Conda-Vendor

Conda Vendor is a tool to create local conda channels and attestetation manifests based on an input conda environment yaml file. This tool is particularly useful when trying to use conda packages inside an air-gapped network.

Installation

To install with pip, run:

pip install conda-vendor

To install with conda, run:

conda install -c conda-forge conda-vendor

Development Installation

Create a conda environment with conda-vendor development dependencies:

mamba env create --file environment.yaml

Install conda-vendor as an editable pip package:

pip install -e .

Running Tests:

pytest tests/ -vvv -s

Usage

Supported Solvers

• Conda
• Mamba
• Micromamba

Vendor dependencies from an environment.yaml into a local channel:

# Use conda as the solver for linux-64
conda-vendor vendor --file environment.yaml --solver conda --platform linux-64

# use mamba as the solver for osx-64
conda-vendor vendor --file environment.yaml --solver mamba --platform osx-64

# use micromamba as the solver for the host platform
conda-vendor vendor --file environment.yaml --solver micromamba

Use Dry-Run install to verify that conda can solve using only the vendored channel:

# NOTE: ensure to use the same solver used to create the vendored channel
# example channel vendored with mamba: that includes micromamba python and pip
mamba create -n some-new-env --offline --channel ./my-vendored-channel --override-channels --dry-run micromamba python pip

Specifying Virtual Packages

Conda uses virtual packages as a way to track system dependencies. For more information about virtual packages see the conda documentation Managing virtual packages. Since these are not real packages they might not exist on the host system, or they might need to be manually specified. conda-vendor allows you to pass in a yaml file describing the virtual packages that are needed and their version.

By default conda-vendor uses a set of packages tied to the --platform options. To override this selection use the --virtual-package-spec options with the path to a spec yaml. The yaml should look something like

subdirs:
  linux-64:
    packages:
      __glibc: 2.17
      __cuda: 11.4

For more information see the documentation for conda-lock.

In order to assist with a correct choice of virtual packages, conda-vendor now includes a convenience command virtual-packages that will create an appropriately formatted virtual package spec for the system that conda vendor is running on.

$ conda-vendor virtual-packages

Which might output something like:

subdirs:
    osx-64:
        packages:
            __archspec: '1'
            __osx: 10.15.7
            __unix: '0'