[COMMITTED] gas: testsuite: fix regression in cfi-common-10.d

[Indu Bhagat]

On 8/20/25 4:59 PM, Alan Modra wrote:
> On Thu, Aug 21, 2025 at 09:09:18AM +0930, Alan Modra wrote:
>> On Wed, Aug 20, 2025 at 02:40:47PM -0700, Indu Bhagat via Binutils wrote:
>>> On 8/20/25 1:42 PM, H.J. Lu wrote:
>>>> Error: bad or irreducible absolute expression>, no expected output
>>>> FAIL: CFI common 10
>>>>
>>>> on Linux/i386.
>>>>
>>>> Isn't
>>>>
>>>> cfi_def_cfa_offset 1099511627808
>>>>
>>>> too big for 32-bit targets?
>>>>
>>>
>>> For 32-bit targets, true, this offset should not realistically occur.
>>> Perhaps it is best to move this test as cfi-x86-64-3.s.
>>>
>>> (I do have i386-darwin i386-lynxos in my regression script. I wonder why
>>> this didnt show up as failure in my testing...)
>>
>> If both host and target are 32-bit, and binutils is not configured
>> with --enable-64-bit-bfd, then bfd_vma will be 32 bits.  Likely you
>> are compiling on a 64-bit host.
> 
> This works for me.  I haven't checked what run_dump_test does for
> remote host testing.
> 

Yes, this does work.  Tested on a i386 cfarm machine.

I tried looking at dejagnu doc and exp files, but cannot evaluate the 
open question of whether run_dump_test is doing the right thing for 
remote host testing..

> diff --git a/gas/testsuite/gas/cfi/cfi.exp b/gas/testsuite/gas/cfi/cfi.exp
> index d89a102c885..60094fe06f7 100644
> --- a/gas/testsuite/gas/cfi/cfi.exp
> +++ b/gas/testsuite/gas/cfi/cfi.exp
> @@ -140,7 +140,9 @@ if { ![istarget "hppa64*-*"] } then {
>     run_dump_test "cfi-common-7"
>     run_dump_test "cfi-common-8"
>     run_dump_test "cfi-common-9"
> -  run_dump_test "cfi-common-10"
> +  if { [is_elf64 tmpdir/cfi-common-9.o] } {
> +      run_dump_test "cfi-common-10"
> +  }
>   
>     run_dump_test "cie-version-0"
>     run_dump_test "cie-version-1"
>