Ok, push done.

Even after full rebuild those tests are still UNRESOLVED on my system.

Yes, I also noticed that I could remove this check. I'll propose it later.

François

On 02/06/2023 09:43, Jonathan Wakely wrote:
> On Fri, 2 Jun 2023 at 08:33, François Dumont wrote:
>
>     I haven't been able to reproduce so far.
>
>     Here is however a patch that I think will fix the problem. At
>     least failing tests are UNRESOLVED on my system.
>
>         libstdc++: Fix broken _GLIBCXX_PARALLEL mode
>
>         Add missing <parallel/search.h> include in <parallel/algobase.h>.
>
>
> This fixes the broken parallel mode.
>
>
>         Detect availability of <omp.h> in tests needing it to make
>     them UNSUPPORTED
>         rather than PASS when header is missing.
>
>         libstdc++-v3/ChangeLog:
>
>                 * include/parallel/algobase.h: Include
>     <parallel/search.h>.
>                 * testsuite/lib/libstdc++.exp
>     (check_effective_target_omp): New.
>                 * testsuite/17_intro/headers/c++2011/parallel_mode.cc:
>                 Add { dg-require-effective-target omp }.
>                 * testsuite/17_intro/headers/c++2014/parallel_mode.cc:
>     Likewise.
>                 * testsuite/17_intro/headers/c++2017/parallel_mode.cc:
>     Likewise.
>
>     Ok to commit ?
>
>
> Please just add the #include to parallel/algobase.h for now.
>
> The effective-target keyword seems reasonable, but "omp" is not a good 
> name. And if we add that dg-require-effective-target to those tests 
> then they don't need to repeat the check in the test itself:
> #if __has_include(<omp.h>)
>
> So please just add the #include and then we can revisit the 
> effective-target separately.
>
>
>
>     On 01/06/2023 23:57, Jonathan Wakely wrote:
>>     On Thu, 1 Jun 2023, 21:37 François Dumont via Libstdc++,
>>     <libstdc++@gcc.gnu.org <mailto:libstdc%2B%2B@gcc.gnu.org>> wrote:
>>
>>         Now I've install OMP and try to rebuild lib to reproduce the
>>         failure.
>>
>>
>>     You shouldn't need to install anything, just build gcc and don't
>>     configure it with --disable-libgomp
>>
>     I haven't used --disable-libgomp. But maybe when I run configure
>     the 1st time it tried to detect OMP install and failed to find it
>     as I just installed it.
>
>
> I don't know what you mean, because GCC doesn't depend on "OMP". GCC 
> includes its own OpenMP implementation, and installs its own libgomp 
> runtime library to support the -fopenmp flag. It doesn't depend on 
> anything else.
>
> Which OS are you testing on?
>