TODO

Essential
---------

- Check if scalar variables are handled correctly; if not, handle them
  specially, either by:
  - Fixing existing VarDiffs / VarInfo to handle scalar variables
  - Or making a separate VarDiffs class for scalar variables

- Compute and print other statistics, as is done for cprnc

- Print header information for each variable that matches fortran: include
  dimension names

- Print header information for the file comparisons, to match fortran
  - files being compared
  - fields present on one file but not the other
  - global attributes? (not essential)
  - dimension sizes? (not essential)
  - anything else?

- Look through Fortran for anything else that may be missing

- Run the Fortran cprnc test suite on the python version; compare output with
  the Fortran. We don't expect identical output, so this will require some
  hand-comparisons
  - First, patch the Fortran code so that its output more closely matches the
    python. The biggest change needed is a clean separation of the
    single-variable info from the variable diff stats.



Nice to have
------------

- Improve performance in other ways?

- Add one or more "system" tests
  - Much of what I was testing before with "system" tests of the Fortran
    (through the run_tests script) is now tested via unit tests. But it would be
    good to have one or a few system test(s), to at least make sure the whole
    thing runs. Maybe make these automated tests that just check the final
    "diff_test" line?


Optional
--------

- Convert some usage info in docstrings to doctest format

- When variables are separated along some dimension (e.g., time), find matching
  indices by examining the coordinate variables (as is done for the fortran
  version)

- Single-file analysis. I envision having a separate wrapper script for this,
  which wraps VarInfo (rather than trying to incorporate this into the same
  script, as is done for the Fortran - this makes the code hard to follow).

- Allow for alternative netCDF packages (e.g., the netCDF4 python package)
  - See if an alternative package would improve performance considerably