This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Destructor support for C++11 thread_local variables
On Wed, 10 Oct 2012 15:51:42 +0000, Joseph wrote:
> On Wed, 10 Oct 2012, Siddhesh Poyarekar wrote:
>
> > ChangeLog:
> >
> > * include/stdlib.h (__cxa_thread_atexit_impl): Declare.
> > (__call_tls_dtors): Likewise.
> > * stdlib/Makefile (routines): Add __cxa_thread_atexit_impl.
> > * stdlib/Versions (GLIBC_2.17): Likewise.
>
> Since you're adding a new public symbol, I'd have expected your
> testing to fail because the patch isn't updating the ABI baselines.
> They should be updated at least for all libc architectures
> (preferably for all architectures).
Ugh, sorry. I had made a mental not for that to be [patch 2/2] but I
simply missed doing it in the end. I'll send it in as a [patch 2/2],
thanks for catching it.
> (Note on the baselines: I thought the symbols for each version were
> in ASCII collation order, meaning that __cxa_thread_atexit_impl
> should be *before* secure_getenv under GLIBC_2.17 for each
> architecture. But
> sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/nptl/libc.abilist is
> listing __ppc_get_timebase_freq *after* secure_getenv. I don't know
> whether that indicates some other order is in use or that the order
> doesn't actually matter, or whether the addition to that ABI baseline
> of __ppc_get_timebase_freq was untested. For the corresponding ports
> update I just followed the libc change mechanically.)
I didn't see any order in it, so I just appended it to the end.
Siddhesh