Use a fixed DateTime in rounding tests instead of now()

[quinnj]

Using now led to inconsistent answers depending on when now() is; make
tests more reliable by using a fixed DateTime.

[Moelf]

but now() was only accessed once, tests failing means there were unexpected behavior (which user will face too) right?

[simeonschaub]

I can't say much about whether the test failures actually represent a problem in the implementation, but it generally seems like a bad idea for tests to depend on the current time.

[Moelf]

I agree but in this case it seems there exist a Time where round(Date, now_, p, r) did something different than first convert now_ to Date(now_) and then do rounding -- which is strictly what user would expect.

[quinnj]

Valid concerns; I dug in a bit further and discovered we were passing the wrong default rounding mode to round, which is RoundNearestTiesUp, but we were passing RoundUp, which led to the test failures. I've pushed the fix here.

[Moelf]

thanks! I must missed that bit