Something not really mentioned until now is how Outcome interoperates with the proposed P0323 std::expected<T, E> , whose design lands in between unchecked<T, E = varies> and checked<T, E = varies> (both of which are type aliases hard coding no-value policies as previously covered in this tutorial).

Expected and Outcome are isomorphic to one another in design intent, but interoperation for code using Expected and Outcome ought to be seamless thanks to the proposed ValueOrError concept framework, a subset of which Outcome implements.

The explicit basic_result(ValueOrError<T, E> &&) and explicit basic_outcome(ValueOrError<T, E> &&) constructors will explicitly construct from any type matching the ValueOrError<T, E> concept, which includes std::expected<A, B>, if A is constructible to X, and B is constructible to Y. The ValueOrError concept in turn is true if and only if the input type has:

1. A .has_value() observer returning a bool.
2. .value() and .error() observers.

Implementation

Outcome’s machinery for implementing ValueOrError conversion is user extensible by injection of specialisations of the value_or_error<T, U> type into the BOOST_OUTCOME_V2_NAMESPACE::convert namespace.

Outcome’s default convert::value_or_error<T, U> implementation explicitly excludes Outcome result and outcome types from the default mechanism as there is a major gotcha: the ValueOrError matched type’s .value() is often not callable in constexpr as it can throw, which makes this conversion mechanism pretty much useless for constexpr code. Besides, outcome has a converting constructor overload for result inputs which is constexpr capable.

Note that if you do enable outcome inputs, a result will match an input outcome, but silently dropping any exception state. This is probably undesirable.

Examples of how to implement your own convert::value_or_error<T, U> converter is demonstrated in the worked example, next.