I suspect that the current failure is due to something changing in Python internals that breaks OPS storage between different versions. Most likely, a change in bytecode. OPS stores the bytecode of CVs, and bytecode can change between Python minor versions. This means that files should only be opened with the same version of Python (or need to be opened in safe mode). We had to make a change based on this when Python 3.6 came out.

from looking at examples/ipynbtests.sh
it might be as easy as making another couple lines in:

case $PYTHON_VERSION in
    "2.7")
        mstis=$dropbox_base_url/1x4ny0c93gvu54n/toy_mstis_1k_OPS1.nc
        mistis=$dropbox_base_url/qaeczkugwxkrdfy/toy_mistis_1k_OPS1.nc
        ;;
    "3.6")
        mstis=$dropbox_base_url/1ulzssv5p4lr61f/toy_mstis_1k_OPS1_py36.nc
        mistis=$dropbox_base_url/76981cbgxm639m3/toy_mistis_1k_OPS1_py36.nc
        ;;
    "3.7")
        mstis=$dropbox_base_url/1ulzssv5p4lr61f/toy_mstis_1k_OPS1_py36.nc
        mistis=$dropbox_base_url/76981cbgxm639m3/toy_mistis_1k_OPS1_py36.nc
        ;;
    *)
        echo "Unsupported Python version: $PYTHON_VERSION"
esac

(As 3.6 and 3.7 already are identical?)

(if that is the case, please also make a note in .github/workflows/tests.yml to remind future developers to add these whenever the testing matrix is changed)

@sroet : Yes, that's pretty much it. Basically, I think all I need to do is re-run the example in each version of Python, and then save the file with 1000 MC steps (which should be enough that TIS analysis has enough overlap), stick that in my Dropbox, and update with the appropriate links.

Adding a comment to the CI yml is a good idea. It isn't that you necessarily need to do this every time (the file from 3.6 also works with 3.7), but it is a problem that can occur and should be straightforward to fix when it comes up.

Probably won't get to that today, though.

Codecov Report

Merging #993 (b9c1c89) into master (5f86032) will increase coverage by 0.52%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master     #993      +/-   ##
==========================================
+ Coverage   80.62%   81.15%   +0.52%     
==========================================
  Files         138      138              
  Lines       14695    15118     +423     
==========================================
+ Hits        11848    12269     +421     
- Misses       2847     2849       +2     

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5f86032...b9c1c89. Read the comment docs.

Summary of changes here:

• Added 3.8 and 3.9 to the testing matrix (including adding new example data files based on 3.8)
• Added script examples/prep_example_data.py to create data for analysis. It turns out that our old setups used a version of the examples that started with trajectories that met the requirements for the outermost ensembles. This ensured overlap in the TIS analysis, but this approach has long since been removed from the actual examples. The actual examples need more than 1000 steps to get overlap for TIS analysis. This new script sets things up using the old approach (and even figures out the Python version and labels the output file correctly).
• Changed tutorial_storage.ipynb to load CV by name, not number. The datafiles we have now seem to go back to before we had "CVs" that take trajectories as input. storage.cv[0] in current files is a trajectory CV giving the maximum value of an order parameter over the trajectory. This does not give a list for output, which caused issues in the notebook.
• Minor change to matrix so that minimal build shows in GitHub as "(3.7, minimal)" instead of "(3.7, true)".
• Removed omnia from channels
• Updated trove classifiers to match our tested versions

This is ready for review and comment. I will leave it open for at least 24 hours, merging no earlier than Wed 07 Apr 20:00 GMT (22:00 local).