Dependencies

[wpbonelli]

Just some thoughts on dependencies for flopy4.

We could make matplotlib (and any other future plotting deps) optional, pandas for instance does this. We could also consider finer-grained dependency groups like they do.

So far we've done some prototyping with the following

• attrs
• cattrs
• boltons
• lark
• mypy
• Jinja2

I think at minimum, per the current design sketch, the core would need

• c/attrs
• Jinja2

Along with numpy and pandas which are existing requirements.

Lark could be used to parse input files but still need to see how performance compares to hand-parsing.

Boltons is a small dependency-free set of utilities I've found generally useful but by no means necessary.

MyPy (or another type-checker) would be a developer dependency. Alternatives could be

• beartype
• pyright

I'm intrigued by beartype since it seems friendly to incremental adoption, and runtime checking lets us verify type hints when we apply them to existing flopy modules.

[wpbonelli]

We'll also need a TOML library. The most popular seem to be

• https://github.com/uiri/toml
• https://github.com/hukkin/tomli
• https://github.com/python-poetry/tomlkit

tomli is fastest, but doesn't preserve comments/style — but really, we shouldn't be relying on configuration file cosmetics, so this is probably the best option.

[wpbonelli]

quick/dirty benchmark, to load all the files in the input spec in serial, tomli takes ~1.5s, tomlkit takes ~4.1s.

[wpbonelli]

collecting some examples of what other libraries do

• like flopy3, pyemu keeps the core small and defines an optional optional group https://github.com/pypest/pyemu/blob/da5735fa008ff0b796eee6dc29a49613d7fd1556/pyproject.toml#L27
• astropy keeps the core small and defines a recommended optional group with plotting deps as well as several other optional groups https://github.com/astropy/astropy/blob/9aedf2b90259ba0d57fe109f75aec1ec7c422599/pyproject.toml#L45
• metpy has a slightly bigger core including e.g. matplotlib and pyproj but keeps fancier stuff e.g. cartopy in optional groups https://github.com/Unidata/MetPy/blob/84aa93a54cf3ed573ffff1dbb591364a68561348/pyproject.toml#L28
• xclim keeps more or less everything a user would likely want in the core https://github.com/Ouranosinc/xclim/blob/ee2e34ed56d04ec2e4941cf480d44d518dac2a97/pyproject.toml#L36