build(c): work around clang making __COUNTER__ a warning

[lidavidm]

Closes #4042.

[lidavidm]

CC @paleolimbot this happens on recent Clang: google/benchmark#2057

I did the hacky thing to get around it but maybe it would be better to fix this in nanoarrow.

[paleolimbot]

Can you wrap auto&& NAME = (RHS) in a scope in the _IMPL macros?

do {
  auto&& NAME = (RHS);
  if (!(NAME).ok()) {
    return (NAME).ToADBC(ERROR);
  }
} while(0);

...should be safe, although won't work for UNWRAP_RESULT(some_type some_var) unless this works:

lhs;
do {
  auto&& NAME = (RHS);
  if (!(NAME).has_value()) {
    return std::move(name).status();
  }
  lhs = std::move((name).value());
} while(0);

In nanoarrow I think I can just define the status name to be a hard-coded long but almost certainly unique value since all return values would happen via side-effect and not the result code.

[lidavidm]

Updated the vendored nanoarrow and updated the macros similarly

[paleolimbot]

Thank you!

It's a tiny bit of a bummer we can't do UNWRAP_RESULT(std::string_view create, create_query.GetString()); anymore but it's also only internal usage.

[lidavidm]

I think the fun time will come when arrow-cpp has to decide how/whether to fix this, since I believe we all derived our macros from that one 😅