* Increase maximum number of open files (the default of 1000 is too less
for some packages)
* Use /etc/resolv.conf from host (otherwise systemd-nspawn uses a config
which does not work on my host setup)
* Now in-line with upstream: Remove --keep-unit since it causes sometimes
errors and does not seem to be required
Anything that requires user input (such as a key unlock or hostkey
verify) will block the terminal and wait for input which will never
come.
When cloning or configuring a repo via ssh we therefore initially
connect to gitlab to warm the connection.
Afterwards users are expected to either have setup a ssh ControlMaster
or use something like a ssh agent.
Fixes#148
Component: pkgctl repo clone/configure
Co-Authored-by: Christian Heusel <christian@heusel.eu>
Signed-off-by: Christian Heusel <christian@heusel.eu>
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
By default the sync operation enables dependency checks during the
preparation of the transaction. This isn't necessary for the print
option, but is done nevertheless, which leads to issues for example
during soname dependency problems. The result is a none functional
automatic repo detection.
Fix this by explicitly disabling dependency version checks. We specify
this option twice to skip all dependency checks.
Fixes#189
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
When releasing a package always generate a .SRCINFO file so we can build
tooling based upon these files instead of having to parse bash in a
non-bash language.
Some tests have shown around 30 seconds or generating the metadata on
packages like thunderbird. The implementation in makepkg becomes semi
slow for some split packages that provide a huge number of individual
packages, but it seems reasonably fast so we can instead have a uniform
state of always providing .SRCINFO files.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Currently we used grep -E, in a way that special characters in the
package name such as the '+' in 'ls++' also became part of the regex.
This commit switches this to become a literal string matching using awk.
Signed-off-by: Christian Heusel <christian@heusel.eu>
As the relative links are relative to the pkgctl invocations PWD and not
necessarily relative to the location where makechrootpkg is invoked from
this fails unexpectedly.
This commit fixes this by just using the full path when passing the
location to makechrootpkg.
Fixes https://gitlab.archlinux.org/archlinux/devtools/-/issues/181
Signed-off-by: Christian Heusel <christian@heusel.eu>
`--keep-unit` is really only for use in services like
`systemd-nspawn@.service`.
The parameter was added in commit 000ea6c7bb because
systemd-nspawn defaults the name of the machine (and thus the generated
scope) to the name of the working directory, which is not unique. Thus
spawning a container from `archbuild/extra-x86_64/foo` while
`archbuild/testing-x86_64/foo` is already running would fail.
We can avoid the unit conflict by giving the container a unique machine
name. Creating a scope also allows us to place the container in a slice
hierarchy for resource control.
This is done so that pkgctl can be better used to build aur packages
which can have arch=(...) settings for which we do not have a clean
chroot builder.
Signed-off-by: Christian Heusel <christian@heusel.eu>
Early exit in case the PKGBUILD is not yet properly under version
control, which can happen for freshly initialized repositories.
Furthermore print an appropriate error message including a hint how to
resolve this.
Fixes#154Fixes#167
Signed-off-by: Christian Heusel <christian@heusel.eu>
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Not being in any official repo does not necessarily mean this is a new
package. One could simply be building an AUR or custom local package.
Make the message less confusing in such case.
Before porting commitpkg to Git, the code has checked the SVN status for
none commited files. During the port this has changed by straight
checking for any passed files if they were under version control or not.
In general the whole logic is very brittle as variables are searched by
regex and directly passed to eval while ignoring any function scoping.
This leads to missing files when they reference the $pkgname inside a
package function but also provide wrong ones when eval simply returns
the first $pkgname while ignoring and function scopes.
In the future this should completely be replaces by .SRCINFO processing.
Fixes#145
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
arch-nspawn, mkarchroot, makerepropkg all call "shift" after getopts
processing. Save the original args and pass on to check_root to prevent
options being discarded.
Fixes 41d4624879Fixes#149
Signed-off-by: Toolybird <toolybird@tuta.io>
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
This is a rather quick and simple implementation to override the current
logic and force clone with HTTPS. Allowing to explicitly clone over HTTPS
is currently required to unblock reproducible builds where no ssh keys
and GitLab user accounts are set up as of now. Hence this quick solution
comes into play to mitigate the regression on reproducible builds
builders.
Revisit the overall auto detection and protocol logic approach for a
later release related to some ideas floating around in pending
merge-requests.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
This currently makes .SRCINFO files opt-in and helps to keep the file up
to date if already present.
Signed-off-by: Morten Linderud <foxboron@archlinux.org>
`diff_pkgs` needs to use a local variable for its options, otherwise
they will accumulate for each package diffed.
Whem comparing split packages this lead to earlier mutated DIFFOPTIONS
containing too many labels which resulted in a failure.
Check git status --porcelain, not --short.
`--short` is influenced by user configuration like `status.branch`
making it non-empty even on a clean tree. Use `--porcelain` to avoid
this.
Run up to N jobs in parallel. By default the number of jobs is equal to the
number of available processing units. For sequential processing this option
needs to be passed with 1.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Bumping the chroot version will result in the chroots checking against
the local version and force recreation in case they do not match.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
We only need to specifically use ssh:// protocol prefix if we want to
specify a special port. As we moved to support pulling directly over
port 22 from out GitLab instance we can change the url scheme to the
simple variant.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Before modularizing the repo layout, we used m4_include to assemble
together sources into a single file. Now, we properly use a library
layout without assembling multiple files, which means we cannot anymore
rely on BASH_SOURCE inside the library file. Hence, pass along the
actual command and argv from the check_root caller.
Fixes: src: modularize repo layout into a library
Signed-off-by: Morten Linderud <foxboron@archlinux.org>
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Allow overriding the worker slot with a dedicated option. Furthermore
detect if the current tty is no pts and fall back to choosing a random
worker slot between 1 and number of available processing units.
Fixes#137
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Git push will print its status, hence switch from a busy line
indicator to a single line message and allow Git to print its status
accordingly.
Otherwise we get:
==> Releasing package...Enumerating objects: 6, done.
Add an option to call the switch command after clone. Switch to a
specified version. The working tree and the index are updated to match
the version.
Signed-off-by: Christian Heusel <christian@heusel.eu>
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
It leads to weird expectations when using --db-update or --message
without --release. Make the behavior more user friendly, by aborting the
operation and explaining that release options only work in conjunction
with the release option.
Fixes#131
Signed-off-by: Christian Heusel <christian@heusel.eu>
Normally the default in Arch is that all home directories are private.
However, this may have been changed locally. To make sure we never
expose secrets, lets use a umask of 0077 when writing the config.
Additionally add some temporary fixup code to migrate the file and
directory permissions of already existing paths.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
This would allow to supply the gitlab tokens via the env var
DEVTOOLS_GITLAB_TOKEN and therefore allow users to choose whatever
program they want to fill this env var.
Closes#113
Signed-off-by: Christian Heusel <christian@heusel.eu>
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Adds a check for the configured Git editor (git config core.editor) in
both commitpkg and build.sh.
Additionally, instead of blindly executing vi when all other options are
exhausted, remove it instead as it is a none standard installed editor
anyway.
Closes#106
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Instead of trying to port this ancient script, which doesn't even seem
to work with community, let's instead remove it. We will be adding a
replacement script in pkgctl soon with a smarter and more convenient UX.
Automatic path conversion is limited to GitLab API v4 and will be
removed in the future. It's expected that the caller does the path
conversion on caller side and only passes a valid path to the API within
its limitations.
Hence convert project names to valid paths:
1. replace single '+' between word boundaries with '-'
2. replace any other '+' with literal 'plus'
3. replace any special chars other than '_', '-' and '.' with '-'
4. replace consecutive '_-' chars with a single '-'
5. replace 'tree' with 'unix-tree' due to GitLab reserved keyword
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
The remote protocol is automatically determined from the author email
address by choosing SSH for all official packager identities and
read-only HTTPS otherwise.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
This ensures the repository we try to commit and release from uses the
latest distro specs for its local git config. The check errors out early
before touching anything and prints a recommendation how to update the
repo.
Signed-off-by: Levente Polyak <anthraxx@archlinux.org>
Martchus
Kevin Mihelich
Levente Polyak
Jelle van der Waa
Christian Heusel
Jan Alexander Steffens (heftig)
Antonio Rojas
Daniel M. Capella
Toolybird
Chih-Hsuan Yen
Morten Linderud
Alexander Epaneshnikov
Campbell Jones