gcc segfault with newly developed variant reflection about glaze HOT 7 CLOSED

@jbbjarnason, I discovered this last night as well, as my company uses gcc 13 in production. I do not currently have a bugzilla account, so feel free to go ahead and report this. I'm going to try to simplify the code this morning so that gcc will build. I was able to simplify the code last night so that the compiler no longer segfaults, but its errors are bogus.

from glaze.

So this is a problem with gcc 13, I added a PR with gcc 13 in github workflows here, #605 which reproduces the error.

@stephenberry do you have an account at gcc bugzilla or should I make one to report this?

from glaze.

Here is a simplified test that causes the segfault:

struct put_action
{
};

struct delete_action
{
};

using tagged_variant = std::variant<put_action, delete_action>;

template <>
struct glz::meta<tagged_variant>
{
   static constexpr std::string_view tag = "action";
   static constexpr auto ids = std::array{"PUT", "DELETE"}; // Defaults to glz::name_v of the type
};

suite tagged_variant_tests = [] {
   "tagged_variant_read_tests"_test = [] {
      tagged_variant var{};
      expect(glz::read_json(var, R"({})") == glz::error_code::none);
   };
};

from glaze.

you two rock. sad about the error, but good finds

from glaze.

I found a workaround, for some reason GCC 13 is corrupting the capture by reference for std::array<std::string_view, get_max_keys<T, N>> data{};

This is for the function: get_combined_keys_from_variant

The fix is that we can just change this to a pointer and GCC will properly capture the pointer:

auto* data_ptr = &data; // This intermediate pointer is necessary for GCC 13 (otherwise segfaults with reflection logic)

from glaze.

Fixed in #605

from glaze.

Great job

from glaze.