Re: Problem with LD-to-LE relaxation in gnu2 tls-dialect (IA32 and X86-64)

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Mon, Jul 28, 2008 at 11:01 AM, Alexandre Oliva <aoliva@redhat.com> wrote:
> On Jul 23, 2008, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>
>> On Wed, Jul 23, 2008 at 5:01 AM, Alexandre Oliva <aoliva@redhat.com> wrote:
>>> On Jul 23, 2008, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>
>>>> Will the run-time tests pass with glibc 2.8 on Fedora 9?
>
>>> No.  These extensions were merged after glibc 2.8.
>
>> Please provide run-time tests and mark them XFAIL. We can
>> remove XFAIL when glibc is updated.
>
> Can anyone please help me find other TLS run-time tests in the GNU ld
> testsuite, so that I can use the same infrastructure and similar
> patterns of testing for availability of features in the compiler and
> libc?
> Actually...  There's no reason why these tests should even be machine
> specific.  Once someone helps me find them (I tried), I'll be happy to
> adapt them so that they cover TLSDESC on x86 too.
>
> That I didn't get any failures before, even with CC='gcc
> -mtls-dialect=gnu2', leads me to the conclusion that what you're
> asking for is an unrelated generic feature, that's missing in the GNU
> testsuite collection.

I think we should add a set of generic TLS run-time. We can
start with creating ld-elf/tls.exp by copying it from ld-elf/dwarf.exp.
Is there a way to tell if TLSDESC is supported by glibc?

>
> The absence of other TLS run-time tests in the GNU ld testsuite, that
> are present elsewhere, is an indication that this is not the best
> place for this kind of test, so I'll refrain from holding up any
> further this obvious fix for an obvious bug.  There's no point in
> conflating these two issues, and no point in delaying the fix until it
> gets something that wasn't demanded of any other TLS implementation.
>

To me, it is an indication that we need some TLS run-time tests
since link time tests may not test the linker support properly.


-- 
H.J.