This page is a snapshot from the LWG issues list, see the Library Active Issues List for more information and the meaning of New status.

4052. Bogus requirements for piecewise_linear_distribution

Section: 28.5.9.6.2 [rand.dist.samp.pconst], 28.5.9.6.3 [rand.dist.samp.plinear] Status: New Submitter: Jonathan Wakely Opened: 2024-02-05 Last modified: 2024-03-12

Priority: 4

View all other issues in [rand.dist.samp.pconst].

View all issues with New status.

Discussion:

In the second constructor of 28.5.9.6.3 [rand.dist.samp.plinear], P1719R2 replaced:

template<class InputIteratorB, class InputIteratorW>
  piecewise_linear_distribution(InputIteratorB firstB, InputIteratorB lastB,
                                InputIteratorW firstW);

Effects: [...] Moreover, the id-expressions iterator_traits<InputIteratorB>::value_type and iterator_traits<InputIteratorW>::value_type shall each denote a type that is convertible to double.

template<class InputIteratorB, class InputIteratorW>
  piecewise_linear_distribution(InputIteratorB firstB, InputIteratorB lastB,
                                InputIteratorW firstW);

Mandates: is_invocable_r_v<double, UnaryOperation&, double> is true.

Preconditions: [...]

Effects: [...]

This was a copy & paste error from the next constructor, and was obviously not intended. There is no UnaryOperation in that constructor.

A less obviously wrong issue is the use of double there in the first place. Shouldn't it be RealType instead? That seems to be incorrect throughout both 28.5.9.6.2 [rand.dist.samp.pconst] and 28.5.9.6.3 [rand.dist.samp.plinear], and was only partially fixed by LWG 1439.

Finally, the preconditions also say:

Preconditions: [...] If firstB == lastB or ++firstB == lastB, let n = 1, ρ₀ = ρ₁ = 1, b₀ = 0, and b₁ = 1. Otherwise, [firstB, lastB) forms a sequence b of length n + 1, the length of the sequence w starting from firstW is at least n, and any w_k for k ≥ n are ignored by the distribution.

These are not preconditions. I think it is trying to say something like:

Preconditions: [...] [firstB, lastB) is a valid range and [firstW, firstW + (lastB - firstB)) is a valid range.

Effects: If firstB == lastB or ++firstB == lastB, let n = 1, ρ₀ = ρ₁ = 1, b₀ = 0, and b₁ = 1. Otherwise, let [firstB, lastB) form a sequence b₀, …, b_n, and let w_k = *firstW++ for k = 0, …, n.

The equivalent constructor for piecewise_constant_distribution has similar problems with its preconditions in terms of n + 1.

[2024-03-12; Reflector poll]

Set priority to 4 after reflector poll. The copy & paste error was fixed editorially.

Proposed resolution: