Thanks David. You call this PR an "attempt", but it looks like it fully addresses the rounding issue for isZero. Are there other things that you want to address here? I think it probably makes sense to address isPositive and isNegative in this same PR, so the three functions keep working in a consistent way.

Hi Jos, thanks for the review! I agree I shouldn't have called it an "attempt" as it indeed solves the issue in my tests.

I was hesitant that I didn't follow the suggestions exactly but I should have phrased it differently.

I agree with you, I will look into isPositive and isNegative and apply a consistent logic.

I will be back during the week with that change.

With this latest commit isPositive and isNegative now works as discussed including some tests.

Thanks, this looks good David!

This originally was to fix #3135. I think it would be good to add a unit test to verify that #3135 is solved now, or would you like to do that in a separate PR?

And, reminder to myself reading #2838: we should publish this as a breaking change.

I included the test for #3135
I don't know if it might be too specific.

Looks good to me. It indeed feels like a quite specific test, but I think it's fine (we can always add more tests if needed in the future).

I'll merge this PR now in a new v13 branch.