Skip to content

Latest commit

 

History

History
178 lines (134 loc) · 6.02 KB

is_clamped.md

File metadata and controls

178 lines (134 loc) · 6.02 KB

Doc. no: P1440R0
Date: 2019-01-19
Reply-to: Johel Guerrero [email protected]
Audience: Library Incubator
Source: https://github.com/johelegp/proposals/blob/master/is_clamped.md

is_clamped

Abstract

Introduce std::is_clamped(v, lo, hi) to mean lo <= v && v <= hi, except that v is only evaluated once.

Now Proposed alternative
Axis::x <= Coord::axis && Coord::axis <= Axis::z std::is_clamped(Coord::axis, Axis::x, Axis::z)
Now Proposed alternative + P0330 [4]
1u <= digits.size() && digits.size() <= 7u std::is_clamped(digits.size(), 1uz, 7uz)

Revision history

  • cv void -> R0
    • 2019y/January/19, pre-Kona.

Problem

Expressions equivalent to lo <= v && v <= hi are common. They are used to check whether v is within the range [lo, hi]. However, because v appears twice, they

  • are error prone to type and update,
  • reduce readability as v gets longer, and
  • may degrade performance due to the repeated evaluation of v.

We propose std::is_clamped(v, lo, hi) as an alternative spelling that better captures intent and is potentially more performant.

Alternatives

Chaining Comparisons (rejected)

P0893 [1] would have allowed us to write lo <= v <= hi. Its rejection made us write this proposal.

There was no consensus on making comparison operators chain, nor was there encouragement for further work in that area.
-- P1018R2 [2]

Initialize intermediate variable with v

It may be undesirable for some of the reasons selection and iteration statements have been getting space for initialization.

In some contexts it is impossible, such as member intializer lists and requires clauses.

Immediately invoked lambda expression

This works, at the cost of much syntactical noise on par with repeating v:

  • [&] { const auto& w{v}; return what we actually care about; }(), or
  • [](const auto& v) { return what we actually care about; }(v).

Safe integral comparisons' std::in_range

Vicente J. Botet Escriba made us aware [6] of std::in_range proposed in P0586 [7]. Its name suggests similarity to our proposed std::is_clamped. However, it checks whether an integral value is in the range of an integral type with safe integral comparisons proposed in the same paper. Neither can subsume the other.

Bikeshedding

The following have been suggested of the proposed std::is_clamped design:

  • Different names.
  • Allowing arguments of different types.
  • Marking a bound as closed or half-open.

However, we believe the proposed design was fixed once ISO/IEC 14882:2017 (C++17) was published with std::clamp. After all, we mean for std::is_clamped to complement std::clamp in the same way std::is_sorted complements std::sort.

Wording

Add the following declarations to [algorithm.syn] after those of clamp.

template<class T>
  constexpr bool is_clamped(const T& v, const T& lo, const T& hi);
template<class T, class Compare>
  constexpr bool is_clamped(const T& v, const T& lo, const T& hi, Compare comp);

namespace ranges {
  template<class T, class Proj = identity,
           IndirectStrictWeakOrder<projected<const T*, Proj>> Comp = ranges::less<>>
    constexpr bool is_clamped(const T& v, const T& lo, const T& hi, Comp comp = {}, Proj proj = {});
}

Add the following detailed specifications to [alg.clamp] after those of clamp.

template<class T>
  constexpr bool is_clamped(const T& v, const T& lo, const T& hi);
template<class T, class Compare>
  constexpr bool is_clamped(const T& v, const T& lo, const T& hi, Compare comp);

namespace ranges {
  template<class T, class Proj = identity,
           IndirectStrictWeakOrder<projected<const T*, Proj>> Comp = ranges::less<>>
    constexpr bool is_clamped(const T& v, const T& lo, const T& hi, Comp comp = {}, Proj proj = {});
}

Expects: The value of lo is no greater than hi. For the first form, type T shall be Cpp17LessThanComparable (Table 24).

Returns: true if

  • v is not less than lo, and
  • hi is not less than v.
    Otherwise, false.

[ Note: If NaN is avoided, T can be a floating-point type. -- end note ]

Complexity: At most two comparisons and three applications of the projection, if any.

Possible implementation

template<class T>
  constexpr bool is_clamped(const T& v, const T& lo, const T& hi)
{
  return !(v < lo) && !(hi < v);
}

template<class T, class Compare>
  constexpr bool is_clamped(const T& v, const T& lo, const T& hi, Compare comp)
{
  return !comp(v, lo) && !comp(hi, v);
}

namespace ranges {
  template<class T, class Proj = identity,
           IndirectStrictWeakOrder<projected<const T*, Proj>> Comp = ranges::less<>>
    constexpr bool is_clamped(const T& v, const T& lo, const T& hi, Comp comp = {}, Proj proj = {})
  {
    auto&& pv{INVOKE(proj, v)};
    return !INVOKE(comp, pv, INVOKE(proj, lo)) &&
           !INVOKE(comp, INVOKE(proj, hi), pv);
  }
}

Related proposals

Paper Title Relation
P1243 [3] Rangify New Algorithms Introduces std::ranges::clamp.

Acknowledgements

Thanks to Ray Hamel, Vicente J. Botet Escriba, and everyone else for their comments in the std-proposals thread [5].

References

[1] Chaining Comparisons, https://wg21.link/P0893
[2] Evolution status after San Diego 2018, https://wg21.link/P1018r2
[3] Rangify New Algorithms, https://wg21.link/P1243
[4] Literal Suffixes for ptrdiff_t and size_t, https://wg21.link/P0330
[5] is_clamped, https://groups.google.com/a/isocpp.org/forum/?fromgroups#!topic/std-proposals/LDhzb-58sV4
[6] Personal communications
[7] Safe integral comparisons, https://wg21.link/P0586