On Wed, Jun 3, 2009 at 8:37 AM, Steven Ross <spreadsort@gmail.com> wrote:
Is insertion_sort to std::sort such a big complicated change that it can't be accepted?
Do you have a proof for it's correctness? If not, perhaps it would be easy to modify the proof(s) for the algorithm you modified.
I've described the changes in string_sort vs. American Flag Sort already,
Is there a publication you can reference? My apologies if you already posted it.
It would get expensive to publish a paper for every little tweak I've stuck in a sorting algorithm, so depending on publications for every improvement is a great way to slow down progress, especially if you demand that people cite said publications.
It's also very expensive to standardize on broken algorithms and interfaces. Publishing each little tweak in a separate paper is certainly a Bad Idea. However, if you think the "package" is good enough for a boost library, then it is certainly good enough for a publication.
People seem to get the impression from his book that comparison-based sorting is the only general way to sort, ...
"People" are wrong all the time. ;-)
a worst-case fallback check (integer_sort/float_sort), falling back to std::sort an optimization for long substrings (string_sort) float to integer cast (float_sort) assorted performance tweaks
Are those changes really so dangerous?
This is Math, not politics... Let's see a proof. ;-) Jon