Mamba does'nt find a solution when mixing conda-forge/defaults and not specifying Python explicitely

[jolos]

on mamba >= 0.15 the following environment file (adapted from the quetz repo) fails to find a solution and progress to installation of the packages

name: quetz
channels:
  - defaults
  - https://repo.mamba.pm/conda-forge
dependencies:
  #- python>=3.7
  - pip
  - fastapi
  - typer
  - authlib
  - psycopg2
  - httpx=0.16.0
  - sqlalchemy<1.4.0
  - sqlalchemy-utils
  - sqlite
  - python-multipart
  - uvicorn
  - zstandard
  - conda-build
  - appdirs
  - toml
  - quetz-client
  - fsspec
  - requests
  - h2<4.0.0
  - pluggy
  - jinja2
  - itsdangerous
  - alembic
  - zchunk
  - s3fs
  - sphinx
  - sphinx-book-theme
  - tenacity
  - xattr
  - aiofiles
  - pyyaml
  - ujson
  - prometheus_client
  - pamela
  - typing_extensions
  - adlfs

Any of the following actions resolves the issue

• removing the defaults channel
• uncommenting the python requirement
• downgrading to mamba 0.14
• reducing the number of requirements in the file.

[mparry]

Have you got channel_priority: strict configured? Under mamba 0.15+ you probably want to change that to channel_priority: flexible, or just remove that config altogether since flexible is the default. We saw similar after updating.

This is the relevant bit of the changelog:

Note that this changes the meaning of strict and flexible priority as the previous implementation did not follow conda's semantics. Mamba now has three modes, just like conda: strict, flexible and disabled. Strict will completely disregard any packages from lower-priority channels if a package of the same name exists in a higher priority channel. Flexible will use packages from lower-priority channels if necessary to fulfill dependencies or explicitly requested (e.g. by version number). Disabled will use the highest version number, irregardless of the channel order.

[jolos]

That doesn't seem to resolve the issue, it persists regardless if I set it to strict/flexible. However setting it to disabled does seem to help. Not sure what the implications of that are.

[adriendelsalle]

Hi @jolos!

Could you please describe the error you get?
I just tried to reproduce but it works locally on linux-64 even if it's terribly long to solve with conda, mamba or micromamba.
Looks like env create doesn't support channel_priority..

Running micromamba create -f env.yml --strict-channel-priority solves ubber fast.

[jolos]

Thanks for checking @adriendelsalle, there's no error showing up, it's just that it indeed it takes terribly long. Perhaps if I would leave it running for a while it would eventually find a solution. But so far I always kill it when I don't see any progress after 20min.
Given that mamba is typically super snappy, to me that feels as a regression compared to 0.14.

I confirm that using micromamba works (both with and without strict-channel-priority) using conda env create work as well.

[TejasAvinashShetty]

similar issue here. Just adding python to the list of packages, without altering strict, etc solves the issue.