boost/hana/fwd/eval_if.hpp
/*! @file Forward declares `boost::hana::eval_if`. Copyright Louis Dionne 2013-2022 Distributed under the Boost Software License, Version 1.0. (See accompanying file LICENSE.md or copy at http://boost.org/LICENSE_1_0.txt) */ #ifndef BOOST_HANA_FWD_EVAL_IF_HPP #define BOOST_HANA_FWD_EVAL_IF_HPP #include <boost/hana/config.hpp> #include <boost/hana/core/when.hpp> namespace boost { namespace hana { //! Conditionally execute one of two branches based on a condition. //! @ingroup group-Logical //! //! Given a condition and two branches in the form of lambdas or //! `hana::lazy`s, `eval_if` will evaluate the branch selected by the //! condition with `eval` and return the result. The exact requirements //! for what the branches may be are the same requirements as those for //! the `eval` function. //! //! //! Deferring compile-time evaluation inside `eval_if` //! -------------------------------------------------- //! By passing a unary callable to `eval_if`, it is possible to defer //! the compile-time evaluation of selected expressions inside the //! lambda. This is useful when instantiating a branch would trigger //! a compile-time error; we only want the branch to be instantiated //! when that branch is selected. Here's how it can be achieved. //! //! For simplicity, we'll use a unary lambda as our unary callable. //! Our lambda must accept a parameter (usually called `_`), which //! can be used to defer the compile-time evaluation of expressions //! as required. For example, //! @code //! template <typename N> //! auto fact(N n) { //! return hana::eval_if(n == hana::int_c<0>, //! [] { return hana::int_c<1>; }, //! [=](auto _) { return n * fact(_(n) - hana::int_c<1>); } //! ); //! } //! @endcode //! //! What happens here is that `eval_if` will call `eval` on the selected //! branch. In turn, `eval` will call the selected branch either with //! nothing -- for the _then_ branch -- or with `hana::id` -- for the //! _else_ branch. Hence, `_(x)` is always the same as `x`, but the //! compiler can't tell until the lambda has been called! Hence, the //! compiler has to wait before it instantiates the body of the lambda //! and no infinite recursion happens. However, this trick to delay the //! instantiation of the lambda's body can only be used when the condition //! is known at compile-time, because otherwise both branches have to be //! instantiated inside the `eval_if` anyway. //! //! There are several caveats to note with this approach to lazy branching. //! First, because we're using lambdas, it means that the function's //! result can't be used in a constant expression. This is a limitation //! of the current language. //! //! The second caveat is that compilers currently have several bugs //! regarding deeply nested lambdas with captures. So you always risk //! crashing the compiler, but this is a question of time before it is //! not a problem anymore. //! //! Finally, it means that conditionals can't be written directly inside //! unevaluated contexts. The reason is that a lambda can't appear in an //! unevaluated context, for example in `decltype`. One way to workaround //! this is to completely lift your type computations into variable //! templates instead. For example, instead of writing //! @code //! template <typename T> //! struct pointerize : decltype( //! hana::eval_if(hana::traits::is_pointer(hana::type_c<T>), //! [] { return hana::type_c<T>; }, //! [](auto _) { return _(hana::traits::add_pointer)(hana::type_c<T>); } //! )) //! { }; //! @endcode //! //! you could instead write //! //! @code //! template <typename T> //! auto pointerize_impl(T t) { //! return hana::eval_if(hana::traits::is_pointer(t), //! [] { return hana::type_c<T>; }, //! [](auto _) { return _(hana::traits::add_pointer)(hana::type_c<T>); } //! ); //! } //! //! template <typename T> //! using pointerize = decltype(pointerize_impl(hana::type_c<T>)); //! @endcode //! //! > __Note__: This example would actually be implemented more easily //! > with partial specializations, but my bag of good examples is empty //! > at the time of writing this. //! //! Now, this hoop-jumping only has to be done in one place, because //! you should use normal function notation everywhere else in your //! metaprogram to perform type computations. So the syntactic //! cost is amortized over the whole program. //! //! Another way to work around this limitation of the language would be //! to use `hana::lazy` for the branches. However, this is only suitable //! when the branches are not too complicated. With `hana::lazy`, you //! could write the previous example as //! @code //! template <typename T> //! struct pointerize : decltype( //! hana::eval_if(hana::traits::is_pointer(hana::type_c<T>), //! hana::make_lazy(hana::type_c<T>), //! hana::make_lazy(hana::traits::add_pointer)(hana::type_c<T>) //! )) //! { }; //! @endcode //! //! //! @param cond //! The condition determining which of the two branches is selected. //! //! @param then //! An expression called as `eval(then)` if `cond` is true-valued. //! //! @param else_ //! A function called as `eval(else_)` if `cond` is false-valued. //! //! //! Example //! ------- //! @include example/eval_if.cpp #ifdef BOOST_HANA_DOXYGEN_INVOKED constexpr auto eval_if = [](auto&& cond, auto&& then, auto&& else_) -> decltype(auto) { return tag-dispatched; }; #else template <typename L, typename = void> struct eval_if_impl : eval_if_impl<L, when<true>> { }; struct eval_if_t { template <typename Cond, typename Then, typename Else> constexpr decltype(auto) operator()(Cond&& cond, Then&& then, Else&& else_) const; }; BOOST_HANA_INLINE_VARIABLE constexpr eval_if_t eval_if{}; #endif }} // end namespace boost::hana #endif // !BOOST_HANA_FWD_EVAL_IF_HPP