        ]);
        let mask = RowAddrMask::from_allowed(RowAddrTreeMap::from_iter(&[0, 2, 46, 100, 104]));
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..1, 2..3, 5..6, 8..9, 10..11]);

        // Test with empty mask (should select everything)
        let sequence = RowIdSequence(vec![U64Segment::Range(0..10)]);
        let mask = RowAddrMask::default();
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![0..10]);

        // Test with allow nothing mask
        let sequence = RowIdSequence(vec![U64Segment::Range(0..10)]);
        let mask = RowAddrMask::allow_nothing();
        let ranges = sequence.mask_to_offset_ranges(&mask);
        assert_eq!(ranges, vec![]);
    }

    #[test]
    fn test_row_id_sequence_rechunk_with_empty_segments() {
        // equal case (segment exactly fills remaining space)
        let input_sequences = vec![
            RowIdSequence::from(0..2),   // [0, 1] - 2 elements
            RowIdSequence::from(20..23), // [20, 21, 22] - 3 elements
        ];
        let chunk_sizes = vec![2, 3]; // First chunk wants 2, second wants 3

        let result = rechunk_sequences(input_sequences, chunk_sizes, false).unwrap();
        assert_eq!(result.len(), 2);
        assert_eq!(result[0].len(), 2);
        assert_eq!(result[1].len(), 3);

        let first_chunk: Vec<u64> = result[0].iter().collect();
        let second_chunk: Vec<u64> = result[1].iter().collect();
        assert_eq!(first_chunk, vec![0, 1]);
        assert_eq!(second_chunk, vec![20, 21, 22]);

        // less case (segment smaller than remaining space)
        let input_sequences = vec![
            RowIdSequence::from(0..2),   // [0, 1] - 2 elements (less than remaining)
            RowIdSequence::from(20..21), // [20] - 1 element (less than remaining)
            RowIdSequence::from(30..32), // [30, 31] - 2 elements (exactly fills remaining)
        ];
        let chunk_sizes = vec![5]; // Request 5 elements, have exactly 5

        let result = rechunk_sequences(input_sequences, chunk_sizes, false).unwrap();
        assert_eq!(result.len(), 1);
        assert_eq!(result[0].len(), 5);

        let elements: Vec<u64> = result[0].iter().collect();
        assert_eq!(elements, vec![0, 1, 20, 30, 31]);

        // empty segment in the middle
        let input_sequences = vec![
            RowIdSequence::from(0..2),   // [0, 1] - 2 elements
            RowIdSequence::from(10..10), // [] - 0 elements (empty)
            RowIdSequence::from(20..22), // [20, 21] - 2 elements
        ];
        let chunk_sizes = vec![3, 1];
        let result = rechunk_sequences(input_sequences, chunk_sizes, false).unwrap();

        assert_eq!(result.len(), 2);
        assert_eq!(result[0].len(), 3);
        assert_eq!(result[1].len(), 1);

        let first_chunk_elements: Vec<u64> = result[0].iter().collect();
        let second_chunk_elements: Vec<u64> = result[1].iter().collect();
        assert_eq!(first_chunk_elements, vec![0, 1, 20]);
        assert_eq!(second_chunk_elements, vec![21]);

        // multiple empty segments
        let input_sequences = vec![
            RowIdSequence::from(0..1),   // [0] - 1 element
            RowIdSequence::from(10..10), // [] - 0 elements (empty)
            RowIdSequence::from(20..20), // [] - 0 elements (empty)
            RowIdSequence::from(30..32), // [30, 31] - 2 elements
        ];
        let chunk_sizes = vec![3];
        let result = rechunk_sequences(input_sequences, chunk_sizes, false).unwrap();

        assert_eq!(result.len(), 1);
        assert_eq!(result[0].len(), 3);

        let elements: Vec<u64> = result[0].iter().collect();
        assert_eq!(elements, vec![0, 30, 31]);

        // empty segment at chunk boundary
        let input_sequences = vec![
            RowIdSequence::from(0..3), // [0, 1, 2] - 3 elements (exactly fills first chunk)
            RowIdSequence::from(10..10), // [] - 0 elements (empty, at boundary)
            RowIdSequence::from(20..22), // [20, 21] - 2 elements (for second chunk)
        ];
        let chunk_sizes = vec![3, 2];
        let result = rechunk_sequences(input_sequences, chunk_sizes, false).unwrap();

        assert_eq!(result.len(), 2);
        assert_eq!(result[0].len(), 3);
        assert_eq!(result[1].len(), 2);

        let first_chunk_elements: Vec<u64> = result[0].iter().collect();
        let second_chunk_elements: Vec<u64> = result[1].iter().collect();
        assert_eq!(first_chunk_elements, vec![0, 1, 2]);
        assert_eq!(second_chunk_elements, vec![20, 21]);

        // empty segments with allow_incomplete = true
        let input_sequences = vec![
            RowIdSequence::from(0..2),   // [0, 1] - 2 elements
            RowIdSequence::from(10..10), // [] - 0 elements (empty)
        ];
        let chunk_sizes = vec![5]; // Request more than available
        let result = rechunk_sequences(input_sequences, chunk_sizes, true).unwrap();

        assert_eq!(result.len(), 1);
        assert_eq!(result[0].len(), 2);

        let elements: Vec<u64> = result[0].iter().collect();
        assert_eq!(elements, vec![0, 1]);
    }
}