Re: [boost] Proposed templated integer_sort

From: Steven Ross (spreadsort_at_[hidden])
Date: 2008-07-11 15:36:27

On Fri, Jul 11, 2008 at 10:43 AM, Phil Endecott <
spam_from_boost_dev_at_[hidden]> wrote:
>
> Are you certain that resizing the bins_cache vector when you do is safe,
> from the point of view of subsequent iterations of the loop which continue
> to use a pointer to the first element from before the resize()?
>

Oops.  Good point.  The bins pointer is no longer safe as soon as a
recursive call is made; it didn't crash because reallocs didn't change the
vector location.   I should use the vector (which will stay valid)  when I
start making recursive calls, not the bins pointer, and the performance
impact should be trivial, as that's a loop over the bins (n/512), not all
items as the other loops are.   I'll fix that.
I'll also fix the minor syntax issues you and David pointed out.  Anything
else?