Re: m68k gcc-3.3 __fixunsdfdi problem (Re: Bug#278135: gawk: FTBFS on m68k: clobber.awk test fails with "internal error")

[Fumitoshi UKAI <ukai@debian.or.jp>]

At Wed, 27 Oct 2004 00:33:32 +0200,
Matthias Klose wrote:
 
> there is a workaround (gcc-3.2), therefore the severity could be
> reduced, otoh it's generating wrong code. please could you check, if
> this one is reproducible in gcc-3.4?

No, unreproducible in gcc-3.4.

> The last awk sucessfully built on m68k is the 3.3 branch 20040728.

Yes, according to buildd log. As I checked, the results are

  gcc version           result
  gcc-3.3_1:3.3.4-6     ok (gawk 1:3.1.4-1's case)

  gcc-3.3_1:3.3.5-1     broken (Bug#278135 FTBFS)
  gcc-3.2_1:3.2.3ds9-9  ok
  gcc-3.4_3.4.2-3       ok

So, what should we do with gawk for sarge?

 - upgrade gawk to 1:3.1.4-1.3 to use gcc-3.2 on m68k arch
 - upgrade gawk to 1:3.1.4-1.3 to use gcc-3.4 on m68k arch
 - wait gcc-3.3 (default version of gcc) is fixed and rebuild gawk 1:3.1.4-1.2

Are there any other packages affected by this bug?

Regards,
Fumitoshi UKAI

> Fumitoshi UKAI writes:
> > clone 278135 -1
> > reassign -1 gcc-3.3
> > retitle -1 gcc-3.3: __fixunsdfdi problem on m68k
> > severity -1 critical
> > retitle 278135 gawk: FTBFS on m68k: clobber.awk test fails with "internal error" with gcc-3.3
> > thanks
> > 
> > I confirm RC Bug#278135 on crest.debian.org's dchroot unstable, so I
> > track down it with help of gotom and we suspect this is because of
> > __fixunsdfdi in libgcc.a of gcc-3.3 (1:3.3.5-1).
> > 
> > I confirm that gawk-3.1.4-1 (was succsessfully built on Aug 7 2004) is 
> > also failed to build from source on m68k with gcc-3.3 now and 
> > gawk-3.1.4-1.2 (was failed to build now with default gcc) can be built 
> > from source on m68k with gcc-3.2 now.
> > 
> > I think it is very critical bug, because all binaries that do casting
> > from double to unsigned int would include __fixunsdfdi code fragment
> > in it instead of just refering shared object, so that we need to
> > recompile such binaries after this bug is fixed. 
> > (and currently m68k's buildd uses gcc-3.3 as /usr/bin/gcc)
> > 
> > This is small example to reproduce this problem.
> > 
> > crest% vi test.c
> > #include <stdio.h>
> > #include <stdint.h>
> > 
> > int main()
> > {
> >         int x;
> >         double y = 831.0;
> >         x = (uintmax_t) y;
> >         return x;
> > }
> > crest% cc -g -o t test.c
> > crest% ./t
> > zsh: bus error  ./t
> > crest% gdb ./t
> > GNU gdb 6.1-debian
> > Copyright 2004 Free Software Foundation, Inc.
> > GDB is free software, covered by the GNU General Public License, and you are
> > welcome to change it and/or distribute copies of it under certain conditions.
> > Type "show copying" to see the conditions.
> > There is absolutely no warranty for GDB.  Type "show warranty" for details.
> > This GDB was configured as "m68k-linux"...Using host libthread_db library "/lib/libthread_db.so.1".
> > 
> > (gdb) run
> > Starting program: /home/ukai/tmp/t
> > 
> > Program received signal SIGBUS, Bus error.
> > 0xc003cebe in __libc_start_main () from /lib/libc.so.6
> > (gdb) bt
> > #0  0xc003cebe in __libc_start_main () from /lib/libc.so.6
> > #1  0x800004ce in __fixunsdfdi ()
> > #2  0x80000466 in main () at test.c:8
> > 
> > Regards,
> > Fumitoshi UKAI