YAML area configuration does not allow to specify the dtype and dtype is ignored in equality comparisons

Today I learned that an areadefinition has a dtype attribute. This seems to be not very well documented and not very well supported, but it's relevant. It is used at least in get_lonlats(), which is used in resampling, so changing the dtype can have considerable memory implications and make the difference between having enough RAM or not.

Code Sample, a minimal, complete, and verifiable piece of code

import pyresample

ar = pyresample.AreaDefinition(
        "area_id", "descr", "proj_id", 4326, 100, 100,
       [-100, -100, 100, 100], dtype="float32")
s = ar.dump()
print(s)
ar2 = pyresample.area_config.load_area_from_string(s)
print("equals?", ar == ar2)
print("dtypes", ar.dtype, ar2.dtype)
s2 = """area_id:
  description: descr
  projection:
    EPSG: 4326
  shape:
    height: 100
    width: 100
  area_extent:
    lower_left_xy: [-100, -100]
    upper_right_xy: [100, 100]
  dtype: float32
"""
ar3 = pyresample.area_config.load_area_from_string(s2)
print(ar3.dtype)

Problem description

Executing the code reveals several problems:

• ar.dump() does not output the dtype
• ar and ar2 are considered equal despite having different dtypes
• ar3, loaded from a string that does encode the dtype, gets float64 despite the YAML definition stating float32.

It does not appear to be possible to specify the dtype in the YAML configuration (nor in pyresample.create_area_def()).

Expected Output

area_id:
  description: descr
  projection:
    EPSG: 4326
  shape:
    height: 100
    width: 100
  area_extent:
    lower_left_xy: [-100, -100]
    upper_right_xy: [100, 100]
  dtype: float32

equals? True
dtypes float32 <class 'numpy.float32'>
<class 'numpy.float32'>

Actual Result, Traceback if applicable

area_id:
  description: descr
  projection:
    EPSG: 4326
  shape:
    height: 100
    width: 100
  area_extent:
    lower_left_xy: [-100, -100]
    upper_right_xy: [100, 100]

equals? True
dtypes float32 <class 'numpy.float64'>
/data/gholl/checkouts/pyresample/pyresample/area_config.py:91: UserWarning: Unused/unexpected area definition parameter(s) for area_id: params={'dtype': 'float32'}
  area_list = parse_area_file(area_file_name, *regions)
<class 'numpy.float64'>

Versions of Python, package at hand and relevant dependencies

pyresample main (v1.28.2-2-g711f354)

Start with pyresample/area_config.py and the AreaDefinition dump/load and equality entry points; trace how dtype is passed by load_area_from_string and create_area_def. Done means YAML serialization preserves dtype, YAML loading and create_area_def accept it, and differing dtypes affect equality, with regression coverage for the sample cases.