Inconsistent test results in github vs legend-base

[jasondet]

github is happy passing all tests. however in legend-base on the lngs server, the tests die on importing orca_packet with:

raw/orca/orca_packet.py:15
ERROR test_orca_packet.py - TypeError: 'numpy._DTypeMeta' object is not subscriptable

I suspect we are missing a required version dependency for annotations?

[gipert]

Might be due to an incompatible numpy version... Will check and fix once I'm back online.

We had to disable testing with the container because our image is too heavy to be downloaded on the GitHub action runners

[gipert]

DSP tests are failing in the container:

$ pytest tests/dsp
ImportError while loading conftest '/tmp/pygama/tests/dsp/conftest.py'.
tests/dsp/conftest.py:3: in <module>
    import pygama.dsp.processors  # noqa: F401
/user/.local/lib/python3.9/site-packages/pygama/dsp/__init__.py:20: in <module>
    from .build_dsp import build_dsp
/user/.local/lib/python3.9/site-packages/pygama/dsp/build_dsp.py:21: in <module>
    from pygama.dsp.processing_chain import build_processing_chain
/user/.local/lib/python3.9/site-packages/pygama/dsp/processing_chain.py:23: in <module>
    from pygama.math.units import Quantity, Unit
/user/.local/lib/python3.9/site-packages/pygama/math/units.py:1: in <module>
    from pint import Measurement, Quantity, Unit, UnitRegistry, set_application_registry
/user/.local/lib/python3.9/site-packages/pint/__init__.py:16: in <module>
    from .context import Context
/user/.local/lib/python3.9/site-packages/pint/context.py:20: in <module>
    from .definitions import Definition, UnitDefinition
/user/.local/lib/python3.9/site-packages/pint/definitions.py:16: in <module>
    from .converters import Converter, LogarithmicConverter, OffsetConverter, ScaleConverter
/user/.local/lib/python3.9/site-packages/pint/converters.py:12: in <module>
    from .compat import HAS_NUMPY, exp, log  # noqa: F401
/user/.local/lib/python3.9/site-packages/pint/compat.py:163: in <module>
    from xarray import DataArray, Dataset, Variable
/opt/anaconda3/lib/python3.9/site-packages/xarray/__init__.py:1: in <module>
    from . import testing, tutorial, ufuncs
/opt/anaconda3/lib/python3.9/site-packages/xarray/testing.py:8: in <module>
    from xarray.core import duck_array_ops, formatting, utils
/opt/anaconda3/lib/python3.9/site-packages/xarray/core/duck_array_ops.py:24: in <module>
    from . import dask_array_compat, dask_array_ops, dtypes, npcompat, nputils
/opt/anaconda3/lib/python3.9/site-packages/xarray/core/dask_array_compat.py:5: in <module>
    from .pycompat import dask_version
/opt/anaconda3/lib/python3.9/site-packages/xarray/core/pycompat.py:55: in <module>
    dsk = DuckArrayModule("dask")
/opt/anaconda3/lib/python3.9/site-packages/xarray/core/pycompat.py:22: in __init__
    duck_array_version = LooseVersion(duck_array_module.__version__)
/opt/anaconda3/lib/python3.9/site-packages/setuptools/_distutils/version.py:53: in __init__
    warnings.warn(
E   DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.

and seems to be related to Pint.

CC @iguinn

[gipert]

The issue has been fixed in xarray since a while now: pydata/xarray#6092. The xarray installation in legend-base seems too old. Running pip install --upgrade xarray in the container fixes the issue.

@oschulz can you reproduce this with the current pygama main? Can you have a look into this?

[oschulz]

can you reproduce this with the current pygama

Do you mean reproduce with the current container, or an updated one with a current xarray?

[gipert]

I mean with the latest container and the latest pygama commit (on main). You would need to first pip install .[test] and then pytest.

[oschulz]

I can make that part of the container build, I think.

[gipert]

I can make that part of the container build, I think.

Ok, so the test suite runs just fine? Can we close this?

[oschulz]

Haven't run it yet, have to rebuild the containers first the next days.

[oschulz]

pygama tests pass now in the legend-software container and are part of the container build.

[gipert]

@oschulz why ignoring the deprecation warning?

https://github.com/legend-exp/legendexp_legend-software_img/blob/master/provisioning/install-sw-scripts/pygama-setup.sh#L20

I think the right solution would be to get a more recent xarray version, as I mentioned earlier in this thread: #327 (comment). Isn't it possible?

[oschulz]

Well, it would be nice to be able to go with regular Anaconda releases in general, and not always have to do specific updates for individual packages. Will this xarray thing result in reduced performance or is it just a pure deprecation warning?

As far as I know, ignore::DeprecationWarning is not uncommon in Python production testing.

[gipert]

No, it's not a big issue. I agree we should stick to Anaconda stable releases, even if I don't like suppressing indiscriminately all deprecation warnings...

Let's hope it will go away by itself at some point.

[oschulz]

Ok :-)