tags:

views:

907

answers:

4

I am bit confused about the difference between the usage of std::remove algorithm. Specifically I am not able to understand what is being removed when I use this algorithm. I wrote a small test code like this:

std::vector<int> a;
a.push_back(1);
a.push_back(2);

std::remove(a.begin(), a.end(), 1);


int s = a.size();

std::vector<int>::iterator iter = a.begin();
std::vector<int>::iterator endIter = a.end();

std::cout<<"Using iter...\n";
for(; iter != endIter; ++iter)
{
 std::cout<<*iter<<"\n";
}

std::cout<<"Using size...\n";
for(int i = 0; i < a.size(); ++i)
{
 std::cout<<a[i]<<"\n";
}

The output was 2,2 in both the cases.

However, if I use erase with the remove something like this:

a.erase(std::remove(a.begin(), a.end(), 1), a.end());

I get the output as 2.

So my questions are:

(1). Is there any use of std::remove other than using it with erase function.

(2). Even after doing std::remove, why a.size() returns 2 and not 1?

I read the item in Scott Meyer's Effective STL book about the erase-remove idiom. But am still having this confusion.

+2  A: 

std::remove does not remove the actual objects, rather, pushes them to the end of the container. Actual deletion and deallocation of memory is done via erase. So:

(1). Is there any use of std::remove other than using it with erase function.

Yes, it helps to get a pair of iterators to a new sequence without having worry about proper de-allocation etc.

(2). Even after doing std::remove, why a.size() returns 2 and not 1?

The container still holds to those objects, you only have a new set of iterators to work with. Hence the size is still what it used to be.

dirkgently
Hmm. Actually std::remove() does not move the deleted elements to the end of the container -- the remaining positions of the container will contain their original values. (It must work this way to preserve O(n) time with forward iterators.)
j_random_hacker
j_random_hacker, hmm actually it looks like after it did the work, the range after the returned iterator also must not contain any "removed" elements - which is an additional requirement i haven't yet considered - so those elements are not entirely indeterminate. So my answer were wrong and i deleted it
Johannes Schaub - litb
it says: "Eliminates all the elements referred to by iterator i in the range [first, last) for which the following corresponding conditions hold: *i == value" which i interpret as that that range does not contain any "removed" value anymore in the end. Interesting, i haven't known that. But i don't think that one can rely on those elements being the same as before. One certainly can't if that range would have contained a "removed" value, at least. So i think cplusplus.com is wrong again imo :D
Johannes Schaub - litb
Still reading the draft -- it gets more interesting with erase which *calls* the dtors. IMO, remove is only about swapping pointers, which is how iterators are implemented, more or less.
dirkgently
curiously enough, GCC seems to not remove any "removed" elements from the range [returned_iterator, last) so i'm not sure how one should interpret the wording? Any idea?
Johannes Schaub - litb
in the end, i think it's just poor wording and i shouldn't have deleted my answer. elements after returned_iterator just have some indeterminate value i think. sigh :( hehe
Johannes Schaub - litb
Note that remove() is not supposed to call the dtors, at least that's not required. This is an efficient implementation.
dirkgently
indeed, it just moves things around. but i think the value of things after the returned iterators are not required to be identical as they were before. forward iterators are multipass. It could just zero out any elements in [returned_iterator, end), even if it read those elements again before to compare them . As far as i can see, there is no requirement that forbids that.
Johannes Schaub - litb
I cannot disagree in principle from what I have found so far. Will update if I find anything concrete.
dirkgently
@litb: Intriguing, I'm certain that that wording is actually a bug in the standard! I believe it should refer to any iterator in the range [first, returned_value) instead of [first, last). On the one hand, it's only practical; but also, if the iterator range contained *only* values that are to be removed, there is *no way* that remove() can meet the guarantee as stated! :)
j_random_hacker
Also my initial comment was actually wrong -- the remaining elements (I believe) could contain anything at all, including the text "j_random_hacker RULZ!!!1!", suitably encoded. :) (My statement is true for implementations of std::vector that I've come across, but it might be more efficient to move deleted elements to the end for std::list -- sorry ceretullis, maybe there was a grain of truth to what you were saying.)
j_random_hacker
@litb: I don't think remove() could just "zero out" the remaining elements, as the expression "*i = 0" (where i is a forward iterator) is not required to be valid.
j_random_hacker
i'm sorry i mean just setting all the elements to some value, like the last one or so. :)
Johannes Schaub - litb
+13  A: 

remove() doesn't actually delete elements from the container -- it only shunts non-deleted elements forwards on top of deleted elements. The key is to realise that remove() is designed to work on not just a container but on any arbitrary forward iterator pair: that means it can't actually delete the elements, because an arbitrary iterator pair doesn't necessarily have the ability to delete elements.

For example, pointers to the beginning and end of a regular C array are forward iterators and as such can be used with remove():

int foo[100];

...

remove(foo, foo + 100, 42);    // Remove all elements equal to 42

Here it's obvious that remove() cannot resize the array!

j_random_hacker
A: 

Simplest I can come up with:

erase() is something you can do to an element in a container. Given an iterator/index into a container, erase( it ) removes the thing the iterator refers to from the container.

remove() is something you can do to a range, it re-arranges that range but doesn't erase anything from the range.

anon
A: 

remove doesn't "really" remove anything, because it can't.

In order to "actually" remove the elements from container you need to access container APIs. Where as remove works only with iterators irrespective of what containers those iterators points to. Hence, even if remove wants an "actual remove", it can't.

Remove overwrite "removed" elements by the following elements that were not removed and then it is up to the caller to decide to use the returned new logical end instead of the original end.

In your case remove logically removed 1 from vector a but size remained to 2 itself. Erase actually deleted the elements from vector. [ from vector new end to old end ]

The main idea of remove is it cannot change the number of elements and it just remove elements from a range as per criteria.

aJ