除非必要，否则如何避免使用 conda-forge 包？

Question

I'm running into a minor but annoying issue with conda when installing individual packages from conda-forge. Typically I'd run something like this:

conda install -c conda-forge somepackage

This has the effect of frequently updating other packages using conda-forge: almost always ca-certificates, certifi and openssl; and sometimes major packages like python, numpy etc (all from conda-forge). I suspect these updates are not strictly necessary; those packages get pulled in using some logic along the lines of "in order to install the latest version of somepackage I need python>=3.8 but the current python is 3.7; I can use conda-forge, so I'll get the latest python from conda-forge which is newer than the one in default".

What I would like to accomplish is: install any version of somepackage (not necessarily latest) while installing as little as possible from conda-forge. In particular, if there is a choice between keeping an already installed package and upgrading, I always want to keep; and if there is a choice between default channel and conda-forge, I always want default (even if those choices result in ending up with older versions of the new packages I'm trying to install, and/or their new dependencies - but not if they result in the install being impossible).

How do I accomplish this?

Answer 1

how do I tell conda "don't upgrade dependencies"

One can use conda install --freeze-installed PACKAGE... ( documentation ) to prevent conda from updating packages that are already installed. This does not, however, seem to prevent updating of packages under the aggressive_update_packages key of conda's config. The default packages there are ca-certificates , certifi , and openssl (see the default configuration ). One can check their own configuration with

conda config --show aggressive_update_packages

install any version of somepackage (not necessarily latest) while installing as little as possible from conda-forge

According to the conda documentation on managing channels , one can do this by putting the conda-forge channel at the bottom of the list of channels and setting channel_priority: strict in conda's config.

With strict channel priority, packages in lower priority channels are not considered if a package with the same name appears in a higher priority channel.

The "manage channels" page recommends setting strict as the default and indicates that conda 5.x will set strict as the default value for channel_priority .

One can write the following to their ~/.condarc file:

channel_priority: strict
channels:
  - defaults
  - conda-forge

However, if one uses the -c/--channel option in conda config , that channel will take precedence over any other channels. So if one sets their conda config as above but uses conda install -c conda-forge numpy , then numpy will be installed from conda-forge .

Answer 2

Generally, I endorse the conclusion in @jakub's answer : Define a reasonable global channel priority and try to avoid using ad hoc specifications via the --channel|-c argument.

There are some additional options that may be worth pointing out:

1. Specify that a particular package should come from a given channel with

    conda install conda-forge::somepackage

   and this will not change the channel priority.

2. Explicitly list channel priorities in the install command:

    conda install -c defaults -c conda-forge somepackage

   which puts defaults with top priority.

3. Define and manipulate environments through YAML files. YAMLs include a channels section, so you can be explicit about the priority. If you require a new package in the environment, it can be added to the YAML - let's say env.yaml - and that edit can be transferred to the environment with

    conda env update -f env.yaml