Why is collections.Counter much slower than ''.count?

Question

I have a simple task: To count how many times every letter occurs in a string. I\'ve used a Counter() for it, but on one forum I saw information that using

Answer 1

Counter() allows you to count any hashable objects, not just substrings. Both solutions are O(n)-time. Your measurements show that the overhead of iterating and hashing individual characters by Counter() is greater than running s.count() 4 times.

Counter() can use C helper to count elements but it seems it doesn't special case strings and uses general algorithm applicable for any other iterable i.e., processing a single character involves multiple Python C API calls to advance the iterator, get previous value (a lookup in the hash table), increment counter, set new value (a lookup in the hash table):

    while (1) {
        key = PyIter_Next(it);
        if (key == NULL)
            break;
        oldval = PyObject_GetItem(mapping, key);
        if (oldval == NULL) {
            if (!PyErr_Occurred() || !PyErr_ExceptionMatches(PyExc_KeyError))
                break;
            PyErr_Clear();
            Py_INCREF(one);
            newval = one;
        } else {
            newval = PyNumber_Add(oldval, one);
            Py_DECREF(oldval);
            if (newval == NULL)
                break;
        }
        if (PyObject_SetItem(mapping, key, newval) == -1)
            break;
        Py_CLEAR(newval);
        Py_DECREF(key);
    }

Compare it to FASTSEARCH() overhead for bytestrings:

    for (i = 0; i < n; i++)
        if (s[i] == p[0]) {
           count++;
           if (count == maxcount)
              return maxcount;
        }
    return count;