[PATCH] x86: cover a.out in recently added tests

[Jan Beulich]

On 09.06.2021 21:10, H.J. Lu wrote:
> On Wed, Jun 9, 2021 at 6:32 AM Jan Beulich <jbeulich@suse.com> wrote:
>>
>> On 08.06.2021 09:58, Jan Beulich via Binutils wrote:
>>> Follow the pattern found elsewhere when relocations are involved. For
>>> wrap32-data also drop a mistakenly left "ELF" from the test name. (Note
>>> that Darwin, for which the wrap32 tests are also failing, is left as-is,
>>> for there being numerous other failures already anyway, and it hence
>>> being questionable whether that target is actually properly maintained.)
>>>
>>> gas/
>>> 2021-06-XX  Jan Beulich  <jbeulich@suse.com>
>>>
>>>       * testsuite/gas/i386/quoted.d, testsuite/gas/i386/wrap32-text.d:
>>>       Adjust expectations.
>>>       * testsuite/gas/i386/wrap32-data.d: Likewise. Correct name.
>>
>> Despite this change I've just now noticed that on a 64-bit build
>> host (i.e. with BFD64 implicitly in use) ...
>>
>>> --- a/gas/testsuite/gas/i386/wrap32-data.d
>>> +++ b/gas/testsuite/gas/i386/wrap32-data.d
>>
>> ... this and ...
>>
>>> --- a/gas/testsuite/gas/i386/wrap32-text.d
>>> +++ b/gas/testsuite/gas/i386/wrap32-text.d
>>
>> ... this still fail (observed with i386-msdos, which previously I
>> had checked only on a 32-bit distro used for building):
>>
>> .../gas/testsuite/gas/i386/wrap32.s:43: Warning: value 0x1000000f4 truncated to 0xf4
>> .../gas/testsuite/gas/i386/wrap32.s:51: Warning: value 0xffffffff700002f4 truncated to 0x700002f4>, no expected output
>>
>> I'm of the opinion that any such truncation warnings are actually
>> bogus when it comes to a 32-bit target. I do have patches mostly
>> ready to address some others, but not these two. I'd be inclined
>> to make this an XFAIL on BFD64 a.out builds. Do you have an
>> opinion one way or the other here? If so, I'm afraid I wouldn't
>> know how to express this with an #xfail at the top of the .d file.
>>
>> It also doesn't help that i386.exp's gas_bfd64_check doesn't work
>> for a.out, so I also couldn't easily skip the test in that case.
>> Since it depends on "as --help" producing a description for the
>> --32 / --64 options, and since that line is absent for a.out
>> irrespective of the BFD64 setting, all a.out builds get taken as
>> if they were !BFD64.
> 
> Just limit these testcases for known good targets.

But that way effectively saying "works only for ELF and PE/COFF" is
a lie. That's what I'd like to avoid, alongside the test starting to
fail on a non-ELF, non-PE/COFF target because of a future change.

Jan