GAS 2.11.1 doesn't work for sh-hitachi-coff

GAS 2.11.1 doesn't work for sh-hitachi-coff

[Schirmer, Hartmut]

Hi,

I can't build a GCC on 

bash-2.05$ uname -a
CYGWIN_NT-4.0 SCHIRMER_H 1.3.2(0.39/3/2) 2001-05-20 23:28 i686 unknown

for target sh-hitachi-coff. GAS from 2.11.1pre1 fails:

/home/SchirmerH/gcc-sh/build-gcc/gcc/xgcc
-B/home/SchirmerH/gcc-sh/build-gcc/gcc
/ -B/gcc-3.0-sh/sh-hitachi-coff/bin/ -B/gcc-3.0-sh/sh-hitachi-coff/lib/
-isystem
 /gcc-3.0-sh/sh-hitachi-coff/include -O2  -DCROSS_COMPILE -DIN_GCC    -W
-Wall -
Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -isystem ./include
-g1
  -DIN_LIBGCC2 -D__GCC_FLOAT_NOT_NEEDED -Dinhibit_libc -I. -I.
-I../../gcc-20010
611/gcc -I../../gcc-20010611/gcc/. -I../../gcc-20010611/gcc/config
-I../../gcc-2
0010611/gcc/../include  -DL_muldi3 -c ../../gcc-20010611/gcc/libgcc2.c -o
libgcc
/./_muldi3.o
/cygdrive/c/TEMP/cc1oUj4A.s: Assembler messages:
/cygdrive/c/TEMP/cc1oUj4A.s:1: Fatal error: dwarf2 is not supported for this
obj
ect file format

2.11.1pre1 build out of the box using

  bash-2.05$ gcc -v
  Reading specs from /usr/lib/gcc-lib/i686-pc-cygwin/2.95.3-4/specs
  gcc version 2.95.3-4 (cygwin special)

Hartmut

Re: GAS 2.11.1 doesn't work for sh-hitachi-coff

[Nick Clifton]

Hi Hartmut,

> for target sh-hitachi-coff. GAS from 2.11.1pre1 fails:
 
> /cygdrive/c/TEMP/cc1oUj4A.s:1: Fatal error: dwarf2 is not supported
> for this object file format

This appears to be a gcc configuration problem.  The COFF file format
does not support the DWARF2 debug format, so gcc should not be
producing it in the first place.

I would suggest that you investigate why the STABS debug format is not
being used by gcc.

Cheers
        Nick

RE: GAS 2.11.1 doesn't work for sh-hitachi-coff

[Schirmer, Hartmut]

> Why is .file in COFF illegal and why did gas 2.10.1 accept it ?
> 
> obj-coff.c has
> 
> void
> c_dot_file_symbol (filename)
>      char *filename;
> {
>   symbolS *symbolP;
> 
>   symbolP = symbol_new (".file",
> 			SEG_DEBUG,
> 			0,
> 			&zero_address_frag);
> ...
> 
> I'll try and block out the two lines in tc-sh.c

It worked !

Hartmut

*** tc-sh.c.orig	Thu Jun  7 03:15:32 2001
--- tc-sh.c	Wed Jun 13 16:14:27 2001
***************
*** 97,104 ****
--- 97,106 ----
    {"2byte", s_uacons, 2},
    {"4byte", s_uacons, 4},
    {"8byte", s_uacons, 8},
+ #ifdef BFD_ASSEMBLER
    {"file", dwarf2_directive_file, 0 },
    {"loc", dwarf2_directive_loc, 0 },
+ #endif
    {0, 0, 0}
  };
  
***************
*** 1940,1946 ****
--- 1942,1950 ----
  	}
      }
  
+ #ifdef BFD_ASSEMBLER
    dwarf2_emit_insn (size);
+ #endif
  }
  
  /* This routine is called each time a label definition is seen.  It

RE: GAS 2.11.1 doesn't work for sh-hitachi-coff

[Schirmer, Hartmut]

> > dwarf2_directive_file gets called
> 
> Oh dear - then gcc really is generating DWARF2 line number
> information.  This should only happen if DWARF2_ASM_LINE_DEBUG_INFO is
> defined, and according to the latest CVS sources that is only defined
> in gcc/config/sh/elf.h.  I assume that this header is not included in
> your sh-coff configuration (but you could check the Makefile in your
> gcc build directory to make sure), so maybe it is defined in another
> header that you are including.  Can you check ?

No time to dig into, sorry. (I'll leave on vacations really soon now ...)

> Just supressing the support for .file and .loc in gas/config/tc-sh.c
> will not work, since they will turn up as unrecognised pseudo ops.  We
> really need to find out why they are being generated by gcc for a COFF
> based configuration.

Why is .file in COFF illegal and why did gas 2.10.1 accept it ?

obj-coff.c has

void
c_dot_file_symbol (filename)
     char *filename;
{
  symbolS *symbolP;

  symbolP = symbol_new (".file",
			SEG_DEBUG,
			0,
			&zero_address_frag);
...

I'll try and block out the two lines in tc-sh.c

Hartmut

Re: GAS 2.11.1 doesn't work for sh-hitachi-coff

[Nick Clifton]

Hi Hartmut,

> this didn't help.

Darn :-(

> I noticed two additional dwarf2 calls in tc-sh.c
> 
> const pseudo_typeS md_pseudo_table[] =
> {
> ...
>   {"file", dwarf2_directive_file, 0 },
>   {"loc", dwarf2_directive_loc, 0 },
> 
> dwarf2_directive_file gets called

Oh dear - then gcc really is generating DWARF2 line number
information.  This should only happen if DWARF2_ASM_LINE_DEBUG_INFO is
defined, and according to the latest CVS sources that is only defined
in gcc/config/sh/elf.h.  I assume that this header is not included in
your sh-coff configuration (but you could check the Makefile in your
gcc build directory to make sure), so maybe it is defined in another
header that you are including.  Can you check ?

Just supressing the support for .file and .loc in gas/config/tc-sh.c
will not work, since they will turn up as unrecognised pseudo ops.  We
really need to find out why they are being generated by gcc for a COFF
based configuration.

Cheers
        Nick

RE: GAS 2.11.1 doesn't work for sh-hitachi-coff

[Schirmer, Hartmut]

Hi Nick,

this didn't help.

I noticed two additional dwarf2 calls in tc-sh.c

const pseudo_typeS md_pseudo_table[] =
{
...
  {"file", dwarf2_directive_file, 0 },
  {"loc", dwarf2_directive_loc, 0 },

dwarf2_directive_file gets called

Hartmut

> -----Original Message-----
> From: Nick Clifton [ mailto:nickc@cambridge.redhat.com ]
> Sent: Tuesday, June 12, 2001 3:06 PM
> To: Schirmer, Hartmut
> Subject: Re: GAS 2.11.1 doesn't work for sh-hitachi-coff
> 
> 
> Hi Hartmut,
> 
> > Seems to be a general problem. Using gas 2.11.1 fails for my target
> > too. I don't use debug information here (no -g to GCC).
> 
> Hmm, in which case please could try the enclosed patch and let me know
> if it works.
> 
> Cheers
>         Nick
> 
> Index: gas/config/tc-sh.c
> ===================================================================
> RCS file: /cvs/src/src/gas/config/tc-sh.c,v
> retrieving revision 1.38
> diff -p -r1.38 tc-sh.c
> *** tc-sh.c	2001/04/05 04:21:36	1.38
> --- tc-sh.c	2001/06/12 13:03:19
> *************** md_assemble (str)
> *** 1945,1951 ****
> --- 1945,1953 ----
>   	}
>       }
>   
> + #ifdef BFD_ASSEMBLER
>     dwarf2_emit_insn (size);
> + #endif
>   }
>   
>   /* This routine is called each time a label definition is seen.  It
> 

RE: GAS 2.11.1 doesn't work for sh-hitachi-coff

[Schirmer, Hartmut]

Hi Nick,

> > for target sh-hitachi-coff. GAS from 2.11.1pre1 fails:
>  
> > /cygdrive/c/TEMP/cc1oUj4A.s:1: Fatal error: dwarf2 is not supported
> > for this object file format
> 
> This appears to be a gcc configuration problem.  The COFF file format
> does not support the DWARF2 debug format, so gcc should not be
> producing it in the first place.

I managed to build GCC using gas 2.10.1 without any problem.

> I would suggest that you investigate why the STABS debug format is not
> being used by gcc.

Seems to be a general problem. Using gas 2.11.1 fails for my target
too. I don't use debug information here (no -g to GCC).

Hartmut