makepkg

makepkg is a script to automate the building of packages. The requirements for using the script are a build-capable Unix platform and a PKGBUILD.

makepkg is provided by the pacman package.

Configuration

See for details on configuration options for makepkg.

The system configuration is available in , but user-specific changes can be made in or . It is recommended to review the configuration prior to building packages.

Packager information

Each package is tagged with metadata identifying amongst others also the packager. By default, user-compiled packages are marked with Unknown Packager. If multiple users will be compiling packages on a system, or if one is otherwise distributing packages to other users, it is convenient to provide real contact. This can be done by setting the PACKAGER variable in .

To check this on an installed package:

To automatically produce signed packages, also set the variable in .

Package output

By default, makepkg creates the package tarballs in the working directory and downloads source data directly to the directory. Custom paths can be configured, for example to keep all built packages in and all sources in .

Configure the following variables if needed:

  • — directory for storing resulting packages
  • — directory for storing source data (symbolic links will be placed to if it points elsewhere)
  • SRCPKGDEST — directory for storing resulting source packages (built with makepkg -S)

Signature checking

If a signature file in the form of .sig or .asc is part of the PKGBUILD source array, makepkg automatically attempts to verify it. In case the user's keyring does not contain the needed public key for signature verification, makepkg will abort the installation with a message that the PGP key could not be verified.

If a needed public key for a package is missing, the PKGBUILD will most likely contain a validpgpkeys entry with the required key IDs. Import it manually, or find it on a keyserver and import it from there. To temporarily disable signature checking, run makepkg with the option.

Usage

Before continuing, install the group. Packages belonging to this group are not required to be listed as build-time dependencies (makedepends) in PKGBUILD files.

To build a package, one must first create a PKGBUILD, or build script, as described in Creating packages. Existing scripts are available from the Arch Build System (ABS) tree or the AUR. Once in possession of a , change to the directory where it is saved and run the following command to build the package:

$ makepkg

If required dependencies are missing, makepkg will issue a warning before failing. To build the package and install needed dependencies, add the flag /:

$ makepkg --syncdeps

Adding the -r/ flag causes makepkg to remove the make dependencies later, which are no longer needed. If constantly building packages, consider using Pacman/Tips and tricks#Removing unused packages (orphans) once in a while instead.

Note:
  • These dependencies must be available in the configured repositories; see pacman#Repositories and mirrors for details. Alternatively, one can manually install dependencies prior to building (pacman -S --asdeps dep1 dep2).
  • Only global values are used when installing dependencies, i.e any override done in a split package's packaging function will not be used.

Once all dependencies are satisfied and the package builds successfully, a package file () will be created in the working directory. To install, use / (same as ):

$ makepkg --install

To clean up leftover files and directories, such as files extracted to the , add the option /. This is useful for multiple builds of the same package or updating the package version, while using the same build directory. It prevents obsolete and remnant files from carrying over to the new builds:

$ makepkg --clean

For more, see .

Tips and tricks

Reduce source download and extraction times

Make use of , especially when building VCS packages, to save time acquiring and unpacking sources in subsequent rebuilds.

Building optimized binaries

A performance improvement of the packaged software can be achieved by enabling compiler optimizations for the host machine. The downside is that binaries compiled for a specific processor architecture will not run correctly on other machines. On x86_64 machines, there are rarely significant enough real world performance gains that would warrant investing the time to rebuild official packages.

However, it is very easy to reduce performance by using "nonstandard" compiler flags. Many compiler optimizations are only useful in certain situations and should not be indiscriminately applied to every package. Unless benchmark data are available to prove that something is faster, there is a very good chance it is not! The Gentoo GCC optimization and Safe CFLAGS wiki articles provide more in-depth information about compiler optimization.

The options passed to a C/C++ compiler (e.g. gcc or ) are controlled by the , , and environment variables. For use in the Arch build system, makepkg exposes these environment variables as configuration options in . The default values are configured to produce generic binaries that can be installed on a wide range of machines.

Note:
  • Keep in mind that not all build systems use the variables configured in makepkg.conf. For example, cmake disregards the preprocessor options environment variable, CPPFLAGS. Consequently, many PKGBUILDs contain workarounds with options specific to the build system used by the packaged software.
  • The configuration provided with the source code in the Makefile or a specific argument in the compilation command line takes precedence and can potentially override the one in makepkg.conf.

GCC can automatically detect and enable safe architecture-specific optimizations. To use this feature, first remove any and flags, then add . For example:

To see what flags this enables, run:

$ gcc -march=native -v -Q --help=target
Note: Specifying different values instead of -march=native, then -Q --help=target will not work as expected. To find out which options are really enabled, go through a compilation. See Gentoo:Safe CFLAGS#Manual for instructions.

Starting in version 5.2.2, also includes overrides for the environment variable, for flags given to the Rust compiler. The Rust compiler can also detect and enable architecture-specific optimizations by adding to the given value:

To see which CPU features this will enable, run:

$ rustc -C target-cpu=native --print cfg

Running without will print the default configuration. The parameter can be changed to , s, or as desired. See The Rust compiler's documentation for details.

Parallel compilation

The build system uses the MAKEFLAGS environment variable to specify additional options for make. The variable can also be set in the file.

Users with multi-core/multi-processor systems can specify the number of jobs to run simultaneously. This can be accomplished with the use of nproc to determine the number of available processors, e.g. . Some PKGBUILDs specifically override this with , because of race conditions in certain versions or simply because it is not supported in the first place. Packages that fail to build because of this should be reported on the bug tracker (or in the case of AUR packages, to the package maintainer) after making sure that the error is indeed being caused by MAKEFLAGS.

See for a complete list of available options.

Building from files in memory

As compiling requires many I/O operations and handling of small files, moving the working directory to a tmpfs may bring improvements in build times.

The variable can be temporarily exported to makepkg to set the build directory to an existing tmpfs. For example:

$ BUILDDIR=/tmp/makepkg makepkg

Persistent configuration can be done in by uncommenting the option, which is found at the end of the section in the default file. Setting its value to e.g. BUILDDIR=/tmp/makepkg will make use of the Arch's default temporary file system.

Using a compilation cache

The use of ccache can improve build times by caching the results of compilations for successive use.

Generate new checksums

Install and run the following command in the same directory as the PKGBUILD file to generate new checksums:

$ updpkgsums

updpkgsums uses to generate the checksums. See this forum discussion for more details.

The checksums can also be obtained with e.g and added to the array by hand.

Use other compression algorithms

To speed up both packaging and installation, with the tradeoff of having larger package archives, change .

For example, the following skips compression of the package file, which will in turn have no need to be decompressed on install:

$ PKGEXT='.pkg.tar' makepkg

As another example, the following uses the LZ4 algorithm, which is focused on speed:

$ PKGEXT='.pkg.tar.lz4' makepkg

To make one of these settings permanent, set in .

Utilizing multiple cores on compression

supports symmetric multiprocessing (SMP) via the  flag to speed up compression. For example, to let makepkg use as many CPU cores as possible to compress packages, edit  array in :
COMPRESSZST=(zstd -c -z -q --threads=0 -)

xz supports symmetric multiprocessing (SMP) via the flag to speed up compression. For example, to let makepkg use as many CPU cores as possible to compress packages, edit array in :

COMPRESSXZ=(xz -c -z --threads=0 -)
is a drop-in, parallel implementation for gzip which by default uses all available CPU cores (the / flag can be used to employ less cores):
COMPRESSGZ=(pigz -c -f -n)
is a drop-in, parallel implementation for  which also uses all available CPU cores by default. The  flag can be used to employ less cores (note: no space between the  and number of cores).
COMPRESSBZ2=(pbzip2 -c -f)
is another drop-in, parallel implementation for  which also uses all available CPU cores by default. The -n flag can be used to employ less cores.
COMPRESSBZ2=(lbzip2 -c -f)
is a multithreaded implementation for  which also uses all available CPU cores by default. The -n/ flag can be used to employ less cores.
COMPERSSLZ=(plzip -c -f)

Show packages with specific packager

is a pacman database extraction utility. This command shows all packages installed on the system with the packager named packagername:
$ expac "%n %p" | grep "packagername" | column -t

This shows all packages installed on the system with the packager set in the variable PACKAGER. This shows only packages that are in a repository defined in .

$ . /etc/makepkg.conf; grep -xvFf <(pacman -Qqm) <(expac "%n\t%p" | grep "$PACKAGER$" | cut -f1)

Build 32-bit packages on a 64-bit system

See 32-bit package guidelines.

Unattended package signing

A person may not be available to provide the passphrase for the gpg private key used to sign with in automated build environments such as Jenkins. It is ill-advised to store a private gpg key on a system without a passphrase.

A resulting zst package made with makepkg can still be be signed after creation:

$ gpg --detach-sign --pinentry-mode loopback --passphrase --passphrase-fd 0 --output NewlyBuilt.pkg.tar.zst.sig --sign NewlyBuilt.pkg.tar.zst 

where the GPG passphrase is securely provided and obscured by your automation suite of choice.

The resulting and file can be referenced by pacman clients expecting a valid signature and repositories created with when hosting your own repo.

Magnet URIs

Support for magnet URIs resources (with magnet:// prefix) in the field can be added using the download agent.

Troubleshooting

Specifying install directory for QMAKE based packages

The makefile generated by qmake uses the environment variable to specify where the program should be installed. Thus this package function should work:

Note, that qmake also has to be configured appropriately. For example put this in the corresponding .pro file:

WARNING: Package contains reference to $srcdir

Somehow, the literal strings contained in the variables or ended up in one of the installed files in the package.

To identify which files, run the following from the makepkg build directory:

$ grep -R "$PWD/src" pkg/

One possible cause would be from the usage of macro in C/C++ code with full path passed to compiler.

Makepkg fails to download dependencies when behind proxy

When makepkg calls dependencies, it calls pacman to install the packages, which requires administrative privileges via sudo. However, sudo does not pass any environment variables to the privileged environment, and includes the proxy-related variables ftp_proxy, , , and no_proxy.

In order to have makepkg working behind a proxy, invoke one of the following methods.

Enable proxy by setting its URL in XferCommand

The XferCommand can be set to use the desired proxy URL in . Add or uncomment the following line in :

Enable proxy via sudoer's env_keep

Alternatively, one may want to use sudoer's option, which enables preserving given variables the privileged environment. See Pacman#Pacman does not honor proxy settings for more details.

Makepkg fails, but make succeeds

If something successfully compiles using make, but fails through makepkg, it is almost certainly because sets an incompatible compilation variable. Try adding these flags to the PKGBUILD array:

, to prevent its default , , , and .

!makeflags, to prevent its default MAKEFLAGS.

, to prevent its default , and DEBUG_CXXFLAGS, in case the PKGBUILD is a debug build.

If any of these fix the problem, this could warrant an upstream bug report assuming the offending flag has been identified.

gollark: Compact machines is neat?
gollark: I don't know. Generally I just look through the entire curseforge modlist in order when making packs.
gollark: CN pack + 29388291 other mods? It may be worth looking at newer versions than 1.12.2 but they are variously bees.
gollark: We "could" construct a modpack for this.
gollark: Oh dear. Try removing your lungs. COVID-19 affects those.

See also

This article is issued from Archlinux. The text is licensed under Creative Commons - Attribution - Sharealike. Additional terms may apply for the media files.