@kernellogger Hitting this in mm-unstable breaking my build - https://lore.kernel.org/all/ZoM6CUhbWuAFuHjP@krava/ - seems nobody is chasing this to get hotfixed, any chance of prodding the right people? Thx :)

@kernellogger I mean more importantly, -rc6 is broken (I just repro'd locally...)

@ljs @kernellogger Look, the guy has just started, and has already broken something. A true junior!

@oleksandr @kernellogger I didn't break it mate ;) it's been broken for weeks, I'm just reporting it.

Big difference. Probably.

@ljs @kernellogger This is what they usually say, yes.

@ljs @oleksandr @kernellogger ‘hey it was already like that when I found it’

@oleksandr @kernellogger am I... an intern??

@ljs @kernellogger Being 40+, you are a midtern.

@oleksandr @ljs @kernellogger yeah
"mom, hobbyists broke my computer again!"

@oleksandr @ljs @kernellogger
A.K.A. "a troublemaker".

If you can't afford it, don't break it!

@lkundrak

I've read:

Yeah, mom hobbyists broke my computer again!

@oleksandr @ljs @kernellogger

@djasa @lkundrak @oleksandr @kernellogger I'm a cat mom if that helps?

@ljs

Thx for letting me know. I'm tracking that regression since two days and with a bit of luck mention this in a email with "pending patches that might be wise to merge directly for -rc7" that I hope to still send to Linus today.

@kernellogger thanks, this underlying issue has been going on for weeks now (the fix patch is nearly a month old).

I'm kind of surprised build bots haven't picked it up, as defconfig include CONFIG_WERROR.

But then again you need CONFIG_BPF_SYSCALL to trigger it and defconfig does NOT set that (but does set CONFIG_BPF).

This may be a check that's in newer gcc + older though (I run arch (btw) so have gcc 14.1.1).

@ljs

I multiple times already consider creating and submitting patches every two years that add targets like "pc-distro-2024" that are based on what Debian, Fedora and/or Ubuntu use as .config that CIs then could use for testing…

@ljs @kernellogger -Werror is only set for allmodconfig by default - it is actively unhelpful for runtime testing (due to pointless bisection blocks). That should still be covered by bots and Linus does test it but if there’s a compiler version dependency that’ll limit coverage of the specific issue (one of the reasons it’s bad for bisection…).

@broonie @kernellogger

$ make mrproper
$ make defconfig
$ grep CONFIG_WERROR .config
CONFIG_WERROR=y

@broonie @kernellogger you can repro the bpf bug by doing the above

Then:

$ scripts/config --enable bpf_syscall
$ make olddefconfig
$ make -j $(nproc)

It's not hard to hit this, and a lot of people do stuff based on defconfig...

Also if you mix up the parameters of a function even if it 'kind of' works anyway by mistake, you should probably be hotfixing that asap...

But I also believe people should run get_maintainers.pl and cc maintainers/reviewers for patches and bpf guys don't think you need to do that either so. I guess they go their own way... 👀

@ljs @kernellogger Oh dear, someone added it for x86 in spite of the whole discussion we had when it was added (there’s a reason why it’s default COMPILE_TEST…). That should be dropped if anyone cares about testing x86. :/

@broonie @kernellogger well it found this bug :P

it is fucking me up though mm-unstable is broken when I use virtme-ng build which annoyingly sets CONFIG_BPF_SYSCALL.

I have to run a fix to make it work.

I mean it's totally trivial, just a pain

@ljs @kernellogger Right, it’s a legit bug and one of the reasons people test allmodconfig is to catch stuff like it (-next checks it).

The reason we don’t want it in defconfig is precisely the sort of disruption you’re seeing here - there’s too many variables that can trigger warnings a developer won’t see themselves and you can end up with huge ranges of commits that some users will need manual intervention to work with which is disproportionately disruptive, especially with bisection.

@broonie @kernellogger yeah I totally agree and I've had massive issues with bisection with newer compilers because of this.

Actually in the kernel, several times.

I literally can't go back very much now without it breaking :/

So get it.

I mean in THEORY I like -werror because it stops people ignoring broken things.

But the changes over compiler versions really hurts you.

@kernellogger @ljs The general expectation is that defconfig will be that already - a good chunk of the CIs are using Debian or Fedora rootfss already so need that to work, and a lot of the required config comes from systemd which is widely used in more custom rootfs images anyway.

People probably wouldn’t be thrilled about adding more builds either.

@broonie @ljs

yeah, you are likely right there and that updating defconfig when it falls behind what Debian and Fedora do by default is likely the better approach.