Avoiding key construction for std::map::find()

c++, c++11

Solution

Your concern is real, and there's no good workaround for C++11.

C++14 fixes this issue by adding a templated overload of `std::map::find` — the relevant proposal is N3657. In C++14, your program would look like this:

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <map>
#include <algorithm>

class std_string {
    char *m_s;
public:
    std_string() { m_s = nullptr; }
    std_string(const char* s) { puts("Oops! A new std_string was constructed!"); m_s = strdup(s); }
    ~std_string() { free(m_s); }
    std_string(std_string&& ss) = delete;
    std_string(const std_string& ss) = delete;
    std_string& operator=(std_string&& ss) = delete;
    std_string& operator=(const std_string& ss) = delete;

    bool operator< (const char* s) const { return strcmp(m_s, s) < 0; }
    bool operator< (const std_string& ss) const { return strcmp(m_s, ss.m_s) < 0; }
    friend bool operator< (const char* s, const std_string& ss) { return strcmp(s, ss.m_s) < 0; }
};

int main()
{
    {
        puts("The C++11 way makes a copy...");
        std::map<std_string, int> m;
        auto it = m.find("Olaf");
    }
    {
        puts("The C++14 way doesn't...");
        std::map<std_string, int, std::less<>> m;
        auto it = m.find("Olaf");
    }
}

(`std::less<>` is the generalized "less-than" comparator, equivalent to `operator<`. C++03 and C++11 have a broken-by-design version of this comparator that forces both arguments to be the same type. C++14 finally does it right.)

Unfortunately the Committee seems to have decided that people should go through all their C++11 code and update every container to use `std::less<>` as the comparator — it doesn't just happen by default. There's no good reason for this decision; it's just The Way It Is. (See my comments above about broken-by-design. C++ has a bad habit of introducing broken versions of things before introducing the "real" versions several years later.)

For C++11, `std::map::find` has only one overload (the one that takes `const Key&`), so any workaround will necessarily involve changing the `Key` type to be less expensive — we can't just fiddle with the comparator, because by the time execution reaches the comparator, we've already promoted `find`'s argument to the `Key` type.

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <map>
#include <algorithm>

class std_string {
    char *m_s;
public:
    std_string() : m_s(nullptr) { }
    std_string(const char* s) : m_s(strdup(s)) { puts("Oops! A new std_string was constructed!"); }
    ~std_string() { free(m_s); }
    std_string(std_string&& ss) : m_s(nullptr) { std::swap(m_s, ss.m_s); }
    std_string(const std_string& ss) : m_s(strdup(ss.data())) { puts("Oops! A new std_string was constructed!"); }
    std_string& operator=(std_string&& ss) = delete;
    std_string& operator=(const std_string& ss) = delete;

    const char* data() const { return m_s; }

    bool operator< (const char* s) const { return strcmp(m_s, s) < 0; }
    bool operator< (const std_string& ss) const { return strcmp(m_s, ss.m_s) < 0; }
    friend bool operator< (const char* s, const std_string& ss) { return strcmp(s, ss.m_s) < 0; }
};

struct string_or_ptr {
    union {
        const char* ptr;
        alignas(std_string) unsigned char str[sizeof (std_string)];
    } m_u;
    bool m_deep;

    char const* & ptr() { return m_u.ptr; }
    std_string& str() { return *reinterpret_cast<std_string*>(m_u.str); }
    char const* const & ptr() const { return m_u.ptr; }
    std_string const& str() const { return *reinterpret_cast<const std_string*>(m_u.str); }

    string_or_ptr() : m_deep(false) { ptr() = ""; }
    string_or_ptr(const char* s) : m_deep(false) { ptr() = s; }
    string_or_ptr(std_string&& s) : m_deep(true) { new ((void*)&str()) std_string(std::move(s)); }
    string_or_ptr(const std_string& s) : m_deep(true) { new ((void*)&str()) std_string(s); }
    ~string_or_ptr() { if (m_deep) str().~std_string(); }
    std_string& operator=(std_string&& ss) = delete;
    std_string& operator=(const std_string& ss) = delete;


    operator const char*() const { return m_deep ? str().data() : ptr(); }

    bool operator< (const char* s) const { return strcmp((const char*)*this, s) < 0; }
    bool operator< (const std_string& ss) const { return (const char*)*this < ss; }
    bool operator< (const string_or_ptr& sp) const { return strcmp((const char*)*this, (const char*)sp) < 0; }
    friend bool operator< (const char* s, const string_or_ptr& sp) { return strcmp(s, (const char*)sp) < 0; }
    friend bool operator< (const std_string& ss, const string_or_ptr& sp) { return ss < (const char*)sp; }
};

int main()
{
    {
        puts("The C++11 way...");
        std::map<std_string, int> m;
        auto it = m.find("Olaf");
    }
    {
        puts("The C++11 way with a custom string-or-pointer Key type...");
        std::map<string_or_ptr, int> m;
        auto it = m.find("Olaf");
    }
}

Problem

Original source