[systemd-devel] [PATCH] __thread --> thread_local for C11 compat

[Lennart Poettering]

On Sun, 15.12.13 14:56, Shawn Landden (shawn at churchofgit.com) wrote:

> On Sun, Dec 15, 2013 at 1:56 PM, Lennart Poettering
> <lennart at poettering.net> wrote:
> > On Sun, 15.12.13 13:19, Shawn Landden (shawn at churchofgit.com) wrote:
> >
> >> also make thread_local available w/o including <threads.h>
> >
> > Hmm, that looks a bit too early, no? Does gcc even support this? glibc
> gcc-4.9 supports _Thread_local
> http://gcc.gnu.org/gcc-4.9/changes.html

Ah, nice. I figure that should be good enough then. (That said, my
Fedora 20 doesn't have gcc 4.9 yet, so I can't test this...)

> switching -std=gnu99 to -std=gnu11 in Makefile I can compile
> systemd w/o the clause that defines _Thread_local as __thread
> and all tests pass.
> 
> Perhaps we should add
> || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9)
> to the check

Hmm, checks for gcc versions are really only the last resorts. If there
are feature test macros, we should use those, and should they break we
an still resort to version checks.

> >> +#if __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__)
> >> +#else
> >> +#  define _Thread_local __thread

I don't really like the "inverted" if thing with nothing on the true
branch and only a false branch I must say.

Also, should we really take the deviation of filling in the c
low-levelism "_Thread_local" here? I think I'd prefer this:

#ifndef thread_local
#if __STDC_VERSION__ >= 201112L && !defined(__STDC_NO_THREADS__)
#define thread_local _Thread_local
#else
#define thread_local __thread
#endif

With that in place we stay away from defining our own low-level
_Thread_local, instead we only define the high-level thread_local, and
do so either to the C11 language feature or the gcc extension...

If that looks OK (and is tested!) I'd be happy to merge such a patch.

Lennart

-- 
Lennart Poettering, Red Hat