Sigh. Actually, it seems that the libvirt crash was probably using fmt-8.1.0, where the ABI is known to be broken.
fmt-8.1.1 fixed the broken ABI and on inspection looks to me like a rebuild is not required.
On Tue, Jan 11, 2022 at 7:33 AM Kaleb Keithley <kkeithle@xxxxxxxxxx> wrote:
ceph-16.2.7 in fedora rawhide — built with fmt-8.0.1 — apparently worked for a few days between 29 Dec. and 7 Jan.,When fmt was updated in fedora rawhide on 7 Jan to fmt-8.1.1, libvirt users reported a crash[1] due to an undefined symbol reference (fmt::v8::detail::error_handler::on_error(char const*)) in libcephfs.so.2.ceph-16.2.7 has since been rebuilt[2] with fmt-8.1.1. We are waiting for the libvirt team to test the new build to confirm.On Mon, Jan 10, 2022 at 9:40 PM Shengjing Zhu <zsj950618@xxxxxxxxx> wrote:Hi,
On Tue, Jan 11, 2022 at 4:53 AM Kaleb Keithley <kkeithle@xxxxxxxxxx> wrote:
>
> the claim over in #fedora-devel @Libera.chat is that fmt-8.1.1 is the fixed version, but AFAICT it requires recompiling ceph w/ fmt-8.1.1. (IOW, you can't just update to fmt-8.1.1.)
>
fmt 8.1.0 does have regression in ABI from 8.0.0, but it was fixed in
8.1.1. See https://github.com/fmtlib/fmt/issues/2695
But could you elaborate why it requires recompiling ceph? (from fmt
7.x to 8.x, or just from 8.0.0 to 8.1.1 without changing the soname).
--
Regards,
Shengjing Zhu
--Kaleb
Kaleb
_______________________________________________ Dev mailing list -- dev@xxxxxxx To unsubscribe send an email to dev-leave@xxxxxxx
