Default fill value of -9999 is incompatible with numpy 2.0 after integration of NEP 50

[lukecampbell]

https://numpy.org/neps/nep-0050-scalar-promotion.html#nep50

pocean-core/pocean/utils.py

Line 305 in acd977b

fill_value=series.dtype.type(CFDataset.default_fill_value),

pocean-core/pocean/cf.py

Line 14 in acd977b

default_fill_value = -9999.9

With the release of numpy 2.0 adopting NEP 50 by default, type promotion has changed and any attempts to coerce python scalar values into data types that can't represent that value now raise an OverflowError.

Before, if you had a dtype of uint8 the fill value would wind up being 241 after overflowing.

I recommend creating a more sophisticated mapping of appropriate fill values based on dtypes:

• uint8 - 255
• int8 - -128
• ...
• float32 - nan
• float64 - nan
• datetime64 - nan (depends on context... numpy supported contexts are NaT, float-only contexts would be NaN)

I'll try to take a look at a PR if I can find the time.

[ocefpaf]

You can use np._set_promotion_state("legacy") for now until this is properly fixed, in case you have broken workflows.