Another 3.0-branch bootstrap failure

Another 3.0-branch bootstrap failure

[Gerald Pfeifer]

...this time for i386-unknown-freebsd4.2.  This really shouldn't happen
that late during a release cycle:

gcc -c  -DIN_GCC    -g  -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wtraditional -pedantic -Wno-long-long  -DHAVE_CONFIG_H    -I. -I. -I/sw/test/gcc/cvs-3.0/gcc -I/sw/test/gcc/cvs-3.0/gcc/. -I/sw/test/gcc/cvs-3.0/gcc/config -I/sw/test/gcc/cvs-3.0/gcc/../include /sw/test/gcc/cvs-3.0/gcc/dbxout.c -o dbxout.o
/sw/test/gcc/cvs-3.0/gcc/dbxout.c: In function `dbxout_init':
/sw/test/gcc/cvs-3.0/gcc/dbxout.c:428: `N_OPT' undeclared (first use in this function)
/sw/test/gcc/cvs-3.0/gcc/dbxout.c:428: (Each undeclared identifier is reported only once
/sw/test/gcc/cvs-3.0/gcc/dbxout.c:428: for each function it appears in.)
gmake[2]: *** [dbxout.o] Error 1
gmake[2]: Leaving directory `/files/pfeifer/OBJ-0417-0838/gcc'
gmake[1]: *** [stage1_build] Error 2
gmake[1]: Leaving directory `/files/pfeifer/OBJ-0417-0838/gcc'
gmake: *** [bootstrap] Error 2

My gut tells me this is related to the following change:

  2001-04-16  Zack Weinberg  <zackw@stanford.edu>

        * toplev.c (output_lang_identify): Delete.
        (compile_file): Don't call ASM_IDENTIFY_GCC or ASM_IDENTIFY_LANGUAGE.
        Don't generate gcc2_compiled. label.  Don't emit a nop if profiling.
        If IDENT_ASM_OP is defined, emit an .ident "GCC (GNU) <version>"
        after calling ASM_FILE_END.

        * dbxout.c (dbxout_init): Don't call ASM_IDENTIFY_GCC_AFTER_SOURCE.
        Instead, unconditionally emit an N_OPT stab with string equal
        to STABS_GCC_MARKER, which defaults to "gcc2_compiled."

        * defaults.h: Don't provide default for ASM_IDENTIFY_LANGUAGE.

        * config/dbxelf.h, config/freebsd.h, config/nextstep.h,
        config/psos.h, config/ptx4.h, config/1750a/1750a.h,
        config/alpha/alpha.h, config/alpha/elf.h, config/arc/arc.h,
        config/arm/aof.h, config/arm/coff.h, config/arm/elf.h,
        config/arm/linux-elf.h, config/avr/avr.h, config/d30v/d30v.h,
        config/dsp16xx/dsp16xx.h, config/h8300/h8300.h,
        config/i370/i370.h, config/i386/beos-elf.h,
        config/i386/osf1elf.h, config/i386/osfrose.h,
        config/ia64/sysv4.h, config/m68k/3b1.h, config/m68k/auxgas.h,
        config/m68k/crds.h, config/m68k/hp320.h, config/m68k/m68kv4.h,
        config/m68k/mot3300.h, config/m68k/news.h, config/m68k/sgs.h,
        config/m68k/tower-as.h, config/mcore/mcore-elf.h,
        config/mips/iris6.h, config/mips/osfrose.h,
        config/mips/sni-svr4.h, config/pa/som.h, config/pdp11/pdp11.h,
        config/sparc/sp64-elf.h:
        Delete definitions of any or all of: ASM_IDENTIFY_GCC,
        ASM_IDENTIFY_GCC_AFTER_SOURCE, and ASM_IDENTIFY_LANGUAGE.

        * config/elfos.h, config/freebsd.h, config/linux.h,
        config/alpha/elf.h, config/arm/linux-elf.h, config/i386/i386elf.h,
        config/i386/ptx4-i.h, config/mips/gnu.h, config/mips/linux.h,
        config/sparc/linux.h, config/sparc/linux64.h: Delete definitions
        of ASM_FILE_START which merely emit a .version directive.  Delete
        definitions of ASM_FILE_END which merely emit an .ident directive.

Re: Another 3.0-branch bootstrap failure

[Zack Weinberg]

On Tue, Apr 17, 2001 at 09:07:35AM +0200, Gerald Pfeifer wrote:
> ...this time for i386-unknown-freebsd4.2.  This really shouldn't happen
> that late during a release cycle:
> 
> gcc -c  -DIN_GCC    -g  -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -Wtraditional -pedantic -Wno-long-long  -DHAVE_CONFIG_H    -I. -I. -I/sw/test/gcc/cvs-3.0/gcc -I/sw/test/gcc/cvs-3.0/gcc/. -I/sw/test/gcc/cvs-3.0/gcc/config -I/sw/test/gcc/cvs-3.0/gcc/../include /sw/test/gcc/cvs-3.0/gcc/dbxout.c -o dbxout.o
> /sw/test/gcc/cvs-3.0/gcc/dbxout.c: In function `dbxout_init':
> /sw/test/gcc/cvs-3.0/gcc/dbxout.c:428: `N_OPT' undeclared (first use in this function)

Loren and I already discussed this.  He's testing a patch and he said
he would apply it if it worked.

The other message you noticed is a related problem but doesn't affect
FreeBSD, which is sensible and defines stab numbers with #define.

zw

PATCH: Re: Another 3.0-branch bootstrap failure

[Loren James Rittle]

Hi Gerald,

> ...this time for i386-unknown-freebsd4.2.  This really shouldn't happen
> that late during a release cycle:  [...missing N_OPT symbol...]

Already on it. Although an obvious fix, I conferred with Zack and
waited until my bootstraps completed.  Patch just installed on both
mainline and 3.0 with Zack's approval.

I bow my head that I didn't look at Zack's patch closely before he
applied it.

Regards,
Loren

2001-04-17  Loren J. Rittle  <ljrittle@acm.org>

	* dbxout.c (N_OPT): Define if not available from <stab.h>

Index: gcc/dbxout.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/dbxout.c,v
retrieving revision 1.77
diff -c -8 -r1.77 dbxout.c
*** dbxout.c	2001/04/16 18:30:31	1.77
--- dbxout.c	2001/04/17 01:30:44
***************
*** 178,193 ****
--- 178,198 ----
  #include "gstab.h" /* If doing DBX on sysV, use our own stab.h.  */
  #else
  #include <stab.h>
  
  /* This is a GNU extension we need to reference in this file.  */
  #ifndef N_CATCH
  #define N_CATCH 0x54
  #endif
+ 
+ /* This is a Solaris extension we need to reference in this file. */
+ #ifndef N_OPT
+ #define N_OPT 0x3c
+ #endif
  #endif
  
  #ifdef __GNU_STAB__
  #define STAB_CODE_TYPE enum __stab_debug_code
  #else
  #define STAB_CODE_TYPE int
  #endif