nss opted out of building with LTO in Fedora in 3.55.0-2: https://src.fedoraproject.org/rpms/nss/c/d4e86043ee0de880c84f0d6ce5267a5ce16c97ef with the message: # This package fails its testsuite with LTO. Disable LTO for now and building it with LTO is currently still disabled in rawhide. It would be good to either: (a) enable LTO for nss in Fedora so that nss can benefit from LTO (smaller/faster machine code, deeper analysis for compiler warnings, etc), or (b) capture what the problem was, in case something needs fixing in nss, or at the toolchain end. Hence this bug. What specific issues do you run into if you remove the %global _lto_cflags %{nil} ? Is this being tracked in upstream nss or upstream gcc? (and do you need help from the toolchain team on this?) Thanks Dave
It looks like Daiki turned if off last year. I'll have to see what the actual failure is. I'm currently running an nss rebuild to handle the mass rebuild issue. Daiki's comments indicated that the nss upstream tests, which are run as part of the the nss build, fails with lto_cflags on. Looks like there wasn't an upstream bug originally, bug umbutu has one now: https://bugzilla.mozilla.org/show_bug.cgi?id=1721995 They had problems with s390. When my current NSS build is complete, I'll do a scratch build to see which tests and platforms fail if I turn lto on.
This bug appears to have been reported against 'rawhide' during the Fedora 35 development cycle. Changing version to 35.
The builds evidentally completed correctly, fix was checked in on 08-10