fix floating point precision in floatyear_to_date

[pat-schmitt]

In this PR I fix a bug which was brought up by @Ruitangtang.

The problem was when calling floatyear_to_date(1+1/12-1/12) you expect to get (1, 1), but actually (0, 12) was returned, due to a floating point precision issue. I now included a check if the given floatyear is inside machine precision to the next higher int, and if so us the next higher int instead.

• Tests added/passed
• Fully documented
• Entry in whats-new.rst

[fmaussion]

This is a bit of an edge case. Arguably the problem is not in OGGM, since OGGM is given a "wrong" number:

>>> 1 + 1/12 - 1/12
0.9999999999999999

I would normally recommend against doing this kind of arithmetic on "OGGM dates" and use utils.monthly_timeseries, but it's okay to round like this (I hope - the only problem I can foresee is when a tiny time step is needed for the dynamical model - but even then, it's tiny)

[pat-schmitt]

@Ruitangtang is currently implementing a way to save the fl_diagnostics at monthly steps, and for the diagnostic variable climatic_mb we want to use the mb of the previous time-step and hence she is calling self.get_mb(surface_h_previous[fl_id], self.yr - 1/12, fl_id=fl_id), here the current version for yearly timesteps https://github.com/OGGM/oggm/blob/master/oggm/core/flowline.py#L1325.

But probably their is a better way around? Maybe directly saving the previous mb instead of calling it again with -1/12.

And yes, the tiny timestep problem could happen with this method, but it is in the order of 1e-13 of a year (~1e-7 seconds).

Further it seems with numpy 2.0 len(np.array(1)) or len(np.int64(1)) is not working anymore. But len(np.array([1])) us still good.

[fmaussion]

But probably their is a better way around?

yes I see. I understand why the current behavior is surprising and I agree to the workaround in OGGM. A probable more robust way would be to not to arithmetics on time but rather to do an array of times and use this instead:

monthly_time = utils.monthly_timeseries(y0, y1)
i = 12
print(monthly_time[i])
print(monthly_time[i-1])  # previous months