[RFC] Cleanup DW_CFA_GNU_args_size handling

[H.J. Lu]

On Sat, Sep 10, 2011 at 4:01 AM, Eric Botcazou <ebotcazou@adacore.com> wrote:
>> I have no problems with binutils in CVS today.  That
>> may be another --with-as/--with-ld issue similar to
>>
>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50237
>
> No, this was a pilot error on my side.  Now I have HAVE_GAS_CFI_DIRECTIVE to 1

One problem with --with-as/--with-ld is we are using nm/objdump/readelf
from the different version of binutils to test as/ld features.  Sometimes, it
doesn't work too well.  For example, we have

gcc_GAS_CHECK_FEATURE([working cfi advance], gcc_cv_as_cfi_advance_working,
  ,,
[       .text
        .cfi_startproc
        .cfi_adjust_cfa_offset 64
        .skip 75040, 0
        .cfi_adjust_cfa_offset 128
        .cfi_endproc],
[[
if $gcc_cv_objdump -Wf conftest.o 2>/dev/null \
    | grep 'DW_CFA_advance_loc[24]:[    ][      ]*75040[        ]'
>/dev/null; then
   gcc_cv_as_cfi_advance_working=yes
fi

Our assembler supports CFI, but objdump may be too old to dump
DW_CFA_advance_loc
and this feature is mis-detected.

> but the bootstrap still fails if --disable-initfini-array isn't specified.
>

Please try:

http://gcc.gnu.org/ml/gcc-patches/2011-09/msg00668.html

-- 
H.J.