-----Original Message-----
From: Greg Ewing
Of course if we had a NaN value for integers, int('NaN'), then we could just set the initial count to it and since NaN - anything = NaN all would be golden.
Or we could use floating-point reference counts... -- Greg [Steve Barnes] I thought of floating-point reference counts but: a) 65535.0 -= 1 is slower than 65535 =- 1 (7.6% on my system quite a bit worse on some embedded systems). b) There comes a point in floats, (for large values of x), where x - 1 == x (about 10**15 which for some scientific & big data calculations is not that big) but unless reference counting uses python integers, rather than C, this will not be an issue. c) of course I like the concept of integer nan, inf, etc. 😊