Is timsort general-purpose or Python-specific?

Question

Timsort is an adaptive, stable, natural mergesort. It has supernatural performance on many kinds of partially ordered arrays (less than lg(N!) comparisons needed, and as few as N-1), yet as fast as Python's previous highly tuned samplesort hybrid on random arrays.

Have you seen timsort used outside of CPython? Does it make sense?

why do you ask? without more context, your question cannot be answered. — , Sep 30 '08 at 19:31
Have you noticed "Have you seen timsort used outside of CPython?" part? — Constantin, Sep 30 '08 at 19:40
i have noticed it, and it still gives us no context. what would you learn from a simple "no" as answer? — , Sep 30 '08 at 20:12
'timsort' is really very specific, although it won't mean much to you unless you actually know what timsort is. — Thomas Wouters, Sep 30 '08 at 21:17

score 31 · Accepted Answer · answered Jun 29 '09 at 20:09

31

Yes, it makes quite a bit of sense to use timsort outside of CPython, in specific, or Python, in general.

There is currently an effort underway to replace Java's "modified merge sort" with timsort, and the initial results are quite positive.

answered Jun 29 '09 at 20:09

plinehan

700
7
17

4

Java SE 7 uses Timsort as it's sort algorithm now. See http://www.docjar.com/docs/api/java/util/Collections.html#sort(List) – Abhinav Sarkar Jul 07 '12 at 17:45

Thomas Wouters · Answer 2 · 2008-09-30T21:15:53.920

23

The algorithm is pretty generic, but the benefits are rather Python-specific. Unlike most sorting routines, what Python's list.sort (which is what uses timsort) cares about is avoiding unnecessary comparisons, because generally comparisons are a lot more expensive than swapping items (which is always just a set of pointer copies) or even allocating some extra memory (because it's always just an array of pointers, and the overhead is small compared to the average overhead in any Python operation.)

If you're under similar constraints, then it may be suitable. I've yet to see any other case where comparisons are really that expensive, though :-)

edited Sep 30 '08 at 21:15

answered Sep 30 '08 at 20:14

Thomas Wouters

130,178
23
148
122

1

If comparisons are expensive, then a data-specific algorithm will usually perform better than a comparison-based one. – Rafał Dowgird Oct 01 '08 at 12:53
1

That is a good observation, and indeed probably the main reason you won't see timsort or anything close to it in the wild. – Thomas Wouters Oct 01 '08 at 13:34
1

The benefits are *not* Python-specific. The same benefit of using Timsort in Python to sort a list of object references exists in any programming language which has pointers or object references. For instance, Java SE 7+, Android and Swift use Timsort to sort objects. – Alan Evangelista Nov 29 '19 at 18:17

score 6 · Answer 3 · answered Sep 30 '08 at 20:02

6

It doesn't look particularly familiar, but "smart" mergesorts are pretty common out in the wide world of software.

As for whether it makes sense, that depends on what you're sorting, and the relative cost of comparisons vs. memory allocation. A sort that requires up to 2*N bytes of extra memory isn't going to be a good choice in a memory-constrained environment.

answered Sep 30 '08 at 20:02

Mark Bessey

19,598
4
47
69

Are you talking about Timsort in your last sentence? It does not take 2*N bytes of extra memory. – Alan Evangelista Nov 29 '19 at 18:01
From the linked description in the original question:” + timsort can require a temp array containing as many as N//2 pointers, which means as many as 2*N extra bytes on 32-bit boxes.” – Mark Bessey Dec 01 '19 at 00:44
Thanks for making it clear. As you assumed a 32-bit arch in the sentence in your answer, it'd be nice to explicitly mention that in it. – Alan Evangelista Dec 01 '19 at 22:10

score 4 · Answer 4 · answered Oct 29 '10 at 07:36

4

Answered now on Wikipedia: timsort will be used in Java 7 who copied it from Android.

answered Oct 29 '10 at 07:36

Douglas Bagnall

796
7
7

score 3 · Answer 5 · answered Jun 09 '11 at 06:53

3

Timsort is also in Android now: http://www.kiwidoc.com/java/l/x/android/android/5/p/java.util/c/TimSort

answered Jun 09 '11 at 06:53

Constantin

27,478
10
60
79

score 0 · Answer 6 · answered Sep 30 '08 at 19:21

0

The description you linked looks completely general.

answered Sep 30 '08 at 19:21

dmckee --- ex-moderator kitten

98,632
24
142
234

Yes, but have you seen timsort used outside of CPython? – Constantin Sep 30 '08 at 19:40

Alan Evangelista · Answer 7 · 2019-11-29T19:05:37.310

Timsort is not Python-specific. The benefits of using Timsort in Python to sort a list of object references exists in any programming language which has pointers or object references. For instance, Java SE 7+, Android and Swift use Timsort to sort objects.

On the other hand, some variation of quicksort (eg introsort, dual-pivot quicksort) usually sorts primitive types faster, due to cache coherence, and therefore it is usually chosen for this task.

Is timsort general-purpose or Python-specific?

7 Answers7

Linked