On 10/7/2020 5:48 AM, Mark Kettenis wrote:
Date: Tue, 6 Oct 2020 13:06:04 +0000
From: Miod Vallat <[email protected]>
Synopsis: lib/csu/init_priority regress test failure on macppc
Category: powerpc
Environment:
System : OpenBSD 6.8
Details : OpenBSD 6.8-current (GENERIC) #3: Tue Oct 6 07:58:23 GMT
2020
[email protected]:/usr/src/sys/arch/macppc/compile/GENERIC
Architecture: OpenBSD.macppc
Machine : macppc
Description:
The lib/csu/init_priority regress test fails when using clang, but
passes when compiled with gcc.
How-To-Repeat:
$ cd /usr/src/regress/lib/csu/init_priority && make
c++ -O2 -pipe -MD -MP -c
/usr/src/regress/lib/csu/init_priority/init_priority
_test.cc
c++ -o init_priority_test init_priority_test.o
./init_priority_test
12597
14117
17071
20840
28373
31319
40956
47220
6407
8742
assertion "log[i] >= log[i - 1]" failed: file
"/usr/src/regress/lib/csu/init_priority/init_priority_test.cc", line 48, function "main"
*** Signal SIGABRT in . (<bsd.regress.mk>:36
'run-regress-init_priority_test')
FAILED
However, building with g++ causes the test to pass:
$ make clean && CXX=g++ make
rm -f a.out [Ee]rrs mklog *.core y.tab.h init_priority_test
init_priority_test.o init_priority_test.d
g++ -O2 -pipe -MD -MP -c
/usr/src/regress/lib/csu/init_priority/init_priority_test.cc
g++ -o init_priority_test init_priority_test.o
./init_priority_test
6407
8742
12597
14117
17071
20840
28373
31319
40956
47220
Fix:
unknown.
On powerpc we're still using ld.bfd as the linker. This issue doesn't
happen on powerpc64, where we use ld.lld instead. It looks as if that
linker doesn't order the sections the same way as ld.lld is.
Ufortunately ld.lld doesn't produce binaries that out ld.so likes on
powerpc. Investigating that, but I need to build llvm first, which
takes some time...
Is this still an issue with the new linker?
