Maybe a race condition in barebox Makefiles? 2025-05-28 01:48:07,246 - oe-selftest - INFO - barebox.BareboxTest.test_boot_barebox (subunit.RemotedTestCase) 2025-05-28 01:48:07,248 - oe-selftest - INFO - ... FAIL ... ERROR: barebox-2025.04.0-r0 do_compile: oe_runmake failed ... | fixdep: error opening file: arch/arm/dts/.imx8mq-zii-ultra-rmb3.dtb.pbl.o.d: No such file or directory | make[2]: *** [/srv/pokybuild/yocto-worker/oe-selftest-armhost/build/build-st-1221065/tmp/work/qemuarm64-poky-linux/barebox/2025.04.0/barebox-2025.04.0/scripts/Makefile.build:187: arch/arm/dts/imx8mq-zii-ultra-rmb3.dtb.pbl.o] Error 2 | make[2]: *** Deleting file 'arch/arm/dts/imx8mq-zii-ultra-rmb3.dtb.pbl.o' | make[1]: *** [/srv/pokybuild/yocto-worker/oe-selftest-armhost/build/build-st-1221065/tmp/work/qemuarm64-poky-linux/barebox/2025.04.0/barebox-2025.04.0/Makefile:1057: arch/arm/dts] Error 2 | make[1]: *** Waiting for unfinished jobs....
oe-selftest-armhost ubuntu2204-vk-arm2 master completed at 2025-05-28T05:27:08Z https://autobuilder.yoctoproject.org/valkyrie/#/builders/23/builds/1787/steps/14/logs/stdio
Seems like race in the Makefile...
Oh, haven't seen this before, thank you for notifying! At least with latest master I wasn't able to reproduce it, yet. But I've contacted the maintainers and we'll try to identify what's going on. Did this happen only once so far? Or multiple times?
Enrico, It's good that you notified upstream. Hopefully they can see and fix the race. I think it happened only once so far but Mathieu will correct me if I'm wrong. This is known as an "AB-INT" / "auto-builder intermittent" bug - fun, eh! The YP autobuilder runs 2 or 3 builds at a time that are either building code or running various tests so the load can be very high. You can try using: $ stress -t TIMEOUT_TO_SAVE_ME -c <2x v cores> or similar in an overnight stress tests while building and cleanall-ing your recipe in a tight loop. The smarter thing to do is to study the Makefile and insert a delay to make the rare race be deterministic. Smarter and harder but it's a very useful skill to develop.
> I think it happened only once so far but Mathieu will correct me if I'm wrong. Yes, only happened once so far.
This looks like a missing makefile dependency to me, which modern GNU Make has options which can help. Copy this class and inherit shufflemake in the recipe and make will randomly re-order task execution (but still keeping the dependencies).
Nice, I didn't know about that feature. It is only ~ 2 years old... make.git on master ❯ git log --oneline -1 621d3196f 621d3196 [SV 62100] Add '--shuffle' option support ❯ git tag --contains 621d3196f 4.3.90 4.3.91 4.3.92 4.4 4.4.0.90 4.4.0.91 4.4.1 There's also a commandline option: --shuffle https://www.gnu.org/software/make/manual/make.html#index-_002d_002dshuffle
Yeah, two years old is too new for a host dependency, thus the class to pull in the native dependencies and pass the option globally.
Enrico, Can you try to reproduce this using the new make features?
I failed to link to the class I was talking about: https://github.com/rossburton/meta-ross/blob/master/classes/shufflemake.bbclass That's just: BASEDEPENDS:append:class-target = ' make-native' EXTRA_OEMAKE:append:class-target = " --shuffle"
Thanks for atttaching the patch. Could you send the patch to the email list ? https://docs.yoctoproject.org/contributor-guide/index.html Please add a test if possible, and ensure that you have a good commit log and Signed-off-by line.
There is now a fix for this issue on the barebox mailing list: https://lore.kernel.org/barebox/20250828093135.1959085-1-s.hauer@pengutronix.de It's expected that this will be part of this month's release. Given that this seems to have broken the Yocto autobuilder only once over many months, I'd suggest just to wait out the two~ weeks or so until there is a v2025.09.0 release. While make --shuffle was not used to fix this issue, it may be useful to early detect such issues before downstreams run into them. I used the occasion of Debian Trixie shipped a new enough GNU make to submit patches against upstream CI[1] that run make for the 87 defconfigs that are built on every push with --shuffle. [1]: https://lore.kernel.org/barebox/20250901153852.2554758-1-a.fatoum@pengutronix.de/
The issue is now fixed in v2025.09.0. I assume Enrico will submit a version bump or should I? I want to extend my thanks for the tip around make --shuffle. All our CI builds now use it and just a few days ago it helped eliminate a new build breakage, but this time before it made it into a release: https://github.com/barebox/barebox/commit/030f0acc230 :)
Not seen in > 6 months. barebox is updated: 2025.12.0 https://git.openembedded.org/openembedded-core/log/?qt=grep&q=barebox