std::constexpr_v
Document #: | P2781R3 |
Date: | 2023-06-10 |
Project: | Programming Language C++ |
Audience: |
LEWG-I LEWG |
Reply-to: |
Matthias Kretz <m.kretz@gsi.de> Zach Laine <whatwasthataddress@gmail.com> |
std::integral_constant<int>
are
bad
constexpr_v
more
usefuloperator()
operator->
?std::integral_constant
std::integral_constant
and
std::constexpr_v
, based on LEWG
feedback.std::constexpr_v::operator()
, as
requested by LEWG.std::constexpr_value
, as
suggested by LEWG.constexpr_value
concept.auto
NTTP parameter
for constexpr_v
, and default its
type template parameter.operator->
.std::integral_constant
.and
,
or
and
not
keywords.constexpr-param
concept.constexpr_v
’s
T
template parameter.This paper is co-authored by the authors of P2725R1
(“std::integral_constant
Literals”) and P2772R0
(“std::integral_constant
literals do not suffice —
constexpr_t
?”). This paper
supersedes both of those previous papers.
std::integral_constant<int>
are
badstd::integral_constant<int>
is
used in lots of places to communicate a constant integral value to a
given interface. The length of its spelling makes it very verbose.
Fortunately, we can do a lot better.
Before
|
After
|
---|---|
|
|
The “after” case above would require that
std::strided_index_range
be
changed; that is not being proposed here. The point of the example is to
show the relative convenience of
std::integral_constant
versus
the proposed
std::constexpr_v
.
std::integral_constant
is not
enoughParameters passed to a
constexpr
function lose their
constexpr
-ness when used inside
the function. Replacing
std::integral_constant
with
std::constexpr_v
has the
potential to improve a lot more uses of compile-time constants than just
integrals; what about all the other
constexpr
-friendly C++
types?
Consider:
template<typename T>
struct my_complex
{
T re, im;};
inline constexpr short foo = 2;
template<typename T>
struct X
{
void f(auto c)
{
// c is to be used as a constexpr value here
}
};
We would like to be able to call
X::f()
with a value, and have
that value keep its
constexpr
-ness. Let’s introduce
a template “constexpr_v
” that
holds a constexpr
value that it
is given as an non-type template parameter.
namespace std {
template<auto X, class T/* = remove_cvref_t<decltype(X)>*/>
struct constexpr_v
{
using value_type = T;
using type = constexpr_v;
constexpr operator value_type() const { return X; }
static constexpr value_type value = X;
// The rest of the members are discussed below ....
};
}
Now we can write this.
template<typename T>
void g(X<T> x)
{
.f(std::constexpr_v<1>{});
x.f(std::constexpr_v<2uz>{});
x.f(std::constexpr_v<3.0>{});
x.f(std::constexpr_v<4.f>{});
x.f(std::constexpr_v<foo>{});
x.f(std::constexpr_v<my_complex(1.f, 1.f)>{});
x}
Let’s now add a constexpr
variable template with a shorter name, say
c_
.
namespace std {
template<auto X>
inline constexpr constexpr_v<X> c_{};
}
And now we can write this.
template<typename T>
void g(X<T> x)
{
.f(std::c_<1>);
x.f(std::c_<2uz>);
x.f(std::c_<3.0>);
x.f(std::c_<4.f>);
x.f(std::c_<foo>);
x.f(std::c_<my_complex(1.f, 1.f)>);
x}
std::constexpr_v
and std::c_
std::c_
takes an
auto
NTTP.
std::constexpr_v
takes an
auto
NTTP
X
, and a type
T
which is defaulted to
decltype(X)
. Why is this? ADL!
Even thought the type of X
is
deduced with our without T
,
without T
some natural uses of
constexpr_v
cease to work. For
instance:
auto f = std::c_<strlit("foo")>; // Using the strlit from later in this paper.
::cout << f << "\n"; std
The stream insertion breaks without the
T
parameter. The
T
parameter is
strlit</*...*/>
, which
pulls strlit
’s
operator<<
into
consideration during ADL. Note that this ADL support is imperfect. The
use op operator<<
above is
due to the way the operator overload is declared:
friend std::ostream & operator<<(std::ostream & os, strlit l) { /* ...*/ }
If it is instead declared as a
non-friend
:
template<size_t N>
::ostream & operator<<(std::ostream & os, strlit<N> l) { /* ...*/ } std
… ADL’s help doesn’t suffice. The deduction of
N
is not possible from a type
that isn’t a strlit<N>
itself (e.g. base class) even if it is implicitly convertible to
strlit<N>
. This is pretty
likely to confuse users. This may make the
T
parameter more trouble than
it’s worth.
constexpr_v
more usefulconstexpr_v
is essentially a
wrapper. It takes a value X
of
some structural type T
, and
represents X
in such a way that
we can continue to use X
as a
compile-time constant, regardless of context. As such,
constexpr_v
should be implicitly
convertible to T
; this is
already reflected in the design presented above. For the same reason,
constexpr_v
should provide all
the operations that the underlying type has. Though we cannot predict
what named members the underlying type
T
has, we can guess at
all the operator overloads it might have.
So, by adding conditionally-defined overloads for all the
overloadable operators, we can make
constexpr_v
as natural to use as
many of the types it might wrap.
namespace std {
template<auto X, class T/* = remove_cvref_t<decltype(X)>*/>
struct constexpr_v {
using value_type = T;
using type = constexpr_v;
constexpr operator value_type() const { return X; }
static constexpr value_type value = X;
// unary -
template<auto Y = X>
constexpr constexpr_v<-Y> operator-() const { return {}; }
// binary + and -
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value + V::value> operator+(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value - V::value> operator-(U, V) { return {}; }
// etc... (full listing later)
};
}
These operators are defined in such a way that they behave just like
the operations on underlying the
T
and
V
values would, including
promotions and coercions. For example:
static_assert(std::is_same_v<
decltype(std::c_<42> - std::c_<13u>),
::constexpr_v<29u>>); std
Each operation is only defined if the underlying operation on
X
and
Y
is defined. Each operation
additionally requires that the result of the underlying operation have a
structural type.
The mutating operations are left out, because none of them makes sense – the type of the mutated value would have to change, since the value is itself part of the type.
All the remaining operations are included, even the index and call operators. The rationale for this is that a user may want to make some sort of compile-time domain-specific embedded language using operator overloading, and having all but a couple of the operators specified would frustrate that effort.
The only downside to adding
std::constexpr_v::operator()()
is that it would represent a break from the design of
std::integral_constant
, making
it an imperfect drop-in replacement for that template.
The operators are designed to interoperate with other types and
templates that have a constexpr static
value
member. This works with
std::constexpr_v
s of course, but
also std::integral_constant
s,
and user-provided types as well. For example:
struct my_type { constexpr static int value = 42; };
void foo()
{
constexpr auto zero = my_type{} - std::c_<42>; // Ok.
// ...
}
Note that the addition of these operators is in line with the poll:
“Add a new robust integral constant type with all the numerical
operators, as proposed in P2772R0, and use that for these literals
instead of
std::integral_constant
”?
SF
|
F
|
N
|
A
|
SA
|
---|---|---|---|---|
4 | 7 | 1 | 1 | 1 |
… taken in the 2023-01-17 Library Evolution telecon.
Note that the one SA said he would not be opposed if the word
“integral” was stricken from the poll, and the design of
std::constexpr_v
is not limited
to integral types.
As pointed out on the reflector,
std::c_<"foo">
does not work, because of language rules. However, it’s pretty easy for
users to add an NTTP-friendly string wrapper type, and then use that
with std::c_<>
.
template<size_t N>
struct strlit
{
constexpr strlit(char const (&str)[N]) { std::copy_n(str, N, value); }
template<size_t M>
constexpr bool operator==(strlit<M> rhs) const
{
return std::ranges::equal(bytes_, rhs.bytes_);
}
friend std::ostream & operator<<(std::ostream & os, strlit l)
{
assert(!l.value[N - 1] && "value must be null-terminated");
return os.write(l.value, N - 1);
}
char value[N];
};
int main()
{
auto f = std::c_<strlit("foo")>;
::cout << f; // Prints "foo".
std}
operator()
The addition of non-arithmetic operators may seem academic at first.
However, consider this
constexpr
-friendly parser
combinator mini-library.
namespace parse {
template<typename L, typename R>
struct or_parser;
template<size_t N>
struct str_parser
{
template<size_t M>
constexpr bool operator()(strlit<M> lit) const
{
return lit == str_;
}
template<typename P>
constexpr auto operator|(P parser) const
{
return or_parser<str_parser, P>{*this, parser};
}
<N> str_;
strlit};
template<typename L, typename R>
struct or_parser
{
template<size_t M>
constexpr bool operator()(strlit<M> lit) const
{
return l_(lit) || r_(lit);
}
template<typename P>
constexpr auto operator|(P parser) const
{
return or_parser<or_parser, P>{*this, parser};
}
L l_;
R r_;};
}
int foo()
{
constexpr parse::str_parser p1{strlit("neg")};
constexpr parse::str_parser p2{strlit("incr")};
constexpr parse::str_parser p3{strlit("decr")};
constexpr auto p = p1 | p2 | p3;
constexpr bool matches_empty = p(strlit(""));
static_assert(!matches_empty);
constexpr bool matches_pos = p(strlit("pos"));
static_assert(!matches_pos);
constexpr bool matches_decr = p(strlit("decr"));
static_assert(matches_decr);
}
(This relies on the strlit
struct shown just previously.)
Say we wanted to use the templates in namespace
parser
along side other values,
like int
s and
float
s. We would want that not
to break our std::constexpr_v
expressions. Having to work around the absence of
std::constexpr_v::operator()
would require us to write a lot more code. Here is the equivalent of the
function foo()
above, but with
all the variables wrapped using
std::c_
.
int bar()
{
constexpr parse::str_parser p1{strlit("neg")};
constexpr parse::str_parser p2{strlit("incr")};
constexpr parse::str_parser p3{strlit("decr")};
constexpr auto p_ = std::c_<p1> | std::c_<p2> | std::c_<p3>;
constexpr bool matches_empty_ = p_(std::c_<strlit("")>);
static_assert(!matches_empty_);
constexpr bool matches_pos_ = p_(std::c_<strlit("pos")>);
static_assert(!matches_pos_);
constexpr bool matches_decr_ = p_(std::c_<strlit("decr")>);
static_assert(matches_decr_);
}
As you can see, everything works as it did before. The presence of
operator()
does not enable any
new functionality, it just keeps code that happens to use it from
breaking.
The operators left out of the code below are the ones that involve
mutation, like operator++
,
operator/=
, etc. These seem at
first that these are nonsensical, since all the operations on a
constexpr_v
must be
nonmutating.
However, some DSLs may wish to use these operations with atypical semantics.
struct weirdo
{
constexpr int operator++() const { return 1; }
};
auto result = ++std::c_<weirdo{}>;
result
is obviously
std::c_<1>
here, and no
mutation occurred. You can imagine a more elaborate use case, say a
library that is used to create expression templates. For example:
auto expr = std::c_<var0> += std::c_<var1>;
In this case, var0
and
var1
would be some terminal
types in the expression template library, and
operator+=
would return a
constexpr
expression tree,
rather than mutating the left side of the
+=
.
Since this is a realtively late addition – after the paper has been through two LEWG reviews, the addition of these operators is being presented as an option. However, we have implemented it, and know that they work.
The optional parts in the design listing and wording are marked with
#if LEWG_SAYS_SO
.
We want to add all overloadable operators to
constexpr_v
, including the ones
that are usually mutating.
operator->
?We’re not proposing it, because of its very specific semantics – it must yield a pointer, or something that eventually does. That’s not a very useful operation during constant evaluation.
std::integral_constant
During the LEWG reviews, some attendees suggested that
inter-conversions between
std::integral_constant
and
std::constexpr_v
would be
useful. The important thing to remember is that we want deduction to
occur when calling functions that take a
std::constexpr_v
, including the
std::constexpr_v
operator
overloads. Conversions and deductions are at odds with one another,
because deducing parameter types disables the conversion rules.
If you look at the operator overloads proposed here, you will see
that they are deduction operations at their most essential. The types of
the parameters do not matter, except that each conveys a value that is a
core constant expression because it is embedded in the type system. The
fact that a std::constexpr_v
conveys that value instead of a
std::integral_constant
is
immaterial, and in fact the operators are written in such a way that
they operate on either template (as long as at least one parameter is a
specialization of
std::constexpr_v
). Users can and
should write their code using these kinds of values-as-types in a
similar way. Relying on conversions is a less-useful way to get
interoperability.
constexpr_v
namespace std {
template<auto X, class T = remove_cvref_t<decltype(X)>>
struct constexpr_v;
template <class T>
concept constexpr-param = // exposition only
requires { typename constexpr_v<T::value>; };
template <class T>
concept derived-from-constexpr = // exposition only
<T, constexpr_v<T::value>>;
derived_fromtemplate <class T, class SelfT>
concept lhs-constexpr-param = // exposition only
<T> && (derived_from<T, SelfT> || !derived-from-constexpr<T>);
constexpr-param
template<auto X, class T>
struct constexpr_v {
using value_type = T;
using type = constexpr_v;
constexpr operator value_type() const { return X; }
static constexpr value_type value = X;
#if LEWG_SAYS_SO
template <constexpr-param U>
constexpr constexpr_v<X = U::value> operator=(U) const { return {}; }
#endif
template<auto Y = X>
constexpr constexpr_v<+Y> operator+() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<-Y> operator-() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<~Y> operator~() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<!Y> operator!() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<&Y> operator&() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<*Y> operator*() const { return {}; }
template<class... Args>
constexpr constexpr_v<X(Args::value...)> operator()(Args... args) const { return {}; }
template<class... Args>
constexpr constexpr_v<X[Args::value...]> operator[](Args... args) const { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value + V::value> operator+(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value - V::value> operator-(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value * V::value> operator*(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value / V::value> operator/(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value % V::value> operator%(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value << V::value)> operator<<(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value >> V::value)> operator>>(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value & V::value> operator&(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value | V::value> operator|(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value ^ V::value> operator^(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value && V::value> operator&&(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value || V::value> operator||(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value <=> V::value)> operator<=>(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value == V::value)> operator==(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value != V::value)> operator!=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value < V::value)> operator<(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value > V::value)> operator>(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value <= V::value)> operator<=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value >= V::value)> operator>=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value, V::value)> operator,(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value ->* V::value)> operator->*(U, V) { return {}; }
#if LEWG_SAYS_SO
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value += V::value)> operator+=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value -= V::value)> operator-=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value *= V::value)> operator*=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value /= V::value)> operator/=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value %= V::value)> operator%=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value &= V::value)> operator&=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value |= V::value)> operator|=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value ^= V::value)> operator^=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value <<= V::value)> operator<<=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value >>= V::value)> operator>>=(U, V) { return {}; }
#endif
};
template<auto X>
inline constexpr constexpr_v<X> c_{};
}
Add a new feature macro,
__cpp_lib_constexpr_v
.
Look up a few lines to see an implementation of
std::constexpr_v
. At the time of
this writing, there is one caveat:
operator[]()
looks correct to
the authors, but does not work in any compiler tested, due to the very
limited multi-variate operator[]
support in even the latest compilers.
Additionally, an
integral_constant
with most of
the operator overloads has been a part of Boost.Hana
since its initial release in May of 2016. Its operations have been used
by many, many users.
std::constexpr_v
:
std::constexpr_wrapper
Pro: The name calls out that this type has a relation to constant expressions.
Pro: Consistency. The name is analoguous to
std::reference_wrapper
(and
std::ref
).
Pro:
constexpr_wrapper<1>
reads
as wrapper for constant
expression with value 1.
Con: It’s a long name. However, if we expect that user typically won’t spell out this type, then it doesn’t matter much.
std::constexpr_t
Pro: The name calls out that this type has a relation to constant expressions.
Pro: Read the name as either “the result of a constant expression identified by a type” or “a type identifying a value usable in constant expressions”.
Pro: The name calls out that it’s a type (or class template) and not a variable template.
Con: _t
in the standard
library typically means “alias template for
::type
member of a trait”. The
use of _t
to identify types is
more of a C thing.
std::constexpr_value
Pro: The name calls out that this type has a relation to constant expressions.
Pro:
constexpr_value<1>
can be
read as “the value 1 as a constant expression”.
Con: The _value
suffix
may mislead readers to expect a variable template.
std::constant_value
Con: The _value
suffix
may mislead readers to expect a variable template.
Con: Nothing in the type hints at the primary use case: enabling constant expressions.
Con: The name constant
hints at const
. While
const
isn’t wrong here, it’s
also irrelevant.
Con:
constant_value<1>
reads
pretty much like const int
with
value 1
, which is
misleading.
std::c_
:
std::cc
short for
constexpr_wrapper
constant (somewhat silly, sure)
quick to type: hit the same key twice
203’614 matches found on codesearch.isocpp.org
std::c_
hardest to type (IMHO and with US keyboard layout, i.e. type “c shift+_ shift+<”; the _ < movement is slowing me down)
10’198 matches found on codesearch.isocpp.org
std::cw
short for
constexpr_
wrapper
or constexpr wrap, i.e. read
cw<1>
as “constexpr wrap
1”
41’311 matches found on codesearch.isocpp.org
std::c
3’869’416 matches found on codesearch.isocpp.org
scarily short
Add the following to [meta.type.synop], after
false_type
:
template<auto X, class T = remove_cvref_t<decltype(X)>>
struct constexpr_v;
template <class T>
concept constexpr-param = // exposition only
!is_member_pointer_v<decltype(&T::value)> && requires { typename constexpr_v<T::value>; }
template <class T>
concept derived-from-constexpr = // exposition only
derived_from<T, constexpr_v<T::value>>;
template <class T, class SelfT>
concept lhs-constexpr-param = // exposition only
constexpr-param<T> && (derived_from<T, SelfT> || !derived-from-constexpr<T>);
template<auto X> inline constexpr constexpr_v<X> c_;
Add the following to [meta.help], after
integral_constant
:
template<auto X, class T>
struct constexpr_v {
using value_type = T;
using type = constexpr_v;
constexpr operator value_type() const { return X; }
static constexpr value_type value = X;
#if LEWG_SAYS_SO
template <constexpr-param U>
constexpr constexpr_v<X = U::value> operator=(U) const { return {}; }
#endif
template<auto Y = X>
constexpr constexpr_v<+Y> operator+() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<-Y> operator-() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<~Y> operator~() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<!Y> operator!() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<&Y> operator&() const { return {}; }
template<auto Y = X>
constexpr constexpr_v<*Y> operator*() const { return {}; }
template<class... Args>
constexpr constexpr_v<X(Args::value...)> operator()(Args... args) const { return {}; }
template<class... Args>
constexpr constexpr_v<X[Args::value...]> operator[](Args... args) const { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value + V::value> operator+(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value - V::value> operator-(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value * V::value> operator*(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value / V::value> operator/(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value % V::value> operator%(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value << V::value)> operator<<(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value >> V::value)> operator>>(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value & V::value> operator&(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value | V::value> operator|(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value ^ V::value> operator^(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value && V::value> operator&&(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<U::value || V::value> operator||(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value <=> V::value)> operator<=>(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value == V::value)> operator==(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value != V::value)> operator!=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value < V::value)> operator<(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value > V::value)> operator>(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value <= V::value)> operator<=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value >= V::value)> operator>=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value, V::value)> operator,(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value ->* V::value)> operator->*(U, V) { return {}; }
#if LEWG_SAYS_SO
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value += V::value)> operator+=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value -= V::value)> operator-=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value *= V::value)> operator*=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value /= V::value)> operator/=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value %= V::value)> operator%=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value &= V::value)> operator&=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value |= V::value)> operator|=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value ^= V::value)> operator^=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value <<= V::value)> operator<<=(U, V) { return {}; }
template <lhs-constexpr-param<type> U, constexpr-param V>
friend constexpr constexpr_v<(U::value >>= V::value)> operator>>=(U, V) { return {}; }
#endif
};
template<auto X> inline constexpr constexpr_v<X> c_{};
2
The class template constexpr_v
aids in metaprogramming by ensuring that the evaluation of expressions
comprised entirely of
constexpr_v
s are core constant
expressions ([expr.const]), regardless of the context in which they
appear. In particular, this enables use of
constexpr_v
values that are
passed as arguments to constexpr
functions to be used as template parameters.
3
The variable template c_
is
provided as a convenient way to nominate
constexpr_v
values.
Add to [version.syn]:
#define __cpp_lib_constexpr_v XXXXXXL // also in <type_traits>