Mercurial > lbo > hg > localmr
view src/shard_merge.rs @ 52:6a51cc42d1a6
Move sort/comparison functions into own module
author | Lewin Bormann <lbo@spheniscida.de> |
---|---|
date | Fri, 05 Feb 2016 21:17:34 +0000 |
parents | 8a2538cbac8a |
children | 90b955067a22 |
line wrap: on
line source
//! Implements a merge tree to merge an arbitrary number of sorted map outputs. See //! https://drive.google.com/open?id=1grB87a0w9fQ2k7i04N3VJvYlw2BldxWNcHublW_ygJs. //! Genericized in order to build arbitrary merge trees. #![allow(dead_code)] use sort; use std::cmp::{Ord, Ordering}; use std::iter; /// See module description. /// This type uses dynamic instead of static dispatch because it realizes an arbitrary structure /// and can therefore not work with a single type signature. pub struct ShardMergeIterator<'a, T: Ord> { left: Box<Iterator<Item = T> + 'a>, right: Box<Iterator<Item = T> + 'a>, left_peeked: Option<T>, right_peeked: Option<T>, comparer: Option<sort::Comparer<T>>, } impl<'a, T: Ord + Clone> Iterator for ShardMergeIterator<'a, T> { type Item = T; fn next(&mut self) -> Option<Self::Item> { // fill up match (self.left_peeked.clone(), self.right_peeked.clone()) { (None, None) => { self.left_peeked = self.left.next(); self.right_peeked = self.right.next() } (Some(_), None) => self.right_peeked = self.right.next(), (None, Some(_)) => self.left_peeked = self.left.next(), (Some(_), Some(_)) => (), } // Consume peeked values match (self.left_peeked.clone(), self.right_peeked.clone()) { (None, None) => return None, (l @ Some(_), None) => { self.left_peeked = None; return l; } (None, r @ Some(_)) => { self.right_peeked = None; return r; } (Some(l), Some(r)) => { let cmp = self.compare(&l, &r); if cmp == Ordering::Less || cmp == Ordering::Equal { self.left_peeked = None; return Some(l); } else { self.right_peeked = None; return Some(r); } } } } } impl<'a, T: Ord + Clone> ShardMergeIterator<'a, T> { fn default() -> ShardMergeIterator<'a, T> where T: 'a { ShardMergeIterator { left: Box::new(iter::empty()), right: Box::new(iter::empty()), left_peeked: None, right_peeked: None, comparer: None, } } pub fn build<It: Iterator<Item = T>, ItIt: Iterator<Item = It>>(sources: &mut ItIt) -> ShardMergeIterator<'a, T> where T: 'a, It: 'a { ShardMergeIterator::_build(sources, None) } pub fn build_with_cmp<It: Iterator<Item = T>, ItIt: Iterator<Item = It>>(sources: &mut ItIt, cmp: sort::Comparer<T>) -> ShardMergeIterator<'a, T> where T: 'a, It: 'a { ShardMergeIterator::_build(sources, Some(cmp)) } /// Takes multiple iterators of type It and generates one ShardedMergeIterator.. /// (yes, iterator over a collection of iterators). fn _build<It: Iterator<Item = T>, ItIt: Iterator<Item = It>>(sources: &mut ItIt, cmp: Option<sort::Comparer<T>>) -> ShardMergeIterator<'a, T> where T: 'a, It: 'a { let mut merged: Vec<ShardMergeIterator<T>> = Vec::new(); // Initial merging: Merge pairs of input iterators together. loop { let src1: It; match sources.next() { None => break, Some(src) => src1 = src, } match sources.next() { None => { merged.push(ShardMergeIterator { left: Box::new(src1), right: Box::new(iter::empty()), comparer: cmp, ..ShardMergeIterator::default() }) } Some(src) => { merged.push(ShardMergeIterator { left: Box::new(src1), right: Box::new(src), comparer: cmp, ..ShardMergeIterator::default() }) } } } // Recursively build the merge tree from the leaves. ShardMergeIterator::merge(merged, cmp) } fn compare(&self, a: &T, b: &T) -> Ordering { match self.comparer { None => a.cmp(b), Some(f) => f(a, b) } } /// Merge multiple ShardMergeIterators, recursively (meaning it will result in a more or less /// balanced merge sort tree). fn merge(mut its: Vec<ShardMergeIterator<'a, T>>, cmp: Option<sort::Comparer<T>>) -> ShardMergeIterator<'a, T> where T: 'a { if its.len() == 0 { ShardMergeIterator::default() } else if its.len() == 1 { ShardMergeIterator { left: Box::new(its.remove(0)), comparer: cmp, ..ShardMergeIterator::default() } } else if its.len() == 2 { let it1 = its.remove(0); let it2 = its.remove(0); ShardMergeIterator { left: Box::new(it1), right: Box::new(it2), comparer: cmp, ..ShardMergeIterator::default() } } else { // its is left part, right is right part let split_at = its.len() / 2; let right = its.split_off(split_at); ShardMergeIterator { left: Box::new(ShardMergeIterator::merge(its, cmp)), right: Box::new(ShardMergeIterator::merge(right, cmp)), comparer: cmp, ..ShardMergeIterator::default() } } } } #[cfg(test)] mod tests { use std::vec; use shard_merge::ShardMergeIterator; fn get_collection_1() -> vec::IntoIter<i32> { vec![1, 4, 5, 5, 6, 9, 11, 15, 15, 17, 18, 20].into_iter() } fn get_collection_2() -> vec::IntoIter<i32> { vec![2, 2, 2, 3, 4, 5, 7, 8, 9, 10, 45, 46, 47].into_iter() } fn get_collection_3() -> vec::IntoIter<i32> { vec![5, 8, 9, 10, 22, 25, 30, 37, 41, 46, 71].into_iter() } fn get_collection_4() -> vec::IntoIter<i32> { vec![111, 112, 113, 155].into_iter() } fn get_collection_5() -> vec::IntoIter<i32> { vec![13, 45, 98, 105, 145].into_iter() } fn get_collection_6() -> vec::IntoIter<i32> { vec![14, 67, 99, 111, 222, 566, 643].into_iter() } #[test] fn test_merge_iterator() { let it = ShardMergeIterator::build(&mut vec![get_collection_1(), get_collection_2(), get_collection_3(), get_collection_4(), get_collection_5(), get_collection_6()] .into_iter()); let mut cmp = 0; let mut cnt = 0; for i in it { assert!(i >= cmp); cmp = i; cnt += 1; } assert_eq!(cnt, get_collection_1().len() + get_collection_2().len() + get_collection_3().len() + get_collection_4().len() + get_collection_5().len() + get_collection_6().len()); } use formats::lines; use std::fmt; use std::io::Write; // Slow test! #[test] fn test_merge_large_files() { let mut files = Vec::with_capacity(11); for i in 0..11 { let name = fmt::format(format_args!("testdata/sorted{}.txt", i)); files.push(lines::new_from_file(&name).unwrap()); } let merge_it = ShardMergeIterator::build_with_cmp(&mut files.into_iter(), sort::dict_string_compare); let mut outfile = lines::LinesWriter::new_to_file(&String::from("testdata/all_sorted.txt")).unwrap(); for line in merge_it { let _ = outfile.write(line.as_bytes()); } } use std::cmp::Ordering; use sort; #[test] fn test_sane_char_compare() { assert_eq!(sort::sane_char_compare('a', 'B'), Ordering::Less); assert_eq!(sort::sane_char_compare('a', 'a'), Ordering::Equal); assert_eq!(sort::sane_char_compare('A', 'a'), Ordering::Greater); assert_eq!(sort::sane_char_compare('B', 'a'), Ordering::Greater); } #[test] fn test_sane_string_compare() { let cnv = String::from; let s1 = &cnv(""); let s2 = &cnv("0abc"); let s3 = &cnv("123"); let s4 = &cnv("abc"); let s5 = &cnv("Abc"); let s6 = &cnv("ABC"); let s7 = &cnv("aBC"); assert_eq!(sort::sane_string_compare(s1, s2), Ordering::Less); assert_eq!(sort::sane_string_compare(s2, s3), Ordering::Less); assert_eq!(sort::sane_string_compare(s3, s2), Ordering::Greater); assert_eq!(sort::sane_string_compare(s2, s2), Ordering::Equal); assert_eq!(sort::sane_string_compare(s4, s5), Ordering::Less); assert_eq!(sort::sane_string_compare(s5, s6), Ordering::Less); assert_eq!(sort::sane_string_compare(s6, s7), Ordering::Greater); } #[inline] fn bogus_fn(o: Ordering) -> bool { if o == Ordering::Equal { panic!("bogus panic") } true } // Slow test! //#[test] fn bench_sane_string_compare() { let cnv = String::from; let s1 = &cnv(""); let s2 = &cnv("0abc"); let s3 = &cnv("123"); let s4 = &cnv("abc"); let s5 = &cnv("Abc"); for _ in 0..50000000 { bogus_fn(sort::sane_string_compare(s1, s2)); bogus_fn(sort::sane_string_compare(s4, s3)); bogus_fn(sort::sane_string_compare(s4, s5)); } } }