On 04.08.2018 03:23, Robert Elz wrote:
>     Date:        Sat, 4 Aug 2018 02:15:15 +0200
>     From:        Kamil Rytarowski <n...@gmx.com>
>     Message-ID:  <ed8830b0-94d2-fe82-745b-e07fffaf6...@gmx.com>
>   | In general there shall not be a relation between -O level and
>   | sanitizers. Sanitizers do not need -O0  or -g for operation.
> That's fine.   But are you doing compiles that way?  (necessary or not)

I haven't changed any optimization or similar flags for the builds.

>   | GCC also enables more warnings for UBSan that have to be addressed in
>   | order to compile the source, as the code would be UB anyway (like
>   | changing the signedness bit with a shift operation).
> Sure, some of those, even though they're not really problems, are easy
> to fix in a totally harmless way.   That's fine.   The UB is a technical C
> issue, not really anything that ever fails in those cases though.

Freshly crashed pmax kernel due to integer overflow is a kernel (or
virtualization) bug, but it's also a definition of UB, that it can crash
the computer.

>   | And regarding utility of the Undefined Behavior Sanitizer and coverage
>   | of new tests.. we have just caught a bug on pmax that an integer
>   | overflow crashed the kernel:
> Sure, no-one is saying that the extra work is not worth while.   Just that
> you are sometimes fixing non-problems (and causing code churn to do
> it - particularly when the code being changed comes from upstream ... in
> the dhcpcd case that is not such an issue, as if needed, Roy will fix that
> as well, but why would anyone expect the openbsd people to alter ssh
> to fix a non problem ?)

Some Undefined Behavior fixes were pulled by FreeBSD (at least one of
them merged McKusick!). Regarding the OpenSSH case, I'm in touch with
some of their developers and concluded to push it to one of their
mailinglist. One of the UB patches was already merged by the OpenBSD
project (in tmux).

I will go the same way with other patches and submit them upstream, even
if some are cautious. The dhcpcd one was fixed first in upstream dhcpcd
before landing it to src/.

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to