The ebuild uses python eclasses, which dropped support for EAPI 6
after the update. Updating EAPI required replacing the use of the
versionator eclass as it's banned in EAPI 7 or later. Fortunately,
nothing in the ebuild seemed to use anything from the eclass.
I changed the way we modify the ebuild by keeping the modifications as
minimal as possible, leaving disabling whatever that can be disabled
to the USE flags.
This is a thing that needs to be done just this once. The necessary
modifications to the python-oem ebuild will be done in follow-up
commit, that can be later cherry-picked if we update the package
again.
This is a thing that needs to be done just this once. The necessary
modifications to the python-oem ebuild will be done in follow-up
commit, that can be later cherry-picked if we update the package
again.
- Drop the init.d files.
- Remove the socket unit's rate limiting.
- Mark the package as stable.
Signed-off-by: Sayan Chowdhury <schowdhury@microsoft.com>
Signed-off-by: Dongsu Park <dpark@linux.microsoft.com>
Since the new USE flag `cet` became the default in profiles of
portage-stable, SDK bootstrap using gcc 11.3.1_p20221209 started
to fail at `sys-libs/glibc`.
```
in function `dl_open_worker_begin':
dl-open.c:(.text+0xab4c): undefined reference to `_dl_cet_open_check'
```
That is because gcc is not correctly configured for CET.
That issue was fixed in recent upstream GCC versions,
e.g. 11.3.1_p20221209-r1 or 11.3.1_p20230120-r1.
https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=016184c289f2cc6c6ade496a700a12f135fbae07https://gitweb.gentoo.org/proj/gcc-patches.git/commit/?id=15daf0510a5fab17cd556261d688a6618391a0c1
However, gcc of Flatcar, 11.3.1_p20221209, does not have the fix.
Update sys-devel/gcc to 11.3.1_p20230120-r1 to fix that issue.
Also accept ~arm64 to keep the same version for both arches.
- remove unecessary files
- drop `pkg_postint`
- create `/etc/ssl` with tmpfiles
- mark openssl as stable for arm64 and amd64
- continue shipping app-misc/c_rehash
Signed-off-by: Mathieu Tortuyaux <mtortuyaux@microsoft.com>
Signed-off-by: Dongsu Park <dpark@linux.microsoft.com>
The boot log only showed the unit descriptions which made it hard to
know what unit was meant.
Switch to the combined unit status reporting that includes the unit
name.
Install some of the symlinks in the baselayout package, instead of
coreos-init. Systemd started installing its private libraries in
/usr/lib64/systemd instead of /usr/lib/systemd which led to the
situation where /usr/lib64/systemd stopped being a symlink to
/usr/lib/systemd and became a separate directory. This could break
compatibility with software that expected things to be in
/usr/lib64/systemd.
Some symlink installation stays in init, though. This is to avoid
having them installed for some oem images where they become dangling.
This pulls in the motdgen change
https://github.com/flatcar/init/pull/88
to not create a broken symlink in the symlinked /run/flatcar/ directory
if the directory symlink /run/coreos already exists.
This reverts commit 285f0646e6dfb3381d2ac5555e991cc482a4d91c.
Now that the ebuild dev-lang/rust does not have a manual check for
versions, it is possible to clean up the sed commands in Github
Actions.
Configure with --without=rust-docs-json-preview to avoid failures.
Remove unnecessary checks for package versions like upstream.
Clean up unnecessary patch, add new patch for 1.67.0.
Make use of the existing helpers provided by `coreos-go.eclass`.
To optimize the binary size of crictl, add "-X" to GO_LDFLAGS,
like "-X $(PROJECT)/pkg/version.Version=$(VERSION)" in the original
Makefile of cri-tools. We cannot follow way of Gentoo ebuilds like
`emake`, because Makefile of cri-tools does not allow users to pass in
${GOARCH}.
Add `arm64` to ACCEPTED_KEYWORDS.
Remove unnecessary files from installation, as well as the shell
completion eclass.
This pulls in
https://github.com/flatcar/bootengine/pull/52
to reduce the size of the unified kernel image by starting Ignition/
Afterburn from the /usr partition.
This also requires to install the Ignition binary to /usr.
This pulls in
https://github.com/flatcar/baselayout/pull/26
to fix the regression in Alpha/Beta (itself a fix for another
regression) where systemd-tmpfiles bailed if /etc/passwd|group didn't
have the core user, now solved by always copying missing entries over
to the databases, not only if the files didn't exist.
We do specify -split-usr in make.defaults for the USE variable but
it's not enough - the base profile in portage-stable forces this flag
and this overrides our defaults. As a workaround, we were using
package.use.force to force-disable split-usr for selected
packages. Now, with addition of more split-usr-using packages in
app-alternatives the list in package.use.force would grow. Instead of
listing all the packages having split-usr in their IUSE, use the
use.force and use.mask file to unforce and mask split-usr for all
packages in generic images.
As a bonus, this also allows us to drop a customization we did in
sys-apps/policycoreutils package.
This is to fix the following error during stage2 SDK build:
[[ (3/3) Emerging packages ]]
!!! The ebuild selected to satisfy "app-alternatives/yacc" has unmet requirements.
- app-alternatives/yacc-1-r2::portage-stable USE="-bison -byacc -reference"
The following REQUIRED_USE flag constraints are unsatisfied:
exactly-one-of ( bison byacc reference )
(dependency required by "sys-devel/binutils-2.39-r4::portage-stable" [ebuild])
(dependency required by "sys-devel/gcc-11.3.1_p20221209::portage-stable" [ebuild])
(dependency required by "sys-libs/glibc-2.36-r5::coreos" [ebuild])
(dependency required by "sys-libs/glibc:2.2" [argument])
