Add a function to find what to install from a `pylock.toml`

[brettcannon]

Something like:

def select(
    lock: Pylock,
    *,
    environment: markers.Environment | None = None,
    tags: Sequence[tags.Tag] | None = None,
    groups: Collection[str] | None = None
) -> ...:
    """Select what to install from the lock file.

    The *environment* and *tags* parameters represent the environment being
    selected for. If unspecified, ``packaging.markers.default_environment()``
    and ``packaging.tags.sys_tags()`` are used.

    The *groups* parameter represents the groups to install. If unspecified,
    the default groups are used.
    """

There's no specified return type as I haven't thought that through yet as there needs to be a way to tell what sort of thing was selected (e.g. wheel or sdist; might just be isinstance) and a way to get attestation identities (might just require to get back to the appropriate entry in [[packages]] by returning a tuple of (thing, entry)).

[brettcannon]

/cc @sbidoul

[sbidoul]

Sounds good. A few thoughts / questions:

• Returning an iterable or collection of (Package, PackageWheel | PackageSdist | PackageVcs | PackageArchive | PackageDirectory) looks good to me. In addition to attestation identities, having the corresponding Package is important so installers can find the sdist for no-binary installs, or select the index, or look at the tools table.
• Using isinstance to discriminate is fine with me.
• I suggest renaming the groups argument to dependency_groups?
• Should we add an extras argument to match the symmetry with dependency_groups reflected in the spec? Or use the extras environment marker. Not sure.

[brettcannon]

FYI I think this should be a method on Pylock.

• Returning an iterable or collection of (Package, PackageWheel | PackageSdist | PackageVcs | PackageArchive | PackageDirectory) looks good to me.

👍 That's what I was also thinking, but I wasn't sure of myself.

• Using isinstance to discriminate is fine with me.

I haven't checked how the classes look in a case statement, but I think that will be something to support to make this work well.

• I suggest renaming the groups argument to dependency_groups?

👍

• Should we add an extras argument to match the symmetry with dependency_groups reflected in the spec?

I think so since it's a top-level thing and I assume the installer will set it in the environment.