Re: [PATCH] eal: fix C++ builds when stdatomics is set

[Bruce Richardson <bruce.richardson@intel.com>]

On Fri, Sep 25, 2026 at 12:13:41PM +0200, David Marchand wrote:
> On Thu, 24 Sept 2026 at 16:13, Bruce Richardson
> <bruce.richardson@intel.com> wrote:
> >
> > When the DPDK build is configured to use stdatomics rather than compiler
> > builtin atomics, the C++ chkincs builds were failing
> >
> > In file included from buildtools/chkincs/staging/generic/rte_atomic.h:18,
> >                  from /home/morten/upstreaming/dpdk-stack-std/lib/eal/x86/include/rte_atomic.h:12,
> >                  from buildtools/chkincs/chkincs-cpp.p/rte_atomic.cpp:1:
> > buildtools/chkincs/staging/rte_stdatomic.h:30:9: error: ‘memory_order’ does not name a type
> >    30 | typedef memory_order rte_memory_order;
> >       |         ^~~~~~~~~~~~
> >
> > The cause is differences in atomics in C (C11) and C++ (e.g. C++17 as
> > supported by current GCC). As explained by AI analysis:
> >
> > " _Atomic(T) is a C11 feature. In C++ mode, GCC 15 does not support it as
> >  a type specifier, and <stdatomic.h> in C++17 doesn't expose
> >  memory_order in the global namespace (that's only in C++23). DPDK
> >  headers also use _Atomic in anonymous unions and under extern "C",
> >  which are incompatible with C++'s std::atomic<T> mapping."
> >
> > To fix this in a resilient manner we need to go from two blocks in
> > rte_atomic.h to three. We have the existing non-standard/builtin
> > atomics path, but the standard atomic path now needs to be split into C
> > compatible and C++ compatible blocks. This fixes the build issues with
> > C++ while keeping existing C builds unchanged.
> >
> > Bugzilla ID: 1985
> > Fixes: 5c381a3587d1 ("eal: provide stdatomic API")
> > Cc: stable@dpdk.org
> >
> > Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> 
> Probably worth squashing this in (I tested in GHA):
> 
> $ git diff
> diff --git a/.ci/linux-build.sh b/.ci/linux-build.sh
> index e0b914a142..b09d72dd87 100755
> --- a/.ci/linux-build.sh
> +++ b/.ci/linux-build.sh
> @@ -111,13 +111,12 @@ OPTS="$OPTS -Dplatform=generic"
>  OPTS="$OPTS -Ddefault_library=$DEF_LIB"
>  if [ "$STDATOMIC" = "true" ]; then
>      OPTS="$OPTS -Denable_stdatomic=true"
> +fi
> +OPTS="$OPTS -Dcheck_includes=true"
> +if [ "${CC%%clang}" != "$CC" ]; then
> +    export CXX=clang++
>  else
> -    OPTS="$OPTS -Dcheck_includes=true"
> -    if [ "${CC%%clang}" != "$CC" ]; then
> -        export CXX=clang++
> -    else
> -        export CXX=g++
> -    fi
> +    export CXX=g++
>  fi
>  if [ "$MINI" = "true" ]; then
>      OPTS="$OPTS -Denable_drivers=net/null"
> 
> 
> Compilation passes fine in my env and GHA.
> 
Sure. Shall I do a respin to include that?

/Bruce