[DPDK/core Bug 1425] enable_stdatomic=true breaks C++ on GCC 11 and earlier

[DPDK/core Bug 1425] enable_stdatomic=true breaks C++ on GCC 11 and earlier

[bugzilla]

[-- Attachment #1: Type: text/plain, Size: 872 bytes --]

https://bugs.dpdk.org/show_bug.cgi?id=1425

            Bug ID: 1425
           Summary: enable_stdatomic=true breaks C++  on GCC 11 and
                    earlier
           Product: DPDK
           Version: 23.11
          Hardware: All
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: Normal
         Component: core
          Assignee: dev@dpdk.org
          Reporter: mattias.ronnblom@ericsson.com
  Target Milestone: ---

On GCC 11 and earlier, configuring enable_stdatomic=true prevents the use of
all DPDK header files that directly or indirectly include <rte_stdatomic.h>
from a C++ translation unit (e.g., app).

<rte_stdatomic.h> includes <stdatomic.h>, which in turn is not necessarily
C++-compatible.

-- 
You are receiving this mail because:
You are the assignee for the bug.

[-- Attachment #2: Type: text/html, Size: 2751 bytes --]

Re: [DPDK/core Bug 1425] enable_stdatomic=true breaks C++ on GCC 11 and earlier

[Tyler Retzlaff]

On Mon, Apr 29, 2024 at 06:21:13AM +0000, bugzilla@dpdk.org wrote:
> https://bugs.dpdk.org/show_bug.cgi?id=1425
> 
>             Bug ID: 1425
>            Summary: enable_stdatomic=true breaks C++  on GCC 11 and
>                     earlier
>            Product: DPDK
>            Version: 23.11
>           Hardware: All
>                 OS: Linux
>             Status: UNCONFIRMED
>           Severity: normal
>           Priority: Normal
>          Component: core
>           Assignee: dev@dpdk.org
>           Reporter: mattias.ronnblom@ericsson.com
>   Target Milestone: ---
> 
> On GCC 11 and earlier, configuring enable_stdatomic=true prevents the use of
> all DPDK header files that directly or indirectly include <rte_stdatomic.h>
> from a C++ translation unit (e.g., app).
> 
> <rte_stdatomic.h> includes <stdatomic.h>, which in turn is not necessarily
> C++-compatible.

This is known but to add some information.

C++ and enable_stdatomic=true for llvm and gcc are not currently
supported. the combination will remain unsupported for C++ compilers
that do not support -std=c++23 which is the first C++ standard that
requires interoperability with C11 stdatomic.h

When enable_stdatomic=true there are bugs/incorrect usages of atomic
qualifier in casts that (even when using C++23) cause compilation
failure. These are a fixable but are low priority without -std=c++23.

Finally, the legacy atomics remain unconverted to stdatomic. This will
cause enable_stdatomic=true not to build when using llvm (but not gcc)
because llvm strictly enforces qualification when using atomic generics.

> 
> -- 
> You are receiving this mail because:
> You are the assignee for the bug.

Re: [DPDK/core Bug 1425] enable_stdatomic=true breaks C++ on GCC 11 and earlier

[Mattias Rönnblom]

On 2024-04-30 01:14, Tyler Retzlaff wrote:
> On Mon, Apr 29, 2024 at 06:21:13AM +0000, bugzilla@dpdk.org wrote:
>> https://bugs.dpdk.org/show_bug.cgi?id=1425
>>
>>              Bug ID: 1425
>>             Summary: enable_stdatomic=true breaks C++  on GCC 11 and
>>                      earlier
>>             Product: DPDK
>>             Version: 23.11
>>            Hardware: All
>>                  OS: Linux
>>              Status: UNCONFIRMED
>>            Severity: normal
>>            Priority: Normal
>>           Component: core
>>            Assignee: dev@dpdk.org
>>            Reporter: mattias.ronnblom@ericsson.com
>>    Target Milestone: ---
>>
>> On GCC 11 and earlier, configuring enable_stdatomic=true prevents the use of
>> all DPDK header files that directly or indirectly include <rte_stdatomic.h>
>> from a C++ translation unit (e.g., app).
>>
>> <rte_stdatomic.h> includes <stdatomic.h>, which in turn is not necessarily
>> C++-compatible.
> 
> This is known but to add some information.
> 

Is it also documented?

> C++ and enable_stdatomic=true for llvm and gcc are not currently
> supported. the combination will remain unsupported for C++ compilers
> that do not support -std=c++23 which is the first C++ standard that
> requires interoperability with C11 stdatomic.h >
> When enable_stdatomic=true there are bugs/incorrect usages of atomic
> qualifier in casts that (even when using C++23) cause compilation
> failure. These are a fixable but are low priority without -std=c++23.
> 
> Finally, the legacy atomics remain unconverted to stdatomic. This will
> cause enable_stdatomic=true not to build when using llvm (but not gcc)
> because llvm strictly enforces qualification when using atomic generics.
>

OK, I see. It'll be a while until enable_stdatomic is usable outside 
Windows then, for generic builds.

Am I right if I say that C++23-capable compilers/run-times are supposed 
to have a <stdatomic.h> which interoperates with C++ even in C++11-mode? 
Or need the application be compiled as C++23.

>>
>> -- 
>> You are receiving this mail because:
>> You are the assignee for the bug.

Re: [DPDK/core Bug 1425] enable_stdatomic=true breaks C++ on GCC 11 and earlier

[Tyler Retzlaff]

On Tue, Apr 30, 2024 at 10:06:08PM +0200, Mattias Rönnblom wrote:
> On 2024-04-30 01:14, Tyler Retzlaff wrote:
> >On Mon, Apr 29, 2024 at 06:21:13AM +0000, bugzilla@dpdk.org wrote:
> >>https://bugs.dpdk.org/show_bug.cgi?id=1425
> >>
> >>             Bug ID: 1425
> >>            Summary: enable_stdatomic=true breaks C++  on GCC 11 and
> >>                     earlier
> >>            Product: DPDK
> >>            Version: 23.11
> >>           Hardware: All
> >>                 OS: Linux
> >>             Status: UNCONFIRMED
> >>           Severity: normal
> >>           Priority: Normal
> >>          Component: core
> >>           Assignee: dev@dpdk.org
> >>           Reporter: mattias.ronnblom@ericsson.com
> >>   Target Milestone: ---
> >>
> >>On GCC 11 and earlier, configuring enable_stdatomic=true prevents the use of
> >>all DPDK header files that directly or indirectly include <rte_stdatomic.h>
> >>from a C++ translation unit (e.g., app).
> >>
> >><rte_stdatomic.h> includes <stdatomic.h>, which in turn is not necessarily
> >>C++-compatible.
> >
> >This is known but to add some information.
> >
> 
> Is it also documented?

we don't document anything about C++ right now including minimum
compiler requirements or -std. i agree we probably should, if i find
cycles i'll see if i can provide some text.

> 
> >C++ and enable_stdatomic=true for llvm and gcc are not currently
> >supported. the combination will remain unsupported for C++ compilers
> >that do not support -std=c++23 which is the first C++ standard that
> >requires interoperability with C11 stdatomic.h >
> >When enable_stdatomic=true there are bugs/incorrect usages of atomic
> >qualifier in casts that (even when using C++23) cause compilation
> >failure. These are a fixable but are low priority without -std=c++23.
> >
> >Finally, the legacy atomics remain unconverted to stdatomic. This will
> >cause enable_stdatomic=true not to build when using llvm (but not gcc)
> >because llvm strictly enforces qualification when using atomic generics.
> >
> 
> OK, I see. It'll be a while until enable_stdatomic is usable outside
> Windows then, for generic builds.
> 
> Am I right if I say that C++23-capable compilers/run-times are
> supposed to have a <stdatomic.h> which interoperates with C++ even
> in C++11-mode? Or need the application be compiled as C++23.

the application will need to be compiled with -std=C++23 though i think
llvm does have a stdatomic.h that may work with lower -std, gcc does
not last i checked. in general i think (as you found with _Generic) the
compiler vendors are limiting leaking of C into C++ and vice versa where
it is not explicitly intended to operate in one language or the other.

i'll be looking into this more beyond 24.07 since there is work to do
even consuming with msvc that is not upstream.

> 
> >>
> >>-- 
> >>You are receiving this mail because:
> >>You are the assignee for the bug.