Fail gracefully if offline and no cache is available

[jaimergp]

Description

Needed for conda/conda#12984

Checklist - did you ...

• Add a file to the news directory (using the template) for the next release's release notes?
• Add / update necessary tests?
• Add / update outdated documentation?

[costrouc]

Tested and can confirm in pytest tests/test_create.py -k test_offline_with_empty_index_cache\[libmamba\] that the test continues but still does eventually fail. The error messages though are more helpful in debugging.

[costrouc]

I will be investigating the tests that fail and why it is happening reliably on windows. https://github.com/conda/conda-libmamba-solver/actions/runs/6513586579/job/17735028419?pr=323#step:9:935

[costrouc]

Triggering a rerun and seeing what happens. I ran this test locally on a windows vm and wasn't able to reproduce.

[costrouc]

Spent a long time trying to debug the windows issue. I see a difference of assert PATH0.lower() == PATH.lower() in left side having /usr/bin in the path and the other has /bin in the path. On line https://github.com/conda/conda/blob/e784a57001b129804f17e10cce4cd31400be8103/tests/test_activate.py#L2696. I'm wondering how we test this. @jaimergp any ideas?

[jaimergp]

I'd say we skip it for now because it's not solver related at all. Can you deselect in the pytest collection plugin? Thanks!

[jaimergp]

tests/conda_env/test_cli.py::test_update_env_no_action_json_output errors are legit but caused by a unique situation. This is the gist:

• Having an env.yml file with python=3, pip and pip: click, we should be able to create it with conda env create.
• A subsequent call to conda env update should do nothing because we just installed everything.

Makes sense, but this is what we observe in this error:

1. conda env create creates an environment with python=3.11 and pip=23.3. It prioritizes having a greater pip version than a Python one, possibly because it comes earlier in the sorting (i beats y).
2. However, conda env update imposes python=3 again (not just python!), which is interpreted as "update python". This time, the solver tries to satisfy the petition more aggressively, even allowing a pip downgrade this time, because the alternative is to say "we are ok".

We didn't see this error earlier because the packaging situation is unique: defaults has Python 3.12 now, but only pip 23.2 is available for that version. The most recent 23.3 is only available for Python 3.11. I'm assuming once the new pip is available for 3.12 this will pass, but the currently observed behavior is a bit surprising and might deserve a fix.

[jaimergp]

Note that changing the environment.yml file to the following makes everything pass as expected:

ENVIRONMENT_PYTHON_PIP_CLICK = f"""
name: {TEST_ENV_NAME_1}
dependencies:
- - python=3
+ - python
  - pip
  - pip:
    - click
channels:
  - defaults
"""

[jaimergp]

This alternative change in solver.py also makes things pass. MatchSpec.merge sorts everything by str, and we used to keep that:

        self.specs_to_add = IndexedSet(MatchSpec.merge(s for s in fixed_specs))

However, if we prio Python, it passes:

        resorted_specs = []
        for spec in MatchSpec.merge(s for s in fixed_specs):
            if spec.name == "python":
                resorted_specs.insert(0, spec)
            else:
                resorted_specs.append(spec)
        self.specs_to_add = IndexedSet(resorted_specs)