[c] Easy way to use variables of enum types as string in C?

Here's what I am trying to do:

typedef enum { ONE, TWO, THREE } Numbers;

I am trying to write a function that would do a switch case similar to the following:

char num_str[10];
int process_numbers_str(Numbers num) {
  switch(num) {
    case ONE:
    case TWO:
    case THREE:
    {
      strcpy(num_str, num); //some way to get the symbolic constant name in here?
    } break;
    default:
      return 0; //no match
  return 1;
}

Instead of defining at every case, is there a way to set it using the enum variable like I am trying to do above?

The use of boost::preprocessor makes possible an elegant solution like the following:

Step 1: include the header file:

#include "EnumUtilities.h"

Step 2: declare the enumeration object with the following syntax:

MakeEnum( TestData,
         (x)
         (y)
         (z)
         );

Step 3: use your data:

Getting the number of elements:

td::cout << "Number of Elements: " << TestDataCount << std::endl;

Getting the associated string:

std::cout << "Value of " << TestData2String(x) << " is " << x << std::endl;
std::cout << "Value of " << TestData2String(y) << " is " << y << std::endl;
std::cout << "Value of " << TestData2String(z) << " is " << z << std::endl;

Getting the enum value from the associated string:

std::cout << "Value of x is " << TestData2Enum("x") << std::endl;
std::cout << "Value of y is " << TestData2Enum("y") << std::endl;
std::cout << "Value of z is " << TestData2Enum("z") << std::endl;

This looks clean and compact, with no extra files to include. The code I wrote within EnumUtilities.h is the following:

#include <boost/preprocessor/seq/for_each.hpp>
#include <string>

#define REALLY_MAKE_STRING(x) #x
#define MAKE_STRING(x) REALLY_MAKE_STRING(x)
#define MACRO1(r, data, elem) elem,
#define MACRO1_STRING(r, data, elem)    case elem: return REALLY_MAKE_STRING(elem);
#define MACRO1_ENUM(r, data, elem)      if (REALLY_MAKE_STRING(elem) == eStrEl) return elem;


#define MakeEnum(eName, SEQ) \
    enum eName { BOOST_PP_SEQ_FOR_EACH(MACRO1, , SEQ) \
    last_##eName##_enum}; \
    const int eName##Count = BOOST_PP_SEQ_SIZE(SEQ); \
    static std::string eName##2String(const enum eName eel) \
    { \
        switch (eel) \
        { \
        BOOST_PP_SEQ_FOR_EACH(MACRO1_STRING, , SEQ) \
        default: return "Unknown enumerator value."; \
        }; \
    }; \
    static enum eName eName##2Enum(const std::string eStrEl) \
    { \
        BOOST_PP_SEQ_FOR_EACH(MACRO1_ENUM, , SEQ) \
        return (enum eName)0; \
    };

There are some limitation, i.e. the ones of boost::preprocessor. In this case, the list of constants cannot be larger than 64 elements.

Following the same logic, you could also think to create sparse enum:

#define EnumName(Tuple)                 BOOST_PP_TUPLE_ELEM(2, 0, Tuple)
#define EnumValue(Tuple)                BOOST_PP_TUPLE_ELEM(2, 1, Tuple)
#define MACRO2(r, data, elem)           EnumName(elem) EnumValue(elem),
#define MACRO2_STRING(r, data, elem)    case EnumName(elem): return BOOST_PP_STRINGIZE(EnumName(elem));

#define MakeEnumEx(eName, SEQ) \
    enum eName { \
    BOOST_PP_SEQ_FOR_EACH(MACRO2, _, SEQ) \
    last_##eName##_enum }; \
    const int eName##Count = BOOST_PP_SEQ_SIZE(SEQ); \
    static std::string eName##2String(const enum eName eel) \
    { \
        switch (eel) \
        { \
        BOOST_PP_SEQ_FOR_EACH(MACRO2_STRING, _, SEQ) \
        default: return "Unknown enumerator value."; \
        }; \
    };  

In this case, the syntax is:

MakeEnumEx(TestEnum,
           ((x,))
           ((y,=1000))
           ((z,))
           );

Usage is similar as above (minus the eName##2Enum function, that you could try to extrapolate from the previous syntax).

I tested it on mac and linux, but be aware that boost::preprocessor may not be fully portable.

If you are using gcc, it's possible to use:

const char * enum_to_string_map[]={ [enum1]='string1', [enum2]='string2'};

Then just call for instance

enum_to_string_map[enum1]

I know you have a couple good solid answers, but do you know about the # operator in the C preprocessor?

It lets you do this:

#define MACROSTR(k) #k

typedef enum {
    kZero,
    kOne,
    kTwo,
    kThree
} kConst;

static char *kConstStr[] = {
    MACROSTR(kZero),
    MACROSTR(kOne),
    MACROSTR(kTwo),
    MACROSTR(kThree)
};

static void kConstPrinter(kConst k)
{
    printf("%s", kConstStr[k]);
}

Making something both a C identifier and a string

C or C++ does not provide this functionality, although I've needed it often.

The following code works, although it's best suited for non-sparse enums.

typedef enum { ONE, TWO, THREE } Numbers;
char *strNumbers[] = {"one","two","three"};
printf ("Value for TWO is %s\n",strNumbers[TWO]);

By non-sparse, I mean not of the form

typedef enum { ONE, FOUR_THOUSAND = 4000 } Numbers;

since that has huge gaps in it.

The advantage of this method is that it put the definitions of the enums and strings near each other; having a switch statement in a function spearates them. This means you're less likely to change one without the other.

If the enum index is 0-based, you can put the names in an array of char*, and index them with the enum value.

Try Converting C++ enums to strings. The comments have improvements that solve the problem when enum items have arbitrary values.

Check out the ideas at Mu Dynamics Research Labs - Blog Archive. I found this earlier this year - I forget the exact context where I came across it - and have adapted it into this code. We can debate the merits of adding an E at the front; it is applicable to the specific problem addressed, but not part of a general solution. I stashed this away in my 'vignettes' folder - where I keep interesting scraps of code in case I want them later. I'm embarrassed to say that I didn't keep a note of where this idea came from at the time.

Header: paste1.h

/*
@(#)File:           $RCSfile: paste1.h,v $
@(#)Version:        $Revision: 1.1 $
@(#)Last changed:   $Date: 2008/05/17 21:38:05 $
@(#)Purpose:        Automated Token Pasting
*/

#ifndef JLSS_ID_PASTE_H
#define JLSS_ID_PASTE_H

/*
 * Common case when someone just includes this file.  In this case,
 * they just get the various E* tokens as good old enums.
 */
#if !defined(ETYPE)
#define ETYPE(val, desc) E##val,
#define ETYPE_ENUM
enum {
#endif /* ETYPE */

   ETYPE(PERM,  "Operation not permitted")
   ETYPE(NOENT, "No such file or directory")
   ETYPE(SRCH,  "No such process")
   ETYPE(INTR,  "Interrupted system call")
   ETYPE(IO,    "I/O error")
   ETYPE(NXIO,  "No such device or address")
   ETYPE(2BIG,  "Arg list too long")

/*
 * Close up the enum block in the common case of someone including
 * this file.
 */
#if defined(ETYPE_ENUM)
#undef ETYPE_ENUM
#undef ETYPE
ETYPE_MAX
};
#endif /* ETYPE_ENUM */

#endif /* JLSS_ID_PASTE_H */

Example source:

/*
@(#)File:           $RCSfile: paste1.c,v $
@(#)Version:        $Revision: 1.2 $
@(#)Last changed:   $Date: 2008/06/24 01:03:38 $
@(#)Purpose:        Automated Token Pasting
*/

#include "paste1.h"

static const char *sys_errlist_internal[] = {
#undef JLSS_ID_PASTE_H
#define ETYPE(val, desc) desc,
#include "paste1.h"
    0
#undef ETYPE
};

static const char *xerror(int err)
{
    if (err >= ETYPE_MAX || err <= 0)
        return "Unknown error";
    return sys_errlist_internal[err];
}

static const char*errlist_mnemonics[] = {
#undef JLSS_ID_PASTE_H
#define ETYPE(val, desc) [E ## val] = "E" #val,
#include "paste1.h"
#undef ETYPE
};

#include <stdio.h>

int main(void)
{
    int i;

    for (i = 0; i < ETYPE_MAX; i++)
    {
        printf("%d: %-6s: %s\n", i, errlist_mnemonics[i], xerror(i));
    }
    return(0);
}

Not necessarily the world's cleanest use of the C pre-processor - but it does prevent writing the material out multiple times.

The technique from Making something both a C identifier and a string? can be used here.

As usual with such preprocessor stuff, writing and understanding the preprocessor part can be hard, and includes passing macros to other macros and involves using # and ## operators, but using it is real easy. I find this style very useful for long enums, where maintaining the same list twice can be really troublesome.

Factory code - typed only once, usually hidden in the header:

enumFactory.h:

// expansion macro for enum value definition
#define ENUM_VALUE(name,assign) name assign,

// expansion macro for enum to string conversion
#define ENUM_CASE(name,assign) case name: return #name;

// expansion macro for string to enum conversion
#define ENUM_STRCMP(name,assign) if (!strcmp(str,#name)) return name;

/// declare the access function and define enum values
#define DECLARE_ENUM(EnumType,ENUM_DEF) \
  enum EnumType { \
    ENUM_DEF(ENUM_VALUE) \
  }; \
  const char *GetString(EnumType dummy); \
  EnumType Get##EnumType##Value(const char *string); \

/// define the access function names
#define DEFINE_ENUM(EnumType,ENUM_DEF) \
  const char *GetString(EnumType value) \
  { \
    switch(value) \
    { \
      ENUM_DEF(ENUM_CASE) \
      default: return ""; /* handle input error */ \
    } \
  } \
  EnumType Get##EnumType##Value(const char *str) \
  { \
    ENUM_DEF(ENUM_STRCMP) \
    return (EnumType)0; /* handle input error */ \
  } \

Factory used

someEnum.h:

#include "enumFactory.h"
#define SOME_ENUM(XX) \
    XX(FirstValue,) \
    XX(SecondValue,) \
    XX(SomeOtherValue,=50) \
    XX(OneMoreValue,=100) \

DECLARE_ENUM(SomeEnum,SOME_ENUM)

someEnum.cpp:

#include "someEnum.h"
DEFINE_ENUM(SomeEnum,SOME_ENUM)

The technique can be easily extended so that XX macros accepts more arguments, and you can also have prepared more macros to substitute for XX for different needs, similar to the three I have provided in this sample.

Comparison to X-Macros using #include / #define / #undef

While this is similar to X-Macros others have mentioned, I think this solution is more elegant in that it does not require #undefing anything, which allows you to hide more of the complicated stuff is in the factory the header file - the header file is something you are not touching at all when you need to define a new enum, therefore new enum definition is a lot shorter and cleaner.

I have created a simple templated class streamable_enum that uses stream operators << and >> and is based on the std::map<Enum, std::string>:

#ifndef STREAMABLE_ENUM_HPP
#define STREAMABLE_ENUM_HPP

#include <iostream>
#include <string>
#include <map>

template <typename E>
class streamable_enum
{
public:
    typedef typename std::map<E, std::string> tostr_map_t;
    typedef typename std::map<std::string, E> fromstr_map_t;

    streamable_enum()
    {}

    streamable_enum(E val) :
        Val_(val)
    {}

    operator E() {
        return Val_;
    }

    bool operator==(const streamable_enum<E>& e) {
        return this->Val_ == e.Val_;
    }

    bool operator==(const E& e) {
        return this->Val_ == e;
    }

    static const tostr_map_t& to_string_map() {
        static tostr_map_t to_str_(get_enum_strings<E>());
        return to_str_;
    }

    static const fromstr_map_t& from_string_map() {
        static fromstr_map_t from_str_(reverse_map(to_string_map()));
        return from_str_;
    }
private:
    E Val_;

    static fromstr_map_t reverse_map(const tostr_map_t& eToS) {
        fromstr_map_t sToE;
        for (auto pr : eToS) {
            sToE.emplace(pr.second, pr.first);
        }
        return sToE;
    }
};

template <typename E>
streamable_enum<E> stream_enum(E e) {
    return streamable_enum<E>(e);
}

template <typename E>
typename streamable_enum<E>::tostr_map_t get_enum_strings() {
    // \todo throw an appropriate exception or display compile error/warning
    return {};
}

template <typename E>
std::ostream& operator<<(std::ostream& os, streamable_enum<E> e) {
    auto& mp = streamable_enum<E>::to_string_map();
    auto res = mp.find(e);
    if (res != mp.end()) {
        os << res->second;
    } else {
        os.setstate(std::ios_base::failbit);
    }
    return os;
}

template <typename E>
std::istream& operator>>(std::istream& is, streamable_enum<E>& e) {
    std::string str;
    is >> str;
    if (str.empty()) {
        is.setstate(std::ios_base::failbit);
    }
    auto& mp = streamable_enum<E>::from_string_map();
    auto res = mp.find(str);
    if (res != mp.end()) {
        e = res->second;
    } else {
        is.setstate(std::ios_base::failbit);
    }
    return is;
}

#endif

Usage:

#include "streamable_enum.hpp"

using std::cout;
using std::cin;
using std::endl;

enum Animal {
    CAT,
    DOG,
    TIGER,
    RABBIT
};

template <>
streamable_enum<Animal>::tostr_map_t get_enum_strings<Animal>() {
    return {
        { CAT, "Cat"},
        { DOG, "Dog" },
        { TIGER, "Tiger" },
        { RABBIT, "Rabbit" }
    };
}

int main(int argc, char* argv []) {
    cout << "What animal do you want to buy? Our offering:" << endl;
    for (auto pr : streamable_enum<Animal>::to_string_map()) {          // Use from_string_map() and pr.first instead
        cout << " " << pr.second << endl;                               // to have them sorted in alphabetical order
    }
    streamable_enum<Animal> anim;
    cin >> anim;
    if (!cin) {
        cout << "We don't have such animal here." << endl;
    } else if (anim == Animal::TIGER) {
        cout << stream_enum(Animal::TIGER) << " was a joke..." << endl;
    } else {
        cout << "Here you are!" << endl;
    }

    return 0;
}

#define stringify( name ) # name

enum MyEnum {
    ENUMVAL1
};
...stuff...

stringify(EnumName::ENUMVAL1);  // Returns MyEnum::ENUMVAL1

Further discussion on this method

Preprocessor directive tricks for newcomers

// Define your enumeration like this (in say numbers.h);
ENUM_BEGIN( Numbers )
    ENUM(ONE),
    ENUM(TWO),
    ENUM(FOUR)
ENUM_END( Numbers )

// The macros are defined in a more fundamental .h file (say defs.h);
#define ENUM_BEGIN(typ) enum typ {
#define ENUM(nam) nam
#define ENUM_END(typ) };

// Now in one and only one .c file, redefine the ENUM macros and reinclude
//  the numbers.h file to build a string table
#undef ENUM_BEGIN
#undef ENUM
#undef ENUM_END
#define ENUM_BEGIN(typ) const char * typ ## _name_table [] = {
#define ENUM(nam) #nam
#define ENUM_END(typ) };
#undef NUMBERS_H_INCLUDED   // whatever you need to do to enable reinclusion
#include "numbers.h"

// Now you can do exactly what you want to do, with no retyping, and for any
//  number of enumerated types defined with the ENUM macro family
//  Your code follows;
char num_str[10];
int process_numbers_str(Numbers num) {
  switch(num) {
    case ONE:
    case TWO:
    case THREE:
    {
      strcpy(num_str, Numbers_name_table[num]); // eg TWO -> "TWO"
    } break;
    default:
      return 0; //no match
  return 1;
}

// Sweet no ? After being frustrated by this for years, I finally came up
//  with this solution for my most recent project and plan to reuse the idea
//  forever

I thought that a solution like Boost.Fusion one for adapting structs and classes would be nice, they even had it at some point, to use enums as a fusion sequence.

So I made just some small macros to generate the code to print the enums. This is not perfect and has nothing to see with Boost.Fusion generated boilerplate code, but can be used like the Boost Fusion macros. I want to really do generate the types needed by Boost.Fusion to integrate in this infrastructure which allows to print names of struct members, but this will happen later, for now this is just macros :

#ifndef SWISSARMYKNIFE_ENUMS_ADAPT_ENUM_HPP
#define SWISSARMYKNIFE_ENUMS_ADAPT_ENUM_HPP

#include <swissarmyknife/detail/config.hpp>

#include <string>
#include <ostream>
#include <boost/preprocessor/cat.hpp>
#include <boost/preprocessor/stringize.hpp>
#include <boost/preprocessor/seq/for_each.hpp>


#define SWISSARMYKNIFE_ADAPT_ENUM_EACH_ENUMERATION_ENTRY_C(                     \
    R, unused, ENUMERATION_ENTRY)                                               \
    case ENUMERATION_ENTRY:                                                     \
      return BOOST_PP_STRINGIZE(ENUMERATION_ENTRY);                             \
    break;                                                                      

/**
 * \brief Adapts ENUM to reflectable types.
 *
 * \param ENUM_TYPE To be adapted
 * \param ENUMERATION_SEQ Sequence of enum states
 */
#define SWISSARMYKNIFE_ADAPT_ENUM(ENUM_TYPE, ENUMERATION_SEQ)                   \
    inline std::string to_string(const ENUM_TYPE& enum_value) {                 \
      switch (enum_value) {                                                     \
      BOOST_PP_SEQ_FOR_EACH(                                                    \
          SWISSARMYKNIFE_ADAPT_ENUM_EACH_ENUMERATION_ENTRY_C,                   \
          unused, ENUMERATION_SEQ)                                              \
        default:                                                                \
          return BOOST_PP_STRINGIZE(ENUM_TYPE);                                 \
      }                                                                         \
    }                                                                           \
                                                                                \
    inline std::ostream& operator<<(std::ostream& os, const ENUM_TYPE& value) { \
      os << to_string(value);                                                   \
      return os;                                                                \
    }

#endif

The old answer below is pretty bad, please don't use that. :)

Old answer:

I've been searching a way which solves this problem without changing too much the enums declaration syntax. I came to a solution which uses the preprocessor to retrieve a string from a stringified enum declaration.

I'm able to define non-sparse enums like this :

SMART_ENUM(State, 
    enum State {
        RUNNING,
        SLEEPING, 
        FAULT, 
        UNKNOWN
    })

And I can interact with them in different ways:

// With a stringstream
std::stringstream ss;
ss << State::FAULT;
std::string myEnumStr = ss.str();

//Directly to stdout
std::cout << State::FAULT << std::endl;

//to a string
std::string myStr = State::to_string(State::FAULT);

//from a string
State::State myEnumVal = State::from_string(State::FAULT);

Based on the following definitions :

#define SMART_ENUM(enumTypeArg, ...)                                                     \
namespace enumTypeArg {                                                                  \
    __VA_ARGS__;                                                                         \
    std::ostream& operator<<(std::ostream& os, const enumTypeArg& val) {                 \
            os << swissarmyknife::enums::to_string(#__VA_ARGS__, val);                   \
            return os;                                                                   \
    }                                                                                    \
                                                                                     \
    std::string to_string(const enumTypeArg& val) {                                      \
            return swissarmyknife::enums::to_string(#__VA_ARGS__, val);                  \
    }                                                                                    \
                                                                                     \
    enumTypeArg from_string(const std::string &str) {                                    \
            return swissarmyknife::enums::from_string<enumTypeArg>(#__VA_ARGS__, str);   \
    }                                                                                    \
}                                                                                        \


namespace swissarmyknife { namespace enums {

    static inline std::string to_string(const std::string completeEnumDeclaration, size_t enumVal) throw (std::runtime_error) {
        size_t begin = completeEnumDeclaration.find_first_of('{');
        size_t end = completeEnumDeclaration.find_last_of('}');
        const std::string identifiers = completeEnumDeclaration.substr(begin + 1, end );

        size_t count = 0;
        size_t found = 0;
        do {
            found = identifiers.find_first_of(",}", found+1);

            if (enumVal == count) {
                std::string identifiersSubset = identifiers.substr(0, found);
                size_t beginId = identifiersSubset.find_last_of("{,");
                identifiersSubset = identifiersSubset.substr(beginId+1);
                boost::algorithm::trim(identifiersSubset);
                return identifiersSubset;
            }

            ++count;
        } while (found != std::string::npos);

        throw std::runtime_error("The enum declaration provided doesn't contains this state.");
    }                                                  

    template <typename EnumType>
    static inline EnumType from_string(const std::string completeEnumDeclaration, const std::string &enumStr) throw (std::runtime_error) {
        size_t begin = completeEnumDeclaration.find_first_of('{');
        size_t end = completeEnumDeclaration.find_last_of('}');
        const std::string identifiers = completeEnumDeclaration.substr(begin + 1, end );

        size_t count = 0;
        size_t found = 0;
        do {
            found = identifiers.find_first_of(",}", found+1);

            std::string identifiersSubset = identifiers.substr(0, found);
            size_t beginId = identifiersSubset.find_last_of("{,");
            identifiersSubset = identifiersSubset.substr(beginId+1);
            boost::algorithm::trim(identifiersSubset);

            if (identifiersSubset == enumStr) {
                return static_cast<EnumType>(count);
            }

            ++count;
        } while (found != std::string::npos);

        throw std::runtime_error("No valid enum value for the provided string");
    }                      

}}

When I'll need support for sparse enum and when I'll have more time I'll improve the to_string and from_string implementations with boost::xpressive, but this will costs in compilation time because of the important templating performed and the executable generated is likely to be really bigger. But this has the advantage that it will be more readable and maintanable than this ugly manual string manipulation code. :D

Otherwise I always used boost::bimap to perform such mappings between enums value and string, but it has to be maintained manually.

The technique from Making something both a C identifier and a string? can be used here.

As usual with such preprocessor stuff, writing and understanding the preprocessor part can be hard, and includes passing macros to other macros and involves using # and ## operators, but using it is real easy. I find this style very useful for long enums, where maintaining the same list twice can be really troublesome.

Factory code - typed only once, usually hidden in the header:

enumFactory.h:

// expansion macro for enum value definition
#define ENUM_VALUE(name,assign) name assign,

// expansion macro for enum to string conversion
#define ENUM_CASE(name,assign) case name: return #name;

// expansion macro for string to enum conversion
#define ENUM_STRCMP(name,assign) if (!strcmp(str,#name)) return name;

/// declare the access function and define enum values
#define DECLARE_ENUM(EnumType,ENUM_DEF) \
  enum EnumType { \
    ENUM_DEF(ENUM_VALUE) \
  }; \
  const char *GetString(EnumType dummy); \
  EnumType Get##EnumType##Value(const char *string); \

/// define the access function names
#define DEFINE_ENUM(EnumType,ENUM_DEF) \
  const char *GetString(EnumType value) \
  { \
    switch(value) \
    { \
      ENUM_DEF(ENUM_CASE) \
      default: return ""; /* handle input error */ \
    } \
  } \
  EnumType Get##EnumType##Value(const char *str) \
  { \
    ENUM_DEF(ENUM_STRCMP) \
    return (EnumType)0; /* handle input error */ \
  } \

Factory used

someEnum.h:

#include "enumFactory.h"
#define SOME_ENUM(XX) \
    XX(FirstValue,) \
    XX(SecondValue,) \
    XX(SomeOtherValue,=50) \
    XX(OneMoreValue,=100) \

DECLARE_ENUM(SomeEnum,SOME_ENUM)

someEnum.cpp:

#include "someEnum.h"
DEFINE_ENUM(SomeEnum,SOME_ENUM)

The technique can be easily extended so that XX macros accepts more arguments, and you can also have prepared more macros to substitute for XX for different needs, similar to the three I have provided in this sample.

Comparison to X-Macros using #include / #define / #undef

While this is similar to X-Macros others have mentioned, I think this solution is more elegant in that it does not require #undefing anything, which allows you to hide more of the complicated stuff is in the factory the header file - the header file is something you are not touching at all when you need to define a new enum, therefore new enum definition is a lot shorter and cleaner.

By merging some of the techniques over here I came up with the simplest form:

#define MACROSTR(k) #k

#define X_NUMBERS \
       X(kZero  ) \
       X(kOne   ) \
       X(kTwo   ) \
       X(kThree ) \
       X(kFour  ) \
       X(kMax   )

enum {
#define X(Enum)       Enum,
    X_NUMBERS
#undef X
} kConst;

static char *kConstStr[] = {
#define X(String) MACROSTR(String),
    X_NUMBERS
#undef X
};

int main(void)
{
    int k;
    printf("Hello World!\n\n");

    for (k = 0; k < kMax; k++)
    {
        printf("%s\n", kConstStr[k]);
    }

    return 0;
}

Because I prefer not to use macros for all the usual reasons, I used a more limited macro solution that has the advantage of keeping the enum declaration macro free. Disadvantages include having to copy paste the macro defintion for each enum, and having to explicitly add a macro invocation when adding values to the enum.

std::ostream& operator<<(std::ostream& os, provenance_wrapper::CaptureState cs)
{
#define HANDLE(x) case x: os << #x; break;
    switch (cs) {
    HANDLE(CaptureState::UNUSED)
    HANDLE(CaptureState::ACTIVE)
    HANDLE(CaptureState::CLOSED)
    }
    return os;
#undef HANDLE
}

Here is a solution using macros with the following features:

1. only write each value of the enum once, so there are no double lists to maintain

2. don't keep the enum values in a separate file that is later #included, so I can write it wherever I want

3. don't replace the enum itself, I still want to have the enum type defined, but in addition to it I want to be able to map every enum name to the corresponding string (to not affect legacy code)

4. the searching should be fast, so preferably no switch-case, for those huge enums

https://stackoverflow.com/a/20134475/1812866

There is definitely a way to do this -- use X() macros. These macros use the C preprocessor to construct enums, arrays and code blocks from a list of source data. You only need to add new items to the #define containing the X() macro. The switch statement would expand automatically.

Your example can be written as follows:

 // Source data -- Enum, String
 #define X_NUMBERS \
    X(ONE,   "one") \
    X(TWO,   "two") \
    X(THREE, "three")

 ...

 // Use preprocessor to create the Enum
 typedef enum {
  #define X(Enum, String)       Enum,
   X_NUMBERS
  #undef X
 } Numbers;

 ...

 // Use Preprocessor to expand data into switch statement cases
 switch(num)
 {
 #define X(Enum, String) \
     case Enum:  strcpy(num_str, String); break;
 X_NUMBERS
 #undef X

     default: return 0; break;
 }
 return 1;

There are more efficient ways (i.e. using X Macros to create an string array and enum index), but this is the simplest demo.

KISS. You will be doing all sorts of other switch/case things with your enums so why should printing be different? Forgetting a case in your print routine isn't a huge deal when you consider there are about 100 other places you can forget a case. Just compile -Wall, which will warn of non-exhaustive case matches. Don't use "default" because that will make the switch exhaustive and you wont get warnings. Instead, let the switch exit and deal with the default case like so...

const char *myenum_str(myenum e)
{
    switch(e) {
    case ONE: return "one";
    case TWO: return "two";
    }
    return "invalid";
}