math.sumprod() missing one case for combined int/float inputs.

[rhettinger]

Bug report

Bug description:

The sumprod function has high precision paths for int/int, int/float, and float/float inputs. However it missed a case when an int/int pair follows another pair that contained a float. In other words, it assumes that once a float is encountered, that subsequent inputs won't contain a pair of ints.

>>> import math
>>> data = [-0.01, 1, -1, 0.01]
>>> math.sumprod(data,  [1, 1, 1, 1])
-8.673617379884035e-18
>>> math.sumprod(data,  [1.0, 1.0, 1.0, 1.0])
0.0

CPython versions tested on:

3.13, 3.15, 3.14

Operating systems tested on:

macOS

Linked PRs

• gh-139074: Fix missing high precision case in sumprod() #139075
• [3.14] gh-139074: Fix missing high precision case in sumprod() (gh-139075) #139085
• [3.13] gh-139074: Fix missing high precision case in sumprod() (gh-139075) #139086
• Revert gh-139075. The sumprod fix needs more work. #139092

[skirpichev]

I think this still valid.

[rhettinger]

Yes, it is still valid. I'm putting in on the back burner for a while because 1) it is a minor and rare corner case and 2) I want to experiment to find the cleanest solution.