Skip to content

fix8mt/conjure_enum

Repository files navigation

Lightweight header-only C++20 enum and typename reflection


clang gcc

1. Quick links

1 conjure_enum API and examples
2 enum_bitset Enhanced enum aware std::bitset
3 conjure_type Any type string extractor
4 fixed_string Statically stored null terminated fixed string
5 Building How to build or include
6 vcpkg For vcpkg package
7 Notes Notes on the implementation, limits, etc
8 Benchmarks Benchmarking
9 Compilers Supported compilers
10 Compiler issues Workarounds for various compiler issues
11 Results of std::source_location For implementation specific std::source_location results

Tip

Use the built-in table of contents to navigate this guide. Even better in full read view of this page.

For the latest cutting edge changes, see the dev branch. You can view the changes (if any) here.


2. Introduction

a) Supercharge Your C++ Enums with This Lightweight Reflection Library!

Based on the awesome work in magic_enum1 and boost::describe, this library offers a streamlined and powerful way to add reflection capabilities to your C++ enums and other types. We've optimized the core functionality, focusing on the main features developers usually want. We've also added general purpose typename reflection for any type.

conjure_enum2 takes full advantage of recently added C++20 features. We've leveraged the convenience of std::source_location and unlocked the potential of constexpr algorithms and concepts.

b) Highlights

  • Single Header-Only: No external dependencies, simplifying integration into your project
  • Modern C++20: Entirely constexpr for compile-time safety, efficiency and performance; no macros
  • Broad Support: Works with:
    • scoped and unscoped enums
    • enums with aliases and gaps
    • anonymous and named namespaced enums and types
    • custom enum ranges
  • Easy to Use: Class-based approach with intuitive syntax
  • Convenient: enum_bitset provides an enhanced enum aware std::bitset (see 2 above)
  • Useful: conjure_type gives you the type string of any type! (see 3 above)
  • Wide Compiler Compatibility: Support for: (see 9 above)
    • GCC
    • Clang
    • MSVC
    • XCode/Apple Clang
  • Confidence: Includes comprehensive unit test suite for reliable functionality and profiling
  • Expanded: Enhanced API:
    • add_scope
    • remove_scope
    • unscoped_string_to_enum
    • for_each_n
    • dispatch
    • iterators and more!
  • Transparency: Compiler implementation variability fully documented, verifiable and reportable (see 11 above)

3. conjure_enum

All examples refer to the following enums:

enum class component { scheme, authority, userinfo, user, password, host, port, path=12, test=path, query, fragment };
enum component1 { scheme, authority, userinfo, user, password, host, port, path=12, query, fragment };
enum class numbers { zero, one, two, three, four, five, six, seven, eight, nine };

Important

Your type must be an enum, satisfying:

template<typename T>
concept valid_enum = requires(T)
{
  requires std::same_as<T, std::decay_t<T>>;
  requires std::is_enum_v<T>;
};

a) enum_to_string

static constexpr std::string_view enum_to_string(T value, bool noscope=false);
template<T e>
static constexpr std::string_view enum_to_string();

Returns a std::string_view or empty if not found. Optionally passing true will remove scope in result if present. noscope option .

auto name { conjure_enum<component>::enum_to_string(component::path) };
auto name_trim { conjure_enum<component>::enum_to_string(component::path, true) }; // optionally remove scope in result
auto alias_name { conjure_enum<component>::enum_to_string(component::test) }; // alias
auto noscope_name { conjure_enum<component1>::enum_to_string(path) };
std::cout << name << '\n' << name_trim << '\n' << alias_name << '\n' << noscope_name << '\n';
std::cout << conjure_enum<numbers>::enum_to_string<numbers::two>() << '\n';

output

component::path
path
component::path
path
numbers::two

Aliases

Because all methods in conjure_enum are defined within a class instead of individual template functions in a namespace, you can reduce your typing with standard aliases:

using ec = FIX8::conjure_enum<component>;
std::cout << std::format("\"{}\"\n", ec::enum_to_string(component::authority));
std::cout << std::format("\"{}\"\n", ec::enum_to_string(static_cast<component>(100)));

output

"component::authority"
""

Also supplied is a template version of enum_to_string.

std::cout << std::format("\"{}\"\n", ec::enum_to_string<component::scheme>());
std::cout << std::format("\"{}\"\n", ec::enum_to_string<scheme>());

output

"component::scheme"
"scheme"

b) string_to_enum

static constexpr std::optional<T> string_to_enum(std::string_view str);

Returns a std::optional<T>. Empty if string was not valid. Use std::optional<T>::value_or() to set an error value and avoid throwing an exception.

int value { static_cast<int>(conjure_enum<component>::string_to_enum("component::path").value()) };
int noscope_value { static_cast<int>(conjure_enum<component1>::string_to_enum("path").value()) };
int bad_value { static_cast<int>(conjure_enum<component>::string_to_enum("bad_string").value_or(component(100))) };
std::cout << value << '\n' << noscope_value << '\n' << bad_value << '\n';

output

12
12
100 <-- invalid, error value

c) unscoped_string_to_enum

static constexpr std::optional<T> unscoped_string_to_enum(std::string_view str);

Same as string_to_enum except works with unscoped strings. Returns a std::optional<T>. Empty if string was not valid. Use std::optional<T>::value_or() to set an error value and avoid throwing an exception.

int value { static_cast<int>(conjure_enum<component>::unscoped_string_to_enum("path").value()) };
int noscope_value { static_cast<int>(conjure_enum<component1>::string_to_enum("path").value()) };
int bad_value { static_cast<int>(conjure_enum<component>::string_to_enum("bad_string").value_or(component(100))) };
std::cout << value << '\n' << noscope_value << '\n' << bad_value << '\n';

output

12
12
100 <-- invalid, error value

d) int_to_enum, enum_cast

static constexpr std::optional<T> int_to_enum(int value);
static constexpr T enum_cast(int value);

Returns a std::optional<T>. Empty if value was not valid. Use std::optional<T>::value_or() to set an error value and avoid throwing an exception. enum_cast will cast to the enum type regardless of whether the value is a valid enum.

int value { static_cast<int>(conjure_enum<component>::int_to_enum(12).value()) };
int noscope_value { static_cast<int>(conjure_enum<component1>::int_to_enum(12).value()) };
int bad_value { static_cast<int>(conjure_enum<component>::int_to_enum(100).value_or(component(100))) };
std::cout << value << '\n' << noscope_value << '\n' << bad_value << '\n';
std::cout << static_cast<int>(conjure_enum<component>::enum_cast(150)) << '\n';

output

12
12
100 <-- invalid, error value
150 <-- invalid, but still casted

e) enum_to_int, enum_to_underlying

static constexpr int enum_to_int(T value);
static constexpr std::underlying_type_t<T> enum_to_underlying(T value);

Returns an int or the underlying value for the given enum value. These are added for completeness. For unscoped enums you can always just use the value like an int, or for scoped enums just static_cast<int> it first.

std::cout << conjure_enum<component>::enum_to_int(component::path) << '\n';
std::cout << conjure_enum<component>::enum_to_underlying(component::path) << '\n';
std::cout << conjure_enum<component1>::enum_to_int(path) << '\n';
std::cout << conjure_enum<component1>::enum_to_underlying(path) << '\n';

output

12
12
12
12

f) count

static constexpr std::size_t count();

Returns number of enumerations.

std::cout << conjure_enum<component>::count()  << '\n';

output

10

g) names

static constexpr std::array<std::string_view, std::size_t> names;

This static member is generated for your type. It is a std::array of the std::string_view strings in enum order.

for(const auto ev : conjure_enum<component>::names) // scoped enum
   std::cout << ev << '\n';
for(const auto ev : conjure_enum<component1>::names) // unscoped enum
   std::cout << ev << '\n';

output

component::scheme
component::authority
component::userinfo
component::user
component::password
component::host
component::port
component::path
component::query
component::fragment
scheme
authority
userinfo
user
password
host
port
path
query
fragment

h) unscoped_names

static constexpr std::array<std::string_view, std::size_t> unscoped_names;

This static member is generated for your type. It is a std::array of the std::string_view unscoped strings in enum order. For unscoped enums is the same as names above.

for(const auto ev : conjure_enum<component>::unscoped_names) // scoped enum
   std::cout << ev << '\n';
std::cout << '\n';
for(const auto ev : conjure_enum<component1>::unscoped_names) // unscoped enum
   std::cout << ev << '\n';

output

scheme
authority
userinfo
user
password
host
port
path
query
fragment

scheme
authority
userinfo
user
password
host
port
path
query
fragment

i) values

static constexpr std::array<T, std::size_t> values;

This static member is generated for your type. It is a std::array of the T values in enum order.

for(const auto ev : conjure_enum<component>::values) // scoped enum
   std::cout << static_cast<int>(ev) << '\n';

output

0
1
2
3
4
5
6
12
13
14

j) entries, sorted_entries

static constexpr std::array<std::tuple<T, std::string_view>, std::size_t> entries;
static constexpr std::array<std::tuple<T, std::string_view>, std::size_t> sorted_entries;

These static members are generated for your type. They are std::array of tuples of T and std::string_view. sorted_entries is the same as entries except the array is sorted by the std::string_view name.

using ec = conjure_enum<component>;
for(const auto [value, str] : ec::entries) // scoped enum
   std::cout << std::format("{:<2} {}\n", static_cast<int>(value), str);
std::cout << '\n';
for(const auto [value, str] : ec::sorted_entries) // scoped enum
   std::cout << std::format("{:<2} {}\n", static_cast<int>(value), str);

output

0  component::scheme
1  component::authority
2  component::userinfo
3  component::user
4  component::password
5  component::host
6  component::port
12 component::path
13 component::query
14 component::fragment

1  component::authority
14 component::fragment
5  component::host
4  component::password
12 component::path
6  component::port
13 component::query
0  component::scheme
3  component::user
2  component::userinfo

k) scoped_entries, unscoped_entries

static constexpr std::array<std::tuple<std::string_view, std::string_view>, std::size_t> scoped_entries;

This static member is generated for your type. It is a std::array of a tuple of std::string_view pairs in enum order. It contains pairs of unscoped and their scoped string version. This array is sorted by unscoped name. For unscoped enums, these are identical.

unscoped_entries is the same except the pair is reversed.

for(const auto [a, b] : conjure_enum<component>::scoped_entries)
   std::cout << std::format("{:9} {}\n", a, b);

output

authority component::authority
fragment  component::fragment
host      component::host
password  component::password
path      component::path
port      component::port
query     component::query
scheme    component::scheme
user      component::user
userinfo  component::userinfo

l) rev_scoped_entries

static constexpr std::array<std::tuple<std::string_view, std::string_view>, std::size_t> rev_scoped_entries;

Same as scoped_entries except reversed, sorted by scoped name. Use to lookup unscoped name.

m) index

static constexpr std::optional<size_t> index(T value);
template<T e>
static constexpr std::optional<size_t> index();

Returns the index (position in 0 based array of values) of the supplied enum value as an std::optional<size_t>. Empty if value was not valid. Use std::optional<T>::value_or() to set an error value and avoid throwing an exception.

std::cout << conjure_enum<component>::index(component::password).value() << '\n';
std::cout << conjure_enum<component>::index(component(100)).value_or(100) << '\n';
std::cout << conjure_enum<component>::index<component::password>().value() << '\n';
std::cout << conjure_enum<component>::index<component(100)>().value_or(100) << '\n';

output

4
100 <-- invalid, error value
4
100 <-- invalid, error value

n) contains, is_valid

static constexpr bool contains(T value);
static constexpr bool contains(std::string_view str);
template<T e>
static constexpr bool contains();
template<T e>
static constexpr bool is_valid();

Returns true if the enum contains the given value or string.

std::cout << std::format("{}\n", conjure_enum<component>::contains(component::path));
std::cout << std::format("{}\n", conjure_enum<component1>::contains("nothing"));
std::cout << std::format("{}\n", conjure_enum<component>::contains<component::path>());
std::cout << std::format("{}\n", conjure_enum<component>::is_valid<component::path>());

output

true
false
true
true

o) for_each, for_each_n

template<typename Fn, typename... Args>
requires std::invocable<Fn&&, T, Args...>
[[maybe_unused]] static constexpr auto for_each(Fn&& func, Args&&... args);

template<typename Fn, typename C, typename... Args> // specialisation for member function with object
requires std::invocable<Fn&&, C, T, Args...>
[[maybe_unused]] static constexpr auto for_each(Fn&& func, C *obj, Args&&... args);

template<typename Fn, typename... Args>
requires std::invocable<Fn&&, T, Args...>
[[maybe_unused]] static constexpr auto for_each_n(int n, Fn&& func, Args&&... args);

template<typename Fn, typename C, typename... Args> // specialisation for member function with object
requires std::invocable<Fn&&, C, T, Args...>
[[maybe_unused]] static constexpr auto for_each_n(int n, Fn&& func, C *obj, Args&&... args);

Call supplied invocable for each enum value. Similar to std::for_each except the first parameter of your invocable must accept an enum value (passed by for_each). Optionally provide any additional parameters. You can limit the number of calls to your invocable by using the for_each_n version with the first parameter being the maximum number to call. The second version of for_each and for_each_n is intended to be used when using a member function - the second parameter passed by your call must be the this pointer of the object. If you wish to pass a reference parameter, you must wrap it in std::ref.

Works with lambdas, member functions, functions etc, compatible with std::invoke.

Returns

std::bind(std::forward<Fn>(func), std::placeholders::_1, std::forward<Args>(args)...);
// or
std::bind(std::forward<Fn>(func), obj, std::placeholders::_1, std::forward<Args>(args)...);

which can be stored or immediately invoked.

See enum_bitset::for_each to iterate through a bitset.

conjure_enum<component>::for_each([](component val, int other)
{
   std::cout << static_cast<int>(val) << ' ' << other << '\n';
}, 10);

output

0 10
1 10
2 10
3 10
4 10
5 10
6 10
12 10
13 10
14 10

Above example using for_each_n, limiting to 3:

conjure_enum<component>::for_each_n(3, [](component val, int other)
{
   std::cout << static_cast<int>(val) << ' ' << other << '\n';
}, 10);

output

0 10
1 10
2 10

Example using returned object and additional reference parameter:

int total{};
auto myfunc { conjure_enum<component>::for_each([](component val, int other, int& tot)
{
   std::cout << static_cast<int>(val) << ' ' << other << '\n';
   tot += static_cast<int>(val);
}, 10, std::ref(total)) };
myfunc(component::fragment);
std::cout << total << '\n';

output

0 10
1 10
2 10
3 10
4 10
5 10
6 10
12 10
13 10
14 10
14 10 <== invoked with returned object
74

Example with pointer to member function with additional parameters:

struct foo
{
   void process(component val, int offset, int& tot)
   {
      tot += offset + static_cast<int>(val);
   }
};
int total{};
foo bar;
conjure_enum<component>::for_each(&foo::process, &bar, 10, std::ref(total));
std::cout << total << '\n';

output

160

p) dispatch

template<typename Fn>
static constexpr bool tuple_comp(const std::tuple<T, Fn>& pl, const std::tuple<T, Fn>& pr);

template<std::size_t I, typename R, typename Fn, typename... Args> // with not found value(nval) for return
requires std::invocable<Fn&&, T, Args...>
[[maybe_unused]] static constexpr R dispatch(T ev, R nval, const std::array<std::tuple<T, Fn>, I>& disp, Args&&... args);

template<std::size_t I, typename R, typename Fn, typename C, typename... Args> // specialisation for member function with not found value(nval) for return
requires std::invocable<Fn&&, C, T, Args...>
[[maybe_unused]] static constexpr R dispatch(T ev, R nval, const std::array<std::tuple<T, Fn>, I>& disp, C *obj, Args&&... args);

template<std::size_t I, typename Fn, typename... Args> // void func with not found call to last element
requires (std::invocable<Fn&&, T, Args...> && I > 0)
static constexpr void dispatch(T ev, const std::array<std::tuple<T, Fn>, I>& disp, Args&&... args);

template<std::size_t I, typename Fn, typename C, typename... Args> // specialisation for void member function with not found call to last element
requires (std::invocable<Fn&&, C, T, Args...> && I > 0)
static constexpr void dispatch(T ev, const std::array<std::tuple<T, Fn>, I>& disp, C *obj, Args&&... args);

With a given enum, search and call user supplied invocable. A typical use case would be where you want to demux a complex event, allowing you to easily declare predefined invocable actions for different enum values.

  • Where invocable returns a value, return this value or a user supplied "not found" value.
  • Where invocable is void, call user supplied invocable or "not found" invocable (last in supplied array).

The first parameter of your invocable must accept an enum value (passed by dispatch). Optionally provide any additional parameters.

Works with lambdas, member functions, functions etc, compatible with std::invoke.

There are two versions of dispatch - the first takes an enum value, a 'not found' value, and a std::array of std::tuple of enum and invocable. The second version takes an enum value, and a std::array of std::tuple of enum and invocable. The last element of the array is called if the enum is not found. This version is intended for use with void return invocables.

The second version of each of the above is intended to be used when using a member function - the first parameter passed after your array must be the this pointer of the object. You can also use std::bind to bind the this pointer and any parameter placeholders when declaring your array. If you wish to pass a reference parameter, you must wrap it in std::ref.

Tip

If you wish to provide a constexpr array, you will need to use a simple function prototype, since std::function is not constexpr (see unit tests for examples).

Important

Your std::array of std::tuple should be sorted by enum. The dispatch method performs a binary search on the array. Complexity for a sorted array is at most  $2log_2(N)+O(1)$  comparisons. If the array is not sorted, complexity is linear.

The following example uses a static constexpr array of pointers to functions. For brevity they all point to the same function except the last which is a lambda.

enum class directions { left, right, up, down, forward, backward, notfound=-1 };
static constexpr auto prn([](directions ev) { std::cout << conjure_enum<directions>::enum_to_string(ev) << '\n'; });
static constexpr auto tarr
{
   std::to_array<std::tuple<directions, void(*)(directions)>>
   ({
      { directions::left, prn },
      { directions::right, prn },
      { directions::up, prn },
      { directions::down, prn },
      { directions::backward, prn },
      { directions::notfound, [](directions ev) { std::cout << "not found: "; prn(ev); } }, // not found func
   })
};
conjure_enum<directions>::dispatch(directions::right, tarr);
conjure_enum<directions>::dispatch(directions::down, tarr);
conjure_enum<directions>::dispatch(directions::forward, tarr);
std::cout << conjure_enum<directions>::enum_to_int(directions::notfound) << '\n';

output

directions::right
directions::down
not found: directions::forward
-1

This example uses lambdas:

const auto dd1
{
   std::to_array<std::tuple<component, std::function<int(component, int)>>>
   ({
      { component::scheme, [](component ev, int a) { return a * 100 + conjure_enum<component>::enum_to_int(ev); } },
      { component::port, [](component ev, int a) { return a * 200 + conjure_enum<component>::enum_to_int(ev); } },
      { component::fragment, [](component ev, int a) { return a * 300 + conjure_enum<component>::enum_to_int(ev); } },
   })
};
std::cout << conjure_enum<component>::dispatch(component::port, -1, dd1, 10) << '\n';

output

2006

This example uses member functions:

struct foo
{
   int process(component val, int aint)
   {
      return aint * static_cast<int>(val);
   }
   int process1(component val, int aint)
   {
      return aint + static_cast<int>(val);
   }
};
constexpr auto tarr1
{
   std::to_array<std::tuple<component, int (foo::*)(component, int)>>
   ({
      { component::scheme, &foo::process },
      { component::port, &foo::process },
      { component::fragment, &foo::process1 },
   })
};
foo bar;
for (auto val : { component::scheme, component::path, component::port, component::fragment })
   std::cout << conjure_enum<component>::dispatch(val, -1, tarr1, &bar, 1000) << '\n';

output

0
-1
6000
1015

q) is_scoped

struct is_scoped : std::bool_constant<requires
   { requires !std::convertible_to<T, std::underlying_type_t<T>>; }>{};

Returns true if the specified enum type is scoped.

std::cout << std::format("{}\n", conjure_enum<component>::is_scoped());
std::cout << std::format("{}\n", conjure_enum<component1>::is_scoped());

output

true
false

r) is_continuous

static constexpr bool is_continuous();

Returns true if enum range is continuous (no gaps).

std::cout << std::format("{}\n", conjure_enum<numbers>::is_continuous());
std::cout << std::format("{}\n", conjure_enum<component>::is_continuous());

output

true
false

s) type_name

static constexpr std::string_view type_name();

Returns a std::string_view of T.

std::cout << conjure_enum<component>::type_name() << '\n';
std::cout << conjure_enum<component1>::type_name() << '\n';

output

component
component1

t) remove_scope

static constexpr std::string_view remove_scope(std::string_view what);

Returns a std::string_view with scope removed; for unscoped returns unchanged

std::cout << conjure_enum<component>::remove_scope("component::path"sv) << '\n';
std::cout << conjure_enum<component>::remove_scope("path"sv) << '\n';

output

path
path

u) add_scope

static constexpr std::string_view add_scope(std::string_view what);

Returns a std::string_view with scope added to the enum if the supplied enum string is valid but missing scope; for unscoped returns unchanged

std::cout << conjure_enum<component>::add_scope("path"sv) << '\n';
std::cout << conjure_enum<component1>::add_scope("path"sv) << '\n';

output

component::path
path

v) has_scope

static constexpr bool has_scope(std::string_view what);

Returns true if the supplied string representation is scoped (and is valid).

std::cout << std::format("{}\n", conjure_enum<component>::has_scope("component::scheme"));
std::cout << std::format("{}\n", conjure_enum<component>::has_scope("scheme"));
std::cout << std::format("{}\n", conjure_enum<component1>::has_scope("scheme"));

output

true
false
false

w) iterators

static constexpr auto cbegin();
static constexpr auto cend();
static constexpr auto crbegin();
static constexpr auto crend();

These methods return const_iterator and const_reverse_iterator respectively all from entries defined above.

using en = conjure_enum<numbers>;
for (auto itr{en::cbegin()}; itr != en::cend(); ++itr)
   std::cout << static_cast<int>(std::get<0>(*itr)) << ' ' << std::get<1>(*itr) << '\n';

output

0 numbers::zero
1 numbers::one
2 numbers::two
3 numbers::three
4 numbers::four
5 numbers::five
6 numbers::six
7 numbers::seven
8 numbers::eight
9 numbers::nine

x) iterator_adaptor

template<valid_enum T>
struct iterator_adaptor;

This class wraps conjure_enum<T>::entries allowing it to be used in range based for loops:

for (const auto pp : iterator_adaptor<numbers>())
   std::cout << static_cast<int>(std::get<0>(pp)) << '\n';

output

0
1
2
3
4
5
6
7
8
9

y) front, back

static constexpr auto front();
static constexpr auto back();

These methods return *cbegin() and *std::prev(cend()) respectively all from entries defined above.

for (const auto& [ev,str] : {conjure_enum<numbers>::front(), conjure_enum<numbers>::back()})
   std::cout << static_cast<int>(ev) << ' ' << str << '\n';

output

0 numbers::zero
9 numbers::nine

z) ostream_enum_operator

template<typename CharT, typename Traits=std::char_traits<CharT>, valid_enum T>
constexpr std::basic_ostream<CharT, Traits>& operator<<(std::basic_ostream<CharT, Traits>& os, T value);

Provides std::ostream insertion for any enum. You just need to include

using ostream_enum_operator::operator<<;

Examples

using ostream_enum_operator::operator<<;
std::cout << '"' << component::host << '"' << '\n';
std::cout << '"' << component1::host << '"' << '\n';
std::cout << '"' << static_cast<component>(100) << '"' << '\n';

output

"component::host"
"host"
"100"

A) epeek, tpeek

static consteval const char *tpeek();
template<T e>
static consteval const char *epeek();

These functions return std::source_location::current().function_name() as const char* strings for the enum type or enum value. The actual output is implementation dependent. See Results of source_location for implementation specific std::source_location results.

The following code:

std::cout << conjure_enum<component>::tpeek() << '\n';
std::cout << conjure_enum<component>::epeek<component::scheme>() << '\n';

Generates this output with gcc:

static consteval const char* FIX8::conjure_enum<T>::tpeek() [with T = component]
static consteval const char* FIX8::conjure_enum<T>::epeek() [with T e = component::path; T = component]

B) get_enum_min_value, get_enum_max_value, get_actual_enum_min_value and get_actual_enum_max_value

static constexpr int get_enum_min_value();
static constexpr int get_enum_max_value();
static constexpr int get_actual_enum_min_value();
static constexpr int get_actual_enum_max_value();

The first two functions return the min and max enum range for the specified enum. If you have specialised enum_range then these values will be reported (see below).

The second two functions return the actual min and max enum values as ints for the specified enum.

std::cout << conjure_enum<component>::get_enum_min_value() << '/' << conjure_enum<component>::get_enum_min_value() << '\n';
std::cout << conjure_enum<component>::get_actual_enum_min_value() << '/' << conjure_enum<component>::get_actual_enum_min_value() << '\n';

output

-128/127
0/14

C) in_range

static constexpr bool in_range(T value);

Returns true if the given value is within the minimum and maximum defined values for this enum type.

std::cout << std::format("{}\n", conjure_enum<component>::in_range(static_cast<component>(100)));

output

false

4. enum_bitset

enum_bitset is a convenient way of creating bitsets based on std::bitset. It uses your enum (scoped or unscoped) for the bit positions (and names).

Note

  • Your enum sequence must be 0 based
  • Continuous
  • The last value must be less than the count of enumerations

We decided on these restrictions for both simplicity and practicality - bitsets only really make sense when represented in this manner; also...

  • This implementation is limited to 64 bits (arbitrary length impl. soon).

Important

You must include

#include <fix8/conjure_enum.hpp>
#include <fix8/conjure_enum_bitset.hpp>

Your enum must satisfy the following:

template<typename T>
concept valid_bitset_enum = valid_enum<T> and requires(T)
{
  requires conjure_enum<T>::is_continuous();
  requires conjure_enum<T>::get_actual_enum_min_value() == 0;
  requires conjure_enum<T>::get_actual_enum_max_value() < conjure_enum<T>::count();
};

a) Creating an enum_bitset

constexpr enum_bitset() = default;
constexpr enum_bitset(U bits);
constexpr enum_bitset(std::string_view from, bool anyscope=false,
   char sep='|', bool ignore_errors=true);
constexpr enum_bitset(std::bitset<N> from);

template<valid_bitset_enum... E>
constexpr enum_bitset(E... comp);

template<std::integral... I>
constexpr enum_bitset(I... comp);

You can use the enum values directly in your constructor. No need to | them - this is assumed. Just supply them comma separated:

enum class numbers { zero, one, two, three, four, five, six, seven, eight, nine };
enum_bitset<numbers> b(numbers::zero, numbers::one, numbers::two, numbers::three);
std::cout << b << '\n';

output

0000001111

You can use the underlying type as well:

enum_bitset<numbers> b(0,1,2,3);
std::cout << b << '\n';

output

0000001111

You can use an int initialiser too:

enum_bitset<numbers> b(15);
std::cout << b << '\n';

output

0000001111

You can even use a delimited string based on your enum names. Optionally omit the scope and even specify your own delimiter (default is |). Substrings are trimmed of whitespace before lookup.

enum_bitset<numbers> b("numbers::zero|numbers::one|numbers::two|numbers::three");
std::cout << b << '\n';
enum_bitset<numbers> b1("zero,one  ,two,  three", true, ',');
std::cout << b1 << '\n';
enum_bitset<numbers> b2("zero|one|two|three", true);
std::cout << b2 << '\n';

output

0000001111
0000001111
0000001111

A typical use of the above is for parsing configuration bitsets. Here you can tell the constructor to throw an std::invalid_argument if a substring is invalid by specifying false for ignore_errors:

try
{
   enum_bitset<numbers> b("zero,twenty,two,three", true, ',', false);
   std::cout << b << '\n';
}
catch(const std::invalid_argument& e)
{
   std::cerr << "exception: " << e.what() << '\n';
}

output

exception: twenty

You can also create an enum_bitset from a std::bitset of the same number of bits.

std::bitset<10> bs{1 << 1 | 1 << 3 | 1 << 6};
enum_bitset<numbers> ed(bs);
std::cout << ed << '\n';

output

0001001010

b) Standard bit operators

All of the standard operators are supported. Assignment operators return a enum_bitset&, non-assignment operators return a enum_bitset.

Operator Description
& binary AND
| binary OR
^ binary XOR
~ binary NOT (ones' complement)
<< left shift
>> right shift
<<= left shift assign
>>= right shift assign
&= AND assign
|= OR assign
^= XOR assign

Operators work with enum values or integers:

enum_bitset<numbers> b(numbers::zero, numbers::one, numbers::two, numbers::three);
std::cout << b << '\n';
std::cout << (b & 0b111) << '\n';
b ^= numbers::two;
std::cout << b << '\n';

output

0000001111
0000000111
0000001011

c) Standard accessors and mutators

All of the standard accessors and mutators are supported.

Method Description
test test for bit(s)
set set all or 1 bit, optionally set to off
reset reset bits(s)
flip flip bits(s) (ones' complement)
to_ulong convert to unsigned long
to_ullong convert to unsigned long long
count count of bits on
size number of bits in bitset
operator[] set or test bit at position or enum value
any return true if any bit is on
all return true if all bits are on
none return true if no bits are on

Additional methods

Method Description
set set all specified bits, templated
reset reset all specified bits, templated
rotl rotate left specified times
rotr rotate right specified times
countl_zero counts number of consecutive 0 bits, starting from the most significant bit
countl_one counts number of consecutive 1 bits, starting from the most significant bit
countr_zero counts number of consecutive 0 bits, starting from the least significant bit
countr_one counts number of consecutive 1 bits, starting from the least significant bit
any_of test for one or more bits, templated, function, types and underlyings
all_of test for all specified bits, templated, function, types and underlyings
none_of test for all specified bits set to off, templated, function, types and underlyings
not_count complement of count, count of off bits
has_single_bit return true if bitset is an integral power of two

Note

rotl, rotl, countl* and countr* operate on the used bits of the underlying type.

Take a look at the implementation for more detail on the various API functions available. You can also review the unit test cases for examples of use.

All accessors and mutators work with enum values or integers as with operators. They also work with multiple values, either as template parameters or as variadic arguments:

enum_bitset<numbers> eb;
eb.set<numbers::zero,numbers::two,numbers::five,numbers::nine>();
std::cout << eb << '\n';
std::cout << std::boolalpha << eb.all_of<numbers::zero,numbers::two,numbers::five,numbers::nine>() << '\n';
eb.reset<numbers::five,numbers::two>();
std::cout << std::boolalpha << eb.all_of(0, 2, 5, 9) << '\n';
std::cout << std::boolalpha << eb.any_of(0, 2, 5, 9) << '\n';
std::cout << std::boolalpha << eb.all_of(numbers::zero,numbers::nine) << '\n';
std::cout << eb << '\n';
eb.reset(numbers::nine)
std::cout << ec << '\n';
eb.reset();
eb[2] = true;
eb[numbers::three] = true;
std::cout << eb << '\n';
std::cout << eb.rotr(1) << '\n';

output

1000100101
true
false
true
true
1000000001
0000000001
0000001100
0000011000

d) Other functions

i. operator bool

constexpr operator bool() const;

Return true if any bits are on.

if (enum_bitset<numbers> ec(15); ec)
   std::cout << ec << '\n';

output

0001001111

ii. operator std::bitset<N>()

constexpr operator std::bitset<N>() const;

Cast an enum_bitset to a std::bitset with the same number of bits.

enum_bitset<numbers> ec(numbers::one,numbers::three,numbers::six);
std::bitset<10> bs{ec};
std::cout << bs << '\n';

output

0001001010

iii. std::ostream& operator<<, to_string, to_hex_string

friend constexpr std::ostream& operator<<(std::ostream& os, const enum_bitset& what);
constexpr std::string to_string(char zero='0', char one='1') const;

template<bool showbase=true, bool uppercase=false>
constexpr std::string to_hex_string() const;

constexpr std::string to_hex_string() const;

Inserts default string representation into std::ostream.
Returns a std::string representation of the bitset. Optionally specify which characters to use for 0 and 1.
Returns a std::string representation of the bitset in hex format. Optionally specify showbase which will prefix the string with 0x or 0X; optionally specify uppercase which will set the case of the hex digits.

enum_bitset<numbers> ec(numbers::one,numbers::three,numbers::six);
std::cout << ec << '\n';
std::cout << ec.to_string('-', '+') << '\n';
std::cout << ec.to_hex_string() << '\n';
std::cout << ec.to_hex_string<true, true>() << '\n';

output

0001001010
---+--+-+-
0x4a
0X4A

iv. for_each, for_each_n

template<typename Fn, typename... Args>
requires std::invocable<Fn&&, T, Args...>
[[maybe_unused]] constexpr auto for_each(Fn&& func, Args&&... args);

template<typename C, typename Fn, typename... Args> // specialisation for member function with object
requires std::invocable<Fn&&, C, T, Args...>
[[maybe_unused]] constexpr auto for_each(Fn&& func, C *obj, Args&&... args);

template<typename Fn, typename... Args>
requires std::invocable<Fn&&, T, Args...>
[[maybe_unused]] constexpr auto for_each_n(int n, Fn&& func, Args&&... args);

template<typename C, typename Fn, typename... Args> // specialisation for member function with object
requires std::invocable<Fn&&, C, T, Args...>
[[maybe_unused]] constexpr auto for_each_n(int n, Fn&& func, C *obj, Args&&... args);

Call supplied invocable for every bit that is on. Similar to std::for_each except first parameter of your invocable must accept an enum value (passed by for_each). Optionally provide any additional parameters. Works with lambdas, member functions, functions etc. You can limit the number of calls to your invocable by using the for_each_n version with the first parameter being the maximum number to call. The second version of for_each and for_each_n is intended to be used when using a member function - the second parameter passed by your call must be the this pointer of the object. If you wish to pass a reference parameter, you must wrap it in std::ref.

Returns std::bind(std::forward<Fn>(func), std::placeholders::_1, std::forward<Args>(args)...) or std::bind(std::forward<Fn>(func), obj, std::placeholders::_1, std::forward<Args>(args)...) which can be stored or immediately invoked.

To iterate over every bit regardless of whether it is on or not, use conjure_enum<T>::for_each.

Example using member function:

struct foo
{
   void printer(numbers val, std::ostream& ostr) const
   {
      ostr << conjure_enum<numbers>::enum_to_string(val) << '\n';
   }
};
enum_bitset<numbers> ec(numbers::zero,numbers::two,numbers::five,numbers::nine);
const foo bar;
ec.for_each(&foo::printer, &bar, std::ref(std::cout));

output

numbers::zero
numbers::two
numbers::five
numbers::nine

Above example using for_each_n, limiting to 3:

ec.for_each_n(3, &foo::printer, &bar, std::ref(std::cout));

output

numbers::zero
numbers::two
numbers::five

v. Using conjure_enum::dispatch with enum_bitset

Using an enum_bitset wth conjure_enum::dispatch can be a convenient way of iterating through a set of bits to call specific functions using for_each. The following demonstrates this:

const auto dd3
{
   std::to_array<std::tuple<numbers, std::function<void(numbers, int)>>>
   ({
      { numbers::one, [](numbers ev, int a)
         { std::cout << 1000 + a + conjure_enum<numbers>::enum_to_int(ev) << '\n'; } },
      { numbers::two, [](numbers ev, int a)
         { std::cout << 2000 + a + conjure_enum<numbers>::enum_to_int(ev) << '\n'; } },
      { numbers::three, [](numbers ev, int a)
         { std::cout << 3000 + a + conjure_enum<numbers>::enum_to_int(ev) << '\n'; } },
      { static_cast<numbers>(-1), [](numbers ev, [[maybe_unused]] int a)
         { std::cout << "not found: " << conjure_enum<numbers>::enum_to_int(ev) << '\n'; } }, // not found func
   })
};
enum_bitset<numbers>(1,2,3,5).for_each([](numbers val, const auto& arr, int num)
{
   conjure_enum<numbers>::dispatch(val, arr, num);
}, dd3, 100);

output

1101
2102
3103
not found: 5

vi. get_underlying

constexpr U get_underlying() const;

Returns the underlying integral value.

vii. get_underlying_bit_size

constexpr int get_underlying_bit_size() const

Returns the number of bits that the underlying integral contains. Will always be a power of 2 and an integral type. The number of bits may be larger than the count of bits.

viii. get_bit_mask,get_unused_bit_mask

constexpr U get_bit_mask() const;
constexpr U get_unused_bit_mask() const;

Returns a bit mask that would mask off the unused bits of the underlying integral.
Returns a bit mask that would mask off the used bits of the underlying integral.

ix. std::hash<enum_bitset<T>>

template<typename T>
struct std::hash<FIX8::enum_bitset<T>>;

Provides a specialization of std::hash for enum_bitset<T>.


5. conjure_type

conjure_type is a general purpose class allowing you to extract a string representation of any typename. The string will be stored statically by the compiler, so you can use the statically generated value name to obtain your type.

Important

You must include

#include <fix8/conjure_enum.hpp>
#include <fix8/conjure_type.hpp>

a) name

This static member is generated for your type. It is a fixed_string but has a built-in std::string_view operator.

template<typename T>
class conjure_type;
static constexpr fixed_string name;
class foo;
std::cout << std::format("\"{}\"\n", conjure_type<foo>::name);

output

"foo"

Works with aliases:

using test = std::map<std::size_t, std::string_view>;
using test1 = std::map<std::size_t, foo>;
std::cout << conjure_type<test>::name << '\n';
std::cout << conjure_type<test1>::name << '\n';
std::cout << conjure_type<std::underlying_type_t<numbers>>::name << '\n';

output

std::map<long unsigned int, std::basic_string_view<char> >
std::map<long unsigned int, foo>
int

Works with its own types too:

std::cout << conjure_type<conjure_type<conjure_enum<numbers>>>::name << '\n';

output

FIX8::conjure_type<FIX8::conjure_enum<numbers> >

If you need to explicitly obtain a std::string_view, use the get() method on name (not windows sorry):

auto fstrv { conjure_type<test>::name };
auto strv { conjure_type<test>::name.get() };
std::cout << conjure_type<decltype(fstrv)>::name << '\n';
std::cout << conjure_type<decltype(strv)>::name << '\n';

output

fixed_string<58>
std::basic_string_view<char>

Alternatively you can use the as_string_view() method:

auto fstrv { conjure_type<test>::as_string_view() };
std::cout << conjure_type<decltype(fstrv)>::name << '\n';

output

std::basic_string_view<char>

b) as_string_view

Return the name as a std::string_view.

static constexpr std::string_view as_string_view();

c) tpeek

static consteval const char *tpeek();

These functions return std::source_location::current().function_name() as const char* strings for type. The actual output is implementation dependent. See Results of source_location for implementation specific std::source_location results.

The following code:

std::cout << conjure_type<test>::tpeek() << '\n';

Generates this output with gcc:

static consteval const char* FIX8::conjure_type<T>::tpeek() [with T = test]

6. fixed_string

fixed_string is a specialisation of std::array that provides statics storage for an ASCII zero (asciiz) string. The purpose of this class is to allow the creation of constexpr strings with specfic storage, adding a trailing 0. It is used by conjure_enum to store all strings. API is described below.

a) Creating a fixed_string

template<std::size_t N>
class fixed_string;
constexpr fixed_string(std::string_view sv);

Constructs a fixed_string from a std::string_view. The source string is copied and a null character is added to the end. Note the size of the source string must be passed as a template parameter.

std::string_view sv{"The rain in Spain"};
constexpr fixed_string<sv.size()> fs{sv};

b) get

constexpr std::string_view get() const;

Returns the string as a std::string_view.

c) c_str

constexpr const char *c_str() const;

Returns the string as a null terminated const char *.

d) operator std::string_view

constexpr operator std::string_view() const;

Provides a std::string_view cast. Returns the string as a std::string_view.

e) operator[]

constexpr char operator[](size_t idx) const;

Returns the character at the specifdined index. It is not range checked.

f) size

constexpr std::size_t size() const;

Returns the size of the fixed_string including the null terminator.

g) std::ostream& operator<<

std::ostream& operator<<(std::ostream& os, const fixed_string& what)

Provides an ostream insertor.


7. Building

This implementation is header only. Apart from standard C++20 includes there are no external dependencies needed in your application. Catch2 is used for the built-in unit tests.

Tip

The unit test source files unittests.cpp and edgetests.cpp contain additional examples for all the APIs.

a) Obtaining the source, building the unittests and examples

i. Build platform

*nix based environments

To clone and default build the test app, unit tests and the benchmark:

$ git clone https://github.com/fix8mt/conjure_enum.git
$ cd conjure_enum
$ mkdir build
$ cd build
$ cmake ..
$ make -j4
$ ctest (or make test)

Windows environments

Create a new console project. Add the repo https://github.com/fix8mt/conjure_enum.git and clone the source. Make sure you set the C++ language to C++20 in the project preferences. The project should build and run the unit tests by default.

The package is also available on vckpg.

ii. Default compiler warnings

By default all warnings are enabled. To prevent this, pass the following to cmake:

$ cmake -DBUILD_ALL_WARNINGS=false ..

iii. Default unit tests

By default the unit tests are built (which will download Catch2). To prevent this, pass the following to cmake:

$ cmake -DBUILD_UNITTESTS=false ..

iv. Default executable stripping

To disable stripping of the executables:

$ cmake -DBUILD_STRIP_EXE=false ..

v. Clang compilation profiling

To enable clang compilation profiling:

$ cmake -DBUILD_CLANG_PROFILER=true ..

b) Using in your application with cmake

In CMakeLists.txt set your include path to:

include_directories([conjure_enum directory]/include)
# e.g.
set(cjedir /home/dd/prog/conjure_enum)
include_directories(${cjedir}/include)

and just include:

#include <fix8/conjure_enum.hpp>

in your application. Everything in this class is within the namespace FIX8, so you can add:

using namespace FIX8;

c) Integrating conjure_enum in your project with cmake FetchContent

You can use cmake FetchContent to integrate conjure_enum with your project. If your project was called myproj with the sourcefile myproj.cpp then...

project(myproj)
add_executable (myproj myproj.cpp)
set_target_properties(myproj PROPERTIES CXX_STANDARD 20 CXX_STANDARD_REQUIRED true)
message(STATUS "Downloading conjure_enum...")
include(FetchContent)
FetchContent_Declare(conjure_enum GIT_REPOSITORY https://github.com/fix8mt/conjure_enum.git)
FetchContent_MakeAvailable(conjure_enum)
target_include_directories(myproj PRIVATE ${conjure_enum_SOURCE_DIR}/include)

d) Reporting issues

Raise an issue on the github page. The executable srcloctest should be built when you build the package by default. This application does not use any of the conjure_enum library and is designed to report how your compiler handles std::source_location. The actual output is implementation dependent. See Results of source_location for implementation specific std::source_location results. You should attach the output of this application with your issue.

Tip

Use the -m switch with srcloctest to generate github markdown which you can paste directly into the issue.

$ ./srcloctest
Compiler: Clang: Ubuntu Clang 16.0.6 (23ubuntu4)
1. scoped enum
static const char *conjure_type<Namespace_Enum_Type>::tpeek() [T = Namespace_Enum_Type]
static const char *conjure_enum<Namespace_Enum_Type>::epeek() [T = Namespace_Enum_Type, e = Namespace_Enum_Type::Value]
static const char *conjure_enum<Namespace_Enum_Type>::epeek() [T = Namespace_Enum_Type, e = (Namespace_Enum_Type)100]

2. unscoped enum
static const char *conjure_type<Namespace_Enum_Type1>::tpeek() [T = Namespace_Enum_Type1]
static const char *conjure_enum<Namespace_Enum_Type1>::epeek() [T = Namespace_Enum_Type1, e = Value]
static const char *conjure_enum<Namespace_Enum_Type1>::epeek() [T = Namespace_Enum_Type1, e = (Namespace_Enum_Type1)100]

3. scoped enum in anonymous namespace
static const char *conjure_type<(anonymous namespace)::Anon_Enum_Type>::tpeek() [T = (anonymous namespace)::Anon_Enum_Type]
static const char *conjure_enum<(anonymous namespace)::Anon_Enum_Type>::epeek() [T = (anonymous namespace)::Anon_Enum_Type, e = (anonymous namespace)::Anon_Enum_Type::Value]
static const char *conjure_enum<(anonymous namespace)::Anon_Enum_Type>::epeek() [T = (anonymous namespace)::Anon_Enum_Type, e = ((anonymous namespace)::Anon_Enum_Type)100]

4. unscoped enum in anonymous namespace
static const char *conjure_type<(anonymous namespace)::Anon_Enum_Type1>::tpeek() [T = (anonymous namespace)::Anon_Enum_Type1]
static const char *conjure_enum<(anonymous namespace)::Anon_Enum_Type1>::epeek() [T = (anonymous namespace)::Anon_Enum_Type1, e = (anonymous namespace)::Value]
static const char *conjure_enum<(anonymous namespace)::Anon_Enum_Type1>::epeek() [T = (anonymous namespace)::Anon_Enum_Type1, e = ((anonymous namespace)::Anon_Enum_Type1)100]

5. scoped enum in namespace
static const char *conjure_type<Namespace::Namespace_Enum_Type>::tpeek() [T = Namespace::Namespace_Enum_Type]
static const char *conjure_enum<Namespace::Namespace_Enum_Type>::epeek() [T = Namespace::Namespace_Enum_Type, e = Namespace::Namespace_Enum_Type::Value]
static const char *conjure_enum<Namespace::Namespace_Enum_Type>::epeek() [T = Namespace::Namespace_Enum_Type, e = (Namespace::Namespace_Enum_Type)100]

6. unscoped enum in namespace
static const char *conjure_type<Namespace::Namespace_Enum_Type1>::tpeek() [T = Namespace::Namespace_Enum_Type1]
static const char *conjure_enum<Namespace::Namespace_Enum_Type1>::epeek() [T = Namespace::Namespace_Enum_Type1, e = Namespace::Value]
static const char *conjure_enum<Namespace::Namespace_Enum_Type1>::epeek() [T = Namespace::Namespace_Enum_Type1, e = (Namespace::Namespace_Enum_Type1)100]

7. types in named and anonymous namespaces
static const char *conjure_type<Foo>::tpeek() [T = Foo]
static const char *conjure_type<Namespace::Namespace_Foo>::tpeek() [T = Namespace::Namespace_Foo]
static const char *conjure_type<(anonymous namespace)::Anon_Foo>::tpeek() [T = (anonymous namespace)::Anon_Foo]

8. other types
static const char *conjure_type<int>::tpeek() [T = int]
static const char *conjure_type<std::basic_string_view<char>>::tpeek() [T = std::basic_string_view<char>]
static const char *conjure_type<std::vector<std::tuple<int, char, std::basic_string_view<char>>>>::tpeek() [T = std::vector<std::tuple<int, char, std::basic_string_view<char>>>]

9. edge enum types
static const char *conjure_type<(anonymous namespace)::NineEnums>::tpeek() [T = (anonymous namespace)::NineEnums]
static const char *conjure_type<(anonymous namespace)::NineEnums1>::tpeek() [T = (anonymous namespace)::NineEnums1]
static const char *conjure_type<TEST::NineEnums>::tpeek() [T = TEST::NineEnums]
static const char *conjure_type<TEST::NineEnums1>::tpeek() [T = TEST::NineEnums1]
static const char *conjure_type<(anonymous namespace)::TEST1::NineEnums>::tpeek() [T = (anonymous namespace)::TEST1::NineEnums]
static const char *conjure_type<(anonymous namespace)::TEST1::NineEnums1>::tpeek() [T = (anonymous namespace)::TEST1::NineEnums1]
static const char *conjure_type<TEST::TEST1::NineEnums>::tpeek() [T = TEST::TEST1::NineEnums]
static const char *conjure_type<TEST::TEST1::NineEnums1>::tpeek() [T = TEST::TEST1::NineEnums1]
$

e) Contributing

Contributions are welcome. Make your changes in your fork on the dev branch and open a pull request from there. PRs to master will not be considered.


8. Notes

a) enum limits

Compilation times increase with the number of enums that use conjure_enum in any compilation unit.

  1. For a simple project with few enums, there is probably no need to set any limits;
  2. Where the enum is defined elsewhere (say if you are using std::errc) then use enum_range or one of the convenience macros;
  3. Where the enum is unscoped then use enum_range or one of the convenience macros;
  4. Where you have defined the enum yourself and it is a scoped enum, use T::ce_first and T::ce_last, or 2.

i. FIX8_CONJURE_ENUM_MIN_VALUE, FIX8_CONJURE_ENUM_MAX_VALUE

These are set by default unless you override them by defining them in your application. They are the global range default for all enums using conjure_enum.

Important

If you want to define these values they must appear before you include conjure_enum.hpp.

The following are the default settings:

#if not defined FIX8_CONJURE_ENUM_MIN_VALUE
# define FIX8_CONJURE_ENUM_MIN_VALUE -128
#endif
#if not defined FIX8_CONJURE_ENUM_MAX_VALUE
# define FIX8_CONJURE_ENUM_MAX_VALUE 127
#endif

These definitions set the minimum and maximum enum values that are supported. You can adjust them to suit your requirements but for most use cases the defaults are sufficient.

Tip

You can reduce compile times in some circumstances by narrowing the range of FIX8_CONJURE_ENUM_MIN_VALUE and FIX8_CONJURE_ENUM_MAX_VALUE. For example if your enums are only within the range of say 0-16 you can set FIX8_CONJURE_ENUM_MIN_VALUE and FIX8_CONJURE_ENUM_MAX_VALUE to 0 and 16 respectively. If the range is too narrow conjure_enum will ignore enum values outside your range.

Tip

If you wish to set ranges on a per enum basis, use enum_range (see below).

ii. using enum_range

You can specialise this class to override the defaults and set your own range on a per enum basis.

template<valid_enum T>
struct enum_range
{
   static constexpr int min{FIX8_CONJURE_ENUM_MIN_VALUE}, max{FIX8_CONJURE_ENUM_MAX_VALUE};
};

The min and max values are used to set the range of enum values for enums in conjure_enum. As shown above, the default values will be FIX8_CONJURE_ENUM_MIN_VALUE and FIX8_CONJURE_ENUM_MAX_VALUE.

enum class range_test { first, second, third, fourth, fifth, sixth, seventh, eighth };
template<>
struct FIX8::enum_range<range_test>
{
   static constexpr int min{0}, max{7};
};
static_assert(conjure_enum<range_test>::get_enum_min_value() == 0);
static_assert(conjure_enum<range_test>::get_enum_max_value() == 7);

ii.a FIX8_CONJURE_ENUM_SET_RANGE_INTS, FIX8_CONJURE_ENUM_SET_RANGE

For convenience, two macros are provided to make it easier to set custom ranges using enum_range.

#define FIX8_CONJURE_ENUM_SET_RANGE_INTS(ec,min_int,max_int)...
#define FIX8_CONJURE_ENUM_SET_RANGE(min_enum,max_enum)...

The first macro takes an enum typename followed by a lower and upper int range value.

The second macro takes a lower and upper enum value. For example:

FIX8_CONJURE_ENUM_SET_RANGE_INTS(numbers, 0, 7)
FIX8_CONJURE_ENUM_SET_RANGE(numbers::first, numbers::eighth)

iii. using T::ce_first and T::ce_last

Another approach to setting a custom range for an enum is to alias the first and last enum values in your enum definition using ce_first and ce_last. conjure_enum will use these values to set the enum range. You can set either, both or neither. A range value not set will default to the FIX8_CONJURE_ENUM_MIN_VALUE or FIX8_CONJURE_ENUM_MAX_VALUE.

Warning

With unscoped enums, there can only be one enum type with ce_first and ce_last defined in any translation unit (ODR). It is therefore recommended to only use this feature with scoped enums.

For example:

enum class range_test
{
   first, second, third, fourth, fifth, sixth, seventh, eighth,
   ce_first=first, ce_last=eighth
};
using rt = conjure_enum<range_test>;
std::cout << rt::get_enum_min_value() << '/' << rt::get_enum_max_value() << '\n';
std::cout << rt::get_actual_enum_min_value() << '/' << rt::get_actual_enum_max_value() << '\n';

output

0/7
0/7

b) Choosing the minimal build

#define FIX8_CONJURE_ENUM_MINIMAL

You can select a minimal version of conjure_enum by defining FIX8_CONJURE_ENUM_MINIMAL before you include conjure_enum.hpp

This limits the API to a more basic set of functionality. This will reduce compile times. Static structures and API calls that will be excluded are:

scoped_entries
unscoped_entries
rev_scoped_entries
names
unscoped_names
remove_scope
add_scope
unscoped_string_to_enum
for_each,for_each_n
type_name
dispatch
iterators
enum_to_string //noscope option not available

These are marked in the API documentation above.

c) Continuous enum optimization

#define FIX8_CONJURE_ENUM_IS_CONTINUOUS

If your enum(s) are continuous (no gaps) you can enable this compiler optimization by defining FIX8_CONJURE_ENUM_IS_CONTINUOUS before you include conjure_enum.hpp. Our testing shows a reduction in overall compile times. All enums using conjure_enum.hpp in the current compilation unit must be continuous.

d) Anonymous enum optimization

#define FIX8_CONJURE_ENUM_NO_ANON

If your enum(s) are not within any anonymous namespaces (rarely used for this purpose), you can enable this compiler optimization by defining FIX8_CONJURE_ENUM_NO_ANON before you include conjure_enum.hpp. Our testing shows a reduction in overall compile times. All enums using conjure_enum.hpp in the current compilation unit must be continuous.

e) Enable all optimizations

#define FIX8_CONJURE_ENUM_ALL_OPTIMIZATIONS

You can enable all optimizations described above by defining FIX8_CONJURE_ENUM_ALL_OPTIMIZATIONS before you include conjure_enum.hpp. This is the equivalent of defining:

#define FIX8_CONJURE_ENUM_MINIMAL
#define FIX8_CONJURE_ENUM_IS_CONTINUOUS
#define FIX8_CONJURE_ENUM_NO_ANON

f) Class conjure_enum is not constructible

All methods in this class are static. You cannot instantiate an object of this type. The same goes for conjure_type.

g) It's not real reflection

This library provides a workaround (hack 😏) to current limitations of C++. There are proposals out there for future versions of the language that will provide proper reflection. See Reflection TS and Reflection for C++26 for examples of some of these.

h) Use of std::string_view

All of the generated static strings and generated static tables obtained by std::source_location use the library defined fixed_string. No string copying is done at runtime, resulting in a single static string in your application. All conjure_enum methods that return strings only return std::string_view. To demonstrate this, lets look at the supplied test application statictest:

#include <iostream>
#define FIX8_CONJURE_ENUM_MINIMAL
#include <fix8/conjure_enum.hpp>
enum class component : int { scheme, authority, userinfo, user, password, host, port, path, query, fragment };
FIX8_CONJURE_ENUM_SET_RANGE(component::scheme, component::fragment)
int main(void)
{
   for(const auto& [a, b] : conjure_enum<component>::entries)
      std::cout << conjure_enum<component>::enum_to_int(a) << ' ' << b << '\n';
   std::cout << static_cast<int>(conjure_enum<component>::string_to_enum("component::path").value()) << '\n';
   std::cout << conjure_enum<component>::get_enum_min_value() << '/' << conjure_enum<component>::get_enum_max_value() << '\n';
   return 0;
}

output

$ ./statictest
0 component::scheme
1 component::authority
2 component::userinfo
3 component::user
4 component::password
5 component::host
6 component::port
7 component::path
8 component::query
9 component::fragment
component::scheme
component::authority
component::userinfo
component::user
component::password
component::host
component::port
component::path
component::query
component::fragment
7
0/9
$

The default build of statictest performs a strip on the executable. Then we run strings on the executable.

shell output

$ strings statictest
/lib64/ld-linux-x86-64.so.2
__gmon_start__
_ITM_deregisterTMCloneTable
_ITM_registerTMCloneTable
_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_c
_ZNSolsEi
_ZSt21ios_base_library_initv
_ZSt16__ostream_insertIcSt11char_traitsIcEERSt13basic_ostreamIT_T0_ES6_PKS3_l
_ZNSo3putEc
_ZSt4cout
__stack_chk_fail
__libc_start_main
__cxa_finalize
libstdc++.so.6
libc.so.6
GLIBC_2.4
GLIBC_2.34
GLIBC_2.2.5
GLIBCXX_3.4.32
GLIBCXX_3.4.9
GLIBCXX_3.4
ATUH
[]A\A]A^
PTE1
u+UH
component::fragment
component::query
component::path
component::port
component::host
component::password
component::user
component::userinfo
component::authority
component::scheme
9*3$"
GCC: (Ubuntu 13.2.0-23ubuntu4) 13.2.0
.shstrtab
.interp
.note.gnu.property
.note.gnu.build-id
.note.ABI-tag
.gnu.hash
.dynsym
.dynstr
.gnu.version
.gnu.version_r
.rela.dyn
.rela.plt
.init
.plt.got
.plt.sec
.text
.fini
.rodata
.eh_frame_hdr
.eh_frame
.init_array
.fini_array
.data.rel.ro
.dynamic
.data
.bss
.comment
$

It can be observed that there is only one copy of the scoped enum value string in the executable.

i) Compilation profiling (Clang)

You can profile the compile time for Clang (other compilers TBA). Firstly install ClangBuildAnalyzer. Then configure conjure_enum with:

cmake -DBUILD_CLANG_PROFILER=true ..

You can follow the instructions given on the ClangBuildAnalyzer page to run. Alternatively after you build the included program cbenchmark, run the included script cbenchmark.sh. The script expects the following environment variables:

Variable Description
ClangBuildAnalyzerLoc directory where ClangBuildAnalyzer can be found
ArtifactLoc directory where conjure_enum is built

For example, if ClangBuildAnalyzer was built in ~/prog/ClangBuildAnalyzer/build and your conjure_enum build was in ./build_clang, then you would run the script from the conjure_enum directory as follows:

ClangBuildAnalyzerLoc=~/prog/ClangBuildAnalyzer/build ArtifactLoc=build_clang examples/cbenchmark.sh

The results will be printed to the screen. For example:

output

Processing all files and saving to 'cbenchmark.dat'...
  done in 0.0s. Run 'ClangBuildAnalyzer --analyze cbenchmark.dat' to analyze it.
Analyzing build trace from 'cbenchmark.dat'...
**** Time summary:
Compilation (2 times):
  Parsing (frontend):            0.4 s
  Codegen & opts (backend):      0.0 s

**** Files that took longest to parse (compiler frontend):
   423 ms: build_clang/CMakeFiles/cbenchmark.dir/examples/cbenchmark.cpp.o

**** Templates that took longest to instantiate:
   187 ms: FIX8::conjure_enum<std::errc> (1 times, avg 187 ms)
   119 ms: FIX8::conjure_enum<std::errc>::_entries<0UL, 1UL, 2UL, 3UL, 4UL, 5UL... (1 times, avg 119 ms)
     8 ms: FIX8::conjure_enum<std::errc>::_sorted_entries (1 times, avg 8 ms)
     6 ms: std::sort<std::tuple<std::errc, std::basic_string_view<char>> *, boo... (1 times, avg 6 ms)
     6 ms: std::__sort<std::tuple<std::errc, std::basic_string_view<char>> *, _... (1 times, avg 6 ms)
     5 ms: std::__introsort_loop<std::tuple<std::errc, std::basic_string_view<c... (1 times, avg 5 ms)
     3 ms: std::array<std::tuple<std::errc, std::basic_string_view<char>>, 72> (1 times, avg 3 ms)
     2 ms: std::tuple<std::basic_string_view<char>, char, std::basic_string_vie... (1 times, avg 2 ms)
     2 ms: std::__unguarded_partition_pivot<std::tuple<std::errc, std::basic_st... (1 times, avg 2 ms)
     2 ms: std::__move_median_to_first<std::tuple<std::errc, std::basic_string_... (1 times, avg 2 ms)
     2 ms: std::iter_swap<std::tuple<std::errc, std::basic_string_view<char>> *... (1 times, avg 2 ms)
     2 ms: std::__partial_sort<std::tuple<std::errc, std::basic_string_view<cha... (1 times, avg 2 ms)
     2 ms: std::tuple<std::errc, std::basic_string_view<char>> (1 times, avg 2 ms)
     2 ms: std::__heap_select<std::tuple<std::errc, std::basic_string_view<char... (1 times, avg 2 ms)
     2 ms: std::__make_heap<std::tuple<std::errc, std::basic_string_view<char>>... (1 times, avg 2 ms)
     1 ms: std::optional<unsigned long> (1 times, avg 1 ms)
     1 ms: std::to_array<std::tuple<std::basic_string_view<char>, char, std::ba... (1 times, avg 1 ms)
     1 ms: std::basic_string<char32_t> (1 times, avg 1 ms)
     1 ms: std::basic_string<char8_t> (1 times, avg 1 ms)
     1 ms: std::__adjust_heap<std::tuple<std::errc, std::basic_string_view<char... (1 times, avg 1 ms)
     1 ms: std::basic_string<char16_t> (1 times, avg 1 ms)
     1 ms: std::basic_string<wchar_t> (1 times, avg 1 ms)
     1 ms: std::basic_string<char> (1 times, avg 1 ms)
     1 ms: std::__and_<std::__is_swappable<std::errc>, std::__is_swappable<std:... (1 times, avg 1 ms)
     1 ms: std::_Tuple_impl<0, std::basic_string_view<char>, char, std::basic_s... (1 times, avg 1 ms)
     1 ms: std::__final_insertion_sort<std::tuple<std::errc, std::basic_string_... (1 times, avg 1 ms)
     1 ms: std::basic_string<char32_t>::_M_construct<const char32_t *> (1 times, avg 1 ms)
     1 ms: std::basic_string<char8_t>::_M_construct<const char8_t *> (1 times, avg 1 ms)
     1 ms: std::basic_string<char16_t>::_M_construct<const char16_t *> (1 times, avg 1 ms)
     1 ms: std::operator+<char, std::char_traits<char>, std::allocator<char>> (1 times, avg 1 ms)

**** Template sets that took longest to instantiate:
   187 ms: FIX8::conjure_enum<$> (1 times, avg 187 ms)
   119 ms: FIX8::conjure_enum<$>::_entries<$> (1 times, avg 119 ms)
    49 ms: FIX8::conjure_enum<$>::_get_name<$> (72 times, avg 0 ms)
    12 ms: std::tuple<$>::tuple<$> (21 times, avg 0 ms)
     8 ms: FIX8::conjure_enum<$>::_sorted_entries (1 times, avg 8 ms)
     7 ms: std::basic_string<$> (5 times, avg 1 ms)
     6 ms: std::sort<$> (1 times, avg 6 ms)
     6 ms: std::__sort<$> (1 times, avg 6 ms)
     5 ms: std::basic_string<$>::_M_construct<$> (5 times, avg 1 ms)
     5 ms: std::tuple<$> (2 times, avg 2 ms)
     5 ms: std::__introsort_loop<$> (1 times, avg 5 ms)
     4 ms: std::array<$> (2 times, avg 2 ms)
     2 ms: std::__unguarded_partition_pivot<$> (1 times, avg 2 ms)
     2 ms: std::__and_<$> (3 times, avg 0 ms)
     2 ms: std::__move_median_to_first<$> (1 times, avg 2 ms)
     2 ms: std::iter_swap<$> (1 times, avg 2 ms)
     2 ms: std::__partial_sort<$> (1 times, avg 2 ms)
     2 ms: std::__heap_select<$> (1 times, avg 2 ms)
     2 ms: std::__make_heap<$> (1 times, avg 2 ms)
     1 ms: std::_Tuple_impl<$> (2 times, avg 0 ms)
     1 ms: std::optional<$> (1 times, avg 1 ms)
     1 ms: std::to_array<$> (1 times, avg 1 ms)
     1 ms: std::__adjust_heap<$> (1 times, avg 1 ms)
     1 ms: std::basic_string<$>::basic_string (2 times, avg 0 ms)
     1 ms: std::__final_insertion_sort<$> (1 times, avg 1 ms)
     1 ms: std::operator+<char, std::char_traits<char>, std::allocator<char>> (1 times, avg 1 ms)
     1 ms: std::__insertion_sort<$> (1 times, avg 1 ms)
     0 ms: FIX8::conjure_enum<$>::_tuple_comp_rev (1 times, avg 0 ms)
     0 ms: std::tuple<std::errc, std::basic_string_view<char>>::operator= (1 times, avg 0 ms)
     0 ms: __gnu_cxx::__to_xstring<$> (1 times, avg 0 ms)

**** Functions that took longest to compile:
     0 3s: test_conjure_enum(std::errc) (/home/davidd/prog/conjure_enum_tclass/examples/cbenchmark.cpp)

**** Function sets that took longest to compile / optimize:

**** Expensive headers:
166 ms: /usr/include/c++/14/system_error (included 1 times, avg 166 ms), included via:
  1x: <direct include>

54 ms: /home/davidd/prog/conjure_enum_tclass/include/fix8/conjure_enum.hpp (included 1 times, avg 54 ms), included via:
  1x: <direct include>

  done in 0.0s.


9. Benchmarks

We have benchmarked compilation times for conjure_enum and magic_enum. For magic_enum we created a separate repo (see here).

Compiler conjure_enum (secs) magic_enum (secs) Notes
MSVC 0.376 0.343 using cl from command prompt
command cl /nologo /MD /std:c++latest /Bt+ /I ..\include ..\examples\cbenchmark.cpp|find "c1xx.dll" cl /nologo /MD /std:c++latest /Bt+ -I build\_deps\magic_enum-src\include cbenchmark.cpp|find "c1xx.dll"
clang 0.3 0.3 using ClangBuildAnalyzer
command make; ClangBuildAnalyzerLoc=~/prog/ClangBuildAnalyzer/build ArtifactLoc=build_clang examples/cbenchmark.sh make; ClangBuildAnalyzerLoc=~/prog/ClangBuildAnalyzer/build ArtifactLoc=build_clang ./cbenchmark.sh

Notes

  • Benchmark run 10 times, best result shown
  • Both benchmarks are using cbenchmark.sh and cbenchmark.cpp
  • MSVC: Windows 11 ThinkCentre 16x 13th Gen Intel i7-13700, 32Gb; MSVC 2022 / 17.11.0.
  • Clang: Ubuntu 24.04 12th Gen Intel i9-12900T, 32Gb; Clang 18.1.3
  • magic_enum: single header only
  • conjure_enum: minimal build

Discussion

For MSVC, magic_enum compilation times a slighly better than conjure_enum (around %9). For clang the results are identical. From a compilation performance perspective, conjure_enum roughly matches the performance of magic_enum.


10. Compiler support

Compiler Version(s) Notes Unsupported
gcc 11, 12, 13, 14 std::format not complete in 11, 12 <= 10
clang 15, 16, 17, 18 Catch2 needs cxx_std_20 in 15 <= 14
msvc 16, 17 Visual Studio 2019,2022, latest 17.11.3 <= 16.9
xcode 15 Apple Xcode Clang 15.0.0 (LLVM 16), some issues with constexpr, workarounds <= 14

11. Compiler issues

Compiler Version(s) Issues Workaround
clang 16, 17, 18 Compiler reports integers outside valid range [x,y] specify underlying type when declaring enum eg. enum class foo : int

Footnotes

  1. © 2019 - 2024 Daniil Goncharov

  2. © 2024 Fix8 Market Technologies Pty Ltd, David L. Dight. Logo by Adrian An.