Skip to content
Merged
311 changes: 311 additions & 0 deletions docs/concepts/package_specifications.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,311 @@
# Package Specifications

When adding packages to your Pixi workspace or global environment, you can use various specifications to control exactly which package version and build you want.
This is particularly important when packages have multiple builds for different hardware configurations (like CPU vs GPU).
For the conda packages Pixi uses the [MatchSpec](https://rattler.prefix.dev/py-rattler/match_spec#matchspec) format to specify package requirements.
For PyPI packages, Pixi uses the standard [PEP440 version specifiers](https://peps.python.org/pep-0440/).

## Quick Examples

=== "Add"
```bash
--8<-- "docs/source_files/shell/package_specifications.sh:quick-add-examples"
```
=== "Global"
```bash
--8<-- "docs/source_files/shell/package_specifications.sh:quick-global-examples"
```
=== "Exec"
```bash
--8<-- "docs/source_files/shell/package_specifications.sh:quick-exec-examples"
```

## Basic Version Specifications
Pixi uses the [**conda MatchSpec**](https://rattler.prefix.dev/py-rattler/match_spec#matchspec) format for specifying package requirements.
A MatchSpec allows you to precisely define which package version, build, and channel you want.

The simplest way to specify a package is by name and [optional version operators](#version-operators):

```toml title="pixi.toml"
--8<-- "docs/source_files/pixi_tomls/package_specifications.toml:basic-version"
```

## Full MatchSpec Syntax

Beyond simple version specifications, you can use the full [MatchSpec](https://rattler.prefix.dev/py-rattler/match_spec#matchspec) syntax to precisely control which package variant you want.

### Command Line Syntax

Pixi supports two syntaxes on the command line:

**1. Equals syntax (compact):**
```shell
# Format: package=version=build
pixi add "pytorch=2.0.*=cuda*"
# Only build string (any version)
pixi add "numpy=*=py311*"
```

**2. Bracket syntax (explicit):**
```shell
# Format: package [key='value', ...]
pixi add "pytorch [version='2.0.*', build='cuda*']"
# Multiple constraints
pixi add "numpy [version='>=1.21', build='py311*', channel='conda-forge']"
# Build number constraint
pixi add "python [version='3.11.0', build_number='>=1']"
```

Both syntaxes are equivalent and can be used interchangeably.

### TOML Mapping Syntax

In your `pixi.toml`, use the mapping syntax for complete control:

```toml title="pixi.toml"
--8<-- "docs/source_files/pixi_tomls/package_specifications.toml:mapping-syntax-full"
```

This syntax allows you to specify:

- **version**: Version constraint using [operators](#version-operators)
- **build**: Build string pattern (see [build strings](#build-strings))
- **build-number**: Build number constraint (e.g., `">=1"`, `"0"`) (see [build number](#build-number))
- **channel**: Specific channel name or full URL (see [channel](#channel))
- **sha256/md5**: Package checksums for verification (see [checksums](#checksums-sha256md5))
- **license**: Expected license type (see [license](#license))
- **file-name**: Specific package file name (see [file name](#file-name))

### Version Operators

Pixi supports various version operators:

| Operator | Meaning | Example |
|----------|---------|---------|
| `==` | Exact match | `==3.11.0` |
| `!=` | Not equal | `!=3.8` |
| `<` | Less than | `<3.12` |
| `<=` | Less than or equal | `<=3.11` |
| `>` | Greater than | `>3.9` |
| `>=` | Greater than or equal | `>=3.9` |
| `~=` | Compatible release | `~=3.11.0` (>= 3.11.0, < 3.12.0) |
| `*` | Wildcard | `3.11.*` (any 3.11.x) |
| `,` | AND | `">=3.9,<3.12"` |
| `|` | OR | `"3.10|3.11"` |

### Build Strings

Build strings identify specific builds of the same package version.
They're especially important for packages that have different builds for:

- **Hardware acceleration**: CPU, GPU/CUDA builds
- **Python versions**: Packages built for different Python interpreters
- **Compiler variants**: Different compiler versions or configurations

A build string typically looks like: `py311h43a39b2_0`

Breaking it down:

- `py311`: Python version indicator
- `h43a39b2`: Hash of the build configuration
- `_0`: Build number

You can use wildcards in build patterns to match multiple builds:

```shell
# Match any CUDA build
pixi add "pytorch=*=cuda*"

# Match Python 3.11 builds
pixi add "numpy=*=py311*"

# Using bracket syntax
pixi add "pytorch [build='cuda*']"
```

### Build Number

The build number is an integer that increments each time a package is rebuilt with the same version.
Use build number constraints when you need a specific rebuild of a package:

```shell
# Specific build number
pixi add "python [version='3.11.0', build_number='1']"

# Build number constraint
pixi add "numpy [build_number='>=5']"
```

```toml title="pixi.toml"
[dependencies.python]
version = "3.11.0"
build-number = ">=1"
```

Build numbers are useful when:

- A package was rebuilt to fix a compilation issue
- You need to ensure you have a specific rebuild with bug fixes
- Working with reproducible environments that require exact builds

### Channel

Channels are repositories where conda packages are hosted.
You can specify which channel to fetch a package from:

```shell
# Specific channel by name
pixi add "pytorch [channel='pytorch']"

# Channel URL
pixi add "custom-package [channel='https://prefix.dev/my-channel']"

# Or use the shorter `::` syntax
pixi add pytorch::pytorch
pixi add https://prefix.dev/my-channel::custom-package
```

```toml title="pixi.toml"
[dependencies.pytorch]
channel = "pytorch"

[dependencies.custom-package]
channel = "https://prefix.dev/my-channel"
```

Note that for `pixi add`, channels must also be listed in your workspace configuration:

```toml title="pixi.toml"
[workspace]
channels = ["conda-forge", "pytorch", "nvidia"]
```

You can also add these using the command line:

```shell
pixi workspace channel add conda-forge
```

### Checksums (SHA256/MD5)

Checksums verify package integrity and authenticity.
Use them for reproducibility and security:

```toml title="pixi.toml"
[dependencies.numpy]
version = "1.21.0"
sha256 = "1234567890abcdef1234567890abcdef1234567890abcdef1234567890abcdef"
md5 = "abcdef1234567890abcdef1234567890"
```

When specified, Pixi will:

- Verify the downloaded package matches the checksum
- Fail the installation if checksums don't match
- Ensure you get the exact package you expect

SHA256 is preferred over MD5 for security reasons.

### License

Specify the expected license of a package.
This is useful for:

- Ensuring compliance with your organization's policies
- Filtering packages by license type
- Documentation purposes

```toml title="pixi.toml"
[dependencies.pytorch]
version = "2.0.*"
license = "BSD-3-Clause"
```

### File Name

Specify the exact package file name to download.
This is rarely needed but useful for:

- Debugging package resolution issues
- Ensuring a specific package artifact
- Advanced use cases with custom package builds

```toml title="pixi.toml"
[dependencies.pytorch]
file-name = "pytorch-2.0.0-cuda.tar.bz2"
```

## Source Packages

!!! warning
`pixi-build` is a preview feature, and will change until it is stabilized.
Please keep that in mind when you use it for your workspaces.

For these packages to be recognized they need to be understood by Pixi as source packages.
Look at the [Pixi Manifest Reference](../reference/pixi_manifest.md#the-package-section) for more information on how to declare source packages in your `pixi.toml`.

### Path-based Source Packages
```toml title="pixi.toml"
--8<-- "docs/source_files/pixi_tomls/package_specifications.toml:path-fields"
```

The path should be relative to the workspace root or an absolute path, but absolute paths are discouraged for portability.


### Git-based Source Packages
```toml title="pixi.toml"
--8<-- "docs/source_files/pixi_tomls/package_specifications.toml:git-fields"
```

For git repositories, you can specify:

- **git**: Repository URL
- **branch**: Git branch name
- **tag**: Git tag
- **rev**: Specific git revision/commit SHA
- **subdirectory**: Path within the repository

## PyPI package specifications

Pixi also supports installing package dependencies from PyPI using `pixi add --pypi` and in your `pixi.toml` and `pyproject.toml`.
Pixi implements the standard [PEP440 version specifiers](https://peps.python.org/pep-0440/) for specifying package versions.

### Command Line Syntax

When using `pixi add --pypi`, you can [specify packages similarly to `pip`](https://pip.pypa.io/en/stable/user_guide/#installing-packages):

```shell
# Simple package
pixi add --pypi requests

# Specific version
pixi add --pypi "requests==2.25.1"

# Version range
pixi add --pypi "requests>=2.20,<3.0"

# Extras
pixi add --pypi "requests[security]==2.25.1"

# URL
pixi add --pypi "requests @ https://files.pythonhosted.org/packages/1e/db/4254e3eabe8020b458f1a747140d32277ec7a271daf1d235b70dc0b4e6e3/requests-2.32.5-py3-none-any.whl#sha256=2462f94637a34fd532264295e186976db0f5d453d1cdd31473c85a6a161affb6"

# Git repository
pixi add --pypi "requests @ git+https://github.com/psf/[email protected]"
pixi add --pypi requests --git https://github.com/psf/requests.git --tag v2.25.1
pixi add --pypi requests --git https://github.com/psf/requests.git --branch main
pixi add --pypi requests --git https://github.com/psf/requests.git --rev 70298332899f25826e35e42f8d83425124f755a
```

### TOML Mapping Syntax
In your `pixi.toml` or `pyproject.toml` (under `[tool.pixi.pypi-dependencies]`), you can specify PyPI packages like this:

```toml title="pixi.toml"
--8<-- "docs/source_files/pixi_tomls/package_specifications.toml:pypi-fields"
```

## Further Reading

- [Pixi Manifest Reference](../reference/pixi_manifest.md#dependencies) - Complete dependency specification options
- [Multi-Platform Configuration](../workspace/multi_platform_configuration.md) - Platform-specific dependencies
- [Conda Package Specification](https://conda.io/projects/conda/en/latest/user-guide/concepts/pkg-specs.html) - Conda's package specification docs
15 changes: 15 additions & 0 deletions docs/reference/cli/pixi/add_extender
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,21 @@ pixi add --git https://github.com/mahmoud/boltons.git boltons --tag v0.1.0 --pyp
25. This will add the `boltons` package with the given `git` url and `v0.1.0` tag as `pypi` dependency.
26. This will add the `boltons` package with the given `git` url, `v0.1.0` tag and the `boltons` folder in the repository as `pypi` dependency.

!!! tip "Need to specify build strings or hardware-specific packages?"
For advanced package specifications including build strings, see the [Package Specifications](../../../concepts/package_specifications.md) guide.

Examples (both syntaxes work):
```shell
# Equals syntax (compact)
pixi add "pytorch=*=*cuda*"
pixi add "numpy=*=py311*"

# Bracket syntax (explicit)
pixi add "pytorch [build='*cuda*']"
pixi add "numpy [build='py311*']"
```


!!! tip
If you want to use a non default pinning strategy, you can set it using [pixi's configuration](../../pixi_configuration.md#pinning-strategy).
```
Expand Down
27 changes: 20 additions & 7 deletions docs/reference/cli/pixi/global/install_extender
Original file line number Diff line number Diff line change
Expand Up @@ -8,8 +8,6 @@ pixi global install ruff
pixi global install starship rattler-build
# Specify the channel(s)
pixi global install --channel conda-forge --channel bioconda trackplot
# Or in a more concise form
pixi global install -c conda-forge -c bioconda trackplot

# Support full conda matchspec
pixi global install python=3.9.*
Expand All @@ -33,14 +31,29 @@ pixi global install --expose "py39=python3.9" "python=3.9.*"

--8<-- [start:description]

!!! tip "Need to specify build strings or hardware-specific packages?"
For advanced package specifications including build strings, see the [Package Specifications](../../../../concepts/package_specifications.md) guide.

Examples (both syntaxes work):
```shell
# Equals syntax (compact)
pixi global install "pytorch=*=*cuda*" --channel pytorch
pixi global install "jax=*=*cuda*"

# Bracket syntax (explicit)
pixi global install "pytorch [build='*cuda*']" --channel pytorch
pixi global install "jax [build='*cuda*']"
```

!!! tip
Running `osx-64` on Apple Silicon will install the Intel binary but run it using [Rosetta](https://developer.apple.com/documentation/apple-silicon/about-the-rosetta-translation-environment)
```
pixi global install --platform osx-64 ruff
```
Running `osx-64` on Apple Silicon will install the Intel binary but run it using [Rosetta](https://developer.apple.com/documentation/apple-silicon/about-the-rosetta-translation-environment)
```
pixi global install --platform osx-64 ruff
```

!!! note
When you pass `--path` with a local `.conda` archive, Pixi copies the file into `PIXI_HOME/conda-files` and installs from that managed copy. Supplying any other kind of path keeps the original location unchanged.
When you pass `--path` with a local `.conda` archive, Pixi copies the file into `PIXI_HOME/conda-files` and installs from that managed copy.
Supplying any other kind of path keeps the original location unchanged.

After using global install, you can use the package you installed anywhere on your system.
--8<-- [end:description]
Loading
Loading