AnotherAURHelper/README.md
Stephen Seo 38c42cd7b4 Update README.md
Improve documentation in README.md.
2024-04-09 14:48:16 +09:00

13 KiB

Another AUR Helper (incomplete)

AUR is the Arch User Repository, where anyone can upload a PKGBUILD and supplementary sources to allow others to build their own packages for use in the Arch Linux distribution.

I made an incomplete AUR Helper in Python, and decided to put it in a public repository. It's messy, and it requires a significant amount of set-up, but it works for me. It always builds in a CHROOT, and it lets the user check the PKGBUILD (by default) prior to building. There is no automatic dependency management. That must be done in the config. An example config is provided.

Note that if a "install=<filename>" is specified in the PKGBUILD, then the configured editor will also open the specified file once the PKGBUILD is approved by the user. This check is necessary because such "install scripts" define hooks that are run when the package is installed.

Things to know before using the helper

Security

Apparently makechrootpkg (provided by devtools pkg and used by this script) sources PKGBUILD files directly, meaning that if a malicious PKGBUILD is attempted to be built, it may cause an RCE kind of exploit with the current user. Thus, it is recommended to run this script in a container (like Docker or LXC) so that even if a malicious PKGBUILD is sourced, it will only affect the container. Though if you do set up a container, you may have to set up a directory mount to access the built packages.

Soft-lock due to multiple possible dependencies

Sometimes if a package prompts a user to select between alternate package dependencies, makechrootpkg will fail to select one by default (it will constantly output "y" to stdin when a selection requires an integer). This means you will need to check the logs as it is building a package to make sure this kind of soft-lock doesn't happen. Use tail -f LOG_FILE for example. If such a soft-lock happens, Ctrl-C the helper, and explicitly set a dependency in the TOML config file in a "other_deps" array for the package like so:

[[entry]]
name = "sway-git"
aur_deps = [
    "wlroots-git",
    "swaybg-git"
]
other_deps = [
    "mesa"
]

Package stdout/stderr size limit

The possible issue of output logs filling up disk space is addressed with a "log_limit" config option. By default, if the output log file reaches the limit, the compilation output is no longer logged to file in the logs dir.

Change "log_limit" in the config to a value in bytes if the default of 1 GiB is too little for your use case (if the size of your output logs extend past 1 GiB somehow).

Error when reaching limit

"error_on_limit" can be set to true/false in the config. If set to true, then the build will fail if the limit is reached. If set to false, then the build will continue even if the limit is reached.

Soft-lock if sccache is preinstalled in chroot

Apparently, some packages automatically use ccache/sccache if it is installed in the chroot, and in some cases, causes a soft-lock during a build. It is recommended to not have ccache/sccache preinstalled in the chroot and to just let the aur-helper-script install it when necessary.

For example, when building tenacity-git with sccache preinstalled, the build will hang after the final build step. Apparently, killing the running sccache process stops the soft-lock in this case.

Preloading ccache/sccache

This script expects ccache and sccache not to be installed in the chroot (for reasons as mentioned in the previous section) and ccache or sccache will be appended to a pkg's "other_deps" if a ccache or sccache directory is configured for it.

Setting up the AUR Helper

The AUR Helper requires several things:

  • A CHROOT to build in.
  • A "checking GNUPG" directory that contains the GPG public keys that will be checked when building the PKGBUILD.
  • A "signing GNUPG" directory that contains the GPG private key that will sign the built packages and repository database.
  • SUDO privileges to be able to use makechrootpkg.
  • /etc/pacman.conf must be configured to use the custom repository's packages if pacman -U <pkgs...> will not be used.

Dependencies

The devtools package is required.

The python-toml package is required for the Python script to run.

Create the CHROOT

Use /usr/bin/mkarchroot to create your CHROOT in a directory.

mkarchroot $HOME/mychroot/root base base-devel cmake ninja

As noted earlier, it is better to NOT preinstall ccache and sccache.

You must refer to the CHROOT as $HOME/mychroot if you used the same name as in the previous example:

mkarchroot $HOME/mychroot/root base base-devel cmake ninja
BUILDCHROOT=$HOME/mychoot

Set up the GNUPG dirs

Checking GNUPG

Just create the directory anywhere, and store it in the config.toml. You must manually add public keys to it if a package requires checking source files with GNUPG.

GNUPGHOME=$HOME/myCheckingGNUPGDir gpg --recv-keys A_DEV_KEYS_FINGERPRINT

Note that gpg may not automatically create the GNUPGHOME directory.

Signing GNUPG

You will need to set up a GPG public/private key pair. GNUPG always respects the GNUPGHOME environment variable as the .gnupg dir, so set the variable first, create the directory, then set up your keys. The keys will be used to sign the packages you build and the custom repository that stores the package metadata.

Set the signing_gpg_key_fp variable in the config to the output fingerprint from of:

GNUPGHOME=mySigningGNUPGDir gpg --fingerprint

Note that you must remove the spaces between each part of the fingerprint, like in the example config.

Keep note of the password you store for this GNUPG key, as you will enter it every time you use the Python script.

Set up the config dir

See the example_config.toml for more configuration. It should be commented enough for figuring out how to use it.

########## MANDATORY VARIABLES
# If you did `mkarchchroot /home/username/mychroot/root base ...`, then the following must be:
chroot = "/home/username/mychroot"
# Location to place built packages.
pkg_out_dir = "/home/username/pkgs"
# It is recommended to put the repo file in the "pkg_out_dir".
# If the tar file doesn't already exist, it will be automatically created.
repo = "/home/username/pkgs/custom.db.tar"
# Location to clone packages from AUR.
clones_dir = "/home/username/aur"
# add keys to checking GPG with:
# `GNUPGHOME=/home/username/checkingGPG gpg --recv-keys <fingerprint>`
gpg_dir = "/home/username/checkingGPG"
logs_dir = "/home/username/aur/logs"
signing_gpg_dir = "/home/username/signingGPG"
# You can find the signing key's fingerprint with `gpg -k`.
# Make sure it lists '[S]' before the fingerprint, as that means that key is a signing key.
# You may have to use `gpg -k --with-subkey-fingerprint` if your signing key is a subkey.
# Make sure the `signing_gpg_dir` is used:
# `GNUPGHOME=/home/username/signingGPG gpg -k`
signing_gpg_key_fp = "04D9E3A2880F6418EC4BA70EA0F3F8FAA2088E62"
# It may be more helpful to set this to nano:
# editor = "/usr/bin/nano"
editor = "/usr/bin/vim"
# if true, all logs are prepended with current time in UTC
is_timed = true
# if true, all output build logs are prepended with current time in UTC
is_log_timed = true
# Default log_limit is 1 GiB
log_limit = 1073741824
# If true, then make the build fail if the limit is reached
error_on_limit = false
# If true, timestamps are in localtime. If false, timestamps are UTC.
datetime_in_local_time = true
# If true, all builds will be done in a tmpfs. Recommended to have a lot of RAM and/or swap.
tmpfs = false
########## END OF MANDATORY VARIABLES
...
[[entry]]
name = "cpufetch-git"
skip_branch_up_to_date = false
aur_deps = []
other_deps = []
#ccache_dir = "/home/username/ccache_dirs/cpufetch_ccache"
#sccache_dir = "/home/username/sccache_dirs/cpufetch_sccache"
link_cargo_registry = false
#repo_path = "https://example.com/mypkgrepo.git"
#pkg_name = "cpufetch-git"
...

Setting up the Repository

Create a directory for where you will store built packages and the repository.

The name of the repo must be similar to the repo specified in the config.

For example, if your repo's name is MyAURRepo, then repo should be set to .../MyAURRepo.db.tar.

You must also create symlinks such that MyAURRepo.db points to MyAURRepo.db.tar and MyAURRepo.files points to MyAURRepo.files.tar.

The Python script should automatically make a relative (not absolute) symlink to MyAURRepo.db.tar.sig with the name MyAURRepo.db.sig after signing (which should happen after each package is built and signed). Note the name doesn't have to be MyAURRepo, but is based on the repo variable set in the config.

To use the repository, you can add an entry to your /etc/pacman.conf with the following:

[MyAURRepo]
SigLevel = Required TrustedOnly
Server = file:///home/MyAURRepoDirectory
# Optionally set a file with `Server = ...` entries
# Include = /etc/pacman.d/my_repo_server_list

Note that SigLevel is set expecting the MyAURRepo.db file to be signed (the Python script usually signs the .db file after a package has been successfully built).

Making your system trust the new Repository

Export the public key from your signingGPGDirectory.

GNUPGHOME=mySigningGNUPGDir gpg --export MySigningKeyName > $HOME/MySigningKey.pub

Use pacman-key to add and trust it.

sudo pacman-key -a $HOME/MySigningKey.pub

First check that the name is unique:

sudo pacman-key --finger MySigningKeyName

Then trust it:

sudo pacman-key --lsign-key MySigningKeyName

After these steps, pacman should now trust the packages and repository signed by the GPG key you set up.

Using the AUR Helper

Typically, you will invoke:

./update.py --config my_config.toml

If you want to build in the CHROOT without updating the CHROOT, add the --no-update flag.

If you want to check only specific packages in the list of packages in the config use something like -p <package-name>. You can use -p <package_name> multiple times if you want to check a handful of packages only.

If you want to not skip a package marked with skip_branch_up_to_date in the config, then use --no-skip <package-name>, and the script will act as if skip_branch_up_to_date was not specified for the named package.

When building, the script will not directly output to the terminal it is run in, but rather appends to log files in the log directory specified in the config. To see the output while building, you can use something like:

tail -f $MY_LOG_DIR/google-chrome_stdout_2022-06-02_05-27-49_UTC

It may be helpful to periodically clear out the logs directory in between invocations of the AUR Helper script.

It is recommended to use the script with a prepared config.

Other Notes

By default, makechrootpkg does not verify integrity of files in the PKGBUILD. Use the makechrootpkg_noskipinteg.hook to modify the makechrootpkg script to not skip integrity checks.

update.py now does integrity checks before building with makechrootpkg. It is no longer necessary to modify the /usr/bin/makechrootpkg because the integrity checks are done separately.

If the hook was used previously, remove it from /etc/pacman.d/hooks and reinstall devtools.

If you have .cargo/registry and .cargo/git in your home directory, and you don't want to re-download the Rust registry every time you update a Rust package, you can specify link_cargo_registry = true for a package in your config (see ion-git in the example_config.toml) and that will bind-mount these two directories into the chroot, which will share your local Rust cache with the chroot.

[[entry]]
name = "ion-git"
link_cargo_registry = true

is_timed and is_log_timed

If is_timed is true in the config, then output logs are prepended with a timestamp.

If is_log_timed is true in the config, then output build logs are prepended with a timestamp.

sccache and Rust

If using sccache causes a build error when building a package compiling Rust, one may specify in the config to only wrap rustc and nothing else by specifying sccache_rust_only:

[[entry]]
name = "helix-git"
link_cargo_registry = true
sccache_dir="/home/user/aur/sccache_helix-git"
sccache_rust_only = true

Signal Handling

The script is set up to handle SIGINT and SIGUSR1. SIGINT (Ctrl-C) will print the known package list and status, and exit. SIGUSR1 will also print the known package list and status, but will not stop the script.