write_restart_file/write_partial_restart_file validate directions axis against number_of_frequencies, not number_of_directions

[cgrudz]

Summary

Both write_restart_file and write_partial_restart_file in src/roguewave/wavewatch3/io.py validate the spectra array's directions axis (shape[2]) against parent_restart_file.number_of_frequencies instead of number_of_directions:

• write_restart_file, lines 177–179
• write_partial_restart_file, lines 254–256

Impact

For any grid where number_of_frequencies != number_of_directions, writing a valid spectra array of the correct shape raises a spurious ValueError. Conversely, if a future change ever produced a genuinely wrong-shaped directions axis that happened to equal the frequency count, this check would silently pass instead of catching it.

Why this was never caught

tests/restart_files/'s fixture restart file has number_of_frequencies == number_of_directions == 36 (see test_number_of_directions/test_number_of_frequencies in test_restart_file.py), so shape[2] != number_of_frequencies and shape[2] != number_of_directions are indistinguishable for that fixture. Neither test_io.py's write_partial_restart_file/reassemble_restart_file_from_parts tests nor anything else in the suite exercises a grid with unequal counts.

Real-world reproduction

Sofar's own esm-run-scripts/preprocess_waves/setup/containerized/grid.sh defines grid configs with frequency_bins=50, direction_bins=36 — writing a restart against one of these would trip this bug today.

Suggested fix

Change shape[2] != parent_restart_file.number_of_frequencies to shape[2] != parent_restart_file.number_of_directions in both functions, and add a test fixture (or a second synthetic grid) with n_freq != n_dir so this class of bug can't hide again.

[pieterbartsmit]

Yyp; just put in a pull request @cgrudz

Aside why do we set frequency_bins=50 in esm? We use 36 in prod