CondaPkg.jl

Add Conda dependencies to your Julia project.

This package is a lot like Pkg from the Julia standard library, except that it is for managing Conda packages.

• Conda dependencies are defined in CondaPkg.toml, which is analogous to Project.toml.
• CondaPkg will install these dependencies into a Conda environment specific to the current Julia project. Hence dependencies are isolated from other projects or environments.
• Functions like add, rm, status exist to edit the dependencies programmatically.
• Or you can do pkg> conda add some_package to edit the dependencies from the Pkg REPL.

pkg> add CondaPkg

The simplest way to specify Conda dependencies is through the Pkg REPL, just like for Julia dependencies. For example:

julia> using CondaPkg
julia> # now press ] to enter the Pkg REPL
pkg> conda status                # see what we have installed
pkg> conda add python perl       # adds conda packages
pkg> conda pip_add build         # adds pip packages
pkg> conda rm perl               # removes conda packages
pkg> conda run python --version  # runs the given command in the conda environment
pkg> conda update                # update conda and pip installed packages

For more information do ? or ?conda from the Pkg REPL.

Note: We recommend against adding Pip packages unless necessary - if there is a corresponding Conda package then use that. Pip does not handle version conflicts gracefully, so it is possible to get incompatible versions.

These functions are intended to be used interactively when the Pkg REPL is not available (e.g. if you are in a notebook):

• status() shows the Conda dependencies of the current project.
• add(pkg; version="", channel="") adds/replaces a dependency or a vector of dependencies.
• rm(pkg) removes a dependency or a vector of dependencies.
• add_channel(channel) adds a channel.
• rm_channel(channel) removes a channel.
• add_pip(pkg; version="") adds/replaces a pip dependency.
• rm_pip(pkg) removes a pip dependency.

Finally, you may edit the CondaPkg.toml file directly. Here is a complete example:

channels = ["anaconda", "conda-forge"]

[deps]
# Conda package names and versions
python = ">=3.5,<4"
pyarrow = "==6.0.0"
perl = ""

[deps.llvmlite]
# Long syntax to specify other fields, such as the channel and build
version = ">=0.38,<0.39"
channel = "numba"
build = "*"

[pip.deps]
# Pip package names and versions
build = "~=0.7.0"
six = ""
some-remote-package = "@ https://example.com/foo.zip"
some-local-package = "@ ./foo.zip"

[pip.deps.nmslib]
# Long syntax to specify other fields
version = "~=2.1"
binary = "no"  # or "only"

• envdir() returns the root directory of the Conda environment.
• withenv(f) returns f() evaluated in the Conda environment.
• which(progname) find the program in the Conda environment.
• resolve(; force=false) resolves dependencies. You don't normally need to call this because the other API functions will automatically resolve first. Pass force=true if you change a CondaPkg.toml file mid-session.
• update() update the conda and pip installed packages.
• gc() removes unused caches to save disk space.

Assuming one of the dependencies in CondaPkg.toml is python then the following runs Python to print its version.

# Simplest version.
CondaPkg.withenv() do
  run(`python --version`)
end
# Guaranteed not to use Python from outside the Conda environment.
CondaPkg.withenv() do
  python = CondaPkg.which("python")
  run(`$python --version`)
end
# Explicitly specifies the path to the executable (this is package-dependent).
CondaPkg.withenv() do
  python = joinpath(CondaPkg.envdir(), Sys.iswindows() ? "python.exe" : "bin/python")
  run(`$python --version`)
end

You can specify the channel to install a particular package from, such as with pkg> conda add some-channel::some-package.

You can also specify a top-level list of channels, from which all other packages are installed, such as with pkg> conda channel_add some-channel.

By default, packages are installed from the conda-forge channel.

Direct references such as foo@http://example.com/foo.zip are allowed. As a special case if the URL starts with . then it is interpreted as a path relative to the directory containing the CondaPkg.toml file.

Additionally the binary mode specifies whether to only use binary distributions ("only") or to never use them ("no").

You can configure this package with a number of preferences. These can be set either as Julia preferences or as environment variables. This table gives an overview of the preferences, and later sections describe them in more detail.

The easiest way to set these preferences is with the PreferenceTools package. For example:

julia> using PreferenceTools
julia> # now press ] to enter the Pkg REPL
pkg> preference add CondaPkg backend=System offline=true

This package has a number of different "backends" which control exactly which implementation of Conda is used to manage the Conda environments. You can explicitly select a backend by setting the backend preference to one of the following values:

• MicroMamba: Uses MicroMamba from the package MicroMamba.jl.
• System: Use a pre-installed Conda. If the exe preference is set, that is used. Otherwise we look for conda, mamba or micromamba in your PATH.
• Current: Use the currently activated Conda environment instead of creating a new one. This backend will only ever install packages, never uninstall. The Conda executable used is the same as for the System backend. Similar to the default behaviour of Conda.jl.
• Null: Don't use CondaPkg to manage dependencies. Use this if you are in a pre-existing Conda environment that already satisfies the dependencies of your project. It is up to you to ensure any required packages are installed.

The default backend is an implementation detail, but is currently MicroMamba.

If you set the exe preference but not the backend preference then the System backend is used.

You may activate "offline mode" by setting the preference offline=true. This will prevent CondaPkg from attempting to download or install new packages. In this case, it is up to you to ensure that any required packages are already available (such as by having previously called CondaPkg.resolve()).

By default, CondaPkg installs Conda packages into the current project, so that different projects can have different dependencies. If you wish to centralize the Conda environment, you can set one of these preferences:

• env=@<name> for a named shared environment, stored in ~/.julia/conda_environments/<name>.
• env=<some absolute path> for a shared environment at the given path.
• backend=Current to use the currently activated Conda environment.

Warning: If you do this, the versions specified in a per-julia-version CondaPkg.toml can become un-synchronized with the packages installed in the shared Conda environment. In this case, you will have to re-resolve the dependencies using resolve(force=true). This restriction might be alleviated in future CondaPkg versions.

You can control the verbosity of any conda or pip commands executed by setting the verbosity preference to a number:

• -1 is quiet mode.
• 0 is normal mode (the default).
• 1, 2, etc. are verbose modes, useful for debugging.

No. The location of the Conda environment is configured purely by the user. Letting packages specify this configuration is not composable - if two packages want to set the location of the environment, then they will be in conflict.

Yes, you can add the following to your startup file (~/.julia/config/startup.jl):

Base.identify_package("CondaPkg") === nothing || Base.require(@__MODULE__, :CondaPkg)

Yes, using the "direct reference" @ version syntax. For example in PKG REPL mode

pkg> pip_add some-package@https://example.com/the/url

or using the API

CondaPkg.add_pip("some-package", version="@https://example.com/the/url")

or in CondaPkg.toml

[pip.deps]
"some-package" = "@https://example.com/the/url"