Comment 17 for bug 1246802

(In reply to Chris Jefferson from comment #12)
> The better (in my opinion) fix is to change __introselect from:
>
> __last - __first > 3
>
> to:
>
> __last - __first > int(_S_threshold)
>
> Like the other call the __unguarded_partition_pivot.

Note that the optimal threshold is probably not the same for sort and for select (which drops to full sorting below that threshold!).