Internal compiler error building m68k.

[Rob Landley]

On Tuesday 30 September 2008 02:12:46 Bernhard Reutner-Fischer wrote:
> On Mon, Sep 29, 2008 at 09:28:57PM -0500, Rob Landley wrote:
> >On Monday 29 September 2008 10:56:04 Bernhard Reutner-Fischer wrote:
> >> On Sun, Sep 28, 2008 at 11:55:53AM -0500, Rob Landley wrote:
> >> >So with the attached .config, building last night's svn snapshot, is
> >> > anybody else seeing this:
> >> >
> >> >libc/inet/addr.c: In function 'inet_ntoa_r':
> >> >libc/inet/addr.c:144: internal compiler error: in output_move_qimode,
> >> > at config/m68k/m68k.c:1827
> >>
> >> Seems to work for me with the current, stable release of gcc.
> >
> >I tried 4.1.2 and 4.2.2.  You seem to be saying that uClibc for m68k is
> > not supported on those.
>
> No, i'm saying that it compiles for me with the current, stable release
> of gcc, i.e. 4.3.2 at the time of this writing. I did not try to use an
> outdated version from the old 4.2.x series.

David Wu emailed me offline to say that he's seen this with -O3.  I traced 
down the appropriate line with V=1:

~/firmware/firmware/build/cross-compiler-m68k/bin/m68k-gcc -c \
libc/inet/inet_ntoa.c -o \
libc/inet/inet_ntoa.os -include ./include/libc-symbols.h -Wall \
-Wstrict-prototypes -fno-strict-aliasing -fno-stack-protector -fno-builtin \
-nostdinc -I./include -I. -I./libc/sysdeps/linux/m68k -fno-asm \
-DUCLIBC_INTERNAL -std=gnu99 -Os -funit-at-a-time \
-fno-tree-loop-optimize -fno-tree-dominator-opts \
-fno-strength-reduce -Wa,--bitwise-or \
-I./libpthread/linuxthreads.old/sysdeps/unix/sysv/linux/m68k \
-I./libpthread/linuxthreads.old/sysdeps/m68k \
-I./libpthread/linuxthreads.old/sysdeps/unix/sysv/linux \
-I./libpthread/linuxthreads.old/sysdeps/pthread \
-I./libpthread/linuxthreads.old -I./libpthread \
-I/home/landley/firmware/firmware/build/cross-compiler-m68k/include \
-I/home/landley/firmware/firmware/build/cross-compiler-m68k/bin/../gcc//include-fixed 
\
-I/home/landley/firmware/firmware/build/cross-compiler-m68k/bin/../gcc/include 
\
-DNDEBUG -Wa,--bitwise-or -fpic -MT libc/inet/inet_ntoa.os \
-MD -MP -MF libc/inet/.inet_ntoa.os.dep 

And lo and behold, there's an -Os in there, and removing it does indeed avoid 
the error.  (I've confirmed that any optimization level, including -O2 
and -O, triggers the bug in this file.)

For the moment, I'm getting by with this patch:

--- uClibc/Rules.mak	2008-09-28 02:10:20.000000000 -0500
+++ alt-uClibc/Rules.mak	2008-09-30 16:56:19.000000000 -0500
@@ -144,7 +144,9 @@
 # Flags in OPTIMIZATION are used only for non-debug builds
 OPTIMIZATION:=
 # Use '-Os' optimization if available, else use -O2, allow Config to override
+ifneq ($(TARGET_ARCH),m68k)
 OPTIMIZATION+=$(call check_gcc,-Os,-O2)
+endif
 # Use the gcc 3.4 -funit-at-a-time optimization when available
 OPTIMIZATION+=$(call check_gcc,-funit-at-a-time,)
 
Which is overkill in several ways but at least gets me past the bug, and on to 
the next error:

libc/libc_so.a(memcpy.os): In function `memcpy':
memcpy.c:(.text+0x6e): undefined reference to `WORD_COPY_FWD'

Which I'll worry about in the morning, I guess...

Rob