Re: Userland build fails with GCC48

[Frank Wille <frank%phoenix.owl.de@localhost>]

On Mon, 26 Jan 2015 12:13:37 +0100
Martin Husemann <martin%duskware.de@localhost> wrote:

> On Mon, Jan 26, 2015 at 12:10:18PM +0100, Frank Wille wrote:
> > I fear there are still some problems with PIC code to resolve, before we
> > can switch m68k to GCC48. A normal build works, but as soon as you add
> > MKDEBUG=YES or DBG=-g
> 
> You can not do the latter, use "DBG=-O1 -g" or similar.

I always thought DBG is an additional option. So it does replace COPTS?


> DBG=-g will turn optimization down to -O0, and then you have to deal
> with the PIC fallout you have seen.

Ok. -O0 explains why the size of the code segments was growing so much. ;)
Tried with "DBG=-O1 -g" now and I'm running into a compiler error:

#   compile  backend/tree-loop-distribution.o
/home/frank/netbsd/current/src/objamiga/tooldir.NetBSD-6.1.5-amd64/bin/m68k--netbsdelf-c++ -O1 -g -Werror   -g  -Wno-stack-protector  -I. -I/home/frank/netbsd/current/src/external/gpl3/gcc/usr.bin/backend/../gcc/arch/m68k -DIN_GCC -DHAVE_CONFIG_H -I/home/frank/netbsd/current/src/external/gpl3/gcc/dist/gcc -I/
[...]
arch/m68k --sysroot=/home/frank/netbsd/current/src/objamiga/destdir.amiga -DLOCALEDIR=\"/usr/share/locale\" -DNETBSD_NATIVE -I. -DENABLE_SHARED_LIBGCC  -c    /home/frank/netbsd/current/src/external/gpl3/gcc/dist/gcc/tree-loop-distribution.c -o tree-loop-distribution.o
/home/frank/netbsd/current/src/external/gpl3/gcc/dist/gcc/tree-loop-distribution.c: In function 'unsigned int tree_loop_distribution()':
/home/frank/netbsd/current/src/external/gpl3/gcc/dist/gcc/tree-loop-distribution.c:1568:1: internal compiler error: in cselib_record_set, at cselib.c:2373
 }
 ^
no stack trace because unwind library not available


> Do you need explicit -g anyway?
> I only do:
> 
> 	-V MKDEBUG=yes -V MKDEBUGLIB=yes
> 
> and that seems to work fine everywhere.

I'm not sure. I have read about DBG somewhere. It worked with gcc45.
Will also try without now. This could also make the internal error go away,
because -O1 is probably a rarerly tested setting. :|

-- 
Frank Wille