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 Wed, Jul 23, 2008 at 3:28 AM, Alexandre Oliva <aoliva@redhat.com> wrote:
> On Jul 15, 2008, "H.J. Lu" <hjl.tools@gmail.com> wrote:
>
>> The only changes are the expected outputs. The only thing I
>> know is they match the linker's outputs :-).
>
> They did that before.  All other cases had been tested with run-time
> tests before.  Unfortunately, I'd missed this one case of relaxation,
> both in the specs and in the tests.  See perf-src in
> http://www.lsd.ic.unicamp.br/~oliva/writeups/TLS/ if you want to know
> what was actually run-time tested, in addition to the coverage
> afforded by the glibc testsuite.
>
> Back then, it didn't seem like such a good idea to add linker tests
> that depend on GCC and glibc features that are not widely deployed.
> The features hadn't even been added to GCC or glibc back then, and
> glibc has gained support for these extensions just a couple of months
> ago.
>
> Even now, I don't think this is the place for integration testing.
> Especially when it's completely redundant with existing tests.
>
> This linker test tests the linker behavior.  Testing for availability
> of specific compiler and libc features seems like overkill (if at all
> doable; how do you tell a linker error from lack of support in libc?),
> and a complete waste of time.

That is nothing new.  Weak symbol tests were failing for a while
for some people before everyone updated to the new glibc:

http://sources.redhat.com/ml/binutils/2003-04/msg00568.html

The point is they have been passing for  binutils developers.
That is what counts.

Will the run-time tests pass with glibc 2.8 on Fedora 9? If yes, please
add the run-time tests.

Thanks.


-- 
H.J.