std::stable_sort()

**cpjust** · 01-10-2008

I've been re-reading Scott Meyer's chapter about STL sorting algorithms, and I was wondering when someone would want to use stable_sort() instead of just sort()?
If two objects are equivalent, what difference does it make which one precedes the other? They are equivalent after all...
So far, the only case I could think of for using stable_sort() is if you need to sort a container twice -- once for one criteria, then again for another criteria and you don't want to lose the sorting done in the first sort.

Can anyone tell me what other situations you would use a stable_sort() for and how often people actually use it in their code when they need to sort something?

**laserlight** · 01-10-2008

So far, the only case I could think of for using stable_sort() is if you need to sort a container twice -- once for one criteria, then again for another criteria and you don't want to lose the sorting done in the first sort.

I believe that is general idea, since stable_sort only makes sense when objects have more than one property. Of course, the container (or rather, range) may not even be sorted by the program to begin with.

**matsp** · 01-10-2008

Consider that you have a "compare" operator that doesn't compare ALL parts of the object - say you have a

Code:

class person
{
    string name;
    string street;
    string city;    
public:
    ....
    bool operator>(person &p) {
        return p->name > name;
    };
};

Now, you may not want to swap the "Steve Jones" that live on "London Road" and the "Steve Jones" that live on "Newport Road"?

--
Mats

**Daved** · 01-10-2008

Imagine a grid or table that is sorted based on whichever column the user clicks. If the user wants to sort by name and then date, first they click on the date column and the data is sorted by that property, then they click on the name column and stable_sort is used to make sure that all the rows with the same name are sorted by date.

I often wish the Windows Explorer had this feature.

**cpjust** · 01-10-2008

Originally Posted by Daved

Imagine a grid or table that is sorted based on whichever column the user clicks. If the user wants to sort by name and then date, first they click on the date column and the data is sorted by that property, then they click on the name column and stable_sort is used to make sure that all the rows with the same name are sorted by date.

I often wish the Windows Explorer had this feature.

OK, that's basically the use-case I came up with.
I don't write GUI's since I have absolutely no artistic style, but I suppose column sorting is one place stable_sort() would (or should) be used a lot.
But are there any cases you'd use it other than sorting based on multiple criteria (i.e. sorting using a single call to stable_sort())?

**Daved** · 01-10-2008

The only reason I think you would use it is if you had an existing sort applied to the data that you wanted to keep for elements that are equivalent in the new sort. That would only happen if there are more than one sorting criteria.

Technically, you can be using more than one sorting criteria but still only call stable_sort once. For example, if you insert data in a particular order and want to sort it but leave equivalent values in the order they were inserted then you would call stable_sort just the one time.

**iMalc** · 01-10-2008

An early version of UT2004 used to use a non-stable sort for sorting the list of servers by ping time every now and then, as more were pinged. The problem was that there might be 3 say that all have a 50ms ping time exactly, and those 3 would constantly change order whilst others were being added to the list and the list resorted. This made it hard to click on the right one. Quite likely this was fixed by using stable_sort. Though they may have simply added disambiguation criteria to the comparison function.

That's the best real-world example I have. Actually it's the example I have on my website.

**cpjust** · 01-10-2008

Originally Posted by iMalc

An early version of UT2004 used to use a non-stable sort for sorting the list of servers by ping time every now and then, as more were pinged. The problem was that there might be 3 say that all have a 50ms ping time exactly, and those 3 would constantly change order whilst others were being added to the list and the list resorted. This made it hard to click on the right one. Quite likely this was fixed by using stable_sort. Though they may have simply added disambiguation criteria to the comparison function.

That's the best real-world example I have. Actually it's the example I have on my website.

Aha! I knew there had to be some other reason other than double-sorting. Thanks.

**anon** · 01-10-2008

imalc, if I understand you correctly that might be solved by using insertion sort? (Which appears to be a stable sort, though.)

I am actually using insertion sort for high-scores in a game. The stableness is important, except that most recently added equivalent scores would get the highest place.

**Daved** · 01-10-2008

Off-topic, but in a high score list the most recently added equivalent score should get the lowest place.

**anon** · 01-10-2008

You are probably right.

However, I recorded only the 10 best results, and I figured that once the highscores get reasonably high, you'd at least get the satisfaction of entering your name (which defaults to currently logged on username to save some trouble) and being congratulated on making it to Top 10 if you are able to repeat the 10th score. Especially once I went through the trouble of highlighting the most recently added line.

I mean, you got to give people motivation to play your game

-------------
But anyway, my point is that highscore tables are another practical example where stable sort is needed.

**iMalc** · 01-10-2008

Originally Posted by anon

imalc, if I understand you correctly that might be solved by using insertion sort? (Which appears to be a stable sort, though.)

I am actually using insertion sort for high-scores in a game. The stableness is important, except that most recently added equivalent scores would get the highest place.

Insertion Sort is stable, so long as you are careful about your use of comparisons (e.g. < vs <=) and get the arguments the right way around.
You can check against the implementation on my website if you want. Link in sig.

Your highscore example is definitely another example of where a stable sort (or part of one) should be used. You can add the new highscore at the end, perform an insertion sort (or just the last pass), and then drop off the last item.

Thread: std::stable_sort()

Thread Tools

Search Thread

Display

std::stable_sort()