JSON Handling in Glaze¶
Glaze is one of the fastest JSON libraries in the world, offering automatic reflection and direct memory serialization/deserialization with incredible performance. Glaze makes JSON handling in C++ simple, fast, and type-safe. Start with basic automatic reflection and gradually adopt advanced features as your needs grow!
Basic Example¶
With Glaze, your structs automatically get JSON serialization without any metadata:
#include "glaze/glaze.hpp"
struct Person {
std::string name = "John Doe";
int age = 30;
std::vector<std::string> hobbies = {"reading", "gaming"};
};
int main() {
Person person{};
// Write to JSON
std::string buffer{};
auto ec = glz::write_json(person, buffer);
if (!ec) {
// buffer now contains: {"name":"John Doe","age":30,"hobbies":["reading","gaming"]}
}
// Read from JSON
std::string json_data = R"({"name":"Jane","age":25,"hobbies":["hiking","cooking"]})";
Person jane{};
ec = glz::read_json(jane, json_data);
if (!ec) {
// jane.name is now "Jane", jane.age is 25, etc.
}
return 0;
}
Alternative API Styles¶
Glaze offers multiple ways to read and write JSON:
Writing JSON:
Person person{};
// Primary method: Write into existing buffer
std::string buffer{};
auto ec = glz::write_json(person, buffer);
if (!ec) {
// buffer contains the JSON string
}
// Alternative: Function returns std::expected<std::string, glz::error_ctx>
std::string json = glz::write_json(person).value_or("error");
Reading JSON:
std::string json_data = R"({"name":"Alice","age":28})";
// Primary method: Read into existing object
Person person{};
auto ec = glz::read_json(person, json_data);
if (!ec) {
// person is now populated from json_data
}
// Alternative: Function returns std::expected<T, glz::error_ctx>
auto result = glz::read_json<Person>(json_data);
if (result) {
const Person& person = result.value();
}
File I/O¶
Person person{};
// Read from file
auto ec = glz::read_file_json(person, "./person.json", std::string{});
if (!ec) {
// person is now populated from the file
}
// Write to file
auto ec = glz::write_file_json(person, "./person.json", std::string{});
if (!ec) {
// person data has been written to the file
}
Automatic Reflection¶
Glaze uses compile-time reflection to automatically serialize your structs. No macros, no manual registration - it just works!
Supported out of the box: - All aggregate-initializable structs - Standard library containers - Primitive types - Nested objects
struct Address {
std::string street;
std::string city;
int zip_code;
};
struct Employee {
std::string name;
int employee_id;
Address address;
std::optional<std::string> department;
std::map<std::string, int> skills_rating;
};
// This struct automatically works with JSON - no additional code needed!
Manual Metadata (When Needed)¶
For non-aggregate structs, private members, or custom behavior, you can provide explicit metadata:
struct CustomStruct {
private:
int value;
std::string name;
friend struct glz::meta<CustomStruct>; // friend only needed to access private members
public:
CustomStruct(int v, const std::string& n) : value(v), name(n) {}
};
template <>
struct glz::meta<CustomStruct> {
using T = CustomStruct;
static constexpr auto value = glz::object(
&T::value, // will use the key "value" automatically
&T::name // will use the key "name" automatically
);
};
Custom Field Names¶
You can override the default field names:
template <>
struct glz::meta<Person> {
using T = Person;
static constexpr auto value = glz::object(
"full_name", &T::name,
"years_old", &T::age,
"interests", &T::hobbies
);
};
// JSON output: {"full_name":"John","years_old":30,"interests":["reading"]}
Error Handling¶
Glaze provides comprehensive error handling with detailed error messages:
std::string invalid_json = R"({"name":"John", "age":})"; // Missing value
Person person{};
auto ec = glz::read_json(person, invalid_json);
if (ec) {
// Get a detailed error message
std::string error_msg = glz::format_error(ec, invalid_json);
std::cout << error_msg << '\n';
}
Input Buffer Requirements¶
For optimal performance, use null-terminated buffers (like std::string
):
// Recommended - null terminated
std::string json_data = "...";
Person person{};
auto ec = glz::read_json(person, json_data);
// If you must use non-null terminated buffers
constexpr auto options = glz::opts{.null_terminated = false};
auto ec = glz::read<options>(person, buffer);
Type Support¶
Containers and Arrays¶
All standard C++ containers are supported, and if your containers match C++ standard APIs they will work as well:
std::vector<int> numbers = {1, 2, 3, 4, 5};
std::array<std::string, 3> colors = {"red", "green", "blue"};
std::set<int> unique_numbers = {1, 2, 3};
std::deque<double> values = {1.1, 2.2, 3.3};
// All serialize to JSON arrays: [1,2,3,4,5], ["red","green","blue"], etc.
Maps and Objects¶
std::map<std::string, int> scores = {{"Alice", 95}, {"Bob", 87}};
std::unordered_map<std::string, std::string> config = {{"host", "localhost"}};
// Serialize to JSON objects: {"Alice":95,"Bob":87}
Optional and Nullable Types¶
struct OptionalData {
std::optional<std::string> nickname;
std::unique_ptr<int> priority;
std::shared_ptr<std::vector<int>> tags;
};
// null values are written as JSON null, valid values are serialized normally
Enums¶
Enums can be serialized as integers (default) or strings:
enum class Status { Active, Inactive, Pending };
// Default: serialize as integers (0, 1, 2)
Status status = Status::Active;
// For string serialization, add metadata:
template <>
struct glz::meta<Status> {
using enum Status;
static constexpr auto value = glz::enumerate(
Active,
Inactive,
Pending
);
};
// Now serializes as: "Active", "Inactive", "Pending"
Variants¶
std::variant<int, std::string, double> data = "hello";
// Glaze automatically handles variant serialization
Advanced Features¶
JSON with Comments (JSONC)¶
Read JSON files with comments:
std::string jsonc_data = R"({
"name": "John", // Person's name
/* Multi-line comment
about age */
"age": 30
})";
Person person{};
auto ec = glz::read_jsonc(person, jsonc_data);
// Or: glz::read<glz::opts{.comments = true}>(person, jsonc_data);
Pretty Printing¶
Person person{};
// Write prettified JSON directly
std::string buffer{};
auto ec = glz::write<glz::opts{.prettify = true}>(person, buffer);
// Or prettify existing JSON
std::string minified = R"({"name":"John","age":30})";
std::string pretty = glz::prettify_json(minified);
Output:
Minification¶
// Write minified JSON (default behavior)
Person person{};
std::string buffer{};
auto ec = glz::write_json(person, buffer); // Already minified by default
// Minify existing JSON text
std::string pretty_json = "..."; // JSON with whitespace
std::string minified = glz::minify_json(pretty_json);
// For reading minified JSON (performance optimization)
auto ec = glz::read<glz::opts{.minified = true}>(person, minified_buffer);
Custom Serialization¶
For complex custom behavior, use glz::custom
:
struct TimestampObject {
std::chrono::system_clock::time_point timestamp;
void read_timestamp(const std::string& iso_string) {
// Parse ISO 8601 string to time_point
}
std::string write_timestamp() const {
// Convert time_point to ISO 8601 string
}
};
template <>
struct glz::meta<TimestampObject> {
using T = TimestampObject;
static constexpr auto value = glz::object(
"timestamp", glz::custom<&T::read_timestamp, &T::write_timestamp>
);
};
Boolean Flags¶
Represent multiple boolean flags as a JSON array:
struct Permissions {
bool read = true;
bool write = false;
bool execute = true;
};
template <>
struct glz::meta<Permissions> {
using T = Permissions;
static constexpr auto value = glz::flags(
"read", &T::read,
"write", &T::write,
"execute", &T::execute
);
};
// JSON output: ["read", "execute"]
Dynamic JSON Objects with glz::obj¶
glz::obj
is a compile-time, stack-allocated JSON object builder that allows you to create JSON objects on the fly without defining structs. It's perfect for logging, custom serialization, temporary data structures, and rapid prototyping.
Basic Usage¶
// Create a JSON object with key-value pairs
auto obj = glz::obj{"name", "Alice", "age", 30, "active", true};
std::string buffer{};
auto ec = glz::write_json(obj, buffer);
// Results in: {"name":"Alice","age":30,"active":true}
Key Features¶
- Stack-allocated: No dynamic memory allocation for better performance
- Type-safe: Compile-time type checking
- Heterogeneous types: Mix different value types in the same object
- Reference semantics: Can store references to existing variables
- Nested support: Can contain other
glz::obj
orglz::arr
instances
Advanced Examples¶
// Nested objects and arrays
auto nested = glz::obj{
"user", glz::obj{"id", 123, "name", "Bob"},
"scores", glz::arr{95, 87, 92},
"metadata", glz::obj{"version", "1.0", "timestamp", 1234567890}
};
// Storing references to existing variables
std::string name = "Charlie";
int score = 100;
auto ref_obj = glz::obj{"player", name, "score", score};
// Changes to 'name' or 'score' will be reflected when serializing ref_obj
// Mixed types in one object
std::vector<int> vec = {1, 2, 3};
std::map<std::string, double> map = {{"pi", 3.14}, {"e", 2.71}};
auto mixed = glz::obj{
"numbers", vec,
"constants", map,
"flag", true,
"message", "Hello"
};
glz::obj_copy¶
When you need value semantics instead of reference semantics, use glz::obj_copy
:
int x = 5;
auto obj_ref = glz::obj{"x", x}; // Stores reference to x
auto obj_val = glz::obj_copy{"x", x}; // Copies the value of x
x = 10;
// obj_ref will serialize as {"x":10}
// obj_val will serialize as {"x":5}
Dynamic JSON Arrays with glz::arr¶
Similar to glz::obj
, glz::arr
creates stack-allocated JSON arrays:
// Basic array
auto arr = glz::arr{1, 2, 3, 4, 5};
// Mixed types
auto mixed_arr = glz::arr{"Hello", 42, true, 3.14};
// Nested structures
auto nested_arr = glz::arr{
glz::obj{"id", 1, "name", "Item1"},
glz::obj{"id", 2, "name", "Item2"},
glz::arr{1, 2, 3}
};
std::string buffer{};
auto ec = glz::write_json(nested_arr, buffer);
// Results in: [{"id":1,"name":"Item1"},{"id":2,"name":"Item2"},[1,2,3]]
glz::arr_copy¶
Just like glz::obj_copy
, use glz::arr_copy
for value semantics:
std::string s = "original";
auto arr_ref = glz::arr{s}; // Reference to s
auto arr_val = glz::arr_copy{s}; // Copy of s
Object Merging¶
glz::merge
allows you to combine multiple JSON objects into a single object. This is useful for composing objects from different sources:
// Merge glz::obj instances
glz::obj metadata{"version", "1.0", "author", "John"};
glz::obj settings{"debug", true, "timeout", 5000};
auto merged1 = glz::merge{metadata, settings};
// Results in: {"version":"1.0","author":"John","debug":true,"timeout":5000}
// Merge different object types
std::map<std::string, int> scores = {{"test1", 95}, {"test2", 87}};
auto user_data = glz::obj{"user_id", 42, "username", "alice"};
auto merged2 = glz::merge{user_data, scores};
// Results in: {"user_id":42,"username":"alice","test1":95,"test2":87}
// Merge with raw_json
glz::raw_json raw{R"({"extra":"data"})"};
auto merged3 = glz::merge{metadata, raw};
// Results in: {"version":"1.0","author":"John","extra":"data"}
Key Collision Behavior¶
When merging objects with duplicate keys, the last value wins:
auto obj1 = glz::obj{"key", "first"};
auto obj2 = glz::obj{"key", "second"};
auto merged = glz::merge{obj1, obj2};
// Results in: {"key":"second"}
Performance Considerations¶
- Stack allocation: Both
glz::obj
andglz::arr
are stack-allocated, avoiding heap allocations - Compile-time optimization: The structure is known at compile time, enabling optimizations
- Zero-copy references: When storing references, no data copying occurs
- Efficient serialization: Direct memory access without intermediate representations
Use Cases¶
-
Logging and Debugging:
-
API Responses:
-
Configuration Objects:
Performance Features¶
Direct Memory Access¶
Glaze reads and writes directly to/from object memory, avoiding intermediate representations:
// No DOM tree, no temporary objects - direct serialization
LargeObject large_object{};
std::string buffer{};
auto ec = glz::write_json(large_object, buffer);
Compile-Time Optimizations¶
- Perfect hash tables for object keys
- Compile-time reflection eliminates runtime overhead
- SIMD and SWAR optimizations where possible
JSON Conformance¶
Glaze is fully RFC 8259 compliant with UTF-8 validation. By default, it uses two optimizations:
validate_skipped = false
: Faster parsing by not fully validating skipped values (does not affect resulting C++ objects)validate_trailing_whitespace = false
: Stops parsing after the target object
For strict compliance, enable these options: