Doc. no.: P0608R3
Date: 2018-10-03
Audience: LWG
Reply-to: Zhihao Yuan <zy at miator dot net>

A sane variant converting constructor

Changes since R1

Problems to solve

  1. variant constructs entirely unintended alternatives.
variant<string, bool> x = "abc";             // holds bool

The above holds string with the proposed fix.

  1. variant prefers constructions with information losses.
variant<char, optional<char16_t>> x = u'\u2043';  // holds char = 'C'
double d = 3.14;
variant<int, reference_wrapper<double>> y = d;    // holds int = 3

The above preserves the input value in optional<char16_t> and reference_wrapper<double>, respectively, with the proposed fix.

  1. variant performs unstable constructions.
using T = variant<float, int>;
T v;
v = 0;    // switches to int
using T = variant<float, long>;
T v;
v = 0;    // error
using T = variant<float, big_int<256>>;
T v;
v = 0;    // holds 0.f

In both cases, the proposed fix consistently constructs with the second alternative.

As shown, the problems equally apply to the converting constructor and the converting assignment operator.

See also LEWG 227.

Proposed resolution

This paper proposes to constrain the variant converting constructor and the converting assignment operator to prevent narrowing conversions and conversions to bool. This section explains what exactly this change brings.

Lemma 1. Let X be a sum type T+U. Let A={T1,T2,T3,...} be a set of types that are convertible to T, and B={U1,U2,U3,...} be a set of types that are convertible to U. Let Y be a set of types that are convertible to X. Y=AB (symmetric difference) rather than AB, because each Ti=Uj causes an ambiguity.

Theorem 1. If A and B are extended to include a type τAB, Y is shrunk.

In short, constraining variant<Ts...> (with Ts¯>1) converting constructor may enable more types to be convertible to a variant.

Definition 1. For type T that is convertible to T, let P be a set of all the possible values for T, and Q be a set of all the possible values for T. If PQ, T is denoted as T+. Otherwise, PQ and T is denoted as T. The conversion from T to T (denoted as TT) is a potentially unrepresentable conversion.

In this paper, narrowing conversions (considering only the types) and boolean conversions assemble the potentially unrepresentable conversions in C++.

Lemma 2. Potentially unrepresentable conversions in C++ have Conversion rank.

The proof is left as an exercise for the reader.

Let X be variant<Ts...>, r be a value of T.

When Ts¯=1, without loss of generality, X is variant<T>. Effects of the proposed resolution can be summarized as follows:

X v = r; before after
variant<float> v = 0; holds .0f ill-formed
variant<float> v = INT_MAX; holds INT_MAX + 1 ill-formed

However, variant<V> is such a rare variant, as you can hardly say that V+ is a sum type.

When Ts¯=2, X is variant<T,U>.

The effects are summarized in the order of the bullets:

X v = r; before after
variant<float, vector<int>> v = 0; holds float ill-formed
variant<float, int> v = 'a'; holds int('a') holds int('a')
variant<float, char> v = 0; ill-formed ill-formed
variant<float, long> v = 0; ill-formed holds long
variant<float, big_int<256>> v = 0; holds float holds big_int

When Ts¯>2, let Ts1=T, S be an overload set {f(τ)τTs}, S=S{f(T)}.

The effects are summarized in the order of the bullets:

X v = r; before after
variant<float, big_int<256>, big_int<128>> v = 0; holds float ill-formed
variant<float, long, double> v = 0; ill-formed holds long
variant<float, vector<int>, big_int<256>> v = 0; holds float holds big_int
variant<float, int, big_int<256>> v = 'a'; holds int holds int

Theorem 2. For variant<Ts...> v = r, where r is a value of R, when there exists one and only one τTs rendering R to be potentially unrepresentable converted to τ, the proposed resolution may cause breaking changes

The first case is easy to fix while the second gives the desired outcome for this paper. Both behaviors to be changed are bugs, not features, as shown in Section 1.

However, the definition of potentially unrepresentable conversions here makes it unable to be fully detected in the library, because while we understand type conversions as mappings, C++ type conversions happen in conversion sequences. A boolean conversion in a standard conversion sequence that follows a user-defined conversion sequence cannot be detected. In short, we cannot distinguish

struct Bad { operator char const*(); };

from

struct LessBad { operator bool(); };

The proposed resolution considers both to be bad. Throughout the library, only two types are affected: true_type and false_type. The most commonly seen cases, types that are contextually converted to bool, are not affected by this compromise, because std::variant is not contextual. The compromise also does not break our proof, because user-defined conversions’ rank is lower than that of narrowing conversions.

Alternative designs

The author came up with and experimented a few other designs, here we list two basic ideas.

  1. Use the alternatives’ order information in determining which one to construct. The idea defeats the purpose of the converting constructor because if the construction is sensitive to the order of the alternatives declared in the variant template argument list, in_place_index would be a better choice. The converting constructor and assignment operator assume unordered alternatives.

  2. Distinguish implicit and explicit conversions. First, the idea doesn’t work well with the converting assignment operator; applying the implicit policy seems to be the only choice to maintain a consistent behavior, but this may be overkill. Second, it is counterintuitive to have an explicit constructor accepting fewer types comparing to an implicit one because of Theorem 1.

An additional suggestion is to allow narrowing but unambiguous conversions. In the following example, if double_double is a user-defined type that does not convert from an integer type,

variant<char, double_double> v = 42;

, it might be not so problematic to construct the char alternative. However, special casing this makes the construction less stable. If double_double is upgraded to support converting from int, the code silently constructs the double_double alternative.

Wording

This wording is relative to N4762.

Modify 19.7.3.1 [variant.ctor]/12 as indicated:

template<class T> constexpr variant(T&& t) noexcept(see below );

Let Tj be a type that is determined as follows: build an imaginary function FUN(Ti) for each alternative type Ti for which Ti x[] = {std::forward<T>(t)}; is well-formed for some invented variable x and, if Ti is cv bool, remove_cvref_t<T> is bool. The overload FUN(Ti) selected by overload resolution for the expression FUN(std::forward<T>(t)) defines the alternative Tj which is the type of the contained value after construction.

[…]

[Drafting note: The above uses Ti x[] = {std::forward<T>(t)}; rather than Ti{std::forward<T>(t)} to test whether the conversion sequence from T to Ti contains a narrowing conversion because whether Ti{std::forward<T>(t)} is well-formed may subject to whether Ti has an initializer_list constructor or whether Ti is an aggregate. Here is an example: https://godbolt.org/z/Ck5w-L –end note]

Modify 19.7.3.3 [variant.assign]/10 as indicated:

template<class T> variant& operator=(T&& t) noexcept(see below );

Let Tj be a type that is determined as follows: build an imaginary function FUN(Ti) for each alternative type Ti for which Ti x[] = {std::forward<T>(t)}; is well-formed for some invented variable x and, if Ti is cv bool, remove_cvref_t<T> is bool. The overload FUN(Ti) selected by overload resolution for the expression FUN(std::forward<T>(t)) defines the alternative Tj which is the type of the contained value after assignment.

[…]