[BUG] fails to compile on windows with clang-18 / clang-19

[kelbon]

  // Offset tests to ensure commonly accessed data is on the first cache line.
  const int cache_line_size = 64;
  static_assert(offsetof(State, error_occurred_) <=
                    (cache_line_size - sizeof(error_occurred_)),
                "");

error:

/src/benchmark.cc:191:17: error: offset of on non-standard-layout type 'State' [-Werror,-Winvalid-offsetof]
  191 |   static_assert(offsetof(State, error_occurred_) <=
      |                 ^               ~~~~~~~~~~~~~~~
E:\llvm_\LLVM\lib\clang\18\include\__stddef_offsetof.h:16:24: note: expanded from macro 'offsetof'
   16 | #define offsetof(t, d) __builtin_offsetof(t, d)
      |                        ^                     ~
1 error generated.

[kelbon]

Please, remove Werror flag from build, because new warnings / new enabled warnings broke compilation

[dmah42]

thanks for the report. which compiler/OS are you on? and which version of benchmark are you building?

the warnings/errors exist to provide information about possible mistakes, so removing it doesn't really help. i'd rather take the opportunity to fix the problem. if folks read warnings we could totally rely on them but...

[dmah42]

in this case, we're getting burned (finally) for using offsetof on a data structure that is very much not standard layout: it has non-static data members with different access control.

we could remove the static_assert and replace it with a well-placed comment to not move things around in State but obviously that's less safe than what we have. we could also disable the warning, but it's not obvious to me how big a deal it is... does it make the static_assert invalid?

[kelbon]

thanks for the report. which compiler/OS are you on? and which version of benchmark are you building?

the warnings/errors exist to provide information about possible mistakes, so removing it doesn't really help. i'd rather take the opportunity to fix the problem. if folks read warnings we could totally rely on them but...

I used 1.71 on windows with clang-18/19, it seems to be fixed in 1.9.0, but i dont think its 'invalid' warning

https://github.com/google/benchmark/pull/1821

[dmah42]

yes i'm not sure it's invalid either, but this at least unblocks us until this becomes an error.

i'm open to hearing ideas for how to capture the spirit of the check in another way.