performance analysis producer_consumer.cc TEST_T5

[Theodore Papadopoulo]

dhruvbird@gmx.net said:
> Why can't the library have a configure option to determine whether the
> size() function will be O(N) or not. It would be easy to provide both
> versions in the library AFAIK. 

And why can't we keep the size and incrementally keep it when it is 
cheap to do so. When it is not (eg for splicing), we invalidate the 
size and reestimate it (in 0(N)) the first time the size of the list 
is requested.

This way people would have an O(N) cost in general and (hopefully often) a 
constant cost between "safe" operations...

Now, this incurs a test and an increment for each addition/deletion 
and I do not know the average cost increase because of that....

--------------------------------------------------------------------
Theodore Papadopoulo
Email: Theodore.Papadopoulo@sophia.inria.fr Tel: (33) 04 92 38 76 01
 --------------------------------------------------------------------