Refactor for libmamba v2 (WIP)

[jaimergp]

Description

Adds compatibility for libmamba v2, which enables a few refactors (see below). Supersedes conda/conda#414.

• Will close conda with libmamba solver does not work in offline mode #522

Progress:

• Refactor index helpers
• Refactor repoquery
• Refactor solver
• Refactor conflict reports
• Pass all tests
• Polish and clean up

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• Add / update outdated documentation?

Summary of changes

• Deprecate CONDA_LIBMAMBA_SOLVER_NO_CHANNELS_FROM_INSTALLED. No longer needed with v2. Closes pip-integration: --dry-run still installs with pip conda#411
• Added typing for all methods and functions
• Added time_recorder decorator for metadata collection and solving loop
• Removed state.BaseIndexHelper. We just have our index.IndexHelper now
• The libmamba v1 pool (collection of repos, where a repo represents a loaded repodata.json) is now a database of RepoInfo objects.
• Do use current_repodata.json if explicitly set in CLI
• Move to Ruff for pre-commit linting & formatting
• Logging from libsolv has a HUGE overhead now because it goes from C to C++ to Python to logging to stdout instead of C -> stdout. To enable libsolv verbosity, an extra environment variable CONDA_LIBMAMBA_SOLVER_DEBUG_LIBSOLV is required.
• allow_uninstall was previously set to false (only true for conda remove), and we set it up for individual solver jobs involving updates or conflicts. With v2, we have individual control over what to "Keep" instead of drop. This required marking important updates as keepers instead. Otherwise they would be uninstalled.
• Other changes in the test suite discussed in Amend test suite for compatibility with libmamba v2 conda#13784

[jaimergp]

pre-commit.ci autofix

[jaimergp]

Since the canary workflow has a token issue for the uploads, I built it locally.

• Direct URL: https://anaconda.org/jaimergp/conda-libmamba-solver/24.7.0.125_ga3d9a3d/download/noarch/conda-libmamba-solver-24.7.0.125_ga3d9a3d-py_0.tar.bz2
• Logs: log.txt.zip
• Command:

conda install -c jaimergp/label/conda-libmamba-solver_dev -c conda-forge/label/mamba_dev -c conda-forge conda-libmamba-solver

[conda-bot]

Review build status

The workflow for building and uploading a canary release for conda-libmamba-solver with the label conda-libmamba-solver-pr-457 in channel conda-canary was started by @jaimergp!

Once it's done, use this command to try it out in a new conda environment:

conda install -c conda-canary/label/conda-libmamba-solver-pr-457 conda-libmamba-solver

• build 10474018722 succeeded on Linux (X64)

[jaimergp]

Those macOS crashes can be reproduced with:

>>> from libmambapy.specs import MatchSpec
>>> MatchSpec.parse("small-executable 1.0.0 0")
[1]    74412 trace trap  python

Apparently this is caused by libcxx 18. PR in mamba: mamba-org/mamba#3372

[jaimergp]

All green CI with the v2rc2, no code changes required.

[jezdez]

Some comments for clarification, but this looks really really good!

[jezdez]

What's needed to make this use main?

[jaimergp]

Merge conda/conda#13784. But it has this weird bootstrapping problem because those adjusted tests need this PR to be merged to pass 😬

[jezdez]

Instant classic! I mean, the real solution would be a third test running repo, right? In the meantime, we can clarify this in the upstream repo and mention this PR to make an exception IMO. Just say the word and I leave a statement about this in the conda PR so we can merge it.

[jezdez]

This would be excellent to clean up somehow

[jaimergp]

Unfortunately we can't get the "explained problems" in a structured way directly. We could get the graph (Unsolvable.problems_graph()) but then we need to redo the logic to obtain the actual conflicts, and that involves several Tree* objects only available in the C++ layer. So the simplest thing here is to parse line by line and trim the hierarchy indicators 😬