TST: SpectralGWCS failures using specutils 1.3.1 and glue-astronomy dev

[pllim]

Example log: https://github.com/spacetelescope/jdaviz/runs/3508608981?check_suite_focus=true

Package versions: 
Numpy: 1.22.0.dev0+949.ga90677a0b
Scipy: 1.7.1
Matplotlib: 3.4.3
h5py: 3.4.0
Pandas: 1.3.2
astropy: 5.0.dev786+g0fd7ae818
pyyaml: 5.4.1
scikit-image: 0.18.3
specutils: 1.3.1
spectral-cube: 0.5.1.dev227+g8d07dc9
asteval: 0.9.25
click: 8.0.1
echo: 0.5
idna: 2.10
traitlets: 5.1.0
bqplot: 0.12.30
bqplot-image-gl: 1.4.4
glue-core: 1.2.1
glue-jupyter: 0.8
glue-astronomy: 0.3.dev15+g312e959
ipyvue: 1.5.0
ipyvuetify: 1.8.1
ipysplitpanes: 0.2.0
ipygoldenlayout: 0.4.0
voila: 0.2.11
vispy: 0.8.1
gwcs: 0.16.2a1.dev32+g464e55f
asdf: 2.9.0.dev11+g9e64fbe

____________________________ test_viewer_axis_link _____________________________

mosviz_app = <jdaviz.configs.mosviz.helper.MosViz object at 0x7f623bc65e20>
spectrum1d = <Spectrum1D(flux=<Quantity [ 9.29172181, 11.92673337, 13.19606653, ..., 16.03141721,
           15.67835212, 17.780023...tralAxis [6000.        , 6001.95503421, 6003.91006843, ..., 7996.08993157,
   7998.04496579, 8000.        ] Angstrom>)>
spectrum2d = SpectralCube with shape=(15, 1, 1024):
 n_x:   1024  type_x: RA---TAN  unit_x: deg    range:     0.999898 deg:   86.79...:     0.055866 deg:    0.999746 deg
 n_s:     15  type_s: WAVE      unit_s: m      range:        1.000 m:      15.000 m

    def test_viewer_axis_link(mosviz_app, spectrum1d, spectrum2d):
        label1d = "Test 1D Spectrum"
        mosviz_app.load_1d_spectra(spectrum1d, data_labels=label1d)
    
        label2d = "Test 2D Spectrum"
        mosviz_app.load_2d_spectra(spectrum2d, data_labels=label2d)
    
        table = mosviz_app.app.get_viewer('table-viewer')
        table.widget_table.vue_on_row_clicked(0)
    
        scale_2d = mosviz_app.app.get_viewer('spectrum-2d-viewer').scales['x']
        scale_1d = mosviz_app.app.get_viewer('spectrum-viewer').scales['x']
    
        scale_2d.min = 200.0
>       assert scale_1d.min == spectrum1d.spectral_axis.value[200]
E       assert 6000.0 == 6391.006842619746
E        +  where 6000.0 = LinearScale(allow_padding=False, max=6000.0, min=6000.0).min

../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/mosviz/tests/test_helper.py:261: AssertionError
________________________ test_get_spectral_regions_unit ________________________

specviz_app = <jdaviz.configs.specviz.helper.SpecViz object at 0x7f623a348eb0>
spectrum1d = <Spectrum1D(flux=<Quantity [12.49671415, 12.30618014, 13.53658529, 14.85657834, 13.5436244 ,
           13.98808527, 1...28024,
                   1.72491783, 0.56228753, 1.01283112, 0.31424733,
                   0.90802408, 1.4123037 ]))>

    def test_get_spectral_regions_unit(specviz_app, spectrum1d):
        # Ensure units we put in are the same as the units we get out
        specviz_app.load_spectrum(spectrum1d)
        specviz_app.app.get_viewer("spectrum-viewer").apply_roi(XRangeROI(1, 3.5))
    
>       subsets = specviz_app.get_spectral_regions()

../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/specviz/tests/test_helper.py:88: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/specviz/helper.py:78: in get_spectral_regions
    regions = self.app.get_subsets_from_viewer("spectrum-viewer", subset_type="spectral")
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = Application(components={'g-viewer-tab': '<template>\n  <component :is="stack.container">\n    <g-viewer-tab\n      v-f...lbackList with 1 elements>
  tool_items: <CallbackList with 3 elements>
  tray_items: <CallbackList with 5 elements>
>)
viewer_reference = 'spectrum-viewer', data_label = None
subset_type = 'spectral'

    def get_subsets_from_viewer(self, viewer_reference, data_label=None, subset_type=None):
        """
        Returns the subsets of a specified viewer converted to astropy regions
        objects.
    
        It should be noted that the subset translation machinery lives in the
        glue-astronomy repository. Currently, the machinery only works on 2D
        data for cases like range selection. For e.g. a profile viewer that is
        ostensibly just a view into a 3D data set, it is necessary to first
        reduce the dimensions of the data, then retrieve the subset information
        as a regions object. This means that the returned y extents in the
        region are not strictly representative of the subset range in y.
    
        Parameters
        ----------
        viewer_reference : str
            The reference to the viewer defined with the ``reference`` key
            in the yaml configuration file.
        data_label : str, optional
            Optionally provide a label to retrieve a specific data set from the
            viewer instance.
        subset_type : str, optional
            Optionally specify either "spectral" or "spatial" to return only
            subsets created in a profile (spectrum) viewer or image viewer,
            respectively.
    
        Returns
        -------
        data : dict
            A dict of the transformed Glue subset objects, with keys
            representing the subset name and values as astropy regions
            objects.
        """
        viewer = self.get_viewer(viewer_reference)
        data = self.get_data_from_viewer(viewer_reference,
                                         data_label,
                                         cls=None)
        regions = {}
    
        if data_label is not None:
            data = {data_label: data}
    
        for key, value in data.items():
            if isinstance(value, Subset):
                # Skip spatial or spectral subsets if only the other is wanted
                if subset_type == "spectral" and isinstance(value.subset_state, RoiSubsetState):
                    continue
                elif subset_type == "spatial" and isinstance(value.subset_state, RangeSubsetState):
                    continue
    
                # Range selection on a profile is currently not supported in
                #  the glue translation machinery for astropy regions, so we
                #  have to do it manually. Only data that is 2d is supported,
                #  therefore, if the data is already 2d, simply use as is.
                if value.data.ndim == 2:
                    region = value.data.get_selection_definition(
                        subset_id=key, format='astropy-regions')
                    regions[key] = region
                    continue
                # There is a special case for 1d data (which is also not
                #  supported currently). We now eschew the use of the
                #  translation machinery entirely and construct the astropy
                #  region ourselves.
                elif value.data.ndim == 1:
                    # Grab the data units from the glue-astronomy spectral axis
                    # TODO: this needs to be much simpler; i.e. data units in
                    #  the glue component objects
>                   unit = value.data.coords.spectral_axis.unit
E                   AttributeError: 'SpectralGWCS' object has no attribute 'spectral_axis'

../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/app.py:523: AttributeError
__________________ test_get_spectral_regions_unit_conversion ___________________

specviz_app = <jdaviz.configs.specviz.helper.SpecViz object at 0x7f623a28c250>
spectrum1d = <Spectrum1D(flux=<Quantity [12.49671415, 12.30618014, 13.53658529, 14.85657834, 13.5436244 ,
           13.98808527, 1...28024,
                   1.72491783, 0.56228753, 1.01283112, 0.31424733,
                   0.90802408, 1.4123037 ]))>

    def test_get_spectral_regions_unit_conversion(specviz_app, spectrum1d):
        # If the reference (visible) data changes via unit conversion,
        # check that the region's units convert too
        specviz_app.load_spectrum(spectrum1d)
        specviz_app.app.get_viewer("spectrum-viewer").apply_roi(XRangeROI(1, 3.5))
    
        # Convert the wavelength axis to microns
        new_spectral_axis = "micron"
        conv_func = uc.UnitConversion.process_unit_conversion
        converted_spectrum = conv_func(specviz_app.app, spectrum=spectrum1d,
                                       new_spectral_axis=new_spectral_axis)
    
        # Add this new data and clear the other, making the converted spectrum our reference
        specviz_app.app.add_data(converted_spectrum, "Converted Spectrum")
        specviz_app.app.add_data_to_viewer("spectrum-viewer",
                                           "Converted Spectrum",
                                           clear_other_data=True)
    
        # Retrieve the Subset
>       subsets = specviz_app.get_spectral_regions()

../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/specviz/tests/test_helper.py:114: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/specviz/helper.py:78: in get_spectral_regions
    regions = self.app.get_subsets_from_viewer("spectrum-viewer", subset_type="spectral")
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = Application(components={'g-viewer-tab': '<template>\n  <component :is="stack.container">\n    <g-viewer-tab\n      v-f...lbackList with 1 elements>
  tool_items: <CallbackList with 3 elements>
  tray_items: <CallbackList with 5 elements>
>)
viewer_reference = 'spectrum-viewer', data_label = None
subset_type = 'spectral'

    def get_subsets_from_viewer(self, viewer_reference, data_label=None, subset_type=None):
        """
        Returns the subsets of a specified viewer converted to astropy regions
        objects.
    
        It should be noted that the subset translation machinery lives in the
        glue-astronomy repository. Currently, the machinery only works on 2D
        data for cases like range selection. For e.g. a profile viewer that is
        ostensibly just a view into a 3D data set, it is necessary to first
        reduce the dimensions of the data, then retrieve the subset information
        as a regions object. This means that the returned y extents in the
        region are not strictly representative of the subset range in y.
    
        Parameters
        ----------
        viewer_reference : str
            The reference to the viewer defined with the ``reference`` key
            in the yaml configuration file.
        data_label : str, optional
            Optionally provide a label to retrieve a specific data set from the
            viewer instance.
        subset_type : str, optional
            Optionally specify either "spectral" or "spatial" to return only
            subsets created in a profile (spectrum) viewer or image viewer,
            respectively.
    
        Returns
        -------
        data : dict
            A dict of the transformed Glue subset objects, with keys
            representing the subset name and values as astropy regions
            objects.
        """
        viewer = self.get_viewer(viewer_reference)
        data = self.get_data_from_viewer(viewer_reference,
                                         data_label,
                                         cls=None)
        regions = {}
    
        if data_label is not None:
            data = {data_label: data}
    
        for key, value in data.items():
            if isinstance(value, Subset):
                # Skip spatial or spectral subsets if only the other is wanted
                if subset_type == "spectral" and isinstance(value.subset_state, RoiSubsetState):
                    continue
                elif subset_type == "spatial" and isinstance(value.subset_state, RangeSubsetState):
                    continue
    
                # Range selection on a profile is currently not supported in
                #  the glue translation machinery for astropy regions, so we
                #  have to do it manually. Only data that is 2d is supported,
                #  therefore, if the data is already 2d, simply use as is.
                if value.data.ndim == 2:
                    region = value.data.get_selection_definition(
                        subset_id=key, format='astropy-regions')
                    regions[key] = region
                    continue
                # There is a special case for 1d data (which is also not
                #  supported currently). We now eschew the use of the
                #  translation machinery entirely and construct the astropy
                #  region ourselves.
                elif value.data.ndim == 1:
                    # Grab the data units from the glue-astronomy spectral axis
                    # TODO: this needs to be much simpler; i.e. data units in
                    #  the glue component objects
>                   unit = value.data.coords.spectral_axis.unit
E                   AttributeError: 'SpectralGWCS' object has no attribute 'spectral_axis'

../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/app.py:523: AttributeError
=========================== short test summary info ============================
FAILED ../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/mosviz/tests/test_helper.py::test_viewer_axis_link
FAILED ../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/specviz/tests/test_helper.py::test_get_spectral_regions_unit
FAILED ../../.tox/py39-test-devdeps/lib/python3.9/site-packages/jdaviz/configs/specviz/tests/test_helper.py::test_get_spectral_regions_unit_conversion

🐱

[pllim]

Maybe specutils dev will fix this naturally but I cannot test it until astropy/specutils#858 is resolved.

[pllim]

@astrofrog said the fix for this falls to jdaviz to support GWCS for the affected code.

[astrofrog]

The easiest fix might be to rely on APE14 API by replacing value.data.coords.spectral_axis.unit with value.data.coords.world_axis_units[-1] (I haven't tested this)

[astrofrog]

Did not mean to auto-close this - I didn't know this worked across repos!

[pllim]

I didn't know this worked across repos!

But you knew back in 2014. 😉

astropy/package-template#81 (comment)

[pllim]

This issue can be closed by a PR to update minversion of glue-astronomy (0.3?) here once @astrofrog releases it upstream. I have opened glue-viz/glue-astronomy#46 to track the proper fix as "future work."