On Oct 28, 2010, at 11:44 AM, Jim Jewett wrote:
It uses an open addressing strategy. Each dict entry holds three pointer-sized fields: key object, value object, and cached hash value of the key. (set entries have only two fields, since they don't hold a value object)
Has anyone benchmarked not storing the hash value here
That would be a small disaster. Either you call PyObject_Hash() for every probe (adding function call overhead for int and str, and adding tons of work for other types) or you can go directly to Py_RichCompareBool() which is never fast. I haven't timed this for dicts, but I did see major speed boosts in the performance of set-to-set operations when the internally stored hash was used instead of calling PyObject_Hash(). Raymond