# Talk:Sorting algorithm

Sorting algorithm is a former featured article candidate. Please view the links under Article milestones below to see why the nomination failed. For older candidates, please check the archive.
 August 30, 2004 Featured article candidate Not promoted
WikiProject Computer science (Rated B-class, Top-importance)
This article is within the scope of WikiProject Computer science, a collaborative effort to improve the coverage of Computer science related articles on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.
B  This article has been rated as B-Class on the project's quality scale.
Top  This article has been rated as Top-importance on the project's importance scale.

## How bad is this code

template <class iterator> void gnome(iterator first, iterator last){ // O(n^2) iterator pos = first; while (pos <= last) { if ((pos == first) || ((*pos) >= *(pos - 1))) ++pos; else { std::iter_swap(pos, pos+1); --pos; } } }

## Slowsort

I'm surprised at not finding slowsort here. While it is one of the less serious kind of algorithms, it is interesting since it is, however slowly, steadily progressing to its goal (as opposed to, say, bogosort). The algorithm:

1. Find the maximum
1. Find the maximum of the first half
2. Find the maximum of the second half
3. Find the largest of those maxima
2. Sort the remainder

It is described at http://c2.com/cgi/wiki?SlowSort and was, according to that page, first published in the satyrical article "Pessimal algorithms and the simplexity of computations" by A. Broder and J. Stolfi, in ACM SIGACT News vol. 16 no. 7, pages 49--53, fall 1984. I haven't located that article yet, though. Both links given there are broken. —Preceding unsigned comment added by SQB (talkcontribs) 20:08, 26 July 2009 (UTC)

## Stupid sort

Stupid sort - This article was copied from the deleted Wikipedia article which sat in the "Stupid sort" page. -- A perfect demonstration how stupidity proliferates in the internet with the help and andorsement of wikipedia :-) 23:25, 6 February 2010 (UTC)

Yes, the algorithm does seem to be a joke. I recall seeing it here before and wondered what had happened to it. But it seems the term is also used to refer variously to bogosort or to gnome sort. Currently, stupid sort redirects to bogosort.... — Smjg (talk) 02:26, 16 December 2011 (UTC)

## Why are worst case time complexities wrong?

For example, the time complexity for bubble sort is listed as n2.

While its computational complexity (time + space) is on the order of n2 (2n2-2n), the worst case time complexity is not n2.

worst case time complexity (comparisons) for bubble sort is (n2/2) - (n/2). But they are all wrong. —Preceding unsigned comment added by 74.140.173.132 (talk) 05:57, 24 February 2010 (UTC)

I don't understand what you say, because Ordo((n2/2) - (n/2)) = Ordo(n2). Any term less than n2 is overshadowed by n2, and the constant 1/2 is irrelevant. The complexity is a scalability function when the function compares to itself dealing with increasingly huge arrays. Rursus dixit. (mbork3!) 19:48, 6 July 2010 (UTC)

the wrong is the thing —Preceding unsigned comment added by 203.110.243.22 (talk) 12:07, 20 November 2010 (UTC)

## There is nothing wrong with worst case time complexitiy

(n2/2) - (n/2) IS O(n2)

## Missing O()?

Should it be O(n log n) instead n log n in comparison table? —Preceding unsigned comment added by Conan (talkcontribs) 00:06, 9 May 2010 (UTC)

## Quick sort error

Comparison table states that its "memory" is log(n). But, as far as I know, quick sort is in-place sort. Even its own article said so: "...Coupled with the fact that quicksort is an in-place sort and uses no temporary memory...". —Preceding unsigned comment added by 85.222.169.174 (talk) 13:44, 4 June 2010 (UTC)

As it says later in the quicksort article, memory is required for the stack frames which are implicit in the recursion. "No temporary memory" is just wrong. 67.162.90.113 (talk) 19:59, 2 November 2010 (UTC)

## Randomized Quiz Sort Algorithm

In the worst case, quicksort sorts a group of items in O(N2), where N is the number of items. If randomization is incorporated into the algorithm, however, the chances of the worst case actually occurring become diminishingly small, and on average, quicksort has a runtime of O(N*Log(N)). Other algorithms guarantee a runtime of O(N*Log(N)), even in the worst case, but they are slower in the average case. Even though both algorithms have a runtime proportional to N*Log(N), quicksort has a smaller constant factor - that is it requires C*N*Log(N) operations, while other algorithms require more like 2*C*N*Log(N) operations —Preceding unsigned comment added by Tarun telang (talkcontribs) 10:31, 11 June 2010 (UTC)

## "Comparison of algorithms" table ordering

I find it rather ironic that this table does itself not appear to be sorted! The "exchanging" methods are kept together and are ordered alphabetically, but other than that the order seems to be pretty random. I am in favour of sorting this table according to Method and then name, but other options such as by average case complexity and then name also seem fine to me. It may be like a bit of a plug that Timsort has been placed at the top of this table. There seems to be less reason for it to appear in the table compared to some algorithms that are currently omitted.

Secondly, I'd also like to see a best case column added to the table. Then it would make sense to add other algorithms such as Binary Insertion Sort whose best case differs from normal Insertion Sort, but the average and worst cases are the same. It would also make sense to hilight the fact that bubble sort (amoung others) can be implemeneted with an O(n) best case. Several other algorithms such as Bitonic Merge Sort, Bingo Sort, Odd-Even Transposition Sort, Strand Sort, Squeeze Sort, and Order Sort to name a few, could also be added.

Malcolm —Preceding unsigned comment added by 118.92.109.220 (talk) 03:01, 13 June 2010 (UTC)

## Issues with attached file: SortingAlgoComp.png

The linked file "File:SortingAlgoComp.png" has a very poor quality of information. There are some spelling mistakes "Tourament" (missing 'n') and "Interger" (extra 'r') The description states: "This chart shows the complexity of different algorithms when sorting one million integers (from 1 to 1000).". However it is not clear where these numbers come from. Are they comparison counts? Are they swap (or assignment) counts? A combination of both of these measures? What was the initial ordering of the data? Over how many runs were these numbers averaged? How is it that so many of the algorithms all have exactly 1.99316x10^7 as the count? What does "Sorting networks" mean in this picture, and how is the result for it able to not be a whole number? How was bogosort measured given the ridiculously high number shown?

I believe that in its current form, this image adds no value to the article and should be removed. If the information is to be explained and kept then it should converted to a textual form instead.

Malcolm 118.92.109.220 (talk) 04:45, 13 June 2010 (UTC)

I've just removed it. It is incoherent, and it is OR. But mostly it is incoherent. Apparently nobody cares for the quality of this article, that this was allowed to linger here for a year and a half at least, and God knows for how long really. Appalling. WillNess (talk) 17:39, 25 December 2011 (UTC)

## Fixing a lede sentence

This sentence makes no sense to me: "Efficient sorting is important to optimizing the use of other algorithms (such as search and merge algorithms) that require sorted lists to work correctly." I can't fix it because I don't know what it's trying to say. So I'm inviting others to work on this. If no fix is forthcoming, I will likely delete this sentence.

By the way, it might be helpful in the summary of each algorithm to describe its performance on various types of datasets (e.g., random, nearly sorted, reversed, few unique elements). MichaelBluejay (talk) 12:56, 22 June 2010 (UTC)

--- How about " ...) which require input data to be in sorted lists."

Your suggestion about how sorts handle pathological cases is a good one, but I suspect it can quickly degrade into confusion. For instance, performance of quicksort on pre-structured data will depend a lot on how the pivot is chosen, and this decision is up to the implementer, and not really part of the quicksort definition. Many implementations have 'tweaks' to protect against pathological cases.

184.147.234.114 (talk) 02:27, 10 February 2013 (UTC)

## Are sorting networks truly stable sorts?

A simple sorting network consisting of four wires and five connectors

Consider this sorting network as included in the Sorting network article. Let's put [2, 1a, 1b, 0] on the wires with 'a' and 'b' as markers for the initial order without them belonging to the sorting key. If this sorting network is stable, we can expect it to return [0, 1a, 1b, 2]. The way I see it, it just doesn't:

2, 1a, 1b, 0 is the initial configuration.
1b, 1a, 2, 0 results from the first comparison. 2 > 1, thus we swap.
1b, 0, 2, 1a results from the second comparison. 1 > 0, thus we swap.
0, 1b, 1a, 2 results from the next two comparisons. We swap both since 1 > 0 and 2 > 1.
0, 1b, 1a, 2 is the result. No swap in the last comparison since 1 = 1.

However, [0, 1b, 1a, 2] does not equal [0, 1a, 1b, 2], which was our expected result. From this counterexample follows that sorting networks are not stable sorts in the general case.

So… am I missing anything here? —Preceding unsigned comment added by 129.13.72.197 (talk) 11:23, 23 June 2010 (UTC)

Agreed - there may be a subclass of networks which are stable, but I've not found any discussion of this; I've looked at various example diagrams and have found it easy to find counter-examples in each. Stability would almost certainly come at a cost in extra levels (and more expensive than just 'tagging' the values to force stability). Table entry "Yes" is either plain wrong, or misleading. 184.147.234.114 (talk) 02:05, 10 February 2013 (UTC)

## Quantum sort

Quantum sorting is not faster than normal sorting (http://arxiv.org/abs/quant-ph/0102078) but might offer better time-space tradeoffs. Thus, Quantum sort does not need to have its own article. I propose merging the brief content from quantum sort into the sorting algorithm article. Comments? Please also comment on the quantum sort talk page so that we can delete that article and incorporate it into this one. --DFRussia (talk) 20:56, 29 June 2010 (UTC)

## Bubble sort

The section Bubble sort claims:

This algorithm is highly inefficient, and is rarely used[citation needed], except as a simplistic example.

Looks like false to me. Because of the code brevity of the algorithm, it instead should be heavily used on pretty short lists in naive Q&D implementations where a sort is needed. Also it is often combined with quicksort and quickersort for when the to-be-sorted sublists are under a certain length, such as 3 or 4 elements. This use to improve the sorting speed. So the statement is dubious. Rursus dixit. (mbork3!) 19:57, 6 July 2010 (UTC)

I don't think that's true. If quick & dirty is required, one can implement Quicksort in a single line as in [1]. For sorting of short sub-lists, I have observed variants on Insertion sort much more often than Bubble sort. For the case of finite lengths on those sub-lists, optimal compare-and-swap sequences are known for lists with a maximum length of at least 47 items, probably more. I think there's even a sequence on Sloane's for that, can't seem to find it right now. —Preceding unsigned comment added by 129.13.72.197 (talk) 16:17, 10 July 2010 (UTC)

## Joshua Kehn blog

Is this sorting demo page an external link that should or should not be included in the article. Baltar, Gaius (talk) 05:06, 6 January 2011 (UTC)

Discussion on this sorting algorithm page and its inclusion in the external links section.

I have undone the addition of this link because, IMO, it is blatant self promotion. Furthermore, I and at least one other editor report problems with the animations there crashing or hanging for substantial periods of time. Reyk YO! 06:48, 10 October 2010 (UTC)

Seeing as this section was already there

If you want to include the links, bring the subject up on the article's talk page. Supply links to the existing discussion. State why you think the links are appropriate for WP. I'd like to see your argument why WP should tolerate links that hang a user's browser or denigrate the use of IE.

I believe these links are appropriate for WP because currently there is not enough diversity in the existing examples. You have one that constrains a viewer to a having the JRE installed, and another is non-interactive through the use of animated GIFs. From this StackOverflow question I found this sorting demo page.

The sorting demo is built on open standards, using HTML5's canvas (with support for IE) and JavaScript to display an interactive animated sort. This demo is compatible with almost all browsers and operating systems, including support for mobile devices such as Apple's iPhone, iPod, iPad and other mobile devices (namely Android based). Open standards allow for community improvement and investigation, rather then simply seeing a few bars move you can download the source and research it yourself. While I will concede that this is less of an issue in sort algorithms that have publicly implementations, it is a good path to follow and more conducive to student learning purposes.

In regards to the hanging a user's browser I find no issue with it. If you do then I'm sure contacting the author with more specific details will yield a fix. As far as denigrating the use of Internet Explorer, the most offensive statement I could find is as follows:

: IE users will notice a overall sluggishness due to IE not implementing native <canvas> support. You have my sympathies.


Internet Explorer does not have native canvas support, and the author supports this browser through the use of a excanvas plugin that will reduce the speed at which the animation can be displayed.

Baltar, Gaius (talk) 20:17, 2 December 2010 (UTC)

• Oppose. The site is anti-IE - a stance that WP should avoid. Suggesting that people switch browsers is not reasonable. The requested EL has crashed other browsers. Author states bubble sort will cause browsers to go to lunch for a while. That other demos have problems/requirements does not justify including this one; two wrongs don't make a right. I have a HUGE ISSUE with the notion that a WP user follows a link and his browser locks up. We should not direct people to bad implementations. Despite JK's good intentions, we are not a source of β-testers. Glrx (talk) 21:14, 2 December 2010 (UTC)

Suggesting that a user support standards compliant browsers is reasonable, and the fact that extra effort seems to be made to support Internet Explorer should be noted. The Author states Some sorts (bubble) are particularly intensive and not recommended for older / slower computers., not that will cause browsers to go to lunch for a while. There is a momentary stall considering the large amount of pre-computation required to smoothly render the bubble sort. Considering the source (JavaScript) one can assume that the browsers with faster JavaScript engines will perform better, this is not bias but rather pertinent information.

The argument that other demos have gaps means this one should not isn't valid. This demo is filling gaps left by others, and as such plays a vital role in offering the most options to users. Suggesting that this is beta software is incorrect. Baltar, Gaius (talk) 14:56, 6 December 2010 (UTC)

Continuing. If there are not more opposed I would like inclusion of this link. Baltar, Gaius (talk) 14:56, 6 December 2010 (UTC)

It does not work that way. You want to add some challenged material, so you must garner a WP:Consensus to make that addition. You don't have a consensus. Furthermore, others have reverted the links. Glrx (talk) 16:27, 6 December 2010 (UTC)
Can you explain what issues you still have? I believe the anti-IE and crashing browsers have both been explained unless you have other issues you would like to bring up. Baltar, Gaius (talk) 19:35, 6 December 2010 (UTC)
I would appreciate at least a semblance of discussion. If you could explain what other issues you have I would like a chance to debate them. One other person reverted the link, I don't see that as a consensus. There are two people in a discussion, how do we involve more? — Preceding unsigned comment added by Baltar, Gaius (talkcontribs) 14:20, 13 December 2010 (UTC)
Support inclusion - In my opinion, it is a highly helpful link which will help users visualize the process of sorting algorithms. I just ran it on IE, and it worked just fine (albeit slower than on FireFox). Reaper Eternal (talk) 02:55, 21 December 2010 (UTC)
I also had to remove the 3O template since three editors: Baltar, Gaius (talk · contribs), Glrx (talk · contribs), and Reyk (talk · contribs) are already involved. 3O requires that only two editors be involved in the dispute. I have, however, offered my opinion on the matter above. Reaper Eternal (talk) 03:00, 21 December 2010 (UTC)
My mistake, I forgot Template:Reyk was included. Baltar, Gaius (talk) 05:34, 21 December 2010 (UTC)
• Weak oppose- I agree with Glrx that we should not be directing people to implementations of sorting algorithms that can hang the browser. I am no longer so concerned about denigrating IE; the warning against it is actually rather mild. Reyk YO! 22:28, 22 December 2010 (UTC)
What sorts have you found that hang the browser? Baltar, Gaius (talk) 02:33, 23 December 2010 (UTC)
Bubble sorts at full size cause Firefox to hang for a while. Ztothefifth (talk) 02:47, 23 December 2010 (UTC)
Basically this. I have sometimes seen the bubble sort at full size grind to a halt and not recover at all, forcing me to shut it down in task manager, though it usually does right itself eventually. I also get random five to ten second hangs occasionally with the other algorithms. Not game to even try it with IE. Reyk YO! 02:59, 23 December 2010 (UTC)
To be fair there is a warning before doing a full sized bubble sort. The default sort set is significantly smaller then a full sized set. Baltar, Gaius (talk) 14:31, 23 December 2010 (UTC)

What other links to animations are there? All I see is http://www.sorting-algorithms.com/. Ztothefifth (talk) 18:50, 23 December 2010 (UTC)

Another possible link is http://sortvis.org/. Ztothefifth (talk) 21:06, 23 December 2010 (UTC)

As far as I know there are not many sorting demo's out there, let along interactive ones such as this, which is why I'm pushing strongly for inclusion. Most of the demos are simple. Either pictures or animated gifs. The few more complex ones are applets designed to only run one sort with a specific set – essentially no better then an animated gif. This demo is highly interactive and adjustable. Baltar, Gaius (talk) 01:10, 24 December 2010 (UTC)

Attempting Consensus Can we concede that this page offers a interactive sort visualization that is not tied to any platform or operating system requirements, promoting open standards, and has minor stability problems on few platforms? If we can, I believe that inclusion is supported. Baltar, Gaius (talk) 15:18, 4 January 2011 (UTC)

No, there is not a consensus for inclusion. Glrx (talk) 17:37, 4 January 2011 (UTC)
What are your requirements for consensus? I would like to work with you to agree on a point where this link can be included. I will contact the author and see if there is anything he is willing to change or alter that would better encourage a consensus. Does anyone else have points (for or against) that they would like to bring up? Baltar, Gaius (talk) 18:29, 5 January 2011 (UTC)
You could also open a request for comment on this article to get more opinions. Reaper Eternal (talk) 19:12, 5 January 2011 (UTC)
Thank you, I have done so. Baltar, Gaius (talk) 03:35, 6 January 2011 (UTC)

Strong Support These arguments are ridiculous; if you run IE, you should not expect anything to work IMO. But jokes aside, this is a great visualization of sorting algorithms that can help the reader understand the differences. In many cases, and especially so with our CS articles, the articles are too technical (for the average reader), involving runtime complexity, which data structures the algorithms are best implemented in, etc. As a Computer Scientist, I understand these are invaluable to the article. However, they are not "newbie friendly" at all. This provides a great top-level introductory view of sorting algorithms. Also, W/R/T to the larger demos taking longer to pre-process, I think that is a fine and normal limitation. To be honest, I'm surprised that such a trivial "inconvenience" perhaps really outweighs the benefit of deeper understanding of algorithms. Sheesh. jheiv talk contribs

Update: I ran this on IE and didn't experience any noticeable slowness. I have a pretty decent setup though, so that may explain the difference. My guess is that the warning is just there as a courtesy for users who have slowers computers and run IE. Certainly this is not a bad thing... jheiv talk contribs 01:40, 7 January 2011 (UTC)

Glrx I would like to come to a consensus with you before adding the link back in. I would ask that you list any remaining objections to inclusion. This has been dragged on long enough. Baltar, Gaius (talk) 18:43, 7 January 2011 (UTC)

You don't come to a WP:Consensus with me; that is something that happens with a group of editors. I clearly oppose the addition. I have stated reasons to support that position. Although I have some other objections, I don't believe they are worth going into right now.
Normally, the primary disputants don't declare a consensus; they let others do that to avoid the appearance of bias. This discussion started a long time ago, nobody declared a consensus, and the matter sort of died on its own. You keep trying to revive it with the hope of a different result. That's OK, but the landscape hasn't changed. If I read between the lines, User:Reaper Eternal believes there was not a consensus for inclusion because he suggested a RFC.
I'm opposed; it hung my IE browser. There have been some improvements, but I remain opposed. My take is that WP should not link to sites where the user can get into trouble. I don't expect the average WP reader to be sophisticated about using websites or avoiding trouble. Certainly, somebody learning about computing complexity for the first time may not appreciate the implications of sorting parameters. User:Reyk has voiced opposition and noted the simulations also hung Firefox. Reyk has moderated that position somewhat, but Reyk reports "random five to ten second hangs" with other simulations. Reyk is understandably reluctant to attempt viewing the simulations in IE. User:Ztothefifth hasn't declared either way, but Ztothefifth has confirmed the bubble sort causes "Firefox to hang for a while". (Ztothefifth's query about other simulations out there suggests he is reluctant to include the link.)
You are coming across as strong advocate for Joshua Kehn. That position troubles me. Instead of addressing the stability issues, you want to overlook them as "minor stability problems on a few platforms". Your statement that the JK link "promot[es] open standards" is irrelevant to this discussion; it suggests you support an agenda to get browser suppliers to support certain open standards. I don't think that is a goal of WP, and WP should not directing its users to sites with the hope they will complain about the limitations of their current browser or switch to a different browser. User:Reaper Eternal has voiced support for the link and did not experience trouble. I wish RE had said more, but that's fine. After the RFC, User:jheiv came in with strong support. Jheiv comes across as dismissive of anything that hangs IE (even with all joking aside), and that would prompt me to give his support little weight. He characterizes the simulations as not "newbie friendly". His second comment comes across as more unbiased.
As it stands, I don't see a consensus. Since the RFC, there's only been one addition for inclusion, and I don't see that addition creating consensus. At this point, with this set of editors, if both Reyk and Ztothefifth were to support inclusion, then I'd be overruled. Even if just one of two supported inclusion, you might have a consensus. If the link were included, I'd put a black box warning warning on it.
Glrx (talk) 21:02, 7 January 2011 (UTC)
PS. See also WP:ELNO numbers 8, 11, and 16. Glrx (talk) 21:20, 7 January 2011 (UTC)
I will address your concerns later this evening. Baltar, Gaius (talk) 21:46, 7 January 2011 (UTC)
First off, WP:ELNO 8 doesn't apply because <canvas> isn't a plugin. 11 doesn't apply because it's a dedicated page, not a personal site or blog. 16 doesn't apply because it is continuously functional, and will remain functional.
Your stated reasons (anti-IE / unstable) have been addressed. The page is not anti-IE. The page is stable. There are conditions that on older machines / inadequate browsers can produce unforeseen problems – but no more so then a Flash plugin crashing or a Applet refusing to load.
My take is that WP should not link to sites where the user can get into trouble. The users aren't getting into trouble. If you believe this is the case, why not remove every link to a Flash enabled or Applet page from WP because there are cases (especially with older machines) where these plugins will cripple a browser.
Certainly, somebody learning about computing complexity for the first time may not appreciate the implications of sorting parameters I agree, and it appears so does the author. The page is set to a default level which produces no reported crashes.
You are coming across as strong advocate for Joshua Kehn. That position troubles me Please enlighten me as to why you are troubled with my position. My position is something worthwhile and decent being so violently opposed by someone for ungrounded reasons.
Your statement that the JK link "promot[es] open standards" is irrelevant to this discussion; it suggests you support an agenda to get browser suppliers to support certain open standards. This may be your take on it but it is not my intent. My intent is to offer an alternative to the current which is widely accepted and in use. Why should there only be an Applet and a set of animated gif's?
I don't think that is a goal of WP, and WP should not directing its users to sites with the hope they will complain about the limitations of their current browser or switch to a different browser Where (anywhere, please) is this implied (by me or JK)? It surely isn't said, so it must be implied.
He characterizes the simulations as not "newbie friendly". He does not. He says In many cases, and especially so with our CS articles, the articles are too technical (for the average reader), involving runtime complexity, which data structures the algorithms are best implemented in, etc.
There is no need for a black box warning given that there is ample warning of any slight instability on the page itself.
I will contact Reyk and Ztothefifth via their talk pages and ask for for them to revisit this. Baltar, Gaius (talk) 02:53, 8 January 2011 (UTC)
• I remain opposed to the addition of this link. I've tested the page on Firefox and it frequently hangs for five to ten seconds regardless of the type of sort or the number of items being sorted. Once or twice it has even caused Firefox to stall completely, forcing me to shut it down in Task Manager. I do not regard this as a "minor instability" but a fairly serious one and I think it would be irresponsible to link to it. Finally I think you're treating the consensus building process as "keep talking and talking until everyone does what I tell them", and that's not how it works. Reyk YO! 05:23, 9 January 2011 (UTC)
Reyk, can you clarify which options you are using to create these hangs? I can't duplicate this behavior at all, even on my slowest machine. — Preceding unsigned comment added by Baltar, Gaius (talkcontribs) 13:05, 9 January 2011 (UTC)

Oppose crashed or stalled IE8. This is supposed to be an encyclopedia not a soapbox for some pinhead to tell me what software to run on my computer. If you think it is such a brilliant demo turn it into a gif and have done with it.Greglocock (talk) 04:27, 11 January 2011 (UTC)

WP:NPA? Reaper Eternal (talk) 23:11, 11 January 2011 (UTC)
It'd be NPA if the editor promoting the website link, and the website author, were the same person. I referred to the website author as a pinhead. Obviously they cannot be the same person as that would be COI, and we know that the wiki-editor is a man of integrity, not a pinhead, and so would not commit COI, or indeed, pinheadedness. Don't we? Greglocock (talk) 02:21, 12 January 2011 (UTC)
Note that the author ("pinhead") is not telling you what software to run on your computer. In fact, he is telling you that there is very few requirements, namely a browser made in the past five years. The demo running applets is telling you what to run on your computer.
Turning it into a GIF would completely destroy the uniqueness of the demo.
I don't particularly like what you are suggesting, but feel correcting you would be a waste of time. I'm thoroughly disheartened with the Wikipedia editing system now. Such violent opposition to anything new. The slightest (and do I mean slightest, I have only made it crash a handful of times) instability should be grounds for complete removal? Fine. I remain supportive of inclusion if this issue comes up again. Baltar, Gaius (talk) 05:14, 13 January 2011 (UTC)

## Bubble sort stability

I think bubble sort is stable (in contrary to what that written in the table) —Preceding unsigned comment added by 79.177.127.22 (talk) 09:55, 2 November 2010 (UTC)

I just reverted some changes that claimed BS stability depends on the implementation. BS should be stable because neighboring elements are not swapped if they are already in order (and two equal elements are in order). Consequently, equal elements will not get out of order. (They can get out of order when non-neighboors are swapped as in Shell short.) Sadly, my statements are WP:NOR, and I don't have a WP:RS citation for BS stability. Glrx (talk) 21:55, 7 November 2010 (UTC)

## Log Base

When it says n*log(n) what is the base of the log function? Is it binary (2), decimal (10), natural (e), or some other base? —Preceding unsigned comment added by 69.91.165.139 (talk) 20:04, 15 November 2010 (UTC)

In big O order notation, the base of the logarithm doesn't matter -- log2, log10, and ln are all related by just a constant. In particular algorithms, there will be a logical base to use: a binary sort would be base 2 and a radix 10 sort would be 10. But again, in terms of big O, the actual base doesn't matter; constant multipliers are discarded. Glrx (talk) 20:49, 15 November 2010 (UTC)

## Manual (Physical) Sorting Algorithms

I came looking for assistance in manually sorting tens of thousands of random Election Ballots into a hundred geographic (Precinct) districts. Similarly the Postal Services machine-sort millions of pieces of mail into millions of buckets (zipcodes, routes and addresses). Carrier personnel manually direct-sort their mail into an array of "pigeon-holes" before delivery distribution.

Unfortunately all emphasis here is towards electronic Data-processing. I've got to give these lots of thought to determine applicability to physical-sorts, using physical motions, requiring real-time, and limited buckets within reach or operator constraints. Does EDP sort-theory apply, or have lessons for large physical sorts?

Are info or comments about best physical-sort algorithms appropriate here? What differences apply due to human or mechanical vs electronic/cyber media and speeds? HalFonts (talk) 07:20, 19 November 2010 (UTC)

## TimSort

The best case for TimSort should be O(n). It is not just a hybrid mergesort but a hybrid natural mergesort. If the data contains runs that are ordered or reverse ordered or a mixture, then it is between O(n) and O(n * log n). See http://bugs.python.org/file4451/timsort.txt

The same can be said for SmoothSort. If the runs are ordered (but not reverse ordered), then it is O(n). The best cases on these should reflect this. Stephen Howe (talk) 22:39, 25 November 2010 (UTC)

## Shell sort

What's happened to this section? Just two curly brackets at the moment. Sam Dutton (talk) 14:01, 26 November 2010 (UTC)

How does it have two different averages? —Tamfang (talk) 07:43, 15 December 2010 (UTC)

Because it is an open problem as to what the asymptotic behaviour of shell sort is. Different gap sequences have different asymptotic behaviour. I will see if better bounds can be found as I believe recent publications by computer theorists have managed to give tighter bounds for shell sort.

Stephen Howe (talk) 21:25, 20 December 2010 (UTC)

## Quicksort can be stable?

I dont think this is true, it is only true for quicksort for linked lists. It is not how the pivot is handled, as the article describes, but the fact that all equal elements within the partition set must retain their original order while partitioning. To do so, requires a stable partition (function in C++ library). stable partition requires auxilary memory and cant be done in O(N) without the memory. Any comments?

Stephen Howe (talk) 21:46, 10 December 2010 (UTC)

See Quicksort#Complex version. Glrx (talk) 05:28, 11 December 2010 (UTC)

Stephen Howe (talk) 22:37, 12 December 2010 (UTC)

## Perfect Sort / Triangulation Sort

Is a kind of un-algorithm. It simply triangulates from what data is so far sorted (discouraged / takes extra time). But implemented (see code) it shows shows some surprising benefits. Is sometimes referred to being "a sort in which data becomes available for streaming rapidly" (and formidable multi-field). Some books refer to such a method but I haven't seen it one of them code it or compare it's benefits. Interesting for just that. Enjoy. User:Sven nestle2/Perfect Sort / Triangulating Sort

You need to cite WP:RS for this algorithm. Who has evaluated this algorithm? Also watch out for WP:NOR. Glrx (talk) 21:40, 6 June 2011 (UTC)

Excuse me Glrx did you have time to see my corrections yet?

• I've completely rewritten please look ***

I am trying to provide a good topic. Please say if there is any reason to hold it back.

Wikipedia:Identifying reliable sources. "~ the reliability of a source depends on context. Each source must be carefully weighed to judge ~" I don't see other sorts showing publisher. Wikipedia:No original research. "~ that includes any analysis or synthesis of published material that serves to advance a position not advanced by the sources ~". I'm sure I didn't make any unfounded conclusions. Triangulation - very basic stuff.

But that's parsing hairs. The original post was bad and I'm not sure if you've looked at the new post.

User:Sven nestle2 has continually edited comments out of order, deleted other editors comments, and changed his comments after they received responses. The record for this thread is a mess. This thread starts with this edit.
Bottom line: Sven has not identified any reliable source for his perfect/triangulation sort.
Glrx (talk) 16:41, 12 June 2011 (UTC)
Ah thanks again for the communicae. Mr Glrx I'm unsure what you mean "edited out of order" as my article was or is a work in progress as is yours. If your implying my wiki editing skills need improving I'm sure to admit it. I'm unsure of your use of "thread" as wiki does not resemble chat forum software. My user page should not be an article page is that it? I was going to spend that time fixing later does it matter?
<< But what about the content as it stands now? Any likes or dislikes there? << ?
Sven nestle2 (talk) 20:18, 12 June 2011 (UTC)
What you put on your user page is (mostly) your own business.
Each section on a Talk page is functionally a conversation thread; the lack of inherent forum-style structure obliges us to carefully maintain conventions, among which perhaps the most important is not to damage old comments, because each participant needs to be able to see (and understand) what has gone before. Therefore: try to to place new remarks where they fit the flow, and do not remove or replace old remarks (unless it's yours and you see that you made a mistake before anyone responded). —Tamfang (talk) 18:47, 13 June 2011 (UTC)
Sven writes, "I don't see other sorts showing publisher." Each of the sort methods listed in the article has its own article, which ought to list references; they are not necessarily repeated in this umbrella article. —Tamfang (talk) 19:06, 13 June 2011 (UTC)
Sven, in your exuberance to rewrite your own remarks you removed mine:
What books mention this algo? Any webpages (other than your own)? —Tamfang (talk) 07:21, 11 June 2011 (UTC) restored—Tamfang (talk) 18:47, 13 June 2011 (UTC)

I haven't tried to read the code at User:Sven nestle2/Perfect Sort / Triangulating Sort, but from the verbal description it seems that you're merely giving an exotic name to multikey comparison (if x.key1==y.key1 then compare x.key2 with y.key2, and so on) which can be incorporated into any sort algo. —Tamfang (talk) 18:53, 13 June 2011 (UTC)

Mr. Tamfang thank you. As to removal of past edits yes I was trying to keep my talk section small but I see your point. I'm sorry if my article is not easily instructive I try to make it clear soon. I note you are not Glrx who I asked and have no response from.

As to your multi-key comment you are somewhat right: what you listed would work in one dimension (but yields no new index in one dimension). However it would not work in two and higher dimensions as some way of indexing must be found. It must specify AND track traversal some way. Many sorts do this with assumed partitions to specify and recursion to track. Recursion invisibly "stores" traversal indexing but with poor efficiency. It is amazing that "Triangulation Sort" keeps up near highest speeds and can beat other sorts at a few things isn't it? Recursion wastes but also looses order (you cannot output until all records are finished with such methods) (also: debugging an unusual orders can be frustrating). A sort cannot waste indexing time and still have it. Triangulations sort's best features are not "raw time" - as stated above and on it's page.

If however you believe sorting two and higher dimensions is not a problem and requires no indexes please say how :) I'd love to hear it. Also I do not see a reference for "multi key comparison" algorithm. Is there one other than "triangulation sort"?

— Preceding unsigned comment added by 68.227.219.57 (talk) 20:45, 14 June 2011 (UTC)

Oh, I misunderstood the point of "dimensions". When is it appropriate to sort in two or more dimensions? —Tamfang (talk) 06:33, 15 June 2011 (UTC)

Who said all data must be one dimensional? Do not believe this person. — Preceding unsigned comment added by 68.227.219.57 (talk) 16:08, 17 June 2011 (UTC)

Do you always answer a question with a non sequitur? —Tamfang (talk) 02:26, 29 June 2011 (UTC)

## Library sort -- 2004 or 2006?

The overview says library sort was first published in 2004, but the library sort page says 2006. Which is correct? John lindgren (talk) 13:49, 10 June 2011 (UTC)

## Spaghetti (Poll sort)

I found the general algorithm for spaghetti sort, as well as linking articles. — Preceding unsigned comment added by Ahshabazz (talkcontribs) 00:11, 27 June 2011 (UTC)

I just removed Spaghettisort from the list, because by the wikipedia article that describes it, it is not a real sorting algorithm. Spaghetti sort requiers that a "massively parallel CPU system" needs to be able to find the longest spaghetti in o(1) time, in order for this to work, but without describing that part of the sorting algorithm, the description is incomplete. There is no such thing as an o(1) sorting algorithm, so the precondition that makes "Spaghetti sort" able to exist as an algorithm, does not exist. In other words "Spaghetti sort algorithm" is not a Computer Science algorithm, but a method with spaghetti, that would require a sorting algorithm on a computer, which is not related to spaghetti. Dybdahl (talk) 06:25, 7 July 2011 (UTC)

• Remove spaghetti sort from the article. In the interim, I moved it to the impractical group requiring special hardware. I also changed the space metric to reflect reality of an analog comparison. Glrx (talk) 18:15, 7 July 2011 (UTC)

I agree with removing it from this article, but note that the requirement is O(1), not o(1). Finding the largest strand in time O(1) is impractical, requiring custom purpose parallel hardware scaling with problem size; finding it in o(1) is impossible. CRGreathouse (t | c) 03:50, 7 October 2011 (UTC)

• Keep. I agree that the spaghetti sort, as described in the article on spaghetti sort, is not useful for real computer programs running on classic binary von Neumann computers. However, as far as I can tell, this "sorting algorithm" article is about all kinds of sorting algorithms, whether or not they are useful for computer programs on classic binary von Neumann computers, so I am inclined to keep at least a mention of spaghetti sort in this article. --DavidCary (talk) 21:16, 10 March 2013 (UTC)

A question - spaghetti sort as currently listed claims O(n^2) memory, but its article claims O(n) memory, which seems far more intuitive to me as each value is only tracked once (on its own processor). Is there a reason for this difference? — Preceding unsigned comment added by 12.184.104.214 (talk) 04:57, 21 December 2012 (UTC)

## Sleep sort

I find this sorting algorithm to be amusing:

http://rosettacode.org/wiki/Sorting_algorithms/Sleep_sort — Preceding unsigned comment added by 174.137.113.43 (talk) 00:50, 20 July 2011 (UTC)

This should be included as a non-comparison sort, as there isn't anything there that compares to the slow comparison sort.. 'slow sort' at the moment.--31.185.238.2 (talk) 00:38, 12 March 2012 (UTC)

## add information about online sorting algorithms

that... you could add in the table if an algorithm is online or not. http://en.wikipedia.org/wiki/Category:Online_sorts Mdtr (talk) 03:26, 3 August 2011 (UTC)

## Clarification of Notation

I would appreciate a clarification, either on this page or the logarithms page, of what is meant by $\log^2(N)$. KlappCK (talk) 15:00, 30 August 2011 (UTC)

The equation has been clarified to be n*(logn)^2, which matches what is stated on STL's webpage and in STL books (Duvavic1 (talk) 01:08, 14 September 2011 (UTC))

## Duplicates-Removing Sorts

According to the definition given at the start of the article, duplicates-removing sorting is not sorting. But that kind of algorithms exist; some languages/run-time systems have built-in sort functions with the possibility to specify whether it is needed to remove the duplicates at the same time while sorting, or to preserve them.

One example: pigeonhole sort with removal of duplicates is essentially what the array-based sieve of Eratosthenes is doing while collecting ⁄ marking the composites, and list-based variant exists which performs duplicates-removing mergesort.

Where and how can this be addressed? Or is it already, somewhere? Any pointers, anyone? Thanks. WillNess (talk) 08:20, 22 September 2011 (UTC)

Duplicate-removing sorts can be done in 2-steps: 1. The sort is done 2. An extra pass is made comparing an element with the next element. If the next element is identical on the key, then it should be overwritten by the next successive element. So for identical elements, the first element is preserved. Stephen Howe (talk) 21:31, 28 January 2012 (UTC)

## A New Sorting Algorithm: Over-sort

A I just published a new sorting algorithm I worked on for the last 18+ months. I think Over-sort is past "worthy of notice", and into ground breaking territory, but I'm probably biased. — Preceding unsigned comment added by 84.229.4.103 (talk) 08:24, 14 October 2011 (UTC)

## standard template library stable_sort() is not always in place merge sort

Standard Template Library includes a stable_sort() function, but does not define how it is implemented. In the case of Microsoft's Visual Studio, stable_sort() uses temp vectors (double the memory space, not an in-place merge sort), uses insertion sort to create chunks of 32 elements, then merges chunks until chunk size equals vector size. The sort algorithms section should not include a reference to Standard Template Libaray as an example of an in place merge sort. A specific implementation of STL's stable_sort() might use an in-place merge sort, but stable_sort() is not defined that way. Rcgldr (talk) 02:06, 9 December 2011 (UTC)

The ISO C++ standard (1998 and now 2011) does give implementation constraints on how stable_sort should be implemented. It says "Complexity: It does at most N log2(N) (where N == last - first) comparisons; if enough extra memory is available, it is N log(N)." It works out that 3 STL algorithms are adaptive: inplace_merge(), stable_partition(), stable_sort(). They are allowed to acquire a temporary buffer and if the memory is available, then the minimum complexity is achieved. If the memory is not available but a smaller amount is available, then the algorithm degrades gracefully in the direction of the maximum complexity. In the case of Visual Studio, it uses a hybrid merge sort where the initial step is various insertion sorts before merging the results back and forth to the temporary buffer. For VS2003, VS2005, VS2008 the temporary buffer is the size of all the elements. For VS2010, the temporary buffer is 0.5 the size of all the elements as the well-known trick of merge-sorting 2 halves and then merging one half in the temporary buffer with the other remaining half is employed. This cuts down the overhead. Any algorithm that meets the complexity and stability requirements of ISO C++ can be used but most STLs use a hybrid adaptive mergesort. Stephen Howe (talk) 22:02, 28 January 2012 (UTC)

## Implementation code improvements for all sort algorithms

Before you start the sorting of the N given elements you should check if the given elements already have the desired sort order (check if (a[i-1] <= a[i]) for all i=(1, .. N-1)). This additional check operation has only a linear run time complexity O(N). That means if the given array is already sorted (this is always the best case) so the run time complexity is always O(N). This additional check operation can be executed at the begin of each sort algorithm. That means in particular that each sort algorithm in the best case always has a linear run time complexity O(N).

Aragorn321 (talk) 05:54, 19 July 2012 (UTC)

## Yellow boxes in the "best case" column.

Currently, this column has only red (n^2 or worse) and green (better than n^2). Should nlog(n) be made yellow, for consistency's sake? 140.247.41.147 (talk) 13:48, 23 July 2012 (UTC)

## BULL

This "page maker" is deleting or mis-representing code for faster sort algorithms. Never stating where they came from or sources. Giving confusing explinations for advanced algorthms and leaving out practical issues such as found in unix sort(1) (ie, stability, tape v. memory, caching). Deleting sumbitted old and new algorthims (ie, mine but others). But has publishes "timsort" which is a joke algorithm. Say he has a tester and all algorithms but has no evidence of having the same. And the sources cited have 0 to do with the origion books and creators of the algorithms.

## Quickersort AND Quickestsort

August 31, 2012

To whom it may concern (at Wikipedia),

Sincerely yours,

JPD 22:15, 31 August 2012 (UTC)

### Sort in random

\$ echo "Wikibooks Wikipedia Wiktionary" | sort --random-sort
Wikipedia
Wikibooks
Wiktionary


### Unix Sort's algorithm

The sort in GNU/free Linux uses merge sorting which is explained in the Sorting algorithm wiki page.

### new development of Sort

Sort uses merge sorting and is speedy to complete 1 column sorting (in a table of rows and colums of words to be selected and sorted).

There is a new sort(1), headsort(1), using an algorthim with opposite speed properties of merge sort that when sorting more than one column can finish in 1/2 the time, or when only needing to begin streaming the sorted data can finish in 1/2 the time (thus head), and can fork sorting jobs better (though neither does by default): but more often looses to sort's merge when sorting 1 column: it has opposite benefits of merge. Streaming and columns headsort(1) can being streaming in less than 1/2 the time. It's used when use columns and piping data are more often than not used.

If you are interested in headsort(1) "1/2 the time" contact johnandsara2@cox.net or sven_nestle2 on wiki with subject "sort algorithm".

It's good to note that for top speed no algorithm is needed just memory: however quite impractically allot is need for any appreciable task especial where used for random tasks; such a thing is useable only within a single program.

## ok, let me try describe Euclid sort

Euclid sort sorts in O(N*log(N)) time n needs O(1) mem first we got two groups of sorted items to b merged lets try to describe backward possible variant: first we c what is the last item (in reverse order) from the second group that remains at its place second we c where is moving up the first contestant , in reverse order, by following in reverse order, the elements of first group third we c how many elements from the second group will come compactly with the contestant that i mentioned fourth we apply some Euclid arange function to put the elements in the rite place when exit the loop we probabily will have at least an Euclid arange (as much, but no more) to get the array sorting done

idd wish credits if the case Florin Matei, Romania 93.118.212.93 (talk) 07:36, 22 February 2013 (UTC)

Please read WP:OR; if this were published, it could be added (provided we can figure out what it means). — Arthur Rubin (talk) 16:55, 25 February 2013 (UTC)

ok no problem, suit urself abt that policy of urs i dont think its wrong or anything i was just trying make a decent buck, noone i know likes to publish my ideas, like i said im a poor encoder also thow i got a few ideas, u might b ceen my ip page thank u... n im not playing the smartes nor the poorest here: i just feel like found a way to speak abt my creativity. Florin 93.118.212.93 (talk) 15:36, 27 February 2013 (UTC)

## an evn more simple sorting algo that challenges O(N*log(n)) time n o(1) mem

its all abt merging two sorted groups: main repeated action is to move same number of elements from group first n group second (group interchange same no of elements) so we'll have all the elements of groups (sorted in the future) but not in place, then well repeat starting with the group of less length 4 not charging that much the stack... as a strategy we can alternate between remained smaller group n the bigger 1:1 alternate that assure O(log(n)) stack with some strategy against hazard. when memorize parameters of a group in the stack , memorize also the "breaking point" also

 93.118.212.93 (talk) 08:28, 4 March 2013 (UTC)


## Is there a name for this sort algorithm?

This sort only works on a sequential set of numbers in some random permutation, using swaps to sort the set of numbers. The worst case number of swaps is n-1. Example code for when the set of number is {0, 1, ..., n-1} stored in an array a[]:

void sort(void)
{
int i;
for(i = 0; i < n; i++)
while(   a[i] != a[a[i]])
swap(a[i],   a[a[i]]);
}


Several

Rcgldr (talk) 09:18, 27 April 2013 (UTC)

Infinite loop? (Made a mental off-by-one error here...) —Ruud 10:38, 28 April 2013 (UTC)
Given the constraint that the array of size n contains a permuted set of numbers {0, 1, ... , n-1}, the worst case number of swaps is n-1. I've since figured out that this algorithm can be expanded to sort an array based on a set of sorted indexes:
//      A is an array of size n
A[] = ... ;

//      generate array of indexes I
for(i = 0; i < n; i++)
I[i] = i;

//      sort I based on values in A
sort(I, A);

//      sort A using the sorted list of indexes in I
//      while also sorting I
for(i = 0; i < n; i++){
while(     I[i] !=  I[I[i]] ){
swap(A[I[i]], A[I[I[i]]]);
swap(  I[i],    I[I[i]] );
}
}


Rcgldr (talk) 01:04, 29 April 2013 (UTC)

Cycle sort would seem like an appropriate name, but it has already been taken, although it does seems to be based on the same underlying principle: "It is based on the idea that the permutation to be sorted can be factored into cycles, which can individually be rotated to give a sorted result." I wouldn't be surprised if this sort doesn't have a name, as it seems to be mostly of theoretical interest: with the given preconditions it shouldn't be too hard to keep the list sorted in memory in the first place. The extension of the sort to work on key-value pairs can be applied to any sorting algorithm. —Ruud 17:10, 29 April 2013 (UTC)
It's a special case of cycle sort: special_cycle_sort in B. K. Haddon, "Cycle-Sort: A Linear Sorting Method", The Computer Journal, 1990. —Ruud 17:16, 29 April 2013 (UTC)
Since it swaps instead of rotating, maybe it should be called swap sort? I forgot to mention that each swap places at least one element in it's "correct" location. Rcgldr (talk) 07:55, 30 April 2013 (UTC)
Most in-place sorting algorithms, including bubble sort and quicksort, work by swapping elements. This specific algorithm, though, goes through each cycle in the permutation one-at-a-time. Thus the number of iterations of the for-loop in which you perform at least one iteration of the nested while-loop is always exactly equal to the number of cycles in the permutation. —Ruud 11:54, 30 April 2013 (UTC)
• One interesting aspect of this small loop is that if you consider sorting I by the values of A as a permutation, then this loop "unpermutes" I, while at the same time it "permutes" A. This occurs because the values in the locations swapped in I are used as indexes for the locations swapped in A. Rcgldr (talk) 23:04, 29 April 2013 (UTC)
So the main enhancement of this algorithm is the part that sorts the array A[] while performing the special cycle sort on I[]. I searched again but wasn't able to to fing any references citing that special cycle sort could be used for this purpose. Rcgldr (talk) 04:01, 1 May 2013 (UTC)

## Adding explanation for in-place sort

There should be a section explaining what an in place sort is just like there is a section for explaining stability of sorting — Preceding unsigned comment added by Nileshrathi01 (talkcontribs) 14:07, 5 October 2013 (UTC)

## 0-1 principle

From my university studies I recall there being something called the 0-1 principle. The statement is something like this: "If an oblivious comparison-based sorting algorithm correctly sorts data consisting of only 0s and 1s, then it correctly sorts any data."

I forget the exact wording, but the idea of it is that, if an algorithm meeting certain criteria works as a sorting algorithm for arrays/lists containing only two distinct values, then it works as a sorting algorithm for lists whose elements are taken from an arbitrarily large totally-ordered set.

Does anybody know anything about this? The search hits I've found are explicitly about sorting networks, but what I was taught referred to sorting algorithms more generally. Obviously not all 0-1 sorting algorithms are general sorting algorithms - counting sort with only two counters is an obvious counter-example (pardon the pun). But what are the "certain criteria"? Something that can be said about most sorting algorithms is

• the only way in which the elements are examined is to use a "less than" or "greater than" operator, which has a boolean result
• elements are moved from place to place, rather than making many copies of an element and using these as the output

But the aforementioned counting sort can be adapted to meet these criteria, so they are clearly not sufficient.

It may be the case that it is only a theorem for sorting networks, and my lecturer just intended us to use it as a heuristic test to see if a sorting algorithm seems correct. Still, can anyone provide any further insight into this, such that we might be able to add something to the article about it? — Smjg (talk) 17:42, 30 November 2013 (UTC)

@Smjg: A little late for this, but if you do manage to find a reliable source explaining this, I'm sure it could be integrated into the article. – FenixFeather (talk)(Contribs) 04:36, 11 April 2014 (UTC)
That's kinda obvious.... — Smjg (talk) 14:52, 12 April 2014 (UTC)

For what it's worth, I've just tried searching again and found this: [2]

"The most interesting proof uses the so-called 0-1 Sorting Lemma. It allows us to restrict our attention to an input of 0’s and 1’s only, and works for any “oblivious comparison-exchange” algorithm. (Oblivious means: Whether you exchange two values must only depend on the relative order of the two values, and not on anything else.)"
"Whenever the algorithm compares a pair of 1’s or 0’s, it is not important whether it exchanges the values or not, so we may simply assume that it does the same as on the input x."

What is important, however, is whether the algorithm has any conditional behaviour, other than just the swapping of the elements, based on the result of the comparison.

Maybe "oblivious comparison-exchange" was the phrasing my lecturer used. I guess that this phrase means an operation of the form "if A > B, swap A and B", and the essence is that the data is in a black box and the algorithm has access to it only by this operation (which returns no value, hence "oblivious") plus knowledge of the length of the array.

This means that, for a given algorithm and input size, the sequence of oblivious comparison-exchanges is a constant. So what we really have is an algorithm for building a sorting network and then applying it. That said, on a given pair A and B, it can equally be either "if A > B, swap A and B" or "if B > A, swap B and A". So if you don't consider sorting networks to be allowed to have reverse comparators, then we have a generalisation of the sorting network 0-1 principle to a wider class of algorithms. Otherwise, it doesn't really generalise the principle at all.

Maybe what we can still say is that the 0-1 principle can be used to prove the correctness of some sorting algorithms. I'll see what I can come up with.... — Smjg (talk) 14:52, 12 April 2014 (UTC)

## Sorting networks section still needs work

"Sorting networks" aren't a singular algorithm that can be described as a single best/worst/average case for time and memory, but they're listed as such in the chart. They're also listed in a special category for sorts that are incredibly impractical or require special hardware. That may be true for the AKS sorting network it's apparently referring to, but small and efficient sorting networks are actually a lot faster than insertion sort, even though insertion sort is credited as being the fastest. In my experience they're anywhere from 2 to 5 times faster than insertion sort.

What should be done about this? My recommendation is to remove sorting networks from the chart entirely and only refer to them as a concept that can be applied to many sorting algorithms (AKS, even-odd, bitonic, bubble and insertion, etc.), either to construct hardware or to make an algorithm for fixed-size data sets. — Preceding unsigned comment added by BonzaiThePenguin (talkcontribs) 03:40, 11 April 2014 (UTC)

It doesn't seem like there is a reference supporting its position in the table, so if you have a good source that describes sorting networks as a strategy that can be applied to many different sorting algorithms, I think you could easily make that change. – FenixFeather (talk)(Contribs) 04:34, 11 April 2014 (UTC)