This makes it possible to build RPMs from a git tree using
`rpmbuild --build-in-place --noprep` and have resulting RPMs
that will preserve the override of the PAM config file.
This needs to commit to HAVE_SELINUX being defined (since there
is no longer an m4 step to make that stanza conditional), but
that should be acceptable since the %build step calls Meson
with -Dselinux=true.
Tested:
- Chdir into a checkout of github.com/systemd/systemd tree and run:
$ rpmbuild -bb --build-in-place --noprep \
--define "gitcommit $(git rev-parse HEAD)" \
--define "_sourcedir $HOME/fedorarpms/systemd" \
~/fedorarpms/systemd/systemd.spec
- Inspect the contents of systemd-user in the generated RPM package:
$ rpm2cpio ~/rpmbuild/RPMS/x86_64/systemd-239-3.git99352de.fc29.x86_64.rpm \
| cpio -i --to-stdout --quiet ./etc/pam.d/systemd-user
...
account include system-auth
...
session include system-auth
Existing patches moved to the systemd-stable tarball, with the exception
of 0991-core-fix-resetting-of-Delegate-and-properly-ignore-i.patch,
which was partially reverted upstream and is just removed.
2018-03-06 23:21:31,835 INFO pylorax.dnfhelper: Performing post-installation setup tasks
2018-03-06 23:22:41,901 WARNING pylorax.dnfhelper: Non-fatal POSTTRANS scriptlet failure in rpm package kernel-core
warning: %posttrans(kernel-core-4.16.0-0.rc4.git0.1.fc29.x86_6 scriptlet failed, exit status 1
Please specify the kernel command line in /etc/kernel/cmdline!
Could not determine the kernel command line parameters
/usr/lib/kernel/install.d/90-loaderentry.install: line 53: /proc/cmdline: No such file or directory
GRUB now has BootLoaderSpec support, the user can choose to use this by
setting GRUB_ENABLE_BLSCFG to true in /etc/default/grub. On this setup,
grubby isn't used to generate the kernel modules dependencies, instead
the depmod install script could be used.
But the BLS support in GRUB uses the config snippets that are generated
at build time and included in the kernel package, so the loaderentry
install script shouldn't be executed when using GRUB.
Signed-off-by: Javier Martinez Canillas <javierm@redhat.com>
If the directory does not exist in %buildroot, rpm very usefully defaults
to access mask of 0. When the directory is created in the file system, of
course it has some other mode, and rpm -V reports the directory as changed.
The directory /var/lib/systemd/journal-upload must not exist.
If both private and non-private directory exist, then
systemd cannot create link to private directory, and
the service fails to start.
This makes the script try to remove the non-private directory.
Previous method was untenable, because rpm doesn't have a concept of
put those files here, and the rest in there. So for every positive
pattern that was added, we had to add an %exclude line somewhere else.
So let's generate the lists using pattern matching. This is a bit messy
too, but should not require updates when files are added or removed.
(Sometimes it'll be necessary to add a new pattern if the new files
should not land in the main binary package.)
There's some intentional changes:
- man pages are more consistently included with the files they describe
- shell completion scripts similarly
- various kernel-install and modules-load related files are moved
to -udev subpackage
This reverts commits 3fb4a15096
and 0e8350ca14.
Either building with meson or other upstream changes was causing
issues with booting, and I didn't have time to debug this properly.
This is needed because an additional property was added and the
tests fail otherwise. I think adding of the property is OK, it's
fully backwards compatible.
Also use %if 0 syntax to make it easy to copy&paste the commands.
Error: Transaction check error:
file /usr/share/polkit-1/rules.d from install of systemd-233-1.fc27.x86_64 conflicts with file from package polkit-0.113-7.fc26.x86_64
What is included:
- .wants directories
- for our own targets listed in systemd.special(7)
This is not supposed to cover all possible .wants and .requires and .d
dirs for all units, or even targets, or even just our targets, but only
the reasonably common ones that we provide.
It still fails in koji, and I'd have to download each package from
koji by hand to find out what is the issue (or wait until tomorrow for
a compose). Let's just nuke the whole thing, all deps should be installed.
Build failed with the following:
Traceback (most recent call last):
File "./hwdb/parse_hwdb.py", line 195, in <module>
groups = parse(fname)
File "./hwdb/parse_hwdb.py", line 136, in parse
parsed = grammar.parseFile(fname)
File "/usr/lib/python3.6/site-packages/pyparsing.py", line 2142, in parseFile
file_contents = f.read()
File "/usr/lib64/python3.6/encodings/ascii.py", line 26, in decode
return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc2 in position 15765: ordinal not in range(128)
In mock, LC_CTYPE is unset, so python3 defaults to ascii, uselessly.
I think "ID_VENDOR_FROM_DATABASE=VENGIT Korlátolt Felelősségű Társaság"
is the issue here.
The header on the patch is fake. git am does not deal well with just
diff output, mumbling the following:
Applying: seccomp: rework seccomp code, to improve compat with some archs
Applying:
fatal: empty ident name (for <>) not allowed
We should really move away from mucking around with text patches.
We don't expect those calls to fail (they should work fine both
with systemd running and without). But if they fail, most likely
some unit is missing or misspelt, which we want to catch.
koschei builds are failing on systemd.rpm:
DEBUG util.py:421: error: unpacking of archive failed on file
/etc/dbus-1/system.d/org.freedesktop.hostname1.conf;581cb002: cpio: open
DEBUG util.py:421: error: systemd-232-2.fc26.x86_64: install failed
Only fall back to "dns" if nss-resolve is not installed (for the
architecture of the calling program). Once it is, we never want to
fall back to "dns" as that breaks enforcing DNSSEC verification and
also pointlessly retries NXDOMAIN failures.
C.f. https://anonscm.debian.org/cgit/pkg-systemd/systemd.git/commit/?id=5e00954
Also drop the removal of nss modules in %postun. If the module cannot be
loaded, it is simply ignored. It seems better to simply leave it there,
then play around with nsswitch.conf, which is always risky.
This reverts commit 46083abe1a.
preset-all really does not work right, disabling all sorts of
stuff it should not disable. This caused RHBZ #1363858 (no tty
on vt1) and probably broke other things too.
