Why converting to float64 instead of float32 in case of masked=True

[remi-braun]

Hello, me again 😅

I was wondering why the data is converted to float64 and not float32 when a dataset is read with masked=True (and original data is uin16 for example)
Isn't it dangerous wrt memory consumption ?
(I am not totally fluent in xarray, so it might be an obvious reason that I don't see 😓)

[remi-braun]

And if I have understood everything correctly, the conversion to float is forced because of the use of nan, is it right ?

[snowman2]

the conversion to float is forced because of the use of nan, is it right ?

Correct

[snowman2]

It currently matches the behavior ofxarray.open_dataset and is the safest datatype to convert to.

Isn't it dangerous wrt memory consumption ?

It hasn't been optimized for memory at this point. PRs are welcome 😄

[remi-braun]

If even xarray converts everything to float64, they may have thought this through.
Maybe this makes no problem as long you don't call load() on the array...

Thanks for your very quick replies !

[snowman2]

Hmmm, maybe I was wrong on that one. Related: pydata/xarray#2304

[snowman2]

Looks like it has been updated: https://github.com/pydata/xarray/blob/e0358e586079c12525ce60c4a51b591dc280713b/xarray/core/dtypes.py#L45-L78

[snowman2]

See #304