Use callback=reset_context in conda.plan

[jaimergp]

Description

• Fixes add_pip_as_python_dependency (and others) ignored when run via conda-build conda-libmamba-solver#393
• Fixes KeyError in some conda-build cases conda-libmamba-solver#386

Thanks to all the research done by @mbargull:

We need to use reset_context instead of reset_context_default to prevent he latter from clearing configurations coming from .condarc files (see conda/conda-libmamba-solver#393 for details). We also need callback instead of stack_callback, because:

1. conda.base.context.ContextStack is broken
   (pop/replace don't call self.apply).
2. conda.plan is the only consumer of ContextStack
   => we know nothing else adds to the stack => no need for it.
3. We know always use default search_path => no need for distinctions.

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• Add / update outdated documentation?

[mbargull]

Another thing I noticed was that SubdirData._cache_ does not use add_pip_as_python_dependency as a key.
We should add it here https://github.com/conda/conda/blob/23.10.0/conda/core/subdir_data.py#L77 .
Plus, the SubdirData.__init__ should then store the value of context.add_pip_as_python_dependency as an instance member (since loading of the actual data is not done during __init__ immediately, i.e., the value from context.add_pip_as_python_dependency could change between calls to __init__ and the value's actual usage in _process_raw_repodata afterwards.)

[jaimergp]

Nothing seems to be breaking. Errors are due to #13360.

[mbargull]

(I had this commit 2b308b1 in here too

Fix ContextStack.pop/replace reset last used args

N.B: After the recent change to conda.plan.install_actions to not use
stack_context_default anymore, conda.base.ContextStack and related
functions are only used for test code.
As such, that code may be marked as deprecated and moved to the tests
entirely.

to call ContextStack's apply in pop/replace.
That broke a lot of tests immediately (probably esp. due to changing replace).
So, let's not change that code for now :D.
But the nota bene still applies: We may want to mark it as deprecated and move it to the tests.)

[mbargull]

Tests fail because they seem broken.
They load the index with add_pip_as_python_dependency=False through conda.testing.helpers but don't set context.add_pip_as_python_dependency accordingly.
Plus there's a comment at https://github.com/conda/conda/blob/23.10.0/conda/testing/helpers.py#L600

    # We need CONDA_ADD_PIP_AS_PYTHON_DEPENDENCY=false here again (it's also in
    # get_index_r_*) to cover solver logics that need to load from disk instead of
    # hitting the SubdirData cache

that kind of foreshadows more issues when fixing the tests...

[mbargull]

Okay, the SubdirData._cache_ add_pip-awareness needs more investigation.
I'm splitting this off into a separate PR so we can at least go ahead with the conda.plan.install_actions change which on its own should unblock (at least many, if not all) conda-libmamba-solver+conda-build issues.

[mbargull]

Opened gh-13365 for the SubdirData._cache_ changes.
It would be good to get those in too since they are related to the index issues we are addressing here.
However, the issues with SubdirData._cache_ will likely not be encountered during normal conda/conda-build operation (add_pip_as_python_dependency should normally not change).

I'm adding this PR to the next milestone since it is effectively a blocker for conda-libmamba-solver (and thus conda>=23.10 with its new solver default) adoption in common conda-build uses (e.g., for conda-forge).

[mbargull]

I'm adding this PR to the next milestone since it is effectively a blocker for conda-libmamba-solver (and thus conda>=23.10 with its new solver default) adoption in common conda-build uses (e.g., for conda-forge).

And a note for other reviewers: The changes here are pretty much confined to conda-build since conda.plan.install_actions is not used within conda itself. Meaning, this is high impact for conda-build but low impact for conda.

[jaimergp]

I think it's time to mark tests/core/test_solve.py::test_conda_downgrade[libmamba] as xfailable (strict=False). See #13367.