I am trying to serialize the following string type with boost:
using CustomString = std::basic_string<char, std::char_traits<char>, CustomAllocator<char>>;
1. Since this type is identical to the definition of the std::string, except for the allocator, my first approach was to do the same as boost does for the std::string.
I added BOOST_CLASS_IMPLEMENTATION(CustomString, boost::serialization::primitive_type) after the using statement.
This seemed to work fine till I tried to serialize a string which had a space in it.
Serialization is done into a boost::archive::text_oarchive and the separator is set to space. So after deserialization only the first part of the string got read from the archive e.g. I wrote "Hello World" to the archive but only got "Hello" after deserialization. For std::string boost adds a length field before the text. This is not the case for the custom string.
The source for the length entry is boost\archive\impl\text_oarchive_impl.ipp
text_oarchive_impl<Archive>::save(const std::string &s)
{
const std::size_t size = s.size();
*this->This() << size;
this->This()->newtoken();
os << s;
}
I found the same problem in the answer to the following question: Can boost::container::strings be serialized using boost serialization?
I extended the example to make the problem visible: https://coliru.stacked-crooked.com/a/84b2cb162d58534a
2. Now I wrote my own serialization functions
namespace boost
{
namespace serialization
{
template<class Archive>
inline void serialize(Archive& ar, CustomString & s, const unsigned int file_version)
{
boost::serialization::split_free(ar, s, file_version);
}
template<typename Archive>
inline void save(
Archive& ar, const CustomString & s, const unsigned int /* file_version */
)
{
ar << s.size();
for (size_t i = 0; i < s.size(); i++)
{
ar << s.c_str()[i];
}
}
template<typename Archive>
inline void load(
Archive& ar, CustomString & s, const unsigned int /* file_version */
)
{
size_t size;
ar >> size;
char c;
for (size_t i=0; i < size;i++)
{
ar >> c;
s.push_back(c);
}
}
} // namespace serialization
} // namespace boost
This code does work but produces a quite long archive, because chars get encoded as a decimal values:
std::string Archive: 22 serialization::archive 19 11 Hello World
CustomString Archive: 22 serialization::archive 19 0 0 11 72 101 108 108 111 32 87 111 114 108 100
I would be grateful for a hint on how to improve one of the two approaches. Thank you!
Yeah, I'm still surprised support for {std,boost::container}::basic_string is not out of the box. Also, pretty suprised nobody noticed the old answer was broken.
I agree that custom serialization is the safer bet. It can be a little simpler and lightweigth:
#include <boost/algorithm/string.hpp>
#include <boost/archive/text_iarchive.hpp>
#include <boost/archive/text_oarchive.hpp>
#include <boost/serialization/binary_object.hpp>
#include <iomanip>
#include <iostream>
namespace Funky {
template <typename T> struct FunkyAlloc : std::allocator<T> {
using std::allocator<T>::allocator;
using std::allocator<T>::operator=;
};
using String = std::basic_string<char, std::char_traits<char>, FunkyAlloc<char>>;
template <typename Ar, typename TCh, typename TChT, typename Allocator>
void serialize(Ar& ar, std::basic_string<TCh, TChT, Allocator>& s, unsigned) {
size_t n = s.length();
ar& n;
if (Ar::is_loading::value)
s.resize(n);
// ar& boost::serialization::make_array(s.data(), n);
ar& boost::serialization::make_binary_object(s.data(), n);
}
}
int main() {
for (auto test_case : {"", " ", " ", "ford mustang", "hallberg rassy"}) {
std::stringstream ss;
Funky::String fs = test_case;
boost::archive::text_oarchive(ss) << fs;
std::cout << "Archive reads: " << quoted(boost::replace_all_copy(ss.str(), "\n", "\\n")) << std::endl;
Funky::String roundtrip;
boost::archive::text_iarchive (ss)>>roundtrip;
std::cout << "Roundtripped Funky::String: " << quoted(roundtrip) << "\n";
assert(roundtrip == test_case);
}
}
Passes the asserts and prints the expected:
g++ -std=c++20 -O2 -Wall -pedantic -pthread main.cpp -lboost_serialization && ./a.out
Archive reads: "22 serialization::archive 20 0 0 0\\n\\n"
Roundtripped Funky::String: ""
Archive reads: "22 serialization::archive 20 0 0 1\\n\\nIA==\\n"
Roundtripped Funky::String: " "
Archive reads: "22 serialization::archive 20 0 0 3\\n\\nICAg\\n"
Roundtripped Funky::String: " "
Archive reads: "22 serialization::archive 20 0 0 12\\n\\nZm9yZCBtdXN0YW5n\\n"
Roundtripped Funky::String: "ford mustang"
Archive reads: "22 serialization::archive 20 0 0 14\\n\\naGFsbGJlcmcgcmFzc3k=\\n"
Roundtripped Funky::String: "hallberg rassy"
If you comment out the make_array line instead you get close to what you had manually coded: (Live)
Archive reads: "22 serialization::archive 20 0 0 0\\n"
Roundtripped Funky::String: ""
Archive reads: "22 serialization::archive 20 0 0 1 32\\n"
Roundtripped Funky::String: " "
Archive reads: "22 serialization::archive 20 0 0 3 32 32 32\\n"
Roundtripped Funky::String: " "
Archive reads: "22 serialization::archive 20 0 0 12 102 111 114 100 32 109 117 115 116 97 110 103\\n"
Roundtripped Funky::String: "ford mustang"
Archive reads: "22 serialization::archive 20 0 0 14 104 97 108 108 98 101 114 103 32 114 97 115 115 121\\n"
Roundtripped Funky::String: "hallberg rassy"