Question

I have a hash map defined as

class KeyType {
    int key;
    mutable bool flag;
    KeyType(int key) : key(key), flag(false) {}
    void setFlag() const { flag = true; }
};

struct KeyType_hasher {
    size_t operator()(const KeyType& s) const {
        return static_cast<size_t> key;
    }
};

struct KeyType_equal {
    size_t operator()(const KeyType& s1, const KeyType& s2) const {
        return s1.key == s2.key;
    }
};

typedef hash_map<KeyType , ValueType, KeyType_hasher, KeyType_equal > KeyValueMap;

Later on in the code I have a place where I have to loop though the map and apply a function to each value I find. Based on the function’s outcome, I have to modify the key at the iterator as well.

KeyValueMap theMap;
// theMap[key1] = value1;
// theMap[key2] = value2;
// theMap[key3] = value3;
for(KeyValueMap::iterator i = theMap.begin(); i != theMap.end(); ++i) {
    if(true == ValueFunction(i->second))
        i->first.setFlag();
}

My question is, would that be the right way of modifying the key, if I have to?
Does it have any bad side effects?

Answer 1

You’d have to remove the element from the container and re-add it with the new key.

None of the C++ associative containers support changing the key in a significant way (where significant means the change alters the results of the hash in a hashed container or the comparrsion in an ordered container).

If you did modify the key (by circumventing the const correctness system in some way) you’d get unpredictable results from lookups.