Tips on creating a plugin to install min requirements

[tlambert03]

Hi @ofek,

If you have a moment, I'm looking for some quick tips before I embark on creating a plugin.

For the sake of testing, I like to be able to create an environment that installs the minimum declared dependencies, to ensure that I've set my pins correctly:

So if pyproject declares

[project]
dependencies = ["package>=0.3.4"]

I'd like it to specifically install package==0.3.4.

I've noticed that hatch-mypyc temporarily hides the pyproject.toml, and I thought something like that might be a good way to achieve this - by inspecting dependencies and temporarily rewriting them if the plugin is enabled via some environment variable or something.

Do you any broad suggestions on how you would go about that? (I'm not looking for code or anything, just a quick verification of the general ideas and perhaps pointers to hooks I should specifically use to accomplish it)

Thanks as always for your fantastic work

[ofek]

I would have your package contain one or two plugins:

• An environment type that subclasses the built-in virtual environment like this plugin does. You could then override the sync_dependencies method. This would work fine but the output of env show would have the original definitions which would be rightly confusing to users so instead I would override the undocumented dependencies_complex property which is what is used internally for everything. Just reference the parent property and return the modified data.
• (optional) An environment collector where every environment could be managed and automatically rewrite the environment dependencies (or based on some option to enable in each if off by default). This is mostly useful for detached environments.

[oprypin]

• I think it'd also be reasonable to make it a metadata hook plugin. To populate project.optional-dependencies.min-versions based on project.dependencies

  like what's being done manually here: https://github.com/mkdocs/mkdocs/blob/d6fcc56a3e2b287a2dc0310a7472ba4cdb177854/pyproject.toml#L35-L72

[tlambert03]

many thanks for taking the time!

[oprypin]

Did anyone go ahead and create such a plugin by chance?

[tlambert03]

and of course... feel free to comment over in the issues on that repo now

[juftin]

hatch-pip-compile recently started supporting this with its integration with uv - although it requires you use a pip-compile style lockfile:

[tool.hatch.env]
requires = [
    "hatch-pip-compile"
]

[tool.hatch.envs.default]
type = "pip-compile"
pip-compile-resolver = "uv"
pip-compile-args = [
    "--resolution=lowest-direct"
]

You can also leverage use uv's --resolution=lowest option to get the lowest compatible versions for all transitive dependencies too (although I've found this dicey sometimes).

[tlambert03]

awesome. good to know, thanks @juftin

[tlambert03]

(I added your comment to my readme over there as well)

[juftin]

Thank you! Very nice work with hatch-min-requirements - this is an awesome problem to solve.