Keeping state

The first thing we are going to need is somewhere to store the stack backtrace. We could take the easier route and simply store it into an allocated block and keep the pointer as a custom payload in a result<T, std::pair<error_code, std::unique_ptr<stack_backtrace>>> (see previous section on Custom payloads). But let us assume that we care so deeply about bounded execution times that ever calling malloc is unacceptable.

We therefore are going to need some completely static and trivially typed storage perhaps kept per-thread to avoid the need to keep mutexes.

/* Outcome's hook mechanism works vis ADL, so we will need a custom namespace
to ensure the hooks apply only to the types declared in this namespace only
namespace error_code_extended
  // The extra error information we will keep
  struct extended_error_info
    std::array<void *, 16> backtrace;  // The backtrace
    size_t items;                      // Items in backtrace array which are valid
  struct mythreadlocaldata_t
    // Keep 16 slots of extended error info as a ringbuffer
    extended_error_info slots[16];
    // The current oldest slot
    uint16_t current{0};

    // Return the oldest slot
    extended_error_info &next() { return slots[(current++) % 16]; }

    // Retrieve a previously stored slot, detecting if it is stale
    extended_error_info *get(uint16_t idx)
      // If the idx is stale, return not found
      if(idx - current >= 16)
        return nullptr;
      return slots + (idx % 16);

  // Meyers' singleton returning a thread local data structure for this thread
  inline mythreadlocaldata_t &mythreadlocaldata()
    static thread_local mythreadlocaldata_t v;
    return v;
View this code on Github

The extended error info is kept in a sixteen item long ring buffer. We continuously increment the current index pointer which is a 16 bit value which will wrap after 65,535. This lets us detect an attempt to access recycled storage, and thus return item-not-found instead of the wrong extended error info.