Also, I think that one of changes in your previous PR should be reverted.

Thanks for a suggestion. I did more efficient rewrite of that part, as discussed in the thread.

How did you write new tests? They look wildly random to me.

How did you write new tests? They look wildly random to me.

I prefer exact results, if possible. Then I try to use arch-independent values (e.g. "big enough" integers to trigger overflows on all platforms. And they should hit mentioned lines. Otherwise, they are random.

This PR combines

• test improvements, which should be backported
• refactoring (dead code removal), which IMO should not be backported, and
• possibly a bugfix, which should be backported together with a previously failing test

To make the backporting easier, would it make sense to split the PR along those lines?

@encukou, sorry, I wasn't aware that tests are systematically backported (based on #102067 and #102523).

possibly a bugfix, which should be backported together with a previously failing test

Probably, you are about typo in sumprod(). This is fixed and backported, see #111342. Or something else?

The backporting is a fairly new policy.

But, another benefit of splitting this would be that the test additions could be merged right now, without waiting for @mdickinson to comment on the removed code :)

Thanks for the typo fix!

@encukou, tests now added in #111930

Please could some kind person remove the request for review for me?

@serhiy-storchaka, I'm going to close this (together with the issue) on same ground as #109642. It seems, CPython devs not welcoming now code coverage improvements.