Fedora 33 LTO Support Is Now In Good Shape For Faster, Smaller Packages

[CommunityMember]

Most code builds fine with -flto, but whenever I compile FFMPEG with -flto the build fails. Builds just fine without -flto

For recent FFmpeg, it was *possible* (I did a shared library build slightly over a month ago) to build versions using lto with the necessary configure flags (which almost certainly vary based on distro, but typically need at least --enable=lto to be added to the FFmpeg configure, typically other flags too) with the latest versions of the compilers. However, embedded FFmpeg (as used in a number of projects) often require some adjustments in the projects configuration processes itself because they do not properly pass along certain flags/options to the FFmpeg configuration. The FFmpeg project will accept bug reports if you are encountering something new, so you should certainly let them know if they broke something.

[vegabook]

Waiting for BTRFS customization on the installer.

Just use Ubuntu with ZFS. Why BTRFS still even exists amazes me when you can get a smoooth as silk ZFS experience, designed by Sun Microsystems, the pinnacle of competence.

Ubuntu "gets" desktop like Red Hat never will, because ultimately IBMHat is beholden to corporations, whose incentives are not the same as those of independent, small, desktop users.

[FPScholten]

For recent FFmpeg, it was *possible* (I did a shared library build slightly over a month ago) to build versions using lto with the necessary configure flags (which almost certainly vary based on distro, but typically need at least --enable=lto to be added to the FFmpeg configure, typically other flags too) with the latest versions of the compilers. However, embedded FFmpeg (as used in a number of projects) often require some adjustments in the projects configuration processes itself because they do not properly pass along certain flags/options to the FFmpeg configuration. The FFmpeg project will accept bug reports if you are encountering something new, so you should certainly let them know if they broke something.

I did file bugs on the issue, so i think it will be taken care of in a future version. Will try again soon. Usually I compile a new version every month or so, just to try it out and if necessary report bugs, might also be because I also use latest GCC 10.2 to test with.

[Candy]

I recently swiitched from F32 to F33 (Rawhide) and the performance impact is huge. The system (at least my aging notebook) is running quite performant now. It's noticable faster in any ways.

Thanks

[Britoid]

Just use Ubuntu with ZFS. Why BTRFS still even exists amazes me when you can get a smoooth as silk ZFS experience, designed by Sun Microsystems, the pinnacle of competence.

Ubuntu "gets" desktop like Red Hat never will, because ultimately IBMHat is beholden to corporations, whose incentives are not the same as those of independent, small, desktop users.

Ubuntu is great if you want an ancient Fedora.

[starshipeleven]

Just use Ubuntu with ZFS.

Fedora is a far better distro than Ubuntu and even if Btrfs is inferior to ZFS in many ways, it's really NOT worth using an inferior distro just because of a filesystem.

Ubuntu "gets" desktop like Red Hat never will, because ultimately IBMHat is beholden to corporations, whose incentives are not the same as those of independent, small, desktop users.

Absolute nonsense. ZFS support has been added to make Ubuntu more competitive in the server segment. They don't expect to get any revenue from desktop, none ever does.

[zerothruster]

If the GCC bug is not resolved soon-ish, a proposed fallback is to move to using LLVM, which will get one the equivalent improvements (and, as I recall, Mozilla is using LLVM for their nightly Firefox builds, so it is expected to just build).

But not necessarily with all the CFLAGS you might want to use.

Current clang, at least on x86_64, does not understand -fstack-clash-protection and bails if that is used.

[CommunityMember]

Current clang, at least on x86_64, does not understand -fstack-clash-protection

LLVM 11 (with that support) is also targeted for F33, so there is a potential path forward that is believed to exist in the case that the GCC developers cannot provide a timely fix. There are understood to be a lot of moving parts which will the package owners will need to choose among.

[CommunityMember]

They don't expect to get any revenue from desktop, none ever does.

There is typically some very small revenue stream for the desktop spin (some people/corporations will buy licenses). But the expenses likely exceed the revenue given the small number of licenses, and the desktop spins do seem to be offered mostly for reasons of being able to claim desktop/server compatibility for certain market segments.

But to the point you are likely making, the majority of desktop users of Ubuntu likely pay nothing to no one (other than, perhaps, their ISP if they pay by the byte downloaded).

It is my understanding that the value to most of the Linux distro vendors from their (free) desktop spins would appear to be the (free) testing they get. That *is* worth something, but it is hard to create an accurate profit/loss model, so that it is a good business choice seems as much to do with faith than financials.

[nranger]

Just use Ubuntu with ZFS. Why BTRFS still even exists amazes me when you can get a smoooth as silk ZFS experience, designed by Sun Microsystems, the pinnacle of competence.

I can't tell if you're being sarcastic. But anyway, btrfs and zfs are not 1:1 comparable. And I'd argue btrfs is more useful on the desktop for the ease of using it's features. I have converted metadata levels, expanded volumes, changed compression levels, etc. all on the fly quite easily without dealing with vdevs, pools, etc.