This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Patch to fix irix6o32 explicit-reloc breakage


Richard Sandiford writes:

> > * passing -mdebug to non-gas assemblers
> 
> Yes, this was one of them.  I hacked around it by redefining
> SUBTARGET_ASM_DEBUGGING_SPEC in iris6-o32-as.h, but I don't
> think it was the proper fix.

Indeed: I had a patch for this last year

	http://gcc.gnu.org/ml/gcc-patches/2002-07/msg01070.html

but rth thought the runtime handling of the debug format issues was overly
complicated.  I'll have to try his suggestion, and will probably assume
that every version of gas used with gcc will have to support -mdebug so we
can avoid the configure-time detection of -mdebug as well.

> Also, gas seems to add the name of the assembly file to the
> mdebug info, leading to spurious bootstrap comparison failures.
> Was this always a problem or is it new?

I never encountered this, neither last year with gas 2.11.2, 2.12.1,
2.12.90, or 2.13 nor now with gcc 3.3 and gas 2.13.91.  At least the latter
combination bootstrapped and regtested successfully (except for a
considerable number of EH failures I'll need to investigate):

	http://gcc.gnu.org/ml/gcc-testresults/2003-05/msg01527.html

> > * improperly (or not at all) passing --gstabs or --gdwarf2 to gas
> 
> Didn't hit this, but I didn't do much more than a bootstrap.

It even broke bootstrap for me, but only if configuring with
--without-stabs to get DWARF 2 debugging.  I needed this for proper
debugging last year since stabs support in general and stabs-in-mdebug in
particular seem rather broken in current versions of gdb.

> > * bootstrap error for mips-sgi-irix6.5o32 --with-gnu-as on mainline:
> > 
> > /vol/gnu/src/gcc/gcc-dist/gcc/dwarf2asm.c: In function `dw2_asm_output_nstring':
> > /vol/gnu/src/gcc/gcc-dist/gcc/dwarf2asm.c:281: error: Attempt to delete prologue/epilogue insn:
> > (insn 513 391 514 0 0 (set (mem:SI (plus:SI (reg/f:SI 29 sp)
> >                 (const_int 80 [0x50])) [0 S4 A32])
> >         (reg:SI 6 a2)) -1 (nil)
> >     (nil))
> > /vol/gnu/src/gcc/gcc-dist/gcc/dwarf2asm.c:281: internal compiler error: in ?, at flow.c:1653
> 
> I was using the native assembler so I didn't see this.  I've seen
> something similar when trying to build gdb for mipsel-linux-gnu.
> 
> A patch I'm working on happens to (accidentally) paper over the
> problem but again isn't a proper fix.  I'd like to rework the
> code in mips_expand_prologue in a way that should fix the
> underlying bug as well.  The patch I mentioned is a prerequisite
> for that.

Fine: I'll probably concentrate on the 3.3 branch for the moment, where
this hopefully won't affect me.

> I also had problems building stdc++.h.gch:
> 
> cc1plus: warning: `-g': unknown or unsupported -g option
> cc1plus: warning: `-g': unknown or unsupported -g option
> /es/scratch/rsandifo/gcc/HEAD/gcc/libstdc++-v3/include/stdc++.h:86: fatal error:
>  error
>    writing to /var/tmp//ccVTF42c.s: Bad file number
> compilation terminated.
> cc1plus: warning: `-g': unknown or unsupported -g option
> cc1plus: warning: `-g': unknown or unsupported -g option
> /es/scratch/rsandifo/gcc/HEAD/gcc/libstdc++-v3/include/stdc++.h:86: fatal error:
>  error
>    writing to /var/tmp//ccSfli2b.s: Bad file number
> 
> Haven't really looked into it.

Since mainline didn't bootstrap with native as, I didn't get that far yet.

	Rainer


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]