Better attr diff for testing.assert_identical

[dcherian]

• User visible changes (including notable bug fixes) are documented in whats-new.rst

This gives us better reprs where only differing attributes are shown in the diff.

On main:

...
Differing coordinates:
L * x        (x) <U1 'a' 'b'
    foo: bar
    same: same
R * x        (x) <U1 'a' 'c'
    source: 0
    foo: baz
    same: same
...

With this PR:

Differing coordinates:
L * x        (x) %cU1 'a' 'b'
    Differing variable attributes:
        foo: bar
R * x        (x) %cU1 'a' 'c'
    Differing variable attributes:
        source: 0
        foo: baz

[benbovy]

I think anything is welcome to improve the experience of seeing what exactly differ between two variables. However if this PR is treating the attributes only (IIUC) I find the diff could be potentially misleading.

For variables with many attributes this is useful but in general I'm concerned that users who are not aware of this behavior will struggle figuring out why the variable reprs shown here differ from their full repr.

Alternatively, couldn't we "highlight" the differing attributes but still show the full variable reprs? E.g.,

Differing coordinates:
L * x        (x) <U1 'a' 'b'
    foo: bar
    ^^^^^^^^
    same: same
R * x        (x) <U1 'a' 'c'
    source: 0
    ^^^^^^^^^
    foo: baz
    ^^^^^^^^
    same: same

It requires a bit more work to implement, but this could nicely be generalized to other differing parts like the name, dimensions, labels, etc.:

Differing coordinates:
L * x        (x) <U1 'a' 'b'
                      ^^^^^
    foo: bar
    ^^^^^^^^
    same: same
R * x        (x) <U1 'a' 'c'
                      ^^^^^
    source: 0
    ^^^^^^^^^
    foo: baz
    ^^^^^^^^
    same: same

(not sure about the best way to highlight those parts, though)

[dcherian]

Alternatively, couldn't we "highlight" the differing attributes but still show the full variable reprs?

Good idea but it doesn't scale to large datasets. Here's an example I'm looking at, with only one data variable:

E       AssertionError: Left and right Dataset objects are not identical
E       
E       Differing coordinates:
E       L * longitude   (longitude) float64 0.0 0.25 0.5 0.75 ... 359.2 359.5 359.8
E           units: degrees_east
E           standard_name: longitude
E           long_name: longitude
E       R * longitude   (longitude) float64 0.0 0.25 0.5 0.75 ... 359.2 359.5 359.8
E           long_name: longitude
E           standard_name: longitude
E           units: degrees_east
E       L   surface     float64 ...
E       R   surface     float64 ...
E       L   time        datetime64[ns] 2023-06-29
E           long_name: initial time of forecast
E           standard_name: forecast_reference_time
E       R   time        datetime64[ns] ...
E           long_name: initial time of forecast
E           standard_name: forecast_reference_time
E       L   valid_time  datetime64[ns] ...
E           standard_name: time
E           long_name: time
E       R   valid_time  datetime64[ns] ...
E           long_name: time
E           standard_name: time
E       L * latitude    (latitude) float64 90.0 89.75 89.5 89.25 ... -89.5 -89.75 -90.0
E           units: degrees_north
E           standard_name: latitude
E           long_name: latitude
E       R * latitude    (latitude) float64 90.0 89.75 89.5 89.25 ... -89.5 -89.75 -90.0
E           long_name: latitude
E           standard_name: latitude
E           units: degrees_north
E       L   step        timedelta64[ns] ...
E           long_name: time since forecast_reference_time
E           standard_name: forecast_period
E       R   step        timedelta64[ns] ...
E           long_name: time since forecast_reference_time
E           standard_name: forecast_period
E       Differing data variables:
E       L   tp          (latitude, longitude) float32 ...
E           GRIB_paramId: 228228
E           GRIB_dataType: fc
E           GRIB_numberOfPoints: 1038240
E           GRIB_typeOfLevel: surface
E           GRIB_stepUnits: 1
E           GRIB_stepType: accum
E           GRIB_gridType: regular_ll
E           GRIB_NV: 0
E           GRIB_Nx: 1440
E           GRIB_Ny: 721
E           GRIB_cfName: unknown
E           GRIB_cfVarName: unknown
E           GRIB_gridDefinitionDescription: Latitude/longitude. Also called equidista...
E           GRIB_iDirectionIncrementInDegrees: 0.25
E           GRIB_iScansNegatively: 0
E           GRIB_jDirectionIncrementInDegrees: 0.25
E           GRIB_jPointsAreConsecutive: 0
E           GRIB_jScansPositively: 0
E           GRIB_latitudeOfFirstGridPointInDegrees: 90.0
E           GRIB_latitudeOfLastGridPointInDegrees: -90.0
E           GRIB_longitudeOfFirstGridPointInDegrees: 0.0
E           GRIB_longitudeOfLastGridPointInDegrees: 359.75
E           GRIB_missingValue: 3.4028234663852886e+38
E           GRIB_name: Total Precipitation
E           GRIB_shortName: tp
E           GRIB_units: kg m**-2
E           long_name: Total Precipitation
E           units: kg m**-2
E           standard_name: unknown
E       R   tp          (latitude, longitude) float64 dask.array<chunksize=(721, 1440), meta=np.ndarray>
E           GRIB_NV: 0
E           GRIB_Nx: 1440
E           GRIB_Ny: 721
E           GRIB_cfName: unknown
E           GRIB_cfVarName: unknown
E           GRIB_dataType: fc
E           GRIB_gridDefinitionDescription: Latitude/longitude. Also called equidista...
E           GRIB_gridType: regular_ll
E           GRIB_iDirectionIncrementInDegrees: 0.25
E           GRIB_iScansNegatively: 0
E           GRIB_jDirectionIncrementInDegrees: 0.25
E           GRIB_jPointsAreConsecutive: 0
E           GRIB_jScansPositively: 0
E           GRIB_latitudeOfFirstGridPointInDegrees: 90.0
E           GRIB_latitudeOfLastGridPointInDegrees: -90.0
E           GRIB_longitudeOfFirstGridPointInDegrees: 0.0
E           GRIB_longitudeOfLastGridPointInDegrees: 359.75
E           GRIB_missingValue: 3.4028234663852886e+38
E           GRIB_name: Total Precipitation
E           GRIB_numberOfPoints: 1038240
E           GRIB_paramId: 228228
E           GRIB_shortName: tp
E           GRIB_stepType: accum
E           GRIB_stepUnits: 1
E           GRIB_typeOfLevel: surface
E           GRIB_units: kg m**-2
E           long_name: Total Precipitation
E           standard_name: unknown
E           units: kg m**-2

I'll update to add "Differing attributes" to make it clear that we aren't printing all attributes:

Differing coordinates:
L * x        (x) %cU1 'a' 'b'
    Differing attributes: foo: bar
R * x        (x) %cU1 'a' 'c'
    Differing attributes: source: 0
                          foo: baz

[max-sixty]

Nice improvement!