On 2/12/2012 7:03 AM, John M. Dlugosz wrote:
On 2/10/2012 11:39 AM, Bill Buklis wrote:
What am I missing? How can count_if be properly expressed as count. To me it seems like it can't.
I agree. "count returns the number of elements x in rng where x == val is true." If you filter the original range using the predicate instead, you don't want to then compare them all against some constant. If you did so, you would need a special value such that every one of them compared true.
You just want to count the number of items in the range you get after filtering. The obvious boost::size requires a bidirectional range, and is probably less efficient then we would want.
Nathan mentioned boost::distance, which is hard to find in the documentation. But I see it listed on the synopsis under Forward Range functions. I guess it's missing from the page doc/range/concepts/forward_range.html
In general, it is true that having two algorithms "do something with or to each element" and "same, but selecting elements via a predicate" can be generalized to using the first (only) with a range filter.
But count and count_if are not such a pair. count also has a built-in predicate, but of the form el==value, rather than acting implicitly on all the elements.
—John
Indeed, the documentation must be wrong. Hopefully it can be appropriately fixed soon. That is an excellent suggestion about using distance. I had not thought of that, nor, come to think of it, did I even know that there was a boost::distance function. But, it makes sense. On the other hand, I do have to wonder if using "distance" for "count" would obfuscate the code. In general, I think I would rather use count_if just for clarity. But, at least the statement that the "_if functions aren't required", still holds true even if it may not be rather obvious which one to use. -- Bill