Occasionally you want to pass software to people who are not (yet!) lucky
enough to be using Guix. You’d tell them to run
guix package -i
something, but that’s not possible in this case. This is where
guix pack comes in.
注: If you are looking for ways to exchange binaries among machines that already run Guix, see 调用guix copy, 调用guix publish, and 调用guix archive.
guix pack command creates a shrink-wrapped pack or
software bundle: it creates a tarball or some other archive containing
the binaries of the software you’re interested in, and all its
dependencies. The resulting archive can be used on any machine that does
not have Guix, and people can run the exact same binaries as those you have
with Guix. The pack itself is created in a bit-reproducible fashion, so
anyone can verify that it really contains the build results that you pretend
to be shipping.
For example, to create a bundle containing Guile, Emacs, Geiser, and all their dependencies, you can run:
$ guix pack guile emacs geiser … /gnu/store/…-pack.tar.gz
The result here is a tarball containing a /gnu/store directory with
all the relevant packages. The resulting tarball contains a profile
with the three packages of interest; the profile is the same as would be
guix package -i. It is this mechanism that is used to
create Guix’s own standalone binary tarball (see 二进制文件安装).
Users of this pack would have to run /gnu/store/…-profile/bin/guile to run Guile, which you may find inconvenient. To work around it, you can create, say, a /opt/gnu/bin symlink to the profile:
guix pack -S /opt/gnu/bin=bin guile emacs geiser
That way, users can happily type /opt/gnu/bin/guile and enjoy.
What if the recipient of your pack does not have root privileges on their machine, and thus cannot unpack it in the root file system? In that case, you will want to use the --relocatable option (see below). This option produces relocatable binaries, meaning they they can be placed anywhere in the file system hierarchy: in the example above, users can unpack your tarball in their home directory and directly run ./opt/gnu/bin/guile.
Alternatively, you can produce a pack in the Docker image format using the following command:
guix pack -f docker -S /bin=bin guile guile-readline
The result is a tarball that can be passed to the
command, followed by
docker load < file docker run -ti guile-guile-readline /bin/guile
where file is the image returned by guix pack, and
guile-guile-readline is its “image tag”. See the
documentation for more information.
Yet another option is to produce a SquashFS image with the following command:
guix pack -f squashfs bash guile emacs geiser
The result is a SquashFS file system image that can either be mounted or
directly be used as a file system container image with the
Singularity container execution
environment, using commands like
singularity shell or
Several command-line options allow you to customize your pack:
Produce a pack in the given format.
The available formats are:
This is the default format. It produces a tarball containing all the specified binaries and symlinks.
This produces a tarball that follows the
Docker Image Specification. The “repository name” as it appears in the
output of the
docker images command is computed from package names
passed on the command line or in the manifest file.
This produces a SquashFS image containing all the specified binaries and symlinks, as well as empty mount points for virtual file systems like procfs.
注: Singularity requires you to provide /bin/sh in the image. For that reason,
guix pack -f squashfsalways implies
-S /bin=bin. Thus, your
guix packinvocation must always start with something like:guix pack -f squashfs bash …
If you forget the
bash(or similar) package,
singularity execwill fail with an unhelpful “no such file or directory” message.
Produce relocatable binaries—i.e., binaries that can be placed anywhere in the file system hierarchy and run from there.
When this option is passed once, the resulting binaries require support for user namespaces in the kernel Linux; when passed twice13, relocatable binaries fall to back to other techniques if user namespaces are unavailable, and essentially work anywhere—see below for the implications.
For example, if you create a pack containing Bash with:
guix pack -RR -S /mybin=bin bash
... you can copy that pack to a machine that lacks Guix, and from your home directory as a normal user, run:
tar xf pack.tar.gz ./mybin/sh
In that shell, if you type
ls /gnu/store, you’ll notice that
/gnu/store shows up and contains all the dependencies of
even though the machine actually lacks /gnu/store altogether! That is
probably the simplest way to deploy Guix-built software on a non-Guix
注: By default, relocatable binaries rely on the user namespace feature of the kernel Linux, which allows unprivileged users to mount or change root. Old versions of Linux did not support it, and some GNU/Linux distributions turn it off.
To produce relocatable binaries that work even in the absence of user namespaces, pass --relocatable or -R twice. In that case, binaries will try user namespace support and fall back to another execution engine if user namespaces are not supported. The following execution engines are supported:
Try user namespaces and fall back to PRoot if user namespaces are not supported (see below).
Try user namespaces and fall back to Fakechroot if user namespaces are not supported (see below).
Run the program through user namespaces and abort if they are not supported.
Run through PRoot. The PRoot program provides the necessary support for file system virtualization. It achieves that by using the
ptracesystem call on the running program. This approach has the advantage to work without requiring special kernel support, but it incurs run-time overhead every time a system call is made.
Run through Fakechroot. Fakechroot virtualizes file system accesses by intercepting calls to C library functions such as
exec, and so on. Unlike PRoot, it incurs very little overhead. However, it does not always work: for example, some file system accesses made from within the C library are not intercepted, and file system accesses made via direct syscalls are not intercepted either, leading to erratic behavior.
When running a wrapped program, you can explicitly request one of the execution engines listed above by setting the
GUIX_EXECUTION_ENGINEenvironment variable accordingly.
Use command as the entry point of the resulting pack, if the
pack format supports it—currently
(Singularity) support it. command must be relative to the profile
contained in the pack.
The entry point specifies the command that tools like
docker run or
singularity run automatically start by default. For example, you can
guix pack -f docker --entry-point=bin/guile guile
The resulting pack can easily be loaded and
docker run with no extra
arguments will spawn
docker load -i pack.tar.gz docker run image-id
Consider the package expr evaluates to.
This has the same purpose as the same-named option in
(see --expression in
Use the packages contained in the manifest object returned by the Scheme code in file. This option can be repeated several times, in which case the manifests are concatenated.
This has a similar purpose as the same-named option in
package (see --manifest) and uses the same
manifest files. It allows you to define a collection of packages once and
use it both for creating profiles and for creating archives for use on
machines that do not have Guix installed. Note that you can specify
either a manifest file or a list of packages, but not both.
Attempt to build for system—e.g.,
i686-linux—instead of the
system type of the build host.
Cross-build for triplet, which must be a valid GNU triplet, such as
"aarch64-linux-gnu" (see GNU
configuration triplets in Autoconf).
Compress the resulting tarball using tool—one of
none for no
Add the symlinks specified by spec to the pack. This option can appear several times.
spec has the form
source=target, where source
is the symlink that will be created and target is the symlink target.
-S /opt/gnu/bin=bin creates a /opt/gnu/bin
symlink pointing to the bin sub-directory of the profile.
Save provenance information for the packages passed on the command line. Provenance information includes the URL and commit of the channels in use (see 通道).
Provenance information is saved in the /gnu/store/…-profile/manifest file in the pack, along with the usual package metadata—the name and version of each package, their propagated inputs, and so on. It is useful information to the recipient of the pack, who then knows how the pack was (supposedly) obtained.
This option is not enabled by default because, like timestamps, provenance information contributes nothing to the build process. In other words, there is an infinity of channel URLs and commit IDs that can lead to the same pack. Recording such “silent” metadata in the output thus potentially breaks the source-to-binary bitwise reproducibility property.
Make file a symlink to the resulting pack, and register it as a garbage collector root.
Include the “local state directory”, /var/guix, in the resulting
pack, and notably the /var/guix/profiles/per-user/root/name
profile—by default name is
guix-profile, which corresponds to
/var/guix contains the store database (see 仓库) as well as garbage-collector roots (see 调用guix gc). Providing it in the pack means that the store is “complete” and manageable by Guix; not providing it pack means that the store is “dead”: items cannot be added to it or removed from it after extraction of the pack.
One use case for this is the Guix self-contained binary tarball (see 二进制文件安装).
Print the name of the derivation that builds the pack.
Use the bootstrap binaries to build the pack. This option is only useful to Guix developers.
guix pack supports all the common build options
(see 普通的构建选项) and all the package transformation options
Here’s a trick to memorize it:
-RR, which adds
PRoot support, can be thought of as the abbreviation of “Really
Relocatable”. Neat, isn’t it?