Re: [boost] Proposed templated integer_sort

From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2008-07-10 12:21:32

> __last and __first were copied from std::sort.  Is there a good reason I
> shouldn't use the same notation?

Yes; you're not writing the standard library.

> I'd like to use a smart pointer, but is there one that will call delete[]?

With the way that you've now structured the code, I think your choice 
is between a boost::scoped_array (the difference between scoped_ptr and 
scoped_array is delete vs. delete[]), or to avoid dynamic allocation by 
using a std::vector.  If the number of bins is knowable at compile time 
then a std::tr1::array or C-style array is another choice, but some 
people might complain about having so much data on the stack.

> Is there a simple way to check if an iterator is a vector iterator?

Something like this (probably full of syntax errors....)

template <typename Iter>
void SpreadSort(Iter begin, Iter end) {
   .... generic code ....
}

template <typename T>
void SpreadSort< std::vector<T>::iterator >( std::vector<T>::iterator begin,
                                              std::vector<T>::iterator end)
{
   SpreadSort(&(*begin), &(*end));
}

I have previously asked, probably on this list, about whether there's a 
good way to detect iterators that point to contiguous storage.  The 
answer seems to be that there isn't, except for explicitly detecting 
those that are guaranteed to do so (as above).

Phil.