An header-only alternative to boost::variant for C 11 and C 14

Variant's basic building blocks are:

• variant<Ts...> - a type-safe representation for sum-types / discriminated unions
• recursive_wrapper<T> - a helper type to represent recursive "tree-like" variants
• apply_visitor(visitor, myVariant) - to invoke a custom visitor on the variant's underlying type
• get<T>() - a function to directly unwrap a variant's underlying type
• .match([](Type){}) - a variant convenience member function taking an arbitrary number of lambdas creating a visitor behind the scenes and applying it to the variant

Suppose you want to represent a HTTP API response which is either a JSON result or an error:

struct Result {
  Json object;
};

struct Error {
  int32_t code;
  string message;
};

You can represent this at type level using a variant which is either an Error or a Result:

using Response = variant<Error, Result>;

Response makeRequest() {
  return Error{501, "Not Implemented"};
}

Response ret = makeRequest();

To see which type the Response holds you pattern match on the variant unwrapping the underlying value:

ret.match([] (Result r) { print(r.object); },
          [] (Error e)  { print(e.message); });

Instead of using the variant's convenience .match pattern matching function you can create a type visitor functor and use apply_visitor manually:

struct ResponseVisitor {
  void operator()(Result r) const {
    print(r.object);
  }

  void operator()(Error e) const {
    print(e.message);
  }
};

ResponseVisitor visitor;
apply_visitor(visitor, ret);

In both cases the compiler makes sure you handle all types the variant can represent at compile.

JSON consists of types String, Number, True, False, Null, Array and Object.

struct String { string value; };
struct Number { double value; };
struct True   { };
struct False  { };
struct Null   { };
struct Array  { vector<?> values; };
struct Object { unordered_map<string, ?> values; };

This works for primitive types but how do we represent recursive types such as Array which can hold multiple elements and Array itself, too?

For these use cases Variant provides a recursive_wrapper helper type which lets you express recursive Variants.

struct String { string value; };
struct Number { double value; };
struct True   { };
struct False  { };
struct Null   { };

// Forward declarations only
struct Array;
struct Object;

using Value = variant<String, Number, True, False, Null, recursive_wrapper<Array>, recursive_wrapper<Object>>;

struct Array {
  vector<Value> values;
};

struct Object {
  unordered_map<string, Value> values;
};

For walking the JSON representation you can again either create a JSONVisitor:

struct JSONVisitor {

  void operator()(Null) const {
    print("null");
  }

  // same for all other JSON types
};

JSONVisitor visitor;
apply_visitor(visitor, json);

Or use the convenience .match pattern matching function:

json.match([] (Null) { print("null"); },
           ...);

To summarize: use recursive_wrapper to represent recursive "tree-like" representations:

struct Empty { };
struct Node;

using Tree = variant<Empty, recursive_wrapper<Node>>;

struct Node {
  uint64_t value;
}

Creating type aliases for variants is a great way to reduce repetition. Keep in mind those type aliases are not checked at type level, though. We recommend creating a new type for all but basic variant usage:

// the compiler can't tell the following two apart
using APIResult = variant<Error, Result>;
using FilesystemResult = variant<Error, Result>;

// new type
struct APIResult : variant<Error, Result> {
  using Base = variant<Error, Result>;
  using Base::Base;
}

Mapbox variant has the same speedy performance of boost::variant but is faster to compile, results in smaller binaries, and has no dependencies.

For example on OS X 10.9 with clang and libc :

(Numbers from an older version of Mapbox variant.)

Mapbox variant has been a very valuable, lightweight alternative for apps that can use c 11 or c 14 but that do not want a boost dependency. Mapbox variant has also been useful in apps that do depend on boost, like mapnik, to help (slightly) with compile times and to majorly lessen dependence on boost in core headers. The original goal and near term goal is to maintain external API compatibility with boost::variant such that Mapbox variant can be a "drop in". At the same time the goal is to stay minimal: Only implement the features that are actually needed in existing software. So being an "incomplete" implementation is just fine.

Currently Mapbox variant doesn't try to be API compatible with the upcoming variant standard, because the standard is not finished and it would be too much work. But we'll revisit this decision in the future if needed.

If Mapbox variant is not for you, have a look at these other implementations.

Want to know more about the upcoming standard? Have a look at our overview.

Most modern high-level languages provide ways to express sum types directly. If you're curious have a look at Haskell's pattern matching or Rust's and Swift's enums.

• Compiler supporting -std=c 11 or -std=c 14

Tested with:

• g -4.7
• g -4.8
• g -4.9
• g -5.2
• clang -3.5
• clang -3.6
• clang -3.7
• clang -3.8
• clang -3.9
• Visual Studio 2015

On Unix systems compile and run the unit tests with make test.

On Windows run scripts/build-local.bat.

• The variant can not hold references (something like variant<int&> is not possible). You might want to try std::reference_wrapper instead.

• The included implementation of optional is deprecated and will be removed in a future version. See issue #64.
• Old versions of the code needed visitors to derive from static_visitor. This is not needed any more and marked as deprecated. The static_visitor class will be removed in future versions.

make bench

make sizes /path/to/boost/variant.hpp