All but one of the PyTuple_SetItem() calls are executed without checking the return value.

Callers:
$ grep -r PyTuple_SetItem Modules/_sqlite 
Modules/_sqlite/connection.c:        PyTuple_SetItem(args, i, cur_py_value);
Modules/_sqlite/cursor.c:        PyTuple_SetItem(row, i, converted);
Modules/_sqlite/cursor.c:    if (PyTuple_SetItem(func_args, 0, Py_NewRef(operation)) != 0) {
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 0, column_name);
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 1, Py_NewRef(Py_None));
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 2, Py_NewRef(Py_None));
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 3, Py_NewRef(Py_None));
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 4, Py_NewRef(Py_None));
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 5, Py_NewRef(Py_None));
Modules/_sqlite/cursor.c:                PyTuple_SetItem(descriptor, 6, Py_NewRef(Py_None));
Modules/_sqlite/cursor.c:                PyTuple_SetItem(self->description, i, descriptor);


All of these are operating on newly created tuples, so I suggest replacing them with PyTuple_SET_ITEM() instead of adding error handling.


For the users in _pysqlite_query_execute() I also suggest to move the tuple creation closer to the code that fills it, in order to minimise the number of decref's needed in case of error.