--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dep/toml11/toml/serializer.hpp	Wed Jun 12 05:25:41 2024 -0400
@@ -0,0 +1,994 @@
+//     Copyright Toru Niina 2019.
+// Distributed under the MIT License.
+#ifndef TOML11_SERIALIZER_HPP
+#define TOML11_SERIALIZER_HPP
+#include <cmath>
+#include <cstdio>
+
+#include <limits>
+
+#if defined(_WIN32)
+#include <locale.h>
+#elif defined(__APPLE__) || defined(__FreeBSD__)
+#include <xlocale.h>
+#elif defined(__linux__)
+#include <locale.h>
+#endif
+
+#include "lexer.hpp"
+#include "value.hpp"
+
+namespace toml
+{
+
+// This function serialize a key. It checks a string is a bare key and
+// escapes special characters if the string is not compatible to a bare key.
+// ```cpp
+// std::string k("non.bare.key"); // the key itself includes `.`s.
+// std::string formatted = toml::format_key(k);
+// assert(formatted == "\"non.bare.key\"");
+// ```
+//
+// This function is exposed to make it easy to write a user-defined serializer.
+// Since toml restricts characters available in a bare key, generally a string
+// should be escaped. But checking whether a string needs to be surrounded by
+// a `"` and escaping some special character is boring.
+template<typename charT, typename traits, typename Alloc>
+std::basic_string<charT, traits, Alloc>
+format_key(const std::basic_string<charT, traits, Alloc>& k)
+{
+    if(k.empty())
+    {
+        return std::string("\"\"");
+    }
+
+    // check the key can be a bare (unquoted) key
+    detail::location loc(k, std::vector<char>(k.begin(), k.end()));
+    detail::lex_unquoted_key::invoke(loc);
+    if(loc.iter() == loc.end())
+    {
+        return k; // all the tokens are consumed. the key is unquoted-key.
+    }
+
+    //if it includes special characters, then format it in a "quoted" key.
+    std::basic_string<charT, traits, Alloc> serialized("\"");
+    for(const char c : k)
+    {
+        switch(c)
+        {
+            case '\\': {serialized += "\\\\"; break;}
+            case '\"': {serialized += "\\\""; break;}
+            case '\b': {serialized += "\\b";  break;}
+            case '\t': {serialized += "\\t";  break;}
+            case '\f': {serialized += "\\f";  break;}
+            case '\n': {serialized += "\\n";  break;}
+            case '\r': {serialized += "\\r";  break;}
+            default: {
+                if (c >= 0x00 && c < 0x20)
+                {
+                    std::array<char, 7> buf;
+                    std::snprintf(buf.data(), buf.size(), "\\u00%02x", static_cast<int>(c));
+                    serialized += buf.data();
+                }
+                else
+                {
+                    serialized += c;
+                }
+                break;
+            }
+        }
+    }
+    serialized += "\"";
+    return serialized;
+}
+
+template<typename charT, typename traits, typename Alloc>
+std::basic_string<charT, traits, Alloc>
+format_keys(const std::vector<std::basic_string<charT, traits, Alloc>>& keys)
+{
+    if(keys.empty())
+    {
+        return std::string("\"\"");
+    }
+
+    std::basic_string<charT, traits, Alloc> serialized;
+    for(const auto& ky : keys)
+    {
+        serialized += format_key(ky);
+        serialized += charT('.');
+    }
+    serialized.pop_back(); // remove the last dot '.'
+    return serialized;
+}
+
+template<typename Value>
+struct serializer
+{
+    static_assert(detail::is_basic_value<Value>::value,
+                  "toml::serializer is for toml::value and its variants, "
+                  "toml::basic_value<...>.");
+
+    using value_type           = Value;
+    using key_type             = typename value_type::key_type            ;
+    using comment_type         = typename value_type::comment_type        ;
+    using boolean_type         = typename value_type::boolean_type        ;
+    using integer_type         = typename value_type::integer_type        ;
+    using floating_type        = typename value_type::floating_type       ;
+    using string_type          = typename value_type::string_type         ;
+    using local_time_type      = typename value_type::local_time_type     ;
+    using local_date_type      = typename value_type::local_date_type     ;
+    using local_datetime_type  = typename value_type::local_datetime_type ;
+    using offset_datetime_type = typename value_type::offset_datetime_type;
+    using array_type           = typename value_type::array_type          ;
+    using table_type           = typename value_type::table_type          ;
+
+    serializer(const std::size_t w              = 80u,
+               const int         float_prec     = std::numeric_limits<toml::floating>::max_digits10,
+               const bool        can_be_inlined = false,
+               const bool        no_comment     = false,
+               std::vector<toml::key> ks        = {},
+               const bool     value_has_comment = false)
+        : can_be_inlined_(can_be_inlined), no_comment_(no_comment),
+          value_has_comment_(value_has_comment && !no_comment),
+          float_prec_(float_prec), width_(w), keys_(std::move(ks))
+    {}
+    ~serializer() = default;
+
+    std::string operator()(const boolean_type& b) const
+    {
+        return b ? "true" : "false";
+    }
+    std::string operator()(const integer_type i) const
+    {
+#if defined(_WIN32)
+        _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
+        const std::string original_locale(setlocale(LC_NUMERIC, nullptr));
+        setlocale(LC_NUMERIC, "C");
+#elif defined(__APPLE__) || defined(__FreeBSD__) || defined(__linux__)
+        const auto c_locale = newlocale(LC_NUMERIC_MASK, "C", locale_t(0));
+        locale_t original_locale(0);
+        if(c_locale != locale_t(0))
+        {
+            original_locale = uselocale(c_locale);
+        }
+#endif
+
+        const auto str = std::to_string(i);
+
+#if defined(_WIN32)
+        setlocale(LC_NUMERIC, original_locale.c_str());
+        _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
+#elif defined(__APPLE__) || defined(__FreeBSD__) || defined(__linux__)
+        if(original_locale != locale_t(0))
+        {
+            uselocale(original_locale);
+        }
+#endif
+        return str;
+    }
+    std::string operator()(const floating_type f) const
+    {
+        if(std::isnan(f))
+        {
+            if(std::signbit(f))
+            {
+                return std::string("-nan");
+            }
+            else
+            {
+                return std::string("nan");
+            }
+        }
+        else if(!std::isfinite(f))
+        {
+            if(std::signbit(f))
+            {
+                return std::string("-inf");
+            }
+            else
+            {
+                return std::string("inf");
+            }
+        }
+
+        // set locale to "C".
+        // To make it thread-local, we use OS-specific features.
+        // If we set process-global locale, it can break other thread that also
+        // outputs something simultaneously.
+#if defined(_WIN32)
+        _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
+        const std::string original_locale(setlocale(LC_NUMERIC, nullptr));
+        setlocale(LC_NUMERIC, "C");
+#elif defined(__APPLE__) || defined(__FreeBSD__) || defined(__linux__)
+        const auto c_locale = newlocale(LC_NUMERIC_MASK, "C", locale_t(0));
+        locale_t original_locale(0);
+        if(c_locale != locale_t(0))
+        {
+            original_locale = uselocale(c_locale);
+        }
+#endif
+
+        const auto fmt = "%.*g";
+        const auto bsz = std::snprintf(nullptr, 0, fmt, this->float_prec_, f);
+        // +1 for null character(\0)
+        std::vector<char> buf(static_cast<std::size_t>(bsz + 1), '\0');
+        std::snprintf(buf.data(), buf.size(), fmt, this->float_prec_, f);
+
+        // restore the original locale
+#if defined(_WIN32)
+        setlocale(LC_NUMERIC, original_locale.c_str());
+        _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
+#elif defined(__APPLE__) || defined(__FreeBSD__) || defined(__linux__)
+        if(original_locale != locale_t(0))
+        {
+            uselocale(original_locale);
+        }
+#endif
+
+        std::string token(buf.begin(), std::prev(buf.end()));
+        if(!token.empty() && token.back() == '.') // 1. => 1.0
+        {
+            token += '0';
+        }
+
+        const auto e = std::find_if(
+            token.cbegin(), token.cend(), [](const char c) noexcept -> bool {
+                return c == 'e' || c == 'E';
+            });
+        const auto has_exponent = (token.cend() != e);
+        const auto has_fraction = (token.cend() != std::find(
+            token.cbegin(), token.cend(), '.'));
+
+        if(!has_exponent && !has_fraction)
+        {
+            // the resulting value does not have any float specific part!
+            token += ".0";
+        }
+        return token;
+    }
+    std::string operator()(const string_type& s) const
+    {
+        if(s.kind == string_t::basic)
+        {
+            if((std::find(s.str.cbegin(), s.str.cend(), '\n') != s.str.cend() ||
+                std::find(s.str.cbegin(), s.str.cend(), '\"') != s.str.cend()) &&
+               this->width_ != (std::numeric_limits<std::size_t>::max)())
+            {
+                // if linefeed or double-quote is contained,
+                // make it multiline basic string.
+                const auto escaped = this->escape_ml_basic_string(s.str);
+                std::string open("\"\"\"");
+                std::string close("\"\"\"");
+                if(escaped.find('\n') != std::string::npos ||
+                   this->width_ < escaped.size() + 6)
+                {
+                    // if the string body contains newline or is enough long,
+                    // add newlines after and before delimiters.
+                    open += "\n";
+                    close = std::string("\\\n") + close;
+                }
+                return open + escaped + close;
+            }
+
+            // no linefeed. try to make it oneline-string.
+            std::string oneline = this->escape_basic_string(s.str);
+            if(oneline.size() + 2 < width_ || width_ < 2)
+            {
+                const std::string quote("\"");
+                return quote + oneline + quote;
+            }
+
+            // the line is too long compared to the specified width.
+            // split it into multiple lines.
+            std::string token("\"\"\"\n");
+            while(!oneline.empty())
+            {
+                if(oneline.size() < width_)
+                {
+                    token += oneline;
+                    oneline.clear();
+                }
+                else if(oneline.at(width_-2) == '\\')
+                {
+                    token += oneline.substr(0, width_-2);
+                    token += "\\\n";
+                    oneline.erase(0, width_-2);
+                }
+                else
+                {
+                    token += oneline.substr(0, width_-1);
+                    token += "\\\n";
+                    oneline.erase(0, width_-1);
+                }
+            }
+            return token + std::string("\\\n\"\"\"");
+        }
+        else // the string `s` is literal-string.
+        {
+            if(std::find(s.str.cbegin(), s.str.cend(), '\n') != s.str.cend() ||
+               std::find(s.str.cbegin(), s.str.cend(), '\'') != s.str.cend() )
+            {
+                std::string open("'''");
+                if(this->width_ + 6 < s.str.size())
+                {
+                    open += '\n'; // the first newline is ignored by TOML spec
+                }
+                const std::string close("'''");
+                return open + s.str + close;
+            }
+            else
+            {
+                const std::string quote("'");
+                return quote + s.str + quote;
+            }
+        }
+    }
+
+    std::string operator()(const local_date_type& d) const
+    {
+        std::ostringstream oss;
+        oss << d;
+        return oss.str();
+    }
+    std::string operator()(const local_time_type& t) const
+    {
+        std::ostringstream oss;
+        oss << t;
+        return oss.str();
+    }
+    std::string operator()(const local_datetime_type& dt) const
+    {
+        std::ostringstream oss;
+        oss << dt;
+        return oss.str();
+    }
+    std::string operator()(const offset_datetime_type& odt) const
+    {
+        std::ostringstream oss;
+        oss << odt;
+        return oss.str();
+    }
+
+    std::string operator()(const array_type& v) const
+    {
+        if(v.empty())
+        {
+            return std::string("[]");
+        }
+        if(this->is_array_of_tables(v))
+        {
+            return make_array_of_tables(v);
+        }
+
+        // not an array of tables. normal array.
+        // first, try to make it inline if none of the elements have a comment.
+        if( ! this->has_comment_inside(v))
+        {
+            const auto inl = this->make_inline_array(v);
+            if(inl.size() < this->width_ &&
+               std::find(inl.cbegin(), inl.cend(), '\n') == inl.cend())
+            {
+                return inl;
+            }
+        }
+
+        // if the length exceeds this->width_, print multiline array.
+        // key = [
+        //   # ...
+        //   42,
+        //   ...
+        // ]
+        std::string token;
+        std::string current_line;
+        token += "[\n";
+        for(const auto& item : v)
+        {
+            if( ! item.comments().empty() && !no_comment_)
+            {
+                // if comment exists, the element must be the only element in the line.
+                // e.g. the following is not allowed.
+                // ```toml
+                // array = [
+                // # comment for what?
+                // 1, 2, 3, 4, 5
+                // ]
+                // ```
+                if(!current_line.empty())
+                {
+                    if(current_line.back() != '\n')
+                    {
+                        current_line += '\n';
+                    }
+                    token += current_line;
+                    current_line.clear();
+                }
+                for(const auto& c : item.comments())
+                {
+                    token += '#';
+                    token += c;
+                    token += '\n';
+                }
+                token += toml::visit(*this, item);
+                if(!token.empty() && token.back() == '\n') {token.pop_back();}
+                token += ",\n";
+                continue;
+            }
+            std::string next_elem;
+            if(item.is_table())
+            {
+                serializer ser(*this);
+                ser.can_be_inlined_ = true;
+                ser.width_ = (std::numeric_limits<std::size_t>::max)();
+                next_elem += toml::visit(ser, item);
+            }
+            else
+            {
+                next_elem += toml::visit(*this, item);
+            }
+
+            // comma before newline.
+            if(!next_elem.empty() && next_elem.back() == '\n') {next_elem.pop_back();}
+
+            // if current line does not exceeds the width limit, continue.
+            if(current_line.size() + next_elem.size() + 1 < this->width_)
+            {
+                current_line += next_elem;
+                current_line += ',';
+            }
+            else if(current_line.empty())
+            {
+                // if current line was empty, force put the next_elem because
+                // next_elem is not splittable
+                token += next_elem;
+                token += ",\n";
+                // current_line is kept empty
+            }
+            else // reset current_line
+            {
+                assert(current_line.back() == ',');
+                token += current_line;
+                token += '\n';
+                current_line = next_elem;
+                current_line += ',';
+            }
+        }
+        if(!current_line.empty())
+        {
+            if(!current_line.empty() && current_line.back() != '\n')
+            {
+                current_line += '\n';
+            }
+            token += current_line;
+        }
+        token += "]\n";
+        return token;
+    }
+
+    // templatize for any table-like container
+    std::string operator()(const table_type& v) const
+    {
+        // if an element has a comment, then it can't be inlined.
+        // table = {# how can we write a comment for this? key = "value"}
+        if(this->can_be_inlined_ && !(this->has_comment_inside(v)))
+        {
+            std::string token;
+            if(!this->keys_.empty())
+            {
+                token += format_key(this->keys_.back());
+                token += " = ";
+            }
+            token += this->make_inline_table(v);
+            if(token.size() < this->width_ &&
+               token.end() == std::find(token.begin(), token.end(), '\n'))
+            {
+                return token;
+            }
+        }
+
+        std::string token;
+        if(!keys_.empty())
+        {
+            token += '[';
+            token += format_keys(keys_);
+            token += "]\n";
+        }
+        token += this->make_multiline_table(v);
+        return token;
+    }
+
+  private:
+
+    std::string escape_basic_string(const std::string& s) const
+    {
+        //XXX assuming `s` is a valid utf-8 sequence.
+        std::string retval;
+        for(const char c : s)
+        {
+            switch(c)
+            {
+                case '\\': {retval += "\\\\"; break;}
+                case '\"': {retval += "\\\""; break;}
+                case '\b': {retval += "\\b";  break;}
+                case '\t': {retval += "\\t";  break;}
+                case '\f': {retval += "\\f";  break;}
+                case '\n': {retval += "\\n";  break;}
+                case '\r': {retval += "\\r";  break;}
+                default  :
+                {
+                    if((0x00 <= c && c <= 0x08) || (0x0A <= c && c <= 0x1F) || c == 0x7F)
+                    {
+                        retval += "\\u00";
+                        retval += char(48 + (c / 16));
+                        retval += char((c % 16 < 10 ? 48 : 55) + (c % 16));
+                    }
+                    else
+                    {
+                        retval += c;
+                    }
+                }
+            }
+        }
+        return retval;
+    }
+
+    std::string escape_ml_basic_string(const std::string& s) const
+    {
+        std::string retval;
+        for(auto i=s.cbegin(), e=s.cend(); i!=e; ++i)
+        {
+            switch(*i)
+            {
+                case '\\': {retval += "\\\\"; break;}
+                // One or two consecutive "s are allowed.
+                // Later we will check there are no three consecutive "s.
+                //   case '\"': {retval += "\\\""; break;}
+                case '\b': {retval += "\\b";  break;}
+                case '\t': {retval += "\\t";  break;}
+                case '\f': {retval += "\\f";  break;}
+                case '\n': {retval += "\n";   break;}
+                case '\r':
+                {
+                    if(std::next(i) != e && *std::next(i) == '\n')
+                    {
+                        retval += "\r\n";
+                        ++i;
+                    }
+                    else
+                    {
+                        retval += "\\r";
+                    }
+                    break;
+                }
+                default  :
+                {
+                    const auto c = *i;
+                    if((0x00 <= c && c <= 0x08) || (0x0A <= c && c <= 0x1F) || c == 0x7F)
+                    {
+                        retval += "\\u00";
+                        retval += char(48 + (c / 16));
+                        retval += char((c % 16 < 10 ? 48 : 55) + (c % 16));
+                    }
+                    else
+                    {
+                        retval += c;
+                    }
+                }
+
+            }
+        }
+        // Only 1 or 2 consecutive `"`s are allowed in multiline basic string.
+        // 3 consecutive `"`s are considered as a closing delimiter.
+        // We need to check if there are 3 or more consecutive `"`s and insert
+        // backslash to break them down into several short `"`s like the `str6`
+        // in the following example.
+        // ```toml
+        // str4 = """Here are two quotation marks: "". Simple enough."""
+        // # str5 = """Here are three quotation marks: """."""  # INVALID
+        // str5 = """Here are three quotation marks: ""\"."""
+        // str6 = """Here are fifteen quotation marks: ""\"""\"""\"""\"""\"."""
+        // ```
+        auto found_3_quotes = retval.find("\"\"\"");
+        while(found_3_quotes != std::string::npos)
+        {
+            retval.replace(found_3_quotes, 3, "\"\"\\\"");
+            found_3_quotes = retval.find("\"\"\"");
+        }
+        return retval;
+    }
+
+    // if an element of a table or an array has a comment, it cannot be inlined.
+    bool has_comment_inside(const array_type& a) const noexcept
+    {
+        // if no_comment is set, comments would not be written.
+        if(this->no_comment_) {return false;}
+
+        for(const auto& v : a)
+        {
+            if(!v.comments().empty()) {return true;}
+        }
+        return false;
+    }
+    bool has_comment_inside(const table_type& t) const noexcept
+    {
+        // if no_comment is set, comments would not be written.
+        if(this->no_comment_) {return false;}
+
+        for(const auto& kv : t)
+        {
+            if(!kv.second.comments().empty()) {return true;}
+        }
+        return false;
+    }
+
+    std::string make_inline_array(const array_type& v) const
+    {
+        assert(!has_comment_inside(v));
+        std::string token;
+        token += '[';
+        bool is_first = true;
+        for(const auto& item : v)
+        {
+            if(is_first) {is_first = false;} else {token += ',';}
+            token += visit(serializer(
+                (std::numeric_limits<std::size_t>::max)(), this->float_prec_,
+                /* inlined */ true, /*no comment*/ false, /*keys*/ {},
+                /*has_comment*/ !item.comments().empty()), item);
+        }
+        token += ']';
+        return token;
+    }
+
+    std::string make_inline_table(const table_type& v) const
+    {
+        assert(!has_comment_inside(v));
+        assert(this->can_be_inlined_);
+        std::string token;
+        token += '{';
+        bool is_first = true;
+        for(const auto& kv : v)
+        {
+            // in inline tables, trailing comma is not allowed (toml-lang #569).
+            if(is_first) {is_first = false;} else {token += ',';}
+            token += format_key(kv.first);
+            token += '=';
+            token += visit(serializer(
+                (std::numeric_limits<std::size_t>::max)(), this->float_prec_,
+                /* inlined */ true, /*no comment*/ false, /*keys*/ {},
+                /*has_comment*/ !kv.second.comments().empty()), kv.second);
+        }
+        token += '}';
+        return token;
+    }
+
+    std::string make_multiline_table(const table_type& v) const
+    {
+        std::string token;
+
+        // print non-table elements first.
+        // ```toml
+        // [foo]         # a table we're writing now here
+        // key = "value" # <- non-table element, "key"
+        // # ...
+        // [foo.bar] # <- table element, "bar"
+        // ```
+        // because after printing [foo.bar], the remaining non-table values will
+        // be assigned into [foo.bar], not [foo]. Those values should be printed
+        // earlier.
+        for(const auto& kv : v)
+        {
+            if(kv.second.is_table() || is_array_of_tables(kv.second))
+            {
+                continue;
+            }
+
+            token += write_comments(kv.second);
+
+            const auto key_and_sep    = format_key(kv.first) + " = ";
+            const auto residual_width = (this->width_ > key_and_sep.size()) ?
+                                        this->width_ - key_and_sep.size() : 0;
+            token += key_and_sep;
+            token += visit(serializer(residual_width, this->float_prec_,
+                /*can be inlined*/ true, /*no comment*/ false, /*keys*/ {},
+                /*has_comment*/ !kv.second.comments().empty()), kv.second);
+
+            if(token.back() != '\n')
+            {
+                token += '\n';
+            }
+        }
+
+        // normal tables / array of tables
+
+        // after multiline table appeared, the other tables cannot be inline
+        // because the table would be assigned into the table.
+        // [foo]
+        // ...
+        // bar = {...} # <- bar will be a member of [foo].
+        bool multiline_table_printed = false;
+        for(const auto& kv : v)
+        {
+            if(!kv.second.is_table() && !is_array_of_tables(kv.second))
+            {
+                continue; // other stuff are already serialized. skip them.
+            }
+
+            std::vector<toml::key> ks(this->keys_);
+            ks.push_back(kv.first);
+
+            auto tmp = visit(serializer(this->width_, this->float_prec_,
+                !multiline_table_printed, this->no_comment_, ks,
+                /*has_comment*/ !kv.second.comments().empty()), kv.second);
+
+            // If it is the first time to print a multi-line table, it would be
+            // helpful to separate normal key-value pair and subtables by a
+            // newline.
+            // (this checks if the current key-value pair contains newlines.
+            //  but it is not perfect because multi-line string can also contain
+            //  a newline. in such a case, an empty line will be written) TODO
+            if((!multiline_table_printed) &&
+               std::find(tmp.cbegin(), tmp.cend(), '\n') != tmp.cend())
+            {
+                multiline_table_printed = true;
+                token += '\n'; // separate key-value pairs and subtables
+
+                token += write_comments(kv.second);
+                token += tmp;
+
+                // care about recursive tables (all tables in each level prints
+                // newline and there will be a full of newlines)
+                if(tmp.substr(tmp.size() - 2, 2) != "\n\n" &&
+                   tmp.substr(tmp.size() - 4, 4) != "\r\n\r\n" )
+                {
+                    token += '\n';
+                }
+            }
+            else
+            {
+                token += write_comments(kv.second);
+                token += tmp;
+                token += '\n';
+            }
+        }
+        return token;
+    }
+
+    std::string make_array_of_tables(const array_type& v) const
+    {
+        // if it's not inlined, we need to add `[[table.key]]`.
+        // but if it can be inlined, we can format it as the following.
+        // ```
+        // table.key = [
+        //   {...},
+        //   # comment
+        //   {...},
+        // ]
+        // ```
+        // This function checks if inlinization is possible or not, and then
+        // format the array-of-tables in a proper way.
+        //
+        // Note about comments:
+        //
+        // If the array itself has a comment (value_has_comment_ == true), we
+        // should try to make it inline.
+        // ```toml
+        // # comment about array
+        // array = [
+        //   # comment about table element
+        //   {of = "table"}
+        // ]
+        // ```
+        // If it is formatted as a multiline table, the two comments becomes
+        // indistinguishable.
+        // ```toml
+        // # comment about array
+        // # comment about table element
+        // [[array]]
+        // of = "table"
+        // ```
+        // So we need to try to make it inline, and it force-inlines regardless
+        // of the line width limit.
+        //     It may fail if the element of a table has comment. In that case,
+        // the array-of-tables will be formatted as a multiline table.
+        if(this->can_be_inlined_ || this->value_has_comment_)
+        {
+            std::string token;
+            if(!keys_.empty())
+            {
+                token += format_key(keys_.back());
+                token += " = ";
+            }
+
+            bool failed = false;
+            token += "[\n";
+            for(const auto& item : v)
+            {
+                // if an element of the table has a comment, the table
+                // cannot be inlined.
+                if(this->has_comment_inside(item.as_table()))
+                {
+                    failed = true;
+                    break;
+                }
+                // write comments for the table itself
+                token += write_comments(item);
+
+                const auto t = this->make_inline_table(item.as_table());
+
+                if(t.size() + 1 > width_ || // +1 for the last comma {...},
+                   std::find(t.cbegin(), t.cend(), '\n') != t.cend())
+                {
+                    // if the value itself has a comment, ignore the line width limit
+                    if( ! this->value_has_comment_)
+                    {
+                        failed = true;
+                        break;
+                    }
+                }
+                token += t;
+                token += ",\n";
+            }
+
+            if( ! failed)
+            {
+                token += "]\n";
+                return token;
+            }
+            // if failed, serialize them as [[array.of.tables]].
+        }
+
+        std::string token;
+        for(const auto& item : v)
+        {
+            token += write_comments(item);
+            token += "[[";
+            token += format_keys(keys_);
+            token += "]]\n";
+            token += this->make_multiline_table(item.as_table());
+        }
+        return token;
+    }
+
+    std::string write_comments(const value_type& v) const
+    {
+        std::string retval;
+        if(this->no_comment_) {return retval;}
+
+        for(const auto& c : v.comments())
+        {
+            retval += '#';
+            retval += c;
+            retval += '\n';
+        }
+        return retval;
+    }
+
+    bool is_array_of_tables(const value_type& v) const
+    {
+        if(!v.is_array() || v.as_array().empty()) {return false;}
+        return is_array_of_tables(v.as_array());
+    }
+    bool is_array_of_tables(const array_type& v) const
+    {
+        // Since TOML v0.5.0, heterogeneous arrays are allowed. So we need to
+        // check all the element in an array to check if the array is an array
+        // of tables.
+        return std::all_of(v.begin(), v.end(), [](const value_type& elem) {
+                return elem.is_table();
+            });
+    }
+
+  private:
+
+    bool        can_be_inlined_;
+    bool        no_comment_;
+    bool        value_has_comment_;
+    int         float_prec_;
+    std::size_t width_;
+    std::vector<toml::key> keys_;
+};
+
+template<typename C,
+         template<typename ...> class M, template<typename ...> class V>
+std::string
+format(const basic_value<C, M, V>& v, std::size_t w = 80u,
+       int fprec = std::numeric_limits<toml::floating>::max_digits10,
+       bool no_comment = false, bool force_inline = false)
+{
+    using value_type = basic_value<C, M, V>;
+    // if value is a table, it is considered to be a root object.
+    // the root object can't be an inline table.
+    if(v.is_table())
+    {
+        std::ostringstream oss;
+        if(!v.comments().empty())
+        {
+            oss << v.comments();
+            oss << '\n'; // to split the file comment from the first element
+        }
+        const auto serialized = visit(serializer<value_type>(w, fprec, false, no_comment), v);
+        oss << serialized;
+        return oss.str();
+    }
+    return visit(serializer<value_type>(w, fprec, force_inline), v);
+}
+
+namespace detail
+{
+template<typename charT, typename traits>
+int comment_index(std::basic_ostream<charT, traits>&)
+{
+    static const int index = std::ios_base::xalloc();
+    return index;
+}
+} // detail
+
+template<typename charT, typename traits>
+std::basic_ostream<charT, traits>&
+nocomment(std::basic_ostream<charT, traits>& os)
+{
+    // by default, it is zero. and by default, it shows comments.
+    os.iword(detail::comment_index(os)) = 1;
+    return os;
+}
+
+template<typename charT, typename traits>
+std::basic_ostream<charT, traits>&
+showcomment(std::basic_ostream<charT, traits>& os)
+{
+    // by default, it is zero. and by default, it shows comments.
+    os.iword(detail::comment_index(os)) = 0;
+    return os;
+}
+
+template<typename charT, typename traits, typename C,
+         template<typename ...> class M, template<typename ...> class V>
+std::basic_ostream<charT, traits>&
+operator<<(std::basic_ostream<charT, traits>& os, const basic_value<C, M, V>& v)
+{
+    using value_type = basic_value<C, M, V>;
+
+    // get status of std::setw().
+    const auto w     = static_cast<std::size_t>(os.width());
+    const int  fprec = static_cast<int>(os.precision());
+    os.width(0);
+
+    // by default, iword is initialized by 0. And by default, toml11 outputs
+    // comments. So `0` means showcomment. 1 means nocommnet.
+    const bool no_comment = (1 == os.iword(detail::comment_index(os)));
+
+    if(!no_comment && v.is_table() && !v.comments().empty())
+    {
+        os << v.comments();
+        os << '\n'; // to split the file comment from the first element
+    }
+    // the root object can't be an inline table. so pass `false`.
+    const auto serialized = visit(serializer<value_type>(w, fprec, no_comment, false), v);
+    os << serialized;
+
+    // if v is a non-table value, and has only one comment, then
+    // put a comment just after a value. in the following way.
+    //
+    // ```toml
+    // key = "value" # comment.
+    // ```
+    //
+    // Since the top-level toml object is a table, one who want to put a
+    // non-table toml value must use this in a following way.
+    //
+    // ```cpp
+    // toml::value v;
+    // std::cout << "user-defined-key = " << v << std::endl;
+    // ```
+    //
+    // In this case, it is impossible to put comments before key-value pair.
+    // The only way to preserve comments is to put all of them after a value.
+    if(!no_comment && !v.is_table() && !v.comments().empty())
+    {
+        os << " #";
+        for(const auto& c : v.comments()) {os << c;}
+    }
+    return os;
+}
+
+} // toml
+#endif// TOML11_SERIALIZER_HPP