See Custom Read/Write in the main README.md for using glz::custom.

Advanced custom serialization/deserialization is achievable by implementing your own from and to specializations within the glz namespace.

Example:

struct date
{
   uint64_t data;
   std::string human_readable;
};

template <>
struct glz::meta<date>
{
   using T = date;
   static constexpr auto value = object("date", &T::human_readable);
};

namespace glz
{
   template <>
   struct from<JSON, date>
   {
      template <auto Opts>
      static void op(date& value, is_context auto&& ctx, auto&& it, auto&& end)
      {
         parse<JSON>::op<Opts>(value.human_readable, ctx, it, end);
         value.data = std::stoi(value.human_readable);
      }
   };

   template <>
   struct to<JSON, date>
   {
      template <auto Opts>
      static void op(date& value, is_context auto&& ctx, auto&& b, auto&& ix) noexcept
      {
         value.human_readable = std::to_string(value.data);
         serialize<JSON>::op<Opts>(value.human_readable, ctx, b, ix);
      }
   };
}

void example() {
  date d{};
  d.data = 55;

  std::string s{};
  expect(not glz::write_json(d, s));

  expect(s == R"("55")");

  d.data = 0;
  expect(not glz::read_json(d, s));
  expect(d.data == 55);
}

Notes:

The templated Opts parameter contains the compile time options.

For reading (from specializations), the parameters are:

• value: The object being parsed into
• ctx: The context containing runtime options (use is_context auto&&)
• it: Iterator to the current position in the input buffer
• end: Iterator to the end of the input buffer

For writing (to specializations), the parameters are:

• value: The object being serialized
• ctx: The context containing runtime options (use is_context auto&&)
• b: The output buffer to write to
• ix: The current index in the output buffer

Bitfields

C++ bitfields cannot be referenced with a pointer-to-member, so they need glz::custom adapters in the glz::meta definition. The adapter lets you expose the bitfield as a regular integer while preserving the in-class bit packing.

struct bitfield_struct_t {
   uint8_t f1 : 4{};
   uint8_t f2 : 4{};
   uint8_t f3{};
};

template <>
struct glz::meta<bitfield_struct_t>
{
   using T = bitfield_struct_t;

   static constexpr auto read_f1  = [](T& self, uint8_t v) { self.f1 = v; };
   static constexpr auto write_f1 = [](const T& self) { return static_cast<uint8_t>(self.f1); };
   static constexpr auto read_f2  = [](T& self, uint8_t v) { self.f2 = v; };
   static constexpr auto write_f2 = [](const T& self) { return static_cast<uint8_t>(self.f2); };

   static constexpr auto value = object(
      "f1", glz::custom<read_f1, write_f1>,
      "f2", glz::custom<read_f2, write_f2>,
      "f3", &T::f3
   );
};

• Ordinary members such as f3 can continue to use direct pointers-to-members alongside the custom fields.

UUID Example

Say we have a UUID library for converting a uuid_t from a std::string_view and to a std::string.

namespace glz
{
   template <>
   struct from<JSON, uuid_t>
   {
      template <auto Opts>
      static void op(uuid_t& uuid, is_context auto&& ctx, auto&& it, auto&& end)
      {
         // Initialize a string_view with the appropriately lengthed buffer
         // Alternatively, use a std::string for any size (but this will allocate)
         std::string_view str = "XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX";
         parse<JSON>::op<Opts>(str, ctx, it, end);
         uuid = uuid_lib::uuid_from_string_view(str);
      }
   };

   template <>
   struct to<JSON, uuid_t>
   {
      template <auto Opts>
      static void op(const uuid_t& uuid, is_context auto&& ctx, auto&& b, auto&& ix) noexcept
      {
         std::string str = uuid_lib::uuid_to_string(uuid);
         serialize<JSON>::op<Opts>(str, ctx, b, ix);
      }
   };
}

Handling Ambiguous Partial Specialization

You may want to custom parse a class that matches an underlying glaze partial specialization for a template like below:

template <class T> requires readable_array_t<T>
struct from<JSON, T>

If your own parsing function desires partial template specialization, then ambiguity may occur:

template <class T> requires std::derived_from<T, vec_t>
struct from<JSON, T>

To solve this problem, glaze will check for custom_read or custom_write values within glz::meta and remove the ambiguity and use the custom parser.

template <class T> requires std::derived_from<T, vec_t>
struct glz::meta<T>
{
   static constexpr auto custom_read = true;
};

Mimicking Standard Types

When a custom type serializes as a standard type (like std::string), you can declare this relationship using mimic. This is useful for:

• Using custom types as map keys without double-quoting
• Using custom types in variants with proper type deduction
• Generating correct JSON schemas

Using mimic

Add using mimic = <type>; to indicate that your type serializes as a standard type:

struct my_key {
   std::string value{};
   auto operator<=>(const my_key&) const = default;
};

template <>
struct glz::meta<my_key> {
   using mimic = std::string;  // Declares string-like behavior
   static constexpr auto value = &my_key::value;
};

// Works correctly as map keys (no double-quoting)
std::map<my_key, int> m{{{"hello"}, 42}};
// Produces: {"hello":42}

// Works in variants
std::variant<int, my_key> v;
glz::read_json(v, "\"hello\"");  // → my_key{"hello"}

Available Concepts

• glz::has_mimic<T> – checks if T has a mimic type defined
• glz::mimic_type<T> – extracts the mimic type from T
• glz::mimics<T, Target> – checks if T mimics exactly Target
• glz::mimics_str_t<T> – checks if T mimics any string type
• glz::mimics_num_t<T> – checks if T mimics any numeric type
• glz::mimics_bool_t<T> – checks if T mimics a boolean type

Auto-Inference for Custom Types

When using glz::custom read/write functions, Glaze automatically infers the JSON type from your read function's parameter:

Read parameter type	Concept satisfied
double, int, etc.	custom_num_t<T>
std::string, std::string_view	custom_str_t<T>
bool	custom_bool_t<T>

struct Amount {
   double value{};
};

template <>
struct glz::meta<Amount> {
   // Glaze infers this is numeric from 'const double&'
   static constexpr auto read_fn = [](Amount& a, const double& input) {
      a.value = input;
   };
   static constexpr auto write_fn = [](const Amount& a) -> const double& {
      return a.value;
   };
   static constexpr auto value = glz::custom<read_fn, write_fn>;
};

// Works automatically in variants - no mimic needed!
std::variant<std::string, Amount> v;
glz::read_json(v, "42.5");  // → Amount{42.5}

See Variant Handling - Custom Types in Variants for more details.