Re: Wording of the rationale Per Bothner 08 Nov 2006 02:53 UTC

Jens Axel Søgaard wrote:
>        * Four varieties of algorithms were provided (quick, heap,
>          insert, merge) even though quick, heap, and insert sorts have
>          no significant advantage over merge-sort.
> ...
> Second: What does "no significant advantage" mean? I were of the
> impression, that the "hidden constants" of O(n log (n)) of
> vector-quick-sort were smaller than that of vector-merge-sort.

My impression is that is non-trivially faster when you're sorting an
array of integers, with no indirection through a "compare" function.

In practice, one (almost) never sorts an array of integers (one sorts
an array of records/objects - which may have integer fields), and one
usually indirects though a compare function.  That changes the "constant
factors" significantly.

It is illustrative that java.util.Arrays uses a "tuned quicksort" for
sorting arrays of "primitive" (unboxed) numbers, but a "modified
mergesort" to sort Object arrays.

 > Also: Is merge-sort the fastest algorithm, if the elements are "almost
 > sorted"?

 From the Java documentation:

   This [modified merge-sort] algorithm offers guaranteed n*log(n)
   performance, and can approach linear performance on nearly sorted
   lists.
--
	--Per Bothner
xxxxxx@bothner.com   http://per.bothner.com/