coverage: 99.186%. remained the same
when pulling 7206a82 on LocutusOfBorg:fix-warning
into 8215dba on nlohmann:develop.

Explained at https://stackoverflow.com/questions/76867698/what-does-ignoring-attributes-on-template-argument-mean-in-this-context

Thanks! So to silence GCC's warning about an ignored attribute, we define a unique pointer deleter type without that attribute and call this a fix, because GCC does not realize that we then instantiate this deleter with a function whose argument has an attribute?

The warning is only on the template declaration, so moving it to the instantiation doesn't trigger that particular warning. They may expand it or add a new one in the future.

As described in the thread, below would be a better solution for size and performance if it were in the main library, but since it's in the tests, we're not as concerned about size and performance.

struct unique_ptr_fclose_deleter {
    void operator()(struct FILE* f) const {
        fclose(f);
    }
};
using FilePtr = std::unique_ptr<FILE, unique_ptr_fclose_deleter>;

Thanks for the clarification! I will merge this once #4701 is merged, because otherwise the CI would fail.

thanks!

Thank you!