Struct PermutationIndex 

pub struct PermutationIndex { /* private fields */ }

A sorted permutation index over a columnar dataset.

permutation[virtual_row] = physical row ID in the source.

Internal storage is either in-memory (UInt32Array) or memory-mapped (temp file), chosen automatically based on row count and feature flags.

Implementations

impl PermutationIndex

pub fn natural(n: u64) -> Result<Self, IndexError>

Identity permutation — natural (unsorted) order: [0, 1, 2, ..., n-1].

Errors

Returns IndexError::TooManyRows if n > u32::MAX.

pub fn len(&self) -> u64

Number of rows in the permutation.

pub fn is_empty(&self) -> bool

Whether the permutation is empty.

pub fn read_range(&self, start: usize, end: usize) -> Vec<u32>

Read row IDs for virtual range [start, end).

This is the primary access method — works for both in-memory and mmap. Clamped to [0, len).

pub fn into_vec(self) -> Vec<u32>

Consume and return all row IDs as a Vec<u32>.

pub fn slice(&self, offset: usize, length: usize) -> Self

Slice the permutation to a subrange.

Returns a new in-memory PermutationIndex covering [offset..offset+length].

Panics

Panics if offset + length > self.len().

pub fn to_physical_selection( &self, start: usize, end: usize, ) -> (PhysicalSelection, Vec<u32>)

Convert a virtual row range to a physical selection for late materialisation.

Given virtual rows [start, end), returns the corresponding physical row IDs as sorted, merged ranges suitable for Parquet RowSelection.

Also returns the virtual-order IDs (the raw read_range output) that callers need for reordering materialised rows back to sort order.

Returns

(physical_selection, virtual_ids) where:

• physical_selection has ranges in ascending physical order
• virtual_ids[i] = physical row ID at virtual position start + i

Trait Implementations

impl Debug for PermutationIndex

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.