[PATCH] ld: Limit "readelf --got-contents" tests

[Jan Beulich]

On 04.09.2025 16:45, H.J. Lu wrote:
> On Thu, Sep 4, 2025 at 7:41 AM Jan Beulich <jbeulich@suse.com> wrote:
>>
>> On 04.09.2025 16:34, H.J. Lu wrote:
>>> On Thu, Sep 4, 2025 at 7:12 AM Rainer Orth <ro@cebitec.uni-bielefeld.de> wrote:
>>>>
>>>> Hi H.J.,
>>>>
>>>>> On Thu, Sep 4, 2025 at 6:40 AM Rainer Orth <ro@cebitec.uni-bielefeld.de> wrote:
>>>>>>
>>>>>> Hi H.J.,
>>>>>>
>>>>>>> Since "readelf --got-contents" tests use -melf_i386, -melf_x86_64 and
>>>>>>> -melf32_x86_64 emulations, limit theses tests to targets with those
>>>>>>> emulations.
>>>>>>>
>>>>>>>       PR ld/33350
>>>>>>>       * testsuite/ld-i386/binutils.exp: Limit tests to targets with
>>>>>>>       -melf_i386 support.
>>>>>>>       * testsuite/ld-x86-64/binutils.exp: Limit tests to targets with
>>>>>>>       -melf_x86_64 and -melf32_x86_64 support.
>>>>>>
>>>>>> TBH, this feels like pure lazyness to me: as reported in the PR, if you
>>>>>> exclude the x32 tests (which are strictly Linux-only), at least
>>>>>> FreeBSD/amd64 has clean results.  Solaris/x86 is partially different,
>>>>>> but some of this is already addressed in the patch attached to the PR.
>>>>>>
>>>>>> FreeBSD/amd64 supports both -melf_i386 and -melf_x86_64, but isn't
>>>>>> included in the list above.  Besides, that patch already handles the
>>>>>> -melf_i386 vs. -melf_i386_sol2 (and x86_64 equivalent) difference: it's
>>>>>> effectively trivial.
>>>>>
>>>>> I'd like to make sure that readelf outputs match exactly for a given
>>>>> emulation.  If Solaris doesn't support such emulation, it should be
>>>>> skipped.
>>>>
>>>> the only difference of those emulations is the additions of a couple of
>>>> symbols as per ld/emultempl/solaris2.em.
>>>
>>> My v2 patch checks the emulation support.
>>>
>>>> I honestly wonder why you ask me to test your patches at all if your
>>>> only response to the slightest issue is to restrict the patches to
>>>> Linux.  After all, binutils isn's a Linux-only project (nor x86-only).
>>>
>>> I want to make sure that --plt-contents displays all relevant
>>> information correctly.  My tests are emulation specific.   There
>>> is nothing Linux specific.
>>
>> Yet still the question stands - if it's just a few extra symbols, can't
>> you cater for them using #?REGEXP constructs in the expectations?
> 
> It sounds wrong.  Are you saying that the same emulation generates
> extra symbols for Solaris?

I'm not saying that, no, but see what Rainer said. Tests shouldn't be
artificially restricted - the wider their coverage, the better.

Jan