/usr/bin/ld.gold: --push-state: unknown option

Status changed to 'Confirmed' because the bug affects multiple users.

This is fallout from https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=797577
It will affect any project that tries to link with ld.gold, such as Spidermonkey and WebkitGtk, since ld.gold does not recognize the "--push-state" flag that is injected by the fix/workaround from the above bug.

The underlying bug http://gcc.gnu.org/PR65913 looks to be fixed upstream and will be included in gcc 5.3

On Mon, Sep 21, 2015 at 01:56:46PM -0000, Launchpad Bug Tracker wrote:
> With gcc 5.2.1-17ubuntu4, we're seeing
> https://launchpadlibrarian.net/218108553/buildlog_ubuntu-wily-amd64
> .ubuntu-ui-toolkit_1.3.1639%2B15.10.20150916.2-0ubuntu1_BUILDING.txt.gz

The error in the log is:

g++ -Wl,-Bsymbolic-functions -Wl,-z,relro -Wl,--as-needed -Wl,-O1 -fuse-ld=gold -o ubuntu-ui-toolkit-launcher .obj/launcher.o .obj/MouseTouchAdaptor.o -L/usr/X11R6/lib64 -lQt5Quick -lQt5Gui -lQt5Qml -lQt5Network -lQt5Test -lQt5Core -lGL -lpthread
/usr/bin/ld.gold: --push-state: unknown option
/usr/bin/ld.gold: use the --help option for usage information
collect2: error: ld returned 1 exit status

But --push-state only ever appears in the error message; it does not appear
in the g++ commandline that the build system claims to be invoking. Has
anyone tracked down the origin of this? I do see '--push-state' in the
output of 'strings /usr/bin/g++', so it's certainly possible that g++ itself
is adding it to the commandline when invoking ld. If so this is definitely
a gcc bug, passing extra arguments to ld regardless of which ld is being
called, and it doesn't matter what libraries are being passed.

The --push-state tracks to a Debian-specific patch in the source, debian/patches/pr65913-workaround.diff. This was added to fix Debian bug #797577, a build failure with a particular boost-using package in Debian.

This new failure was introduced in the latest upload (5.2.1-17ubuntu4) and can be resolved by reverting the change in this version to debian/rules.patch. If this is an urgent problem (and it looks like it might be, if it impacts all qmake-using packages in the archive), I would recommend that someone upload this revert now, until a cleaner fix for the original bug can be found.

Steve, did you miss my comment above?

This has been fixed upstream, so should be possible to replace that debian patch with a cherry-pick of the upstream fix
https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=227878

I'm working on it

This also affects building oxide-qt on wily (see e.g. https://launchpad.net/~phablet-team/+archive/ubuntu/ppa/+build/7930449).

this also affects building ubuntu-ui-toolkit on wily

This also affects building qtpurchasing's ubuntu backend on wily

This bug was fixed in the package gcc-5 - 5.2.1-18ubuntu1

---------------
gcc-5 (5.2.1-18ubuntu1) wily; urgency=medium

  * Merge with Debian; remaining changes:

gcc-5 (5.2.1-18) unstable; urgency=medium

  * Update to SVN 20150922 (r228023, 5.2.1) from the gcc-5-branch.

  [ Matthias Klose ]
  * gcc-5-plugin-dev: Depend on libmpc-dev. Closes: #798997.
  * Fix PR libstdc++/65913, taken from the trunk. Closes: #797577.

  [ YunQiang Su ]
  * Build again the gnat-5-sjlj package. Closes: #798782.
  * Fix gnat cross builds, and cross building gnat.

 -- Matthias Klose <email address hidden> Wed, 23 Sep 2015 03:10:37 +0200

> Steve, did you miss my comment above?

Sure did, sorry Tim!