Sometimes in my code I have to find the first few smallest (or biggest) items of an array, I don't know how many I will need of them, but I know in general I will need only few of them, much less than the whole array. Turning the array into an heap is a slow operation if I will only need few items, and I can't use std.algorithm.partialSort because I don't know the number of items I will need. So I have created this very simple LazySort range, based on partialSort (works with DMD 2.056head): import std.stdio, std.algorithm, std.random, std.array, std.range, std.traits; // Missing still: less and SwapStrategy template arguments. struct LazySort(Range) if (isRandomAccessRange!Range) { Range data; private size_t idx, idxSorted; bool empty() { return idx >= data.length; } ForeachType!Range front() { if (idx >= idxSorted) { immutable oldIdxSorted = idxSorted; idxSorted = min(data.length, idxSorted ? (idxSorted * 2) : 1); partialSort(data[oldIdxSorted .. $], idxSorted - oldIdxSorted); } return data[idx]; } void popFront() { idx++; } } void main() { auto A = array(iota(25)); randomShuffle(A); writeln(A); foreach (x; LazySort!(int[])(A)) write(x, " "); } I have not done benchmarks on it yet. This code seems to work, but it is not efficient, it's just to show the semantics of the idea. A better implementation is welcome. I think a lazy sort will be useful in Phobos. Timon Gehr seems to appreciate the idea.
The canonical solution uses a heap. Creating a heap is cheap and quickly amortized over only a few pops. An input range that creates a heap and then yields one element at a time would be a better idea.
(In reply to comment #1) > The canonical solution uses a heap. Creating a heap is cheap and quickly > amortized over only a few pops. An input range that creates a heap and then > yields one element at a time would be a better idea. If benchmarks show that a range that heapifies the input array is about as efficient as a tailored lazy sorting solution for about 4 to 10 requested max/min items, then I am OK with this idea :-)
(In reply to comment #2) > (In reply to comment #1) > > The canonical solution uses a heap. Creating a heap is cheap and quickly > > amortized over only a few pops. An input range that creates a heap and then > > yields one element at a time would be a better idea. > > If benchmarks show that a range that heapifies the input array is about as > efficient as a tailored lazy sorting solution for about 4 to 10 requested > max/min items, then I am OK with this idea :-) This is odd at a few levels. First, you opened this report. So it's not you who's supposed to be OK, it's the rest of us. You have the burden of proof. Second, there's no need for benchmarking anything - simple complexity analysis shows that partialSort does O(n) log(n-k) at each step, which pretty much makes it bankrupt compared to the heap approach.
THIS ISSUE HAS BEEN MOVED TO GITHUB https://github.com/dlang/phobos/issues/9913 DO NOT COMMENT HERE ANYMORE, NOBODY WILL SEE IT, THIS ISSUE HAS BEEN MOVED TO GITHUB