On Mon, Nov 14, 2016 at 06:47:00PM +0100, Michael Biebl wrote: > Am 13.11.2016 um 19:43 schrieb Michael Biebl: > > Am 13.11.2016 um 18:37 schrieb Sven Joachim: > >> The toolchain has also changed quite a bit in the past four weeks, with > >> gcc having pie enabled by default and binutils at a bleeding edge > >> snapshot. Maybe one of those has triggered the build failure. > > > > That might well be it. Currently mutter still builds fine in stretch. > > The new binutils should migrate to testing soon. > > I can then retry the build on a mips porter machine with > > 2.27.51.20161108-1 > > binutils 2.27.51.20161108-1 just migrated to stretch. mutter still > builds fine in stretch with this version. So I'd say we can cross off > binutils from the list of suspects.
As Sven already mentioned these symbols are still available otherwise
Mesa would FTBFS as we strictly check those symbols in the build with
override_dh_makeshlibs:
dh_makeshlibs -a -- -c4
Furthermore I've manually checked libegl1-mesa_12.0.4-2_mips.deb [1]
with nm that those symbols are still exported.
Mesa could be still affected by a broken binutils. To cross out
binutils or other toolchain bugs we would need to rebuild Mesa
12.0.3-3 (= version in testing) with the current toolchain and build
mutter against this rebuild of Mesa.
Michael, could you check this on a mips porter machine?
If it works with Mesa 12.0.3-3 we need to bisect Mesa. Otherwise it's
not a regression introduced with Mesa 12.0.4.
Thanks,
Andreas
[1] https://packages.debian.org/sid/mips/libegl1-mesa/download
signature.asc
Description: Digital signature
