On Tue, 12 Feb 2019 21:48:23 +0100
Jeremie Courreges-Anglas <[email protected]> wrote:

> fwiw, you may have noticed that the "long double diff" has already been
> added both to base and ports-clang.
> 
> For the other diff I don't know yet what to do. I don't have a powerpc
> machine at hand at the moment, maybe soon (tm).

Thank you for your attention.  I am keeping the other diff (which tries
to prevent SIGBUS from misaligned floats) in my local ports tree.

I have just reported another powerpc bug:
https://bugs.llvm.org/show_bug.cgi?id=40736

I don't have a diff for this bug.  clang and gcc are using different
conventions to return a small struct (of up to 8 bytes).  I describe
the incompatible returns in the linked report.  This bug causes apps
using Qt5 to SIGSEGV at startup after they call libxcb's
xcb_intern_atom(), which returns a cookie as a 4-byte struct.

I built Qt5 with ports-clang, because of COMPILER in
/usr/ports/x11/qt5/qt5; but I got libxcb from a January snapshot.
libxcb is part of xenocara, was built by base-gcc.  (My ports tree
also has local changes in infrastructure/mk/gcc4.port.mk and
devel/llvm/Makefile to use lang/gcc/8 as ports-gcc.)

Landry Breuil was removing ports-clang from COMPILER in bulk builds:
https://marc.info/?l=openbsd-ports&m=154641462704437&w=2

Landry's last powerpc bulk (4 Feb) seems to have built Qt5 using
ports-gcc.  If a future powerpc bulk uses ports-clang, while there
is no fix for this bug, then those Qt5 apps will SIGSEGV at startup.

-- 
George Koehler <[email protected]>

Reply via email to