toml11
======

[![Build Status on GitHub Actions](https://github.com/ToruNiina/toml11/workflows/build/badge.svg)](https://github.com/ToruNiina/toml11/actions)
[![Build status on Appveyor](https://ci.appveyor.com/api/projects/status/m2n08a926asvg5mg/branch/master?svg=true)](https://ci.appveyor.com/project/ToruNiina/toml11/branch/master)
[![Build status on CircleCI](https://circleci.com/gh/ToruNiina/toml11/tree/master.svg?style=svg)](https://circleci.com/gh/ToruNiina/toml11/tree/master)
[![Version](https://img.shields.io/github/release/ToruNiina/toml11.svg?style=flat)](https://github.com/ToruNiina/toml11/releases)
[![License](https://img.shields.io/github/license/ToruNiina/toml11.svg?style=flat)](LICENSE)
[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.1209136.svg)](https://doi.org/10.5281/zenodo.1209136)

toml11 is a C++11 (or later) header-only toml parser/encoder depending only on C++ standard library.

- It is compatible to the latest version of [TOML v1.0.0](https://toml.io/en/v1.0.0).
- It is one of the most TOML standard compliant libraries, tested with [the language agnostic test suite for TOML parsers by BurntSushi](https://github.com/BurntSushi/toml-test).
- It shows highly informative error messages. You can see the error messages about invalid files at [CircleCI](https://circleci.com/gh/ToruNiina/toml11).
- It has configurable container. You can use any random-access containers and key-value maps as backend containers.
- It optionally preserves comments without any overhead.
- It has configurable serializer that supports comments, inline tables, literal strings and multiline strings.
- It supports user-defined type conversion from/into toml values.
- It correctly handles UTF-8 sequences, with or without BOM, both on posix and Windows.

## Example

```cpp
#include <toml.hpp>
#include <iostream>

int main()
{
    // ```toml
    // title = "an example toml file"
    // nums  = [3, 1, 4, 1, 5]
    // ```
    auto data = toml::parse("example.toml");

    // find a value with the specified type from a table
    std::string title = toml::find<std::string>(data, "title");

    // convert the whole array into any container automatically
    std::vector<int> nums = toml::find<std::vector<int>>(data, "nums");

    // access with STL-like manner
    if(!data.contains("foo"))
    {
        data["foo"] = "bar";
    }

    // pass a fallback
    std::string name = toml::find_or<std::string>(data, "name", "not found");

    // width-dependent formatting
    std::cout << std::setw(80) << data << std::endl;

    return 0;
}
```

## Table of Contents

- [Integration](#integration)
- [Decoding a toml file](#decoding-a-toml-file)
  - [In the case of syntax error](#in-the-case-of-syntax-error)
  - [Invalid UTF-8 Codepoints](#invalid-utf-8-codepoints)
- [Finding a toml value](#finding-a-toml-value)
  - [Finding a value in a table](#finding-a-value-in-a-table)
  - [In case of error](#in-case-of-error)
  - [Dotted keys](#dotted-keys)
- [Casting a toml value](#casting-a-toml-value)
- [Checking value type](#checking-value-type)
- [More about conversion](#more-about-conversion)
  - [Converting an array](#converting-an-array)
  - [Converting a table](#converting-a-table)
  - [Getting an array of tables](#getting-an-array-of-tables)
  - [Cost of conversion](#cost-of-conversion)
  - [Converting datetime and its variants](#converting-datetime-and-its-variants)
- [Getting with a fallback](#getting-with-a-fallback)
- [Expecting conversion](#expecting-conversion)
- [Visiting a toml::value](#visiting-a-tomlvalue)
- [Constructing a toml::value](#constructing-a-tomlvalue)
- [Preserving Comments](#preserving-comments)
- [Customizing containers](#customizing-containers)
- [TOML literal](#toml-literal)
- [Conversion between toml value and arbitrary types](#conversion-between-toml-value-and-arbitrary-types)
- [Formatting user-defined error messages](#formatting-user-defined-error-messages)
- [Obtaining location information](#obtaining-location-information)
- [Exceptions](#exceptions)
- [Colorize Error Messages](#colorize-error-messages)
- [Serializing TOML data](#serializing-toml-data)
- [Underlying types](#underlying-types)
- [Unreleased TOML features](#unreleased-toml-features)
- [Breaking Changes from v2](#breaking-changes-from-v2)
- [Running Tests](#running-tests)
- [Contributors](#contributors)
- [Licensing Terms](#licensing-terms)

## Integration

Just include the file after adding it to the include path.

```cpp
#include <toml.hpp> // that's all! now you can use it.
#include <iostream>

int main()
{
    const auto data  = toml::parse("example.toml");
    const auto title = toml::find<std::string>(data, "title");
    std::cout << "the title is " << title << std::endl;
    return 0;
}
```

The convenient way is to add this repository as a git-submodule or to install
it in your system by CMake.

Note for MSVC: We recommend to set `/Zc:__cplusplus` to detect C++ version correctly.

### Example installation

For local installation build & use the provided install target

```bash
git clone https://github.com/ToruNiina/toml11.git
mkdir -p toml11/build
cd toml11/build
cmake .. -DCMAKE_CXX_STANDARD=11
sudo make install
```

In case you want to create a `.deb` you can use `checkinstall`.
```bash
sudo checkinstall
[[ .. skipping for clarity ]]
**********************************************************************

 Done. The new package has been installed and saved to

 /home/user/toml11/build/build_20230728-1_amd64.deb

 You can remove it from your system anytime using:

      dpkg -r build

**********************************************************************
```

You should get a package that you can install with `dpkg -i <myfile>.deb` and remove with `dpkg -r <myfile>.deb`

## Decoding a toml file

To parse a toml file, the only thing you have to do is
to pass a filename to the `toml::parse` function.

```cpp
const std::string fname("sample.toml");
const toml::value data = toml::parse(fname);
```

As required by the TOML specification, the top-level value is always a table.
You can find a value inside it, cast it into a table explicitly, and insert it as a value into other `toml::value`.

If it encounters an error while opening a file, it will throw `std::runtime_error`.

You can also pass a `std::istream` to the  `toml::parse` function.
To show a filename in an error message, however, it is recommended to pass the
filename with the stream.

```cpp
std::ifstream ifs("sample.toml", std::ios_base::binary);
assert(ifs.good());
const auto data = toml::parse(ifs, /*optional -> */ "sample.toml");
```

**Note**: When you are **on Windows, open a file in binary mode**.
If a file is opened in text-mode, CRLF ("\r\n") will automatically be
converted to LF ("\n") and this causes inconsistency between file size
and the contents that would be read. This causes weird error.

### In the case of syntax error

If there is a syntax error in a toml file, `toml::parse` will throw
`toml::syntax_error` that inherits `std::exception`.

toml11 has clean and informative error messages inspired by Rust and
it looks like the following.

```console
terminate called after throwing an instance of 'toml::syntax_error'
  what():  [error] toml::parse_table: invalid line format # error description
 --> example.toml                                         # file name
 3 | a = 42 = true                                        # line num and content
   |        ^------ expected newline, but got '='.        # error reason
```

If you (mistakenly) duplicate tables and got an error, it is helpful to see
where they are. toml11 shows both at the same time like the following.

```console
terminate called after throwing an instance of 'toml::syntax_error'
  what():  [error] toml::insert_value: table ("table") already exists.
 --> duplicate-table.toml
 1 | [table]
   | ~~~~~~~ table already exists here
 ...
 3 | [table]
   | ~~~~~~~ table defined twice
```

When toml11 encounters a malformed value, it tries to detect what type it is.
Then it shows hints to fix the format. An error message while reading one of
the malformed files in [the language agnostic test suite](https://github.com/BurntSushi/toml-test).
is shown below.

```console
what(): [error] bad time: should be HH:MM:SS.subsec
 --> ./datetime-malformed-no-secs.toml
 1 | no-secs = 1987-07-05T17:45Z
   |                     ^------- HH:MM:SS.subsec
   |
Hint: pass: 1979-05-27T07:32:00, 1979-05-27 07:32:00.999999
Hint: fail: 1979-05-27T7:32:00, 1979-05-27 17:32
```

You can find other examples in a job named `output_result` on
[CircleCI](https://circleci.com/gh/ToruNiina/toml11).

Since the error message generation is generally a difficult task, the current
status is not ideal. If you encounter a weird error message, please let us know
and contribute to improve the quality!

### Invalid UTF-8 codepoints

It throws `syntax_error` if a value of an escape sequence
representing unicode character is not a valid UTF-8 codepoint.

```console
  what():  [error] toml::read_utf8_codepoint: input codepoint is too large.
 --> utf8.toml
 1 | exceeds_unicode = "\U0011FFFF example"
   |                              ^--------- should be in [0x00..0x10FFFF]
```

## Finding a toml value

After parsing successfully, you can obtain the values from the result of
`toml::parse` using `toml::find` function.

```toml
# sample.toml
answer  = 42
pi      = 3.14
numbers = [1,2,3]
time    = 1979-05-27T07:32:00Z
```

``` cpp
const auto data      = toml::parse("sample.toml");
const auto answer    = toml::find<std::int64_t    >(data, "answer");
const auto pi        = toml::find<double          >(data, "pi");
const auto numbers   = toml::find<std::vector<int>>(data, "numbers");
const auto timepoint = toml::find<std::chrono::system_clock::time_point>(data, "time");
```

By default, `toml::find` returns a `toml::value`.

```cpp
const toml::value& answer = toml::find(data, "answer");
```

When you pass an exact TOML type that does not require type conversion,
`toml::find` returns a reference without copying the value.

```cpp
const auto  data   = toml::parse("sample.toml");
const auto& answer = toml::find<toml::integer>(data, "answer");
```

If the specified type requires conversion, you can't take a reference to the value.
See also [underlying types](#underlying-types).

**NOTE**: For some technical reason, automatic conversion between `integer` and
`floating` is not supported. If you want to get a floating value even if a value
has integer value, you need to convert it manually after obtaining a value,
like the following.

```cpp
const auto vx = toml::find(data, "x");
double x = vx.is_floating() ? vx.as_floating(std::nothrow) :
           static_cast<double>(vx.as_integer()); // it throws if vx is neither
                                                 // floating nor integer.
```

### Finding a value in a table

There are several way to get a value defined in a table.
First, you can get a table as a normal value and find a value from the table.

```toml
[fruit]
name = "apple"
[fruit.physical]
color = "red"
shape = "round"
```

``` cpp
const auto  data  = toml::parse("fruit.toml");
const auto& fruit = toml::find(data, "fruit");
const auto  name  = toml::find<std::string>(fruit, "name");

const auto& physical = toml::find(fruit, "physical");
const auto  color    = toml::find<std::string>(physical, "color");
const auto  shape    = toml::find<std::string>(physical, "shape");
```

Here, variable `fruit` is a `toml::value` and can be used as the first argument
of `toml::find`.

Second, you can pass as many arguments as the number of subtables to `toml::find`.

```cpp
const auto data  = toml::parse("fruit.toml");
const auto color = toml::find<std::string>(data, "fruit", "physical", "color");
const auto shape = toml::find<std::string>(data, "fruit", "physical", "shape");
```

### Finding a value in an array

You can find n-th value in an array by `toml::find`.

```toml
values = ["foo", "bar", "baz"]
```

``` cpp
const auto data   = toml::parse("sample.toml");
const auto values = toml::find(data, "values");
const auto bar    = toml::find<std::string>(values, 1);
```

`toml::find` can also search array recursively.

```cpp
const auto data = toml::parse("fruit.toml");
const auto bar  = toml::find<std::string>(data, "values", 1);
```

Before calling `toml::find`, you can check if a value corresponding to a key
exists. You can use both `bool toml::value::contains(const key&) const` and
`std::size_t toml::value::count(const key&) const`. Those behaves like the
`std::map::contains` and `std::map::count`.

```cpp
const auto data = toml::parse("fruit.toml");
if(data.contains("fruit") && data.at("fruit").count("physical") != 0)
{
    // ...
}
```

### In case of error

If the value does not exist, `toml::find` throws `std::out_of_range` with the
location of the table.

```console
terminate called after throwing an instance of 'std::out_of_range'
  what():  [error] key "answer" not found
 --> example.toml
 6 | [tab]
   | ~~~~~ in this table
```

----

If the specified type differs from the actual value contained, it throws
`toml::type_error` that inherits `std::exception`.

Similar to the case of syntax error, toml11 also displays clean error messages.
The error message when you choose `int` to get `string` value would be like this.

```console
terminate called after throwing an instance of 'toml::type_error'
  what():  [error] toml::value bad_cast to integer
 --> example.toml
 3 | title = "TOML Example"
   |         ~~~~~~~~~~~~~~ the actual type is string
```

**NOTE**: In order to show this kind of error message, all the toml values have
a pointer to represent its range in a file. The entire contents of a file is
shared by `toml::value`s and remains on the heap memory. It is recommended to
destruct all the `toml::value` classes after configuring your application
if you have a large TOML file compared to the memory resource.

### Dotted keys

TOML v0.5.0 has a new feature named "dotted keys".
You can chain keys to represent the structure of the data.

```toml
physical.color = "orange"
physical.shape = "round"
```

This is equivalent to the following.

```toml
[physical]
color = "orange"
shape = "round"
```

You can get both of the above tables with the same c++ code.

```cpp
const auto physical = toml::find(data, "physical");
const auto color    = toml::find<std::string>(physical, "color");
```

The following code does not work for the above toml file.

```cpp
// XXX this does not work!
const auto color = toml::find<std::string>(data, "physical.color");
```

The above code works with the following toml file.

```toml
"physical.color" = "orange"
# equivalent to {"physical.color": "orange"},
# NOT {"physical": {"color": "orange"}}.
```


## Casting a toml value

### `toml::get`

`toml::parse` returns `toml::value`. `toml::value` is a union type that can
contain one of the following types.

- `toml::boolean` (`bool`)
- `toml::integer` (`std::int64_t`)
- `toml::floating` (`double`)
- `toml::string` (a type convertible to std::string)
- `toml::local_date`
- `toml::local_time`
- `toml::local_datetime`
- `toml::offset_datetime`
- `toml::array` (by default, `std::vector<toml::value>`)
  - It depends. See [customizing containers](#customizing-containers) for detail.
- `toml::table` (by default, `std::unordered_map<toml::key, toml::value>`)
  - It depends. See [customizing containers](#customizing-containers) for detail.

To get a value inside, you can use `toml::get<T>()`. The usage is the same as
`toml::find<T>` (actually, `toml::find` internally uses `toml::get` after casting
a value to `toml::table`).

``` cpp
const toml::value  data    = toml::parse("sample.toml");
const toml::value  answer_ = toml::get<toml::table >(data).at("answer");
const std::int64_t answer  = toml::get<std::int64_t>(answer_);
```

When you pass an exact TOML type that does not require type conversion,
`toml::get` returns a reference through which you can modify the content
(if the `toml::value` is `const`, it returns `const` reference).

```cpp
toml::value   data    = toml::parse("sample.toml");
toml::value   answer_ = toml::get<toml::table >(data).at("answer");
toml::integer& answer = toml::get<toml::integer>(answer_);
answer = 6 * 9; // write to data.answer. now `answer_` contains 54.
```

If the specified type requires conversion, you can't take a reference to the value.
See also [underlying types](#underlying-types).

It also throws a `toml::type_error` if the type differs.

### `as_xxx`

You can also use a member function to cast a value.

```cpp
const std::int64_t answer = data.as_table().at("answer").as_integer();
```

It also throws a `toml::type_error` if the type differs. If you are sure that
the value `v` contains a value of the specified type, you can suppress checking
by passing `std::nothrow`.

```cpp
const auto& answer = data.as_table().at("answer");
if(answer.is_integer() && answer.as_integer(std::nothrow) == 42)
{
    std::cout << "value is 42" << std::endl;
}
```

If `std::nothrow` is passed, the functions are marked as noexcept.

By casting a `toml::value` into an array or a table, you can iterate over the
elements.

```cpp
const auto data = toml::parse("example.toml");
std::cout << "keys in the top-level table are the following: \n";
for(const auto& kv : data.as_table())
{
    std::cout << kv.first << '\n';
}
for(const auto& [k, v] : data.as_table()) // or in C++17
{
    std::cout << k << '\n';
}


const auto& fruits = toml::find(data, "fruits");
for(const auto& v : fruits.as_array())
{
    std::cout << toml::find<std::string>(v, "name") << '\n';
}
```

The full list of the functions is below.

```cpp
namespace toml {
class value {
    // ...
    const boolean&         as_boolean()         const&;
    const integer&         as_integer()         const&;
    const floating&        as_floating()        const&;
    const string&          as_string()          const&;
    const offset_datetime& as_offset_datetime() const&;
    const local_datetime&  as_local_datetime()  const&;
    const local_date&      as_local_date()      const&;
    const local_time&      as_local_time()      const&;
    const array&           as_array()           const&;
    const table&           as_table()           const&;
    // --------------------------------------------------------
    // non-const version
    boolean&               as_boolean()         &;
    // ditto...
    // --------------------------------------------------------
    // rvalue version
    boolean&&              as_boolean()         &&;
    // ditto...

    // --------------------------------------------------------
    // noexcept versions ...
    const boolean&         as_boolean(const std::nothrow_t&) const& noexcept;
    boolean&               as_boolean(const std::nothrow_t&) &      noexcept;
    boolean&&              as_boolean(const std::nothrow_t&) &&     noexcept;
    // ditto...
};
} // toml
```

### `at()`

You can access to the element of a table and an array by `toml::basic_value::at`.

```cpp
const toml::value v{1,2,3,4,5};
std::cout << v.at(2).as_integer() << std::endl; // 3

const toml::value v{{"foo", 42}, {"bar", 3.14}};
std::cout << v.at("foo").as_integer() << std::endl; // 42
```

If an invalid key (integer for a table, string for an array), it throws
`toml::type_error` for the conversion. If the provided key is out-of-range,
it throws `std::out_of_range`.

Note that, although `std::string` has `at()` member function, `toml::value::at`
throws if the contained type is a string. Because `std::string` does not
contain `toml::value`.

### `operator[]`

You can also access to the element of a table and an array by
`toml::basic_value::operator[]`.

```cpp
const toml::value v{1,2,3,4,5};
std::cout << v[2].as_integer() << std::endl; // 3

const toml::value v{{"foo", 42}, {"bar", 3.14}};
std::cout << v["foo"].as_integer() << std::endl; // 42
```

When you access to a `toml::value` that is not initialized yet via
`operator[](const std::string&)`, the `toml::value` will be a table,
just like the `std::map`.

```cpp
toml::value v; // not initialized as a table.
v["foo"] = 42; // OK. `v` will be a table.
```

Contrary, if you access to a `toml::value` that contains an array via `operator[]`,
it does not check anything. It converts `toml::value` without type check and then
access to the n-th element without boundary check, just like the `std::vector::operator[]`.

```cpp
toml::value v; // not initialized as an array
v[2] = 42;     // error! UB
```

Please make sure that the `toml::value` has an array inside when you access to
its element via `operator[]`.

## Checking value type

You can check the type of a value by `is_xxx` function.

```cpp
const toml::value v = /* ... */;
if(v.is_integer())
{
    std::cout << "value is an integer" << std::endl;
}
```

The complete list of the functions is below.

```cpp
namespace toml {
class value {
    // ...
    bool is_boolean()         const noexcept;
    bool is_integer()         const noexcept;
    bool is_floating()        const noexcept;
    bool is_string()          const noexcept;
    bool is_offset_datetime() const noexcept;
    bool is_local_datetime()  const noexcept;
    bool is_local_date()      const noexcept;
    bool is_local_time()      const noexcept;
    bool is_array()           const noexcept;
    bool is_table()           const noexcept;
    bool is_uninitialized()   const noexcept;
    // ...
};
} // toml
```

Also, you can get `enum class value_t` from `toml::value::type()`.

```cpp
switch(data.at("something").type())
{
    case toml::value_t::integer:  /*do some stuff*/ ; break;
    case toml::value_t::floating: /*do some stuff*/ ; break;
    case toml::value_t::string :  /*do some stuff*/ ; break;
    default : throw std::runtime_error(
        "unexpected type : " + toml::stringize(data.at("something").type()));
}
```

The complete list of the `enum`s can be found in the section
[underlying types](#underlying-types).

The `enum`s can be used as a parameter of `toml::value::is` function like the following.

```cpp
toml::value v = /* ... */;
if(v.is(toml::value_t::boolean)) // ...
```

## More about conversion

Since `toml::find` internally uses `toml::get`, all the following examples work
with both `toml::get` and `toml::find`.

### Converting an array

You can get any kind of `container` class from a `toml::array`
except for `map`-like classes.

``` cpp
// # sample.toml
// numbers = [1,2,3]

const auto numbers = toml::find(data, "numbers");

const auto vc  = toml::get<std::vector<int>  >(numbers);
const auto ls  = toml::get<std::list<int>    >(numbers);
const auto dq  = toml::get<std::deque<int>   >(numbers);
const auto ar  = toml::get<std::array<int, 3>>(numbers);
// if the size of data.at("numbers") is larger than that of std::array,
// it will throw toml::type_error because std::array is not resizable.
```

Surprisingly, you can convert `toml::array` into `std::pair` and `std::tuple`.

```cpp
// numbers = [1,2,3]
const auto tp = toml::get<std::tuple<short, int, unsigned int>>(numbers);
```

This functionality is helpful when you have a toml file like the following.

```toml
array_of_arrays = [[1, 2, 3], ["foo", "bar", "baz"]] # toml allows this
```

What is the corresponding C++ type?
Obviously, it is a `std::pair` of `std::vector`s.

```cpp
const auto array_of_arrays = toml::find(data, "array_of_arrays");
const auto aofa = toml::get<
    std::pair<std::vector<int>, std::vector<std::string>>
    >(array_of_arrays);
```

If you don't know the type of the elements, you can use `toml::array`,
which is a `std::vector` of `toml::value`, instead.

```cpp
const auto a_of_a = toml::get<toml::array>(array_of_arrays);
const auto first  = toml::get<std::vector<int>>(a_of_a.at(0));
```

You can change the implementation of `toml::array` with `std::deque` or some
other array-like container. See [Customizing containers](#customizing-containers)
for detail.

### Converting a table

When all the values of the table have the same type, toml11 allows you to
convert a `toml::table` to a `map` that contains the convertible type.

```toml
[tab]
key1 = "foo" # all the values are
key2 = "bar" # toml String
```

```cpp
const auto data = toml::parse("sample.toml");
const auto tab = toml::find<std::map<std::string, std::string>>(data, "tab");
std::cout << tab["key1"] << std::endl; // foo
std::cout << tab["key2"] << std::endl; // bar
```

But since `toml::table` is just an alias of `std::unordered_map<toml::key, toml::value>`,
normally you don't need to convert it because it has all the functionalities that
`std::unordered_map` has (e.g. `operator[]`, `count`, and `find`). In most cases
`toml::table` is sufficient.

```cpp
toml::table tab = toml::get<toml::table>(data);
if(data.count("title") != 0)
{
    data["title"] = std::string("TOML example");
}
```

You can change the implementation of `toml::table` with `std::map` or some
other map-like container. See [Customizing containers](#customizing-containers)
for detail.

### Getting an array of tables

An array of tables is just an array of tables.
You can get it in completely the same way as the other arrays and tables.

```toml
# sample.toml
array_of_inline_tables = [{key = "value1"}, {key = "value2"}, {key = "value3"}]

[[array_of_tables]]
key = "value4"
[[array_of_tables]]
key = "value5"
[[array_of_tables]]
key = "value6"
```

```cpp
const auto data = toml::parse("sample.toml");
const auto aot1 = toml::find<std::vector<toml::table>>(data, "array_of_inline_tables");
const auto aot2 = toml::find<std::vector<toml::table>>(data, "array_of_tables");
```

### Cost of conversion

Although conversion through `toml::(get|find)` is convenient, it has additional
copy-cost because it copies data contained in `toml::value` to the
user-specified type. Of course in some cases this overhead is not ignorable.

```cpp
// the following code constructs a std::vector.
// it requires heap allocation for vector and element conversion.
const auto array = toml::find<std::vector<int>>(data, "foo");
```

By passing the exact types, `toml::get` returns reference that has no overhead.

``` cpp
const auto& tab     = toml::find<toml::table>(data, "tab");
const auto& numbers = toml::find<toml::array>(data, "numbers");
```

Also, `as_xxx` are zero-overhead because they always return a reference.

``` cpp
const auto& tab     = toml::find(data, "tab"    ).as_table();
const auto& numbers = toml::find(data, "numbers").as_array();
```

In this case you need to call `toml::get` each time you access to
the element of `toml::array` because `toml::array` is an array of `toml::value`.

```cpp
const auto& num0 = toml::get<toml::integer>(numbers.at(0));
const auto& num1 = toml::get<toml::integer>(numbers.at(1));
const auto& num2 = toml::get<toml::integer>(numbers.at(2));
```

### Converting datetime and its variants

TOML v0.5.0 has 4 different datetime objects, `local_date`, `local_time`,
`local_datetime`, and `offset_datetime`.

Since `local_date`, `local_datetime`, and `offset_datetime` represent a time
point, you can convert them to `std::chrono::system_clock::time_point`.

Contrary, `local_time` does not represents a time point because they lack a
date information, but it can be converted to `std::chrono::duration` that
represents a duration from the beginning of the day, `00:00:00.000`.

```toml
# sample.toml
date = 2018-12-23
time = 12:30:00
l_dt = 2018-12-23T12:30:00
o_dt = 2018-12-23T12:30:00+09:30
```

```cpp
const auto data = toml::parse("sample.toml");

const auto date = toml::get<std::chrono::system_clock::time_point>(data.at("date"));
const auto l_dt = toml::get<std::chrono::system_clock::time_point>(data.at("l_dt"));
const auto o_dt = toml::get<std::chrono::system_clock::time_point>(data.at("o_dt"));

const auto time = toml::get<std::chrono::minutes>(data.at("time")); // 12 * 60 + 30 min
```

`local_date` and `local_datetime` are assumed to be in the local timezone when
they are converted into `time_point`. On the other hand, `offset_datetime` only
uses the offset part of the data and it does not take local timezone into account.

To contain datetime data, toml11 defines its own datetime types.
For more detail, you can see the definitions in [toml/datetime.hpp](toml/datetime.hpp).

## Getting with a fallback

`toml::find_or` returns a default value if the value is not found or has a
different type.

```cpp
const auto data = toml::parse("example.toml");
const auto num  = toml::find_or(data, "num", 42);
```

It works recursively if you pass several keys for subtables.
In that case, the last argument is considered to be the optional value.
All other arguments between `toml::value` and the optional value are considered as keys.

```cpp
// [fruit.physical]
// color = "red"
auto data  = toml::parse("fruit.toml");
auto color = toml::find_or(data, "fruit", "physical", "color", "red");
//                               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^  ^^^^^
//                               arguments                     optional value
```

Also, `toml::get_or` returns a default value if `toml::get<T>` failed.

```cpp
toml::value v("foo"); // v contains String
const int value = toml::get_or(v, 42); // conversion fails. it returns 42.
```

These functions automatically deduce what type you want to get
from the default value you passed.

To get a reference through this function, take care about the default value.

```cpp
toml::value v("foo"); // v contains String
toml::integer& i = toml::get_or(v, 42); // does not work because binding `42`
                                        // to `integer&` is invalid
toml::integer opt = 42;
toml::integer& i = toml::get_or(v, opt); // this works.
```

## Expecting conversion

By using `toml::expect`, you will get your expected value or an error message
without throwing `toml::type_error`.

```cpp
const auto value = toml::expect<std::string>(data.at("title"));
if(value.is_ok()) {
    std::cout << value.unwrap() << std::endl;
} else {
    std::cout << value.unwrap_err() << std::endl;
}
```

Also, you can pass a function object to modify the expected value.

```cpp
const auto value = toml::expect<int>(data.at("number"))
    .map(// function that receives expected type (here, int)
    [](const int number) -> double {
        return number * 1.5 + 1.0;
    }).unwrap_or(/*default value =*/ 3.14);
```

## Visiting a toml::value

toml11 provides `toml::visit` to apply a function to `toml::value` in the
same way as `std::variant`.

```cpp
const toml::value v(3.14);
toml::visit([](const auto& val) -> void {
        std::cout << val << std::endl;
    }, v);
```

The function object that would be passed to `toml::visit` must be able to
receive all the possible TOML types. Also, the result types should be the same
each other.

## Constructing a toml::value

`toml::value` can be constructed in various ways.

```cpp
toml::value v(true);     // boolean
toml::value v(42);       // integer
toml::value v(3.14);     // floating
toml::value v("foobar"); // string
toml::value v(toml::local_date(2019, toml::month_t::Apr, 1)); // date
toml::value v{1, 2, 3, 4, 5};                                 // array
toml::value v{{"foo", 42}, {"bar", 3.14}, {"baz", "qux"}};    // table
```

When constructing a string, you can choose to use either literal or basic string.
By default, it will be a basic string.

```cpp
toml::value v("foobar", toml::string_t::basic  );
toml::value v("foobar", toml::string_t::literal);
```

Datetime objects can be constructed from `std::tm` and
`std::chrono::system_clock::time_point`. But you need to specify what type
you use to avoid ambiguity.

```cpp
const auto now = std::chrono::system_clock::now();
toml::value v(toml::local_date(now));
toml::value v(toml::local_datetime(now));
toml::value v(toml::offset_datetime(now));
```

Since local time is not equivalent to a time point, because it lacks date
information, it will be constructed from `std::chrono::duration`.

```cpp
toml::value v(toml::local_time(std::chrono::hours(10)));
```

You can construct an array object not only from `initializer_list`, but also
from STL containers. In that case, the element type must be convertible to
`toml::value`.

```cpp
std::vector<int> vec{1,2,3,4,5};
toml::value v(vec);
```

When you construct an array value, all the elements of `initializer_list`
must be convertible into `toml::value`.

If a `toml::value` has an array, you can `push_back` an element in it.

```cpp
toml::value v{1,2,3,4,5};
v.push_back(6);
```

`emplace_back` also works.

## Preserving comments

toml11 v3 or later allows you yo choose whether comments are preserved or not via template parameter

```cpp
const auto data1 = toml::parse<toml::discard_comments >("example.toml");
const auto data2 = toml::parse<toml::preserve_comments>("example.toml");
```

or macro definition.

```cpp
#define TOML11_PRESERVE_COMMENTS_BY_DEFAULT
#include <toml11/toml.hpp>
```

This feature is controlled by template parameter in `toml::basic_value<...>`.
`toml::value` is an alias of `toml::basic_value<...>`.

If template parameter is explicitly specified, the return value of `toml::parse`
will be `toml::basic_value<toml::preserve_comments>`.
If the macro is defined, the alias `toml::value` will be
`toml::basic_value<toml::preserve_comments>`.

Comments related to a value can be obtained by `toml::value::comments()`.
The return value has the same interface as `std::vector<std::string>`.

```cpp
const auto& com = v.comments();
for(const auto& c : com)
{
    std::cout << c << std::endl;
}
```

Comments just before and just after (within the same line) a value are kept in a value.

```toml
# this is a comment for v1.
v1 = "foo"

v2 = "bar" # this is a comment for v2.
# Note that this comment is NOT a comment for v2.

# this comment is not related to any value
# because there are empty lines between v3.
# this comment will be ignored even if you set `preserve_comments`.

# this is a comment for v3
# this is also a comment for v3.
v3 = "baz" # ditto.
```

Each comment line becomes one element of a `std::vector`.

Hash signs will be removed, but spaces after hash sign will not be removed.

```cpp
v1.comments().at(0) == " this is a comment for v1."s;

v2.comments().at(1) == " this is a comment for v1."s;

v3.comments().at(0) == " this is a comment for v3."s;
v3.comments().at(1) == " this is also a comment for v3."s;
v3.comments().at(2) == " ditto."s;
```

Note that a comment just after an opening brace of an array will not be a
comment for the array.

```toml
# this is a comment for a.
a = [ # this is not a comment for a. this will be ignored.
  1, 2, 3,
  # this is a comment for `42`.
  42, # this is also a comment for `42`.
  5
] # this is a comment for a.
```

You can also append and modify comments.
The interfaces are the same as `std::vector<std::string>`.

```cpp
toml::basic_value<toml::preserve_comments> v(42);
v.comments().push_back(" add this comment.");
// # add this comment.
// i = 42
```

Also, you can pass a `std::vector<std::string>` when constructing a
`toml::basic_value<toml::preserve_comments>`.

```cpp
std::vector<std::string> comments{"comment 1", "comment 2"};
const toml::basic_value<toml::preserve_comments> v1(42, std::move(comments));
const toml::basic_value<toml::preserve_comments> v2(42, {"comment 1", "comment 2"});
```

When `toml::discard_comments` is chosen, comments will not be contained in a value.
`value::comments()` will always be kept empty.
All the modification on comments would be ignored.
All the element access in a `discard_comments` causes the same error as accessing
an element of an empty `std::vector`.

The comments will also be serialized. If comments exist, those comments will be
added just before the values.

__NOTE__: Result types from `toml::parse(...)` and
`toml::parse<toml::preserve_comments>(...)` are different.

## Customizing containers

Actually, `toml::basic_value` has 3 template arguments.

```cpp
template<typename Comment, // discard/preserve_comment
         template<typename ...> class Table = std::unordered_map,
         template<typename ...> class Array = std::vector>
class basic_value;
```

This enables you to change the containers used inside. E.g. you can use
`std::map` to contain a table object instead of `std::unordered_map`.
And also can use `std::deque` as a array object instead of `std::vector`.

You can set these parameters while calling `toml::parse` function.

```cpp
const auto data = toml::parse<
    toml::preserve_comments, std::map, std::deque
    >("example.toml");
```

Needless to say, the result types from `toml::parse(...)` and
`toml::parse<Com, Map, Cont>(...)` are different (unless you specify the same
types as default).

Note that, since `toml::table` and `toml::array` is an alias for a table and an
array of a default `toml::value`, so it is different from the types actually
contained in a `toml::basic_value` when you customize containers.
To get the actual type in a generic way, use
`typename toml::basic_type<C, T, A>::table_type` and
`typename toml::basic_type<C, T, A>::array_type`.

## TOML literal

toml11 supports `"..."_toml` literal.
It accept both a bare value and a file content.

```cpp
using namespace toml::literals::toml_literals;

// `_toml` can convert a bare value without key
const toml::value v = u8"0xDEADBEEF"_toml;
// v is an Integer value containing 0xDEADBEEF.

// raw string literal (`R"(...)"` is useful for this purpose)
const toml::value t = u8R"(
    title = "this is TOML literal"
    [table]
    key = "value"
)"_toml;
// the literal will be parsed and the result will be contained in t
```

The literal function is defined in the same way as the standard library literals
such as `std::literals::string_literals::operator""s`.

```cpp
namespace toml
{
inline namespace literals
{
inline namespace toml_literals
{
toml::value operator"" _toml(const char* str, std::size_t len);
} // toml_literals
} // literals
} // toml
```

Access to the operator can be gained with `using namespace toml::literals;`,
`using namespace toml::toml_literals`, and `using namespace toml::literals::toml_literals`.

Note that a key that is composed only of digits is allowed in TOML.
And, unlike the file parser, toml-literal allows a bare value without a key.
Thus it is difficult to distinguish arrays having integers and definitions of
tables that are named as digits.
Currently, literal `[1]` becomes a table named "1".
To ensure a literal to be considered as an array with one element, you need to
add a comma after the first element (like `[1,]`).

```cpp
"[1,2,3]"_toml;   // This is an array
"[table]"_toml;   // This is a table that has an empty table named "table" inside.
"[[1,2,3]]"_toml; // This is an array of arrays
"[[table]]"_toml; // This is a table that has an array of tables inside.

"[[1]]"_toml;     // This literal is ambiguous.
                  // Currently, it becomes an empty array of table named "1".
"1 = [{}]"_toml;  // This is a table that has an array of table named 1.
"[[1,]]"_toml;    // This is an array of arrays.
"[[1],]"_toml;    // ditto.
```

NOTE: `_toml` literal returns a `toml::value`  that does not have comments.

## Conversion between toml value and arbitrary types

You can also use `toml::get` and other related functions with the types
you defined after you implement a way to convert it.

```cpp
namespace ext
{
struct foo
{
    int         a;
    double      b;
    std::string c;
};
} // ext

const auto data = toml::parse("example.toml");

// to do this
const foo f = toml::find<ext::foo>(data, "foo");
```

There are 3 ways to use `toml::get` with the types that you defined.

The first one is to implement `from_toml(const toml::value&)` member function.

```cpp
namespace ext
{
struct foo
{
    int         a;
    double      b;
    std::string c;

    void from_toml(const toml::value& v)
    {
        this->a = toml::find<int        >(v, "a");
        this->b = toml::find<double     >(v, "b");
        this->c = toml::find<std::string>(v, "c");
        return;
    }
};
} // ext
```

In this way, because `toml::get` first constructs `foo` without arguments,
the type should be default-constructible.

The second is to implement `constructor(const toml::value&)`.

```cpp
namespace ext
{
struct foo
{
    explicit foo(const toml::value& v)
        : a(toml::find<int>(v, "a")), b(toml::find<double>(v, "b")),
          c(toml::find<std::string>(v, "c"))
    {}

    int         a;
    double      b;
    std::string c;
};
} // ext
```

Note that implicit default constructor declaration will be suppressed
when a constructor is defined. If you want to use the struct (here, `foo`)
in a container (e.g. `std::vector<foo>`), you may need to define default
constructor explicitly.

The third is to implement specialization of `toml::from` for your type.

```cpp
namespace ext
{
struct foo
{
    int         a;
    double      b;
    std::string c;
};
} // ext

namespace toml
{
template<>
struct from<ext::foo>
{
    static ext::foo from_toml(const value& v)
    {
        ext::foo f;
        f.a = find<int        >(v, "a");
        f.b = find<double     >(v, "b");
        f.c = find<std::string>(v, "c");
        return f;
    }
};
} // toml
```

In this way, since the conversion function is defined outside of the class,
you can add conversion between `toml::value` and classes defined in another library.

In some cases, a class has a templatized constructor that takes a template, `T`.
It confuses `toml::get/find<T>` because it makes the class "constructible" from
`toml::value`. To avoid this problem, `toml::from` and `from_toml` always
precede constructor. It makes easier to implement conversion between
`toml::value` and types defined in other libraries because it skips constructor.

But, importantly, you cannot define `toml::from<T>` and `T.from_toml` at the same
time because it causes ambiguity in the overload resolution of `toml::get<T>` and `toml::find<T>`.

So the precedence is `toml::from<T>` == `T.from_toml()` > `T(toml::value)`.

If you want to convert any versions of `toml::basic_value`,
you need to templatize the conversion function as follows.

```cpp
struct foo
{
    template<typename C, template<typename ...> class M, template<typename ...> class A>
    void from_toml(const toml::basic_value<C, M, A>& v)
    {
        this->a = toml::find<int        >(v, "a");
        this->b = toml::find<double     >(v, "b");
        this->c = toml::find<std::string>(v, "c");
        return;
    }
};
// or
namespace toml
{
template<>
struct from<ext::foo>
{
    template<typename C, template<typename ...> class M, template<typename ...> class A>
    static ext::foo from_toml(const basic_value<C, M, A>& v)
    {
        ext::foo f;
        f.a = find<int        >(v, "a");
        f.b = find<double     >(v, "b");
        f.c = find<std::string>(v, "c");
        return f;
    }
};
} // toml
```

----

The opposite direction is also supported in a similar way. You can directly
pass your type to `toml::value`'s constructor by introducing `into_toml` or
`toml::into<T>`.

```cpp
namespace ext
{
struct foo
{
    int         a;
    double      b;
    std::string c;

    toml::value into_toml() const // you need to mark it const.
    {
        return toml::value{{"a", this->a}, {"b", this->b}, {"c", this->c}};
    }
};
} // ext

ext::foo    f{42, 3.14, "foobar"};
toml::value v(f);
```

The definition of `toml::into<T>` is similar to `toml::from<T>`.

```cpp
namespace ext
{
struct foo
{
    int         a;
    double      b;
    std::string c;
};
} // ext

namespace toml
{
template<>
struct into<ext::foo>
{
    static toml::value into_toml(const ext::foo& f)
    {
        return toml::value{{"a", f.a}, {"b", f.b}, {"c", f.c}};
    }
};
} // toml

ext::foo    f{42, 3.14, "foobar"};
toml::value v(f);
```

Any type that can be converted to `toml::value`, e.g. `int`, `toml::table` and
`toml::array` are okay to return from `into_toml`.

You can also return a custom `toml::basic_value` from `toml::into`.

```cpp
namespace toml
{
template<>
struct into<ext::foo>
{
    static toml::basic_value<toml::preserve_comments> into_toml(const ext::foo& f)
    {
        toml::basic_value<toml::preserve_comments> v{{"a", f.a}, {"b", f.b}, {"c", f.c}};
        v.comments().push_back(" comment");
        return v;
    }
};
} // toml
```

But note that, if this `basic_value` would be assigned into other `toml::value`
that discards `comments`, the comments would be dropped.

### Macro to automatically define conversion functions

There is a helper macro that automatically generates conversion functions `from` and `into` for a simple struct.

```cpp
namespace foo
{
struct Foo
{
    std::string s;
    double      d;
    int         i;
};
} // foo

TOML11_DEFINE_CONVERSION_NON_INTRUSIVE(foo::Foo, s, d, i)

int main()
{
    const auto file = toml::parse("example.toml");
    auto f = toml::find<foo::Foo>(file, "foo");
}
```

And then you can use `toml::find<foo::Foo>(file, "foo");`

**Note** that, because of a slight difference in implementation of preprocessor between gcc/clang and MSVC, [you need to define `/Zc:preprocessor`](https://github.com/ToruNiina/toml11/issues/139#issuecomment-803683682) to use it in MSVC (Thank you @glebm !).

## Formatting user-defined error messages

When you encounter an error after you read the toml value, you may want to
show the error with the value.

toml11 provides you a function that formats user-defined error message with
related values. With a code like the following,

```cpp
const auto value = toml::find<int>(data, "num");
if(value < 0)
{
    std::cerr << toml::format_error("[error] value should be positive",
                                    data.at("num"), "positive number required")
              << std::endl;
}
```

you will get an error message like this.

```console
[error] value should be positive
 --> example.toml
 3 | num = -42
   |       ~~~ positive number required
```

When you pass two values to `toml::format_error`,

```cpp
const auto min = toml::find<int>(range, "min");
const auto max = toml::find<int>(range, "max");
if(max < min)
{
    std::cerr << toml::format_error("[error] max should be larger than min",
                                    data.at("min"), "minimum number here",
                                    data.at("max"), "maximum number here");
              << std::endl;
}
```

you will get an error message like this.

```console
[error] max should be larger than min
 --> example.toml
 3 | min = 54
   |       ~~ minimum number here
 ...
 4 | max = 42
   |       ~~ maximum number here
```

You can print hints at the end of the message.

```cpp
std::vector<std::string> hints;
hints.push_back("positive number means n >= 0.");
hints.push_back("negative number is not positive.");
std::cerr << toml::format_error("[error] value should be positive",
                                data.at("num"), "positive number required", hints)
          << std::endl;
```

```console
[error] value should be positive
 --> example.toml
 2 | num = 42
   |       ~~ positive number required
   |
Hint: positive number means n >= 0.
Hint: negative number is not positive.
```

## Obtaining location information

You can also format error messages in your own way by using `source_location`.

```cpp
struct source_location
{
    std::uint_least32_t line()      const noexcept;
    std::uint_least32_t column()    const noexcept;
    std::uint_least32_t region()    const noexcept;
    std::string const&  file_name() const noexcept;
    std::string const&  line_str()  const noexcept;
};
// +-- line()       +--- length of the region (here, region() == 9)
// v            .---+---.
// 12 | value = "foo bar" <- line_str() returns the line itself.
//              ^-------- column() points here
```

You can get this by
```cpp
const toml::value           v   = /*...*/;
const toml::source_location loc = v.location();
```

## Exceptions

The following `exception` classes inherits `toml::exception` that inherits
`std::exception`.

```cpp
namespace toml {
struct exception      : public std::exception  {/**/};
struct syntax_error   : public toml::exception {/**/};
struct type_error     : public toml::exception {/**/};
struct internal_error : public toml::exception {/**/};
} // toml
```

`toml::exception` has `toml::exception::location()` member function that returns
`toml::source_location`, in addition to `what()`.

```cpp
namespace toml {
struct exception : public std::exception
{
    // ...
    source_location const& location() const noexcept;
};
} // toml
```

It represents where the error occurs.

`syntax_error` will be thrown from `toml::parse` and `_toml` literal.
`type_error` will be thrown from `toml::get/find`, `toml::value::as_xxx()`, and
other functions that takes a content inside of `toml::value`.

Note that, currently, from `toml::value::at()` and `toml::find(value, key)`
may throw an `std::out_of_range` that does not inherits `toml::exception`.

Also, in some cases, most likely in the file open error, it will throw an
`std::runtime_error`.

## Colorize Error Messages

By defining `TOML11_COLORIZE_ERROR_MESSAGE`, the error messages from
`toml::parse` and `toml::find|get` will be colorized. By default, this feature
is turned off.

With the following toml file taken from `toml-lang/toml/tests/hard_example.toml`,

```toml
[error]
array = [
         "This might most likely happen in multiline arrays",
         Like here,
         "or here,
         and here"
        ]     End of array comment, forgot the #
```

the error message would be like this.

![error-message-1](https://github.com/ToruNiina/toml11/blob/misc/misc/toml11-err-msg-1.png)

With the following,

```toml
[error]
# array = [
#          "This might most likely happen in multiline arrays",
#          Like here,
#          "or here,
#          and here"
#         ]     End of array comment, forgot the #
number = 3.14  pi <--again forgot the #
```

the error message would be like this.

![error-message-2](https://github.com/ToruNiina/toml11/blob/misc/misc/toml11-err-msg-2.png)

The message would be messy when it is written to a file, not a terminal because
it uses [ANSI escape code](https://en.wikipedia.org/wiki/ANSI_escape_code).

Without `TOML11_COLORIZE_ERROR_MESSAGE`, you can still colorize user-defined
error message by passing `true` to the `toml::format_error` function.
If you define `TOML11_COLORIZE_ERROR_MESSAGE`, the value is `true` by default.
If not, the default value would be `false`.

```cpp
std::cerr << toml::format_error("[error] value should be positive",
                                data.at("num"), "positive number required",
                                hints, /*colorize = */ true) << std::endl;
```

Note: It colorizes `[error]` in red. That means that it detects `[error]` prefix
at the front of the error message. If there is no `[error]` prefix,
`format_error` adds it to the error message.

Compared to the `TOML11_COLORIZE_ERROR_MESSAGE` macro that enables colorization
statically, toml11 provides `toml::color::enable` & `toml::color::disable`
functions to dynamically change the color mode. This feature overwrites
`TOML11_COLORIZE_ERROR_MESSAGE` and the `colorize` argument of
`toml::format_error` when you call `enable`.

Note: If either `TOML11_COLORIZE_ERROR_MESSAGE` is defined or the `colorize`
argument is used, it takes precedence, meaning that `disable` won't work.
Accordingly, we highly recommend using only one of them.

```cpp
toml::color::enable();  // enable colorization
toml::color::disable(); // disable colorization
```

If you use user-defined error message, you can manage the setting as follows:

```cpp
toml::color::enable();
std::cerr << toml::format_error("[error] value should be positive",
                                data.at("num"), "positive number required",
                                hints) << std::endl; // colorized

toml::color::disable();
std::cerr << toml::format_error("[error] value should be positive",
                                data.at("num"), "positive number required",
                                hints) << std::endl; // NOT colorized
```

Or you may use toml11 in your application like:

```cpp
std::vector<std::string> args(argv + 1, argv + argc);
auto result = std::find(args.begin(), args.end(), "--color");
if (result != args.end()) {
    toml::color::enable();
} else {
    toml::color::disable();
}

// use toml11 ...
```

## Opting out of the default `[error]` prefix

toml11 prints error messages with the `[error]` prefix by default.
Defining `TOML11_NO_ERROR_PREFIX` will let toml11 omit the prefix regardless of colorized or not in case you would use a custom prefix, such as `Error:`.

```cpp
#define TOML11_NO_ERROR_PREFIX
```

## Serializing TOML data

toml11 enables you to serialize data into toml format.

```cpp
const toml::value data{{"foo", 42}, {"bar", "baz"}};
std::cout << data << std::endl;
// bar = "baz"
// foo = 42
```

toml11 automatically makes a small table and small array inline.
You can specify the width to make them inline by `std::setw` for streams.

```cpp
const toml::value data{
    {"qux",    {{"foo", 42}, {"bar", "baz"}}},
    {"quux",   {"small", "array", "of", "strings"}},
    {"foobar", {"this", "array", "of", "strings", "is", "too", "long",
                "to", "print", "into", "single", "line", "isn't", "it?"}},
};

// the threshold becomes 80.
std::cout << std::setw(80) << data << std::endl;
// foobar = [
// "this","array","of","strings","is","too","long","to","print","into",
// "single","line","isn't","it?",
// ]
// quux = ["small","array","of","strings"]
// qux = {bar="baz",foo=42}


// the width is 0. nothing become inline.
std::cout << std::setw(0) << data << std::endl;
// foobar = [
// "this",
// ... (snip)
// "it?",
// ]
// quux = [
// "small",
// "array",
// "of",
// "strings",
// ]
// [qux]
// bar = "baz"
// foo = 42
```

It is recommended to set width before printing data. Some I/O functions changes
width to 0, and it makes all the stuff (including `toml::array`) multiline.
The resulting files becomes too long.

To control the precision of floating point numbers, you need to pass
`std::setprecision` to stream.

```cpp
const toml::value data{
    {"pi", 3.141592653589793},
    {"e",  2.718281828459045}
};
std::cout << std::setprecision(17) << data << std::endl;
// e = 2.7182818284590451
// pi = 3.1415926535897931
std::cout << std::setprecision( 7) << data << std::endl;
// e = 2.718282
// pi = 3.141593
```

There is another way to format toml values, `toml::format()`.
It returns `std::string` that represents a value.

```cpp
const toml::value v{{"a", 42}};
const std::string fmt = toml::format(v);
// a = 42
```

Note that since `toml::format` formats a value, the resulting string may lack
the key value.

```cpp
const toml::value v{3.14};
const std::string fmt = toml::format(v);
// 3.14
```

To control the width and precision, `toml::format` receives optional second and
third arguments to set them. By default, the width is 80 and the precision is
`std::numeric_limits<double>::max_digit10`.

```cpp
const auto serial = toml::format(data, /*width = */ 0, /*prec = */ 17);
```

When you pass a comment-preserving-value, the comment will also be serialized.
An array or a table containing a value that has a comment would not be inlined.

## Underlying types

The toml types (can be used as `toml::*` in this library) and corresponding `enum` names are listed in the table below.

| TOML type      | underlying c++ type                | enum class                       |
| -------------- | ---------------------------------- | -------------------------------- |
| Boolean        | `bool`                             | `toml::value_t::boolean`         |
| Integer        | `std::int64_t`                     | `toml::value_t::integer`         |
| Float          | `double`                           | `toml::value_t::floating`        |
| String         | `toml::string`                     | `toml::value_t::string`          |
| LocalDate      | `toml::local_date`                 | `toml::value_t::local_date`      |
| LocalTime      | `toml::local_time`                 | `toml::value_t::local_time`      |
| LocalDatetime  | `toml::local_datetime`             | `toml::value_t::local_datetime`  |
| OffsetDatetime | `toml::offset_datetime`            | `toml::value_t::offset_datetime` |
| Array          | `array-like<toml::value>`          | `toml::value_t::array`           |
| Table          | `map-like<toml::key, toml::value>` | `toml::value_t::table`           |

`array-like` and `map-like` are the STL containers that works like a `std::vector` and
`std::unordered_map`, respectively. By default, `std::vector` and `std::unordered_map`
are used. See [Customizing containers](#customizing-containers) for detail.

`toml::string` is effectively the same as `std::string` but has an additional
flag that represents a kind of a string, `string_t::basic` and `string_t::literal`.
Although `std::string` is not an exact toml type, still you can get a reference
that points to internal `std::string` by using `toml::get<std::string>()` for convenience.
The most important difference between `std::string` and `toml::string` is that
`toml::string` will be formatted as a TOML string when outputted with `ostream`.
This feature is introduced to make it easy to write a custom serializer.

`Datetime` variants are `struct` that are defined in this library.
Because `std::chrono::system_clock::time_point` is a __time point__,
not capable of representing a Local Time independent from a specific day.

## Unreleased TOML features

After TOML v1.0.0 has been released, some features are added to the main branch
of the TOML spec repository. (see: [CHANGELOG.md in toml-lang/toml repository](https://github.com/toml-lang/toml/blob/main/CHANGELOG.md)).

The following list shows available "unreleased" features that can be activated
by defining a macro named `TOML11_USE_UNRELEASED_FEATURES`.

- Add new `\e` shorthand for the escape character.

## Note about heterogeneous arrays

Although `toml::parse` allows heterogeneous arrays, constructor of `toml::value`
does not. Here the reason is explained.

```cpp
// this won't be compiled
toml::value v{
    "foo", 3.14, 42, {1,2,3,4,5}, {{"key", "value"}}
}
```

There is a workaround for this. By explicitly converting values into
`toml::value`, you can initialize `toml::value` with a heterogeneous array.
Also, you can first initialize a `toml::value` with an array and then
`push_back` into it.

```cpp
// OK!
toml::value v{
    toml::value("foo"), toml::value(3.14), toml::value(42),
    toml::value{1,2,3,4,5}, toml::value{{"key", "value"}}
}

// OK!
toml::value v(toml::array{});
v.push_back("foo");
v.push_back(3.14);

// OK!
toml::array a;
a.push_back("foo");
a.push_back(3.14);
toml::value v(std::move(a));
```

The reason why the first example is not allowed is the following.
Let's assume that you are initializing a `toml::value` with a table.

```cpp
                    // # expecting TOML table.
toml::value v{      // [v]
    {"answer", 42}, // answer = 42
    {"pi",   3.14}, // pi = 3.14
    {"foo", "bar"}  // foo = "bar"
};
```

This is indistinguishable from a (heterogeneous) TOML array definition.

```toml
v = [
    ["answer", 42],
    ["pi",   3.14],
    ["foo", "bar"],
]
```

This means that the above C++ code makes constructor's overload resolution
ambiguous. So a constructor that allows both "table as an initializer-list" and
"heterogeneous array as an initializer-list" cannot be implemented.

Thus, although it is painful, we need to explicitly cast values into
`toml::value` when you initialize heterogeneous array in a C++ code.

```cpp
toml::value v{
    toml::value("foo"), toml::value(3.14), toml::value(42),
    toml::value{1,2,3,4,5}, toml::value{{"key", "value"}}
};
```

## Breaking Changes from v2

Although toml11 is relatively new library (it's three years old now), it had
some confusing and inconvenient user-interfaces because of historical reasons.

Between v2 and v3, those interfaces are rearranged.

- `toml::parse` now returns a `toml::value`, not `toml::table`.
- `toml::value` is now an alias of `toml::basic_value<discard_comment, std::vector, std::unordered_map>`.
  - See [Customizing containers](#customizing-containers) for detail.
- The elements of `toml::value_t` are renamed as `snake_case`.
  - See [Underlying types](#underlying-types) for detail.
- Supports for the CamelCaseNames are dropped.
  - See [Underlying types](#underlying-types) for detail.
- `(is|as)_float` has been removed to make the function names consistent with others.
  - Since `float` is a keyword, toml11 named a float type as `toml::floating`.
  - Also a `value_t` corresponds to `toml::floating` is named `value_t::floating`.
  - So `(is|as)_floating` is introduced and `is_float` has been removed.
  - See [Casting a toml::value](#casting-a-tomlvalue) and [Checking value type](#checking-value-type) for detail.
- An overload of `toml::find` for `toml::table` has been dropped. Use `toml::value` version instead.
  - Because type conversion between a table and a value causes ambiguity while overload resolution
  - Since `toml::parse` now returns a `toml::value`, this feature becomes less important.
  - Also because `toml::table` is a normal STL container, implementing utility function is easy.
  - See [Finding a toml::value](#finding-a-toml-value) for detail.
- An overload of `operator<<` and `toml::format` for `toml::table`s are dropped.
  - Use `toml::value` instead.
  - See [Serializing TOML data](#serializing-toml-data) for detail.
- Interface around comments.
  - See [Preserving Comments](#preserving-comments) for detail.
- An ancient `from_toml/into_toml` has been removed. Use arbitrary type conversion support.
  - See [Conversion between toml value and arbitrary types](#conversion-between-toml-value-and-arbitrary-types) for detail.

Such a big change will not happen in the coming years.

## Running Tests

After cloning this repository, run the following command (thank you @jwillikers
for automating test set fetching!).

```sh
$ mkdir build
$ cd build
$ cmake .. -Dtoml11_BUILD_TEST=ON
$ make
$ make test
```

To run the language agnostic test suite, you need to compile
`tests/check_toml_test.cpp` and pass it to the tester.

## Contributors

I appreciate the help of the contributors who introduced the great feature to this library.

- Guillaume Fraux (@Luthaf)
  - Windows support and CI on Appvayor
  - Intel Compiler support
- Quentin Khan (@xaxousis)
  - Found & Fixed a bug around ODR
  - Improved error messages for invalid keys to show the location where the parser fails
- Petr Beneš (@wbenny)
  - Fixed warnings on MSVC
- Ivan Shynkarenka (@chronoxor)
  - Fixed Visual Studio 2019 warnings
  - Fix compilation error in `<filesystem>` with MinGW
- Khoi Dinh Trinh (@khoitd1997)
  - Fixed warnings while type conversion
- @KerstinKeller
  - Added installation script to CMake
- J.C. Moyer (@jcmoyer)
  - Fixed an example code in the documentation
- Jt Freeman (@blockparty-sh)
  - Fixed feature test macro around `localtime_s`
  - Suppress warnings in Debug mode
- OGAWA Kenichi (@kenichiice)
  - Suppress warnings on intel compiler
  - Fix include path in README
- Jordan Williams (@jwillikers)
  - Fixed clang range-loop-analysis warnings
  - Fixed feature test macro to suppress -Wundef
  - Use cache variables in CMakeLists.txt
  - Automate test set fetching, update and refactor CMakeLists.txt
- Scott McCaskill
  - Parse 9 digits (nanoseconds) of fractional seconds in a `local_time`
- Shu Wang (@halfelf)
  - fix "Finding a value in an array" example in README
- @maass-tv and @SeverinLeonhardt
  - Fix MSVC warning C4866
- Mohammed Alyousef (@MoAlyousef)
  - Made testing optional in CMake
- Alex Merry (@amerry)
  - Add missing include files
- sneakypete81 (@sneakypete81)
  - Fix typo in error message
- Oliver Kahrmann (@founderio)
  - Fix missing filename in error message if parsed file is empty
- Karl Nilsson (@karl-nilsson)
  - Fix many spelling errors
- ohdarling88 (@ohdarling)
  - Fix a bug in a constructor of serializer
- estshorter (@estshorter)
  - Fix MSVC warning C26478
- Philip Top (@phlptp)
  - Improve checking standard library feature availability check
- Louis Marascio (@marascio)
  - Fix free-nonheap-object warning
- Axel Huebl (@ax3l)
  - Make installation optional if the library is embedded
- Ken Matsui (@ken-matsui)
  - Support user-defined error message prefix
  - Support dynamic color mode
- Giel van Schijndel (@muggenhor)
  - Remove needless copy in `parse` function
- Lukáš Hrázký (@lukash)
  - Add a `parse(FILE *)` interface and improve file-related error messages
- spiderman idog (@spiderman-idog)
  - Fix typo in README
- Jajauma's GitHub (@Jajauma)
  - Avoid possible lexer truncation warnings
- Moritz Klammler (@ctcmkl)
  - Many patches in (#200) including:
  - Improve CMake scripts, build process, and test file handling
  - Detect error when `discard_comments` is accessed
  - And more.
- Chris White (@cxw42)
  - Fix address-sanitizer error when parsing literal strings having invalid UTF-8 characters
  - Fix function name in error messages
- offa (@offa)
  - Update checkout action to v3
  - Update Required CMake version
  - Cleanup old CI settings
- Sergey Vidyuk (@VestniK)
  - Fix for case when vector iterator is raw pointer
- Kfir Gollan (@kfirgollan)
  - Add installation example with checkinstall and cmake
- Martin Tournoij (@arp242)
  - Escape control characters in keys
- @DavidKorczynski
  - Add fuzzing test based on ClusterFuzzLite
- Esonhugh Skyworship (@Esonhugh)
  - Fix function signature of `strerror_r` on macos

## Licensing terms

This product is licensed under the terms of the [MIT License](LICENSE).

- Copyright (c) 2017-2024 Toru Niina

All rights reserved.