On Tue, Jun 07, 2022 at 08:12:26AM +0200, Jan Beulich via Gcc-patches wrote:
> > This regressed
> > Executing on host: /home/jakub/src/gcc/obj44/gcc/xgcc
> > -B/home/jakub/src/gcc/obj44/gcc/ -fdiagnostics-plain-output -flto -g
> > -gsplit-dwarf -c -o c_lto_pr83719_0.o
> > /home/jakub/src/gcc/gcc/testsuite/gcc.dg/lto/pr83719_0.c (timeout = 300)
> > spawn -ignore SIGHUP /home/jakub/src/gcc/obj44/gcc/xgcc
> > -B/home/jakub/src/gcc/obj44/gcc/ -fdiagnostics-plain-output -flto -g
> > -gsplit-dwarf -c -o c_lto_pr83719_0.o
> > /home/jakub/src/gcc/gcc/testsuite/gcc.dg/lto/pr83719_0.c
> > cc1: note: '-gsplit-dwarf' is not supported with LTO, disabling
> > /home/jakub/src/gcc/obj44/gcc/objcopy: line 120: exec: --: invalid option
> > exec: usage: exec [-cl] [-a name] [command [argument ...]] [redirection ...]
> > compiler exited with status 1
> > FAIL: gcc.dg/lto/pr83719 c_lto_pr83719_0.o assemble, -flto -g
> > -gsplit-dwarf
> > for me, both on x86_64-linux and i686-linux.
>
> Hmm, it surely worked for me for both, with and without in-tree binutils
> (you don't say which variant you saw the failure with).
System binutils.
grep ORIGINAL_ gcc/Makefile
ORIGINAL_AS_FOR_TARGET = /usr/bin/as
ORIGINAL_LD_FOR_TARGET = /usr/bin/ld
ORIGINAL_NM_FOR_TARGET = /usr/bin/nm
ORIGINAL_OBJCOPY_FOR_TARGET =
ls -l /usr/bin/{as,ld,nm,objcopy}
-rwxr-xr-x. 1 root root 439192 Mar 10 12:51 /usr/bin/as
lrwxrwxrwx. 1 root root 20 May 20 13:28 /usr/bin/ld -> /etc/alternatives/ld
-rwxr-xr-x. 1 root root 47928 Mar 10 12:51 /usr/bin/nm
-rwxr-xr-x. 1 root root 184304 Mar 10 12:51 /usr/bin/objcopy
(but ditto grep ORIGINAL_ stage1-gcc/Makefile
or grep ORIGINAL_ prev-gcc/Makefile).
> > For some reason, I have
> > grep OBJCOPY *gcc/Makefile
> > gcc/Makefile:ORIGINAL_OBJCOPY_FOR_TARGET =
> > prev-gcc/Makefile:ORIGINAL_OBJCOPY_FOR_TARGET =
> > stage1-gcc/Makefile:ORIGINAL_OBJCOPY_FOR_TARGET =
>
> What about the corresponding ORIGINAL_NM_FOR_TARGET? And could you provide
> one of the config.log instances?
config.log has:
configure:23317: checking what linker to use
configure:23351: result: /usr/bin/ld
configure:23379: checking for nm
configure:23397: found /usr/bin/nm
configure:23409: result: /usr/bin/nm
configure:23420: checking what nm to use
configure:23428: result: /usr/bin/nm
configure:23498: checking what objcopy to use
configure:23506: result:
configure:23536: checking for objdump
configure:23554: found /usr/bin/objdump
configure:23566: result: /usr/bin/objdump
configure:23577: checking what objdump to use
configure:23587: result: /usr/bin/objdump
It is a bootstrapped compiler:
../configure --enable-languages=default,obj-c++,lto,go,d
--enable-checking=yes,rtl,extra --enable-libstdcxx-backtrace=yes && make -j32
bootstrap > LOG 2>&1
on Fedora 36 x86_64-linux (ada left out because it is currently broken).
Comparing the toplevel Makefile, I see some differences:
grep NM_FOR_TARGET Makefile
NM_FOR_TARGET="$(NM_FOR_TARGET)"; export NM_FOR_TARGET; \
NM="$(COMPILER_NM_FOR_TARGET)"; export NM; \
NM_FOR_TARGET=$(NM)
COMPILER_NM_FOR_TARGET=$$r/$(HOST_SUBDIR)/gcc/nm
"NM_FOR_TARGET=$(NM_FOR_TARGET)" \
'NM=$(COMPILER_NM_FOR_TARGET)' \
grep OBJCOPY_FOR_TARGET Makefile
OBJCOPY_FOR_TARGET="$(OBJCOPY_FOR_TARGET)"; export OBJCOPY_FOR_TARGET; \
OBJCOPY="$(OBJCOPY_FOR_TARGET)"; export OBJCOPY; \
OBJCOPY_FOR_TARGET=$(OBJCOPY)
"OBJCOPY_FOR_TARGET=$(OBJCOPY_FOR_TARGET)" \
'OBJCOPY=$$(OBJCOPY_FOR_TARGET)' \
E.g. the COMPILER_*_FOR_TARGET line is missing completely for OBJCOPY
and the last line is different too.
Also:
grep ^NM[[:space:]]*= Makefile; echo end
NM = nm
end
grep ^OBJCOPY[[:space:]]*= Makefile; echo end
end
Note, I see
S["OBJDUMP"]="objdump"
S["OBJCOPY"]="objcopy"
S["WINDMC"]="windmc"
S["WINDRES"]="windres"
S["STRIP"]="strip"
S["RANLIB"]="ranlib"
S["NM"]="nm"
in toplevel status, it is just that toplevel:
grep ^NM Makefile.tpl
NM_FOR_BUILD = @NM_FOR_BUILD@
NM = @NM@
NM_FOR_TARGET=@NM_FOR_TARGET@
grep ^OBJCOPY Makefile.tpl
OBJCOPY_FOR_TARGET=@OBJCOPY_FOR_TARGET@
doesn't have the OBJCOPY = @OBJCOPY@ line, and perhaps the
COMPILER_OBJCOPY_FOR_TARGET stuff.
Jakub
