This patch adds PySet_Next(), PySet_Update(), and PySet_Clear(), three very useful API calls when you're dealing with concrete PySet objects. Patch includes tests and documentation.
Logged In: YES user_id=80475 Guido, please reject this patch. I laid-out my reasoning in these two posts: http://mail.python.org/pipermail/python-dev/2006- March/062652.html http://mail.python.org/pipermail/python-dev/2006- March/062737.html In particular, I've found PySet_Next() to be disasterously error-prone (and have had to fix several hard-to-find bugs directly attributable this function). There was some need for it in the PyDict API because it fetched key and value pointers at the same time and allowed direct modification of values. This of course does not apply to sets which do not have a value field. The PySet_Clear() function should be PyObject_Clear() like other functions that have tp_slots. There is no more of a case for this function than there would be for PySet_Repr, PySet_Len, PySet_IsTrue, or PySet_Print. The PySet_Update() function gets in the way of my future development plans for this module. None of these functions are necessary. They simply reflect Barry's programming style and views on having fat, overly-concrete APIs. This patch is against my wishes for the module. As the module author and principal maintainer, if I still have a say in the module's development, I strongly prefer rejection of this patch.
Logged In: YES user_id=80475 FWIW, I'm open to PyObject_Clear() for an abstract access to the tp_clear slot. Also, for Barry's use only, I'm open to a semi-private _PySet_Next() that isn't part of the official API but gives a Barry-specific hook. It should not be a public function because of all the problems with the approach. Likewise, _PySet_Update() could be done just for Barry's group. I need to keep PySet_Update() open until Py2.6 since the API may evolve to allow a variable length argument list.
