topN using a heap

[Andrei Alexandrescu via Digitalmars-d]

On 1/16/16 9:37 PM, Timon Gehr wrote:
> Ivan's analysis suggests that even something significantly larger, like
> n/log(n)² might work as an upper bound for k.

I'm not clear on how you got to that boundary. There are a few 
implementation details to be minded as well (quickly and carefully 
approximating the log etc). Could you please make a commented pull 
request against my own? Thanks! -- Andrei