Packed relative ELF relocations (DT_RELR) enabled by default in rawhide

[Florian Weimer]

This changed was originally planned and approved for Fedora 39.  It
reduces startup time somewhat for large objects with lots of function
and object pointers in global data.

It should be a transparent change, internal to the toolchain.  Within
glibc itself, we started using this linker feature in glibc 2.36 (Fedora
37), and no issues surfaced.  A few dozen test rebuilds of core packages
did not show any issues, either.

Thanks,
Florian

[Dan Horák]

On Wed, 03 Jan 2024 10:47:12 +0100
Florian Weimer <fweimer(a)redhat.com&gt; wrote:

>  This changed was originally planned and approved for Fedora 39.  It
>  reduces startup time somewhat for large objects with lots of function
>  and object pointers in global data.
>  
>  It should be a transparent change, internal to the toolchain.  Within
>  glibc itself, we started using this linker feature in glibc 2.36 (Fedora
>  37), and no issues surfaced.  A few dozen test rebuilds of core packages
>  did not show any issues, either. 
is
/usr/bin/ld: warning: -z pack-relative-relocs ignored
related? It seems to appear in s390x builds.


		Dan

[Florian Weimer]

* Dan Horák:

>  On Wed, 03 Jan 2024 10:47:12 +0100
>  Florian Weimer <fweimer(a)redhat.com&gt; wrote:
>
> > This changed was originally planned and approved for Fedora 39.  It
> > reduces startup time somewhat for large objects with lots of function
> > and object pointers in global data.
> > 
> > It should be a transparent change, internal to the toolchain.  Within
> > glibc itself, we started using this linker feature in glibc 2.36 (Fedora
> > 37), and no issues surfaced.  A few dozen test rebuilds of core packages
> > did not show any issues, either.
>
>  is
>  /usr/bin/ld: warning: -z pack-relative-relocs ignored
>  related? It seems to appear in s390x builds. 
Yeah, it is, it did not occur to me to check that it was actually
enabled in the link editor on all architectures because the format is
portable.  I'll raise the gap with IBM.

Thanks,
Florian

[Alessandro Astone]

It also appears for aarch64.
It is an issue because software may use `-Wl,--fatal-warnings` so this warning breaks the
build.
As an example, pretty much the entire KDE software collection now fails to compile on
aarch64 and s390x:
https://invent.kde.org/frameworks/extra-cmake-modules/-/blob/04affb76a6af...

[Florian Weimer]

* Alessandro Astone:

>  It also appears for aarch64.
>  It is an issue because software may use `-Wl,--fatal-warnings` so this warning breaks the
> build.
>  As an example, pretty much the entire KDE software collection now fails to compile on
> aarch64 and s390x:
> https://invent.kde.org/frameworks/extra-cmake-modules/-/blob/04affb76a6af...
The redhat-rpm-config-274-1.fc40 update should have a workaround for
that.  It skips -z pack-relative-relocs on aarch64 and s390x.

Thanks,
Florian