remove_if-like behavior maps?

Started by
6 comments, last by SiCrane 10 years, 9 months ago

So is there an idiomatic way to remove all items from a map for which some predicate is true?

std::remove_if "removes" items by moving them to the end of a collection and then returns an iterator to the first of the "removed" items so you can erase them or do whatever you want to do. This doesn't work for maps because maps are ordered and therefore you can't just arbitrarily move their contents around.

Is there some way of doing this with standard library algorithms? I just wrote a function but it seemed weird that I had to.

Also does anyone else find remove_if broken? I mean why not just get rid of it and have erase_if? Or at least have erase_if and remove_if?

Advertisement

So is there an idiomatic way to remove all items from a map for which some predicate is true?

std::remove_if "removes" items by moving them to the end of a collection and then returns an iterator to the first of the "removed" items so you can erase them or do whatever you want to do. This doesn't work for maps because maps are ordered and therefore you can't just arbitrarily move their contents around.

Is there some way of doing this with standard library algorithms? I just wrote a function but it seemed weird that I had to.

There are no immediate functions in the standard library to do that as far as I am aware. It should, however, be fairly trivial to just make a simple loop and erase elements yourself. Since map iterators are not invalidated on erase, you don't need much effort to keep the loop valid like when you iterate over and erase from, say, a vector.

Also does anyone else just find remove_if broken? I mean why not just get rid of it and have erase_if? Or at least have erase_if and remove_if?

remove_if serves a purpose when the container itself cannot be modified. For example, you can call remove_if on a static array to partition its content into a part with the remaining part and the removed part, and the return value is the pointer to where the second partition starts. An erase_if equivalent would not work on a static array.

Since erase_if can easily be implemented in terms of remove_if and erase, the function itself is not really something fundamental on the same level as remove_if and erase. Although I could definitely agree that it would be nice to have, since remove_if is often called with erase.


For example, you can call remove_if on a static array to partition its content into a part with the remaining part and the removed part, and the return value is the pointer to where the second partition starts.

Couldn't you use std::stable_partition instead?

Unless I'm missing something fundamental, they do the same thing in this case.

Tristam MacDonald. Ex-BigTech Software Engineer. Future farmer. [https://trist.am]

No, partition() or stable_partition() will leave the part at and after the returned iterator as being completely filled with elements not satisfying the predicate. However, remove_if() gives no guarantee whatsoever about the contents of those elements. They may or may not satisfy the predicate.

I use this function in my code:


//Removes all occurances of elements in the map that have a value that returns true for 'predicate'.
template<typename MapType, typename PredicateFunc>
void MapRemoveValues(MapType &map, PredicateFunc predicate)
{
	for(auto it = std::begin(map); it != std::end(map); )
	{
		if(predicate(it->second))
		{
			it = map.erase(it);
		}
		else
		{
			++it;
		}
	}
}

Just make the predicate take *it instead of it->second.

[Edit:] Whoops, skimming too much. Didn't see you said you already wrote a function.


Since erase_if can easily be implemented in terms of remove_if and erase, the function itself is not really something fundamental on the same level as remove_if and erase. Although I could definitely agree that it would be nice to have, since remove_if is often called with erase.

After thinking this all over, my opinion is that erase_if -- or something equivalent -- is fundamental precisely because you cannot implement erase_if-type functionality in terms of erasing items that you moved around unless the underlying data structure you are working with is vector-like.

Fixed sized arrays wouldn't work -- without broadening the semantics of a legal implementation of erase_if --, yes, but then maps don't work with the remove/erase idiom. It is like the authors of the standard library give primacy to vector-like data structures; that when we say such-and-such is fundamental there is a tacit clause like such-and-such is fundamental for vectors and lists.

No, partition() or stable_partition() will leave the part at and after the returned iterator as being completely filled with elements not satisfying the predicate. However, remove_if() gives no guarantee whatsoever about the contents of those elements. They may or may not satisfy the predicate.

Hang on...

remove_if() doesn't guarantee that all the items after the returned iterator satisfy the predicate? Or it doesn't guarantee that all items which satisfy the predicate will be after the returned iterator? Either definition seems problematic.

I'm assuming you mean that it doesn't actually swap the items into the slots beyond the returned iterator, and the contents of those items may or may not be valid?

Tristam MacDonald. Ex-BigTech Software Engineer. Future farmer. [https://trist.am]

std::remove_if() has no guarantees whatsoever about whether or not elements at or after the iterator returned by std::remove_if() satisfy or don't satisfy the predicate passed to std::remove_if(). The contents of those elements will be valid in the sense they can be properly destroyed (assuming that the copy assignment/move assignment operator are implemented correctly).

This topic is closed to new replies.

Advertisement