Update to Yocto 5.0 - Scarthgap #137
Conversation
The seccomp PACKECONFIG is already enabled via our DISTRO_FEATURES, so there is no need to have it here explicitly. Signed-off-by: Leonard Göhrs <[email protected]>
These packageconfigs are enabled by default, so there is no need to list them here explicitly. Signed-off-by: Leonard Göhrs <[email protected]>
Starting with version 1.5.0 the conflicting binaries moved to the libexec directory, so there is no need anymore to rename files in bin. Signed-off-by: Leonard Göhrs <[email protected]>
| # Re-enable LLMNR (but not mDNS, which is handled by avahi), | ||
| # which is disabled by the upstream recipe when the zeroconf DISTRO_FEATURE | ||
| # is set. | ||
| EXTRA_OEMESON:remove = "-Ddefault-llmnr=no" |
There was a problem hiding this comment.
Do we actually use 'zeroconf'?
There was a problem hiding this comment.
I use LLMNR in my setup to resolve LXA TACs in my local network (and other may as well without noticing it by using e.g. ssh root@lxatac-00010 and having it just work), but others may use mDNS (ssh [email protected]), which means zeroconf.
systemd-resolved can do both (with some drawbacks on the mDNS front though, when it comes to service discovery), while avahi (which is enabled by the zeroconf DISTRO_FEATURE) only does mDNS but has the better D-Bus interface.
Right now we do not use any more advanced zeroconf features, like announcing services or detecting other TACs in the network, which means systemd-resolved would be perfectly fine for us. Because of that I initially threw out avahi support and used resolved for both.
But @jluebbe (rightfully) objected, because there are plans to do service announcements and discovery in the tacd in the future, which avahi currently does better.
| @@ -0,0 +1,3 @@ | |||
| # The terminal user interface is the one we document in the manual. | |||
| # Make sure it is installed. | |||
| PACKAGECONFIG:append = "nmtui" | |||
There was a problem hiding this comment.
Seems as if the upstream change left -Dnmtui=true in the EXTRA_OEMESON . Thus potentially a patch opportunity. This change is fine however.
|
I am currently tracking down another issue that needs fixing before this can be merged. For some reason most of the $ bitbake lxatac-core-bundle-base
…
* Solver encountered 1 problem(s):
* Problem 1/1:
* - package python3-labgrid-git-r0.cortexa7t2hf-neon-vfpv4 requires python3-autobahn, but none of the providers can be installed
* - package python3-autobahn-23.6.2-r0.cortexa7t2hf-neon-vfpv4 requires python3-twisted, but none of the providers can be installed
* - conflicting requests
* - nothing provides python3-twisted-conch needed by python3-twisted-24.3.0-r0.cortexa7t2hf-neon-vfpv4
*
* Solution 1:
* - do not ask to install a package providing python3-labgrid
… |
|
Sorry, pressed the wrong button ("Close with comment" looks an awful lot like "discard this comment"). I think I've found the change that has cause the issue ( I think I'll backport the fix like I did for the npm caching issue. |
|
I have added 79ccd1d now, which is not the most elegant solution, but one that at least makes the build work until the openembedded/meta-openembedded scathgap branch is updated again (which is when we can drop it again). |
This is a routine update from yocto nanbield to scarthgap, the next version. Signed-off-by: Leonard Göhrs <[email protected]>
Signed-off-by: Leonard Göhrs <[email protected]>
This is a backport of meta-oe master commit bd1b5cde3
("python3-twisted: prepend split PACKAGES"),
which fixes an issue where the `python3-twisted` package would gobble up
all of the files in the split stage (because it matches broadly and comes
before the other packages in the list) and most of the other packages
ended up empty.
Since this workaround is only required until that commit makes it into
the scarthgap branch go the easy route and just hardcode the package
order achieved by bd1b5cde3.
Signed-off-by: Leonard Göhrs <[email protected]>
Starting with commit f15676592b ("bmaptool: now part of Yocto Project")
bmaptool is part of the Yocto project and has changed name from bmap-tool
to bmaptool.
Signed-off-by: Leonard Göhrs <[email protected]>
The current version of `bottom` does not built with the rust compiler in scarthgap. This is fixed in newer `bottom` versions, so it can be re-enabled after an update of `bottom`. Signed-off-by: Leonard Göhrs <[email protected]>
…yocto This workaround was required when building perf for Linux 6.7 and newer on yocto nanbield (and likely older) but is no longer required on scarthgap. Signed-off-by: Leonard Göhrs <[email protected]>
Starting with meta-openembedded commit c331f59e4 ("networkmanager:
fix some missing pkgconfig") the nmtui user interface for networkmanager
is not enabled by default anymore but must instead be enabled via a
pkgconfig.
Signed-off-by: Leonard Göhrs <[email protected]>
Starting with poky commit c152a1e25c ("systemd: disable mdns feature in
resolved for zeroconf") the mDNS and LLMNR features in systemd-resolved
are disabled when the zeroconf DISTRO_FEATURE is enabled, because the
avahi mDNS service (enabled by the zeroconf feature) and the
systemd-resolved mDNS may collide.
Avahi does however not provide LLMNR, so the system will no longer
respond to LLMNR requests at all.
Re-Enable LLMNR to be able to resolve bare hostnames via multicast DNS.
Signed-off-by: Leonard Göhrs <[email protected]>
Systemd assumes the root users home directory to be at /root and does not support other configurations. (See the documentation around ROOT_HOME for more information). Our migration scripts do/did however assume roots home to be at /home/root. Add a symlink from /home/root to /root to stay backward compatible to existing bundles. Signed-off-by: Leonard Göhrs <[email protected]>
|
I've updated meta-oe to the most recent I've also noticed that the current version of With that I think we are ready to merge this. What do you think @ejoerns / @jluebbe? |
…efile The largefile feature is enabled by default since Yocto 4.3 nanbield > Code on 32 bit platforms is now compiled with largefile support > and 64 bit time_t, to avoid the Y2038 time overflow issue. > This breaks the ABI and could break existing programs in > untested layers. [1] [1]: https://docs.yoctoproject.org/singleindex.html#compiling-changes Signed-off-by: Leonard Göhrs <[email protected]>
Oh, neat. I've added a commit to remove it. |
There was a problem hiding this comment.
Fine for me.
Just to not get it lost: While checking the distro I came across another change that happened in poky (in nanbield) that we might want to adapt to (but not as part of this PR): We could also inherit the create-spdx class in tacos.
https://git.yoctoproject.org/poky/commit/?id=581edf20120cd383e8dea0693239629e7547bb7e
hnez
dismissed
Emantor’s stale review
I've dropped the largefile DISTRO_FEATURE in a separate commit.
This updates meta-lxatac to the yocto scarthgap release.
As part of the update we got rid of a few workarounds we needed for successful nanbield build but gained a few new workarounds needed to build for scarthgap.
The zeroconf/avahi change introduced in poky commit yoctoproject/poky@c152a1e25c may fix the avahi daemon issues we see in #128, but I would like to test that for a bit before closing the issue.