Package Management

Install, update, and manage packages in conda environments

Overview

Conda packages are compressed archives containing compiled software, libraries, and metadata. This guide covers how to install, update, and manage packages in your conda environments.

Conda handles package dependencies automatically, ensuring all required packages are installed together.

Package Formats

Conda supports two package formats:

Package Extensions

CONDA_PACKAGE_EXTENSION_V1 = ".tar.bz2"  # Legacy format
CONDA_PACKAGE_EXTENSION_V2 = ".conda"    # Modern format (faster)

The .conda format is significantly faster to extract and is preferred for new packages.

Installing Packages

Basic Installation

Use conda install to add packages to your environment:

Single Package

# Install into active environment
conda install numpy

# Install specific version
conda install numpy=1.24.3

# Install with version constraint
conda install "numpy>=1.24,<2.0"

Conda installs the newest compatible versions of requested packages by default. Use version specifiers to control which version is installed.

Installation Options

Install from Specific Channel

# Install from conda-forge
conda install -c conda-forge package_name

# Install from multiple channels
conda install -c conda-forge -c bioconda biopython

Install from File

# Install from requirements file
conda install --file requirements.txt

# Install from environment file
conda install --file environment.yml

Install Without Dependencies

# Skip dependency checking (use with caution)
conda install --no-deps package_name

Using --no-deps may result in an environment with incompatible packages. Use this option only when you’re certain dependencies are already satisfied.

Install Explicit Package Files

# Install from local file
conda install ./package-1.0.0-py311.tar.bz2

# Install from URL
conda install https://example.com/package-1.0.0-py311.conda

Installing explicit filenames implies --no-deps and cannot be mixed with package specifications.

Advanced Installation Options

Freeze Installed Packages

# Prevent updates to existing packages
conda install --freeze-installed new_package

Package Specifications

MatchSpec Syntax

Package specifications in conda use the MatchSpec format:

Version Constraints

# Exact version
numpy==1.24.3

# Greater than or equal
numpy>=1.24

# Range
"numpy>=1.24,<2.0"

# Compatible release
numpy~=1.24.0  # equivalent to >=1.24.0,<1.25.0

Updating Packages

Update Individual Packages

Update Single Package

# Update to latest version
conda update numpy

# Update in specific environment
conda update -n myenv numpy

# Update to specific version
conda install numpy=1.25.0

Update All Packages

# Update all packages in active environment
conda update --all

# Update all in specific environment
conda update -n myenv --all

Updating all packages may change many package versions. Consider using --dry-run first to preview changes.

Aggressive Update Packages

Certain packages are updated aggressively for security:

Always Updated Packages

DEFAULT_AGGRESSIVE_UPDATE_PACKAGES = (
    "ca-certificates",
    "certifi",
    "openssl",
)

These security-related packages are prioritized for updates to ensure your environment stays secure.

Update Modifiers

Control how dependencies are updated:

Update Strategies

# Update only specified packages (default)
conda install --update-specs numpy

# Update dependencies of specified packages
conda install --update-deps numpy

# Update all packages possible
conda install --update-all numpy

# Freeze installed packages (don't update deps)
conda install --freeze-installed numpy

Removing Packages

Remove Individual Packages

Remove Single Package

# Remove from active environment
conda remove numpy

# Remove from specific environment
conda remove -n myenv numpy

# Remove from path
conda remove -p /path/to/myenv numpy

Remove with Options

Remove Without Dependencies

# Remove package but keep dependencies
conda remove --no-deps numpy

Prune Unused Dependencies

# Remove packages that are no longer required
conda install --prune

Use --prune after removing packages to clean up orphaned dependencies.

Remove All Packages (Delete Environment)

# Remove all packages from environment
conda remove -n myenv --all

This deletes the entire environment. Use conda env remove instead for clarity.

Protected Packages

Some packages in the base environment cannot be removed:

Protected in Base Environment

ROOT_NO_RM = (
    "python",
    "pycosat",
    "ruamel.yaml",
    "conda",
    "openssl",
    "requests",
)

These packages are essential for conda’s operation and are protected from removal in the base environment.

Listing Packages

Basic Listing

List All Packages

# List packages in active environment
conda list

# List in specific environment
conda list -n myenv

# List in reverse order
conda list --reverse

Custom Output Formats

Export Format

# Export package list for reinstallation
conda list --export > package-list.txt

# Reinstall from export
conda create -n newenv --file package-list.txt

Explicit Format

# Export with full URLs
conda list --explicit > explicit-spec.txt

# Create exact replica
conda create -n replica --file explicit-spec.txt

Custom Fields

# Show only name and version
conda list --fields name,version

# Show with channel information
conda list --fields name,version,channel

Available fields:

List Fields

DEFAULT_CONDA_LIST_FIELDS = (
    "name",
    "version",
    "build",
    "channel_name",
)

# Other available fields:
# arch, build_number, channel, constrains, depends,
# dist_str, features, fn, license, license_family,
# md5, noarch, package_type, requested_spec, sha256,
# size, subdir, timestamp, track_features, url

Canonical Names

# Output canonical package names only
conda list --canonical

JSON Output

# Output as JSON for scripting
conda list --json

# Specific package in JSON
conda list numpy --json

Package Dependencies

Understanding Dependencies

Every package declares its dependencies in metadata. Conda automatically resolves and installs these dependencies.

Dependency Modifiers

class DepsModifier(Enum):
    NOT_SET = "not_set"      # Default: handle normally
    NO_DEPS = "no_deps"      # Skip dependencies
    ONLY_DEPS = "only_deps"  # Install only dependencies

Install Only Dependencies

# Install dependencies but not the package itself
conda install --only-deps numpy

Useful when you’re building the main package from source but need its dependencies.

Package Revisions

View Revision History

Conda tracks all changes to your environment:

# List all revisions
conda list --revisions

# Show specific revision
conda list --revision 5

Revert to Previous Revision

# Revert to specific revision number
conda install --revision 5

# View revision before reverting
conda install --revision 5 --dry-run

Reverting to a previous revision may downgrade packages or remove packages that were installed later.

Package Conflicts and Resolution

Path Conflicts

When packages try to install files to the same location:

Path Conflict Options

class PathConflict(Enum):
    clobber = "clobber"    # Overwrite files
    warn = "warn"          # Warn but continue
    prevent = "prevent"    # Stop installation

# Allow overwriting files
conda install --clobber package_name

Safety Checks

Safety Check Levels

class SafetyChecks(Enum):
    disabled = "disabled"  # No safety checks
    warn = "warn"          # Warn about issues
    enabled = "enabled"    # Enforce safety

Solver Selection

Conda can use different solvers for dependency resolution:

Available Solvers

DEFAULT_SOLVER = "libmamba"  # Modern, fast solver
CLASSIC_SOLVER = "classic"   # Legacy solver

# Use specific solver
conda install --solver=classic numpy

# Use default (libmamba)
conda install numpy

The libmamba solver is significantly faster than the classic solver and is now the default.

Package Cache Management

View Cache

# Show cache information
conda info

# List cached packages
conda clean --dry-run --packages

Clean Cache

# Remove unused cached packages
conda clean --packages

# Remove all cached data
conda clean --all

# Preview what would be removed
conda clean --all --dry-run

Regularly cleaning the package cache can free up significant disk space.

Partial Downloads

During downloads, partial files are marked:

Partial File Extension

PARTIAL_EXTENSION = ".partial"

If a download is interrupted, the .partial file remains and the download can be resumed.

Best Practices

Pin Important Packages

Prevent critical packages from being updated:

echo "python 3.11.*" >> $CONDA_PREFIX/conda-meta/pinned
echo "numpy 1.24.*" >> $CONDA_PREFIX/conda-meta/pinned

Use Dry Run Before Major Changes

Preview changes before committing:

conda update --all --dry-run
conda remove large_package --dry-run

Prefer Conda Packages Over Pip

When available, use conda packages instead of pip:

# Preferred
conda install numpy

# Fallback
pip install numpy

Conda packages include compiled dependencies and are better integrated.

Keep Base Environment Minimal

Install project-specific packages in separate environments:

# In base: only conda and essential tools
conda activate base
conda install conda-build

# Project packages in separate env
conda create -n project1 numpy pandas

Troubleshooting

Solving Environment Takes Forever

Try using the libmamba solver:

conda install --solver=libmamba package_name

Or reduce the search space:

conda install --freeze-installed package_name

Package Not Found

Check which channels are being searched:

conda config --show channels

Try a different channel:

conda install -c conda-forge package_name

Conflicts Between Packages

Create a new environment for conflicting requirements:

conda create -n separate_env package_with_conflict

Or use a constraints file to guide the solver.

Package Version Not Available

Search for available versions:

conda search package_name

# Search in specific channel
conda search -c conda-forge package_name

Next Steps

Channel Management

Learn how to configure and prioritize package channels

Environment Management

Manage environments for your packages