feat: separate out remaining dependencies and improve tests

[jakelorocco]

Misc PR

Type of PR

• Bug Fix
• New Feature
• Documentation
• Other

Description

• Link to Issue: Fixes feat: improve package import handling and testing #629

We no longer need to explicitly version click. Typer is better about its click versioning in its own pyproject but the error is also gone.

Finishes up the dependency isolation tests; splits cli into it's own group; and fixes up import errors and dependency separation.

CI doesn't run the dependency isolation tests; ensured they pass locally.

All nice import error messages function like our current ones. The m cli works this way as well:

(mellea) ~/code/mellea4 ✓ % m --help
The 'm' CLI requires extra dependencies. Please install them with: pip install "mellea[cli]"

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code as added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

[github-actions]

The PR description has been updated. Please fill out the template for your PR to be reviewed.

[jakelorocco]

@frreiss, I believe you are getting pulled in just for the changes in mellea/formatters/granite/retrievers/util.py. This is a quality of life change but let me know if you see issues with it. Thanks!

[planetf1]

initial comments

[planetf1]

Tested the CLI install (following the original comments):

Scenario	Result
m (no extras)	Friendly error: pip install "mellea[cli]" ✓
m --help with [cli]	Works ✓
m serve script.py with [cli] only	Friendly error: pip install "mellea[server]" ✓
m alora train ... with [cli] only	Friendly error: pip install "mellea[hf]" ✓
m alora add-readme ... with [cli] only	Raw ModuleNotFoundError: No module named 'huggingface_hub' ✗

This PR didn't touch the code in the last scenario - so opened as a followup #855

[psschwei]

from copilot:

CI is failing in test/stdlib/components/docs/test_richdocument.py during fixture setup with:

omegaconf.errors.UnsupportedValueType: Value 'PosixPath' is not a supported primitive type
full_key: Global.model_root_dir

Root cause: Somewhere in the RichDocument/docling initialization path, Global.model_root_dir is being set as a pathlib.Path (PosixPath). OmegaConf configs only accept primitive types (str/int/float/bool/None), so it errors before tests run.

Fix: Cast Path to str before it goes into the OmegaConf dict:

from pathlib import Path

if model_root_dir is not None:
    model_root_dir = str(Path(model_root_dir))

cfg_dict["Global"]["model_root_dir"] = model_root_dir

If there may be multiple Path values, defensively stringify paths recursively before calling OmegaConf.create():

from pathlib import Path

def _stringify_paths(obj):
    if isinstance(obj, Path):
        return str(obj)
    if isinstance(obj, dict):
        return {k: _stringify_paths(v) for k, v in obj.items()}
    if isinstance(obj, list):
        return [_stringify_paths(v) for v in obj]
    return obj

cfg = OmegaConf.create(_stringify_paths(cfg_dict))

This should unblock test_richdocument_basics, test_richdocument_markdown, test_richdocument_save, and test_table.

Failing test file reference: test/stdlib/components/docs/test_richdocument.py at ref 733cc0c80cbc6590acc23b229db8210d24e66864.

---

[jakelorocco]

from copilot:

CI is failing in test/stdlib/components/docs/test_richdocument.py during fixture setup with:

...

It's an issue with package dependency requirements. Docling requires both rapidocr and omegaconf. The newest version of rapidocr is incompatible with older versions of omegaconf (when running on linux). I was able to test and setting omegaconf>= 2.1.0 works.

I'm not sure why only my PR seems to be hitting this issue though. I will see if the github action runners can succeed with this versioning change.

[psschwei]

from copilot:
CI is failing in test/stdlib/components/docs/test_richdocument.py during fixture setup with:
...

It's an issue with package dependency requirements.
...

not a ringing endorsement of the built-in "explain the error" functionality is it? 🤣

[jakelorocco]

Unless there are objections before tomorrow 9:00am EST, I will force merge this PR. It requires reviews from two groups that are unlikely to give reviews (since I'm usually the one reviewing for them) due to small import changes (not functional changes).

[ajbozarth]

It requires reviews from two groups that are unlikely to give reviews

This actually came up in our sync this AM, we may want to reevaluate blocking merging of PRs in a whole dir of the project on just two reviewers, especially if those two are often the ones doing the work, thus making it block on one person in those cases.

The easiest fix would be to expand those reviewer groups, the lazier version would be to disable blocking on sub teams and stick to main maintainers

[planetf1]

LGTM for improving dependency tests