[libcxx-dev] Test failures for system with no threads

Louis Dionne via libcxx-dev libcxx-dev at lists.llvm.org
Tue Sep 1 10:35:44 PDT 2020


Hi,

Sorry for the delayed reply. The best way to solve this problem (and all other problems too) is to move away from the classic (legacy) config, and to define how you want to run the test suite yourself from first principles. This should be a lot easier than trying to force the legacy config into doing what you want. You can see an example of doing that here: libcxx/test/configs/libcxx-trunk-shared.cfg.in.

Please reach out if you need help on specifics.

Louis

> On Aug 13, 2020, at 11:45, Hafiz Abid Qadeer via libcxx-dev <libcxx-dev at lists.llvm.org> wrote:
> 
> Hi,
> I test libcxx for baremetal targets and I recently saw many test
> failures. Upon investigation, it seems that recent config changes have
> altered when available_features is populated. This is causing wrong set
> of libraries to be linked. Here are more details:
> 
> Previously features like 'libcpp-has-no-threads' were added in the
> available_features in parse_config_site_and_add_features
> which was called before configure_link_flags. So when control reached
> LinuxLocalTI.add_cxx_link_flags (called from configure_link_flags),
> available_features was already populated. The add_cxx_link_flags could
> check whether feature like 'libcpp-has-no-threads' is present or not and
> link -lpthread accordingly
> 
> After recent changes, the 'available_features' is populated in
> newconfig.configure which is called after configure_link_flags. As a
> result, when add_cxx_link_flags is called, the 'available_features' is
> empty and add_cxx_link_flags always calculates enable_threads to be true.
> 
> I assume that enable_threads is true on most systems so thats why this
> error was not caught. I was testing with LIBCXX_ENABLE_THREADS set to
> false so noticed this error.
> 
> I am working on a fix unless someone beats me to it. But would
> appreciate if somebody who is more familiar with this area can confirm
> that my analysis is correct.
> 
> Thanks,
> -- 
> Hafiz Abid Qadeer
> Mentor Embedded/CodeSourcery
> _______________________________________________
> libcxx-dev mailing list
> libcxx-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/libcxx-dev



More information about the libcxx-dev mailing list