Fix Failing Actions

[simmsa]

Windows Failing Tests

• river functions that call delft_3d_get_all_time, on python 3.10, all matlab versions, mhkit-python 0.7.0.

Running River_TestIO_Delft
  ...
  ================================================================================
  Error occurred in River_TestIO_Delft/testGetAllTime and it did not run to completion.
      ---------
      Error ID:
      ---------
      'MATLAB:Python:ConverterError'
      --------------
      Error Details:
      --------------
      Error using py.numpy.ndarray/double
      Conversion of Python 'ndarray' type to MATLAB 'double' is only supported for real numbers and logicals.

      Error in delft_3d_get_all_time (line 26)
          result = double(python_result);

      Error in River_TestIO_Delft/testGetAllTime (line 37)
                  time = delft_3d_get_all_time(data);

Attempted fixes:

• Update numpy to 1.26.4
  • No change
• Convert output of d3d.get_all_time to a python list then convert to double

Working fix:

• Convert np masked array to np array in delft_3d_get_all_time: f7b6e7c

Unix Failing Tests

Error using ssl><module> (line 99)
      Python Error: ImportError:
      dlopen(/Users/runner/miniconda3/envs/mhkit_conda_env/lib/python3.9/lib-dynload/_ssl.cpython-39-darwin.so,
      0x0002): Symbol not found: _X509_STORE_get1_objects
        Referenced from: <DEADD0A1-08C8-3D77-A19C-08827189D194>
        /Users/runner/miniconda3/envs/mhkit_conda_env/lib/python3.9/lib-dynload/_ssl.cpython-39-darwin.so
            Expected in:     <F5E6CE2E-D06E-3F50-8C70-5B9602FE418A>
        /Users/runner/hostedtoolcache/MATLAB/2024.1.999/x64/MATLAB.app/bin/maci64/libcrypto.3.dylib

Starting point:

• Failing:
  • Macos:
    • 3.9, R2023b
    • 3.9 R2024a
    • 3.10 R2023b
    • 3.10 R2024a
    • 3.11 R2023b
    • 3.11 R2024a

Attempted Fixes:

• Modify dynamic library paths: b7ce780

Working Fix:

• Install specific openssl version, 3.0.* with conda and add it as the first entry in DYLD_LIBRARY_PATH: 2ae226e

[simmsa]

This is working. Summary of changes:

• Handle edge case in Windows, Python 3.10, MATLAB where a np masked array cannot be parsed
  • Converts np masked array to np array
• Update MacOS actions to use a specific version of OpenSSL installed by conda that is expected by MATLAB

[hivanov-nrel]

I think this makes sense to me. My only comments which probably stem from me not knowing the finer details:

• Fixing the github workflows helps our tests pass, but im curious if actual users would experience these issues in their environments. My hope is that things like having an up to date OpenSSL version should be something they already have and is just a weird edge case in our setup.
• The masked array fix makes sense, my only question if whether we are losing some type of data when forcing the conversion? I'm not sure why it's being used in the first place.

[simmsa]

@hivanov-nrel, good questions.

1. High level, there is a relatively high likelihood that testing these OS/MATLAB/Python version combinations via GitHub Actions won't be completely reliable due to the complexity of the software stack being tested. While I've been patching issues as they arise, I acknowledge this might not be the optimal strategy. We're balancing between testing all possible user configurations and maintaining reliable CI. For the macOS Actions changes here, my educated guess is that this is isolated to the GitHub Actions environment and wouldn't affect users' machines.

2. I'm not sure why the masked array is being used either. It could be an artifact of the output from the Delft3D model. However, since our tests are consistent between MHKiT-Python and MHKiT-MATLAB, we should catch any discrepancies in the calculations.

Given that these changes fix the immediate issues with the tests, I'll go ahead and merge this PR.