quickstart.cpp

// ryml: quickstart

/** @addtogroup doc_quickstart
 *
 * This file does a quick tour of ryml. It has multiple self-contained
 * and well-commented samples that illustrate how to use ryml, and how
 * it works.
 *
 * Although this is not a unit test, the samples are written as a
 * sequence of actions and predicate checks to better convey what is
 * the expected result at any stage. And to ensure the code here is
 * correct and up to date, it's also run as part of the CI tests.
 *
 * If something is unclear, please open an issue or send a pull
 * request at https://github.com/biojppm/rapidyaml . If you have an
 * issue while using ryml, it is also encouraged to try to reproduce
 * the issue here, or look first through the relevant section.
 *
 * Happy ryml'ing!
 *
 * ### Some guidance on building
 *
 * The directories that exist side-by-side with this file contain
 * several examples on how to build this with cmake, such that you can
 * hit the ground running. See [the relevant section of the main
 * README](https://github.com/biojppm/rapidyaml/tree/v0.10.0?tab=readme-ov-file#quickstart-samples)
 * for an overview of the different choices. I suggest starting first
 * with the `add_subdirectory` example, treating it just like any
 * other self-contained cmake project.
 *
 * Or very quickly, to build and run this sample on your PC, start by
 * creating this `CMakeLists.txt`:
 * ```cmake
 * cmake_minimum_required(VERSION 3.13)
 * project(ryml-quickstart LANGUAGES CXX)
 * include(FetchContent)
 * FetchContent_Declare(ryml
 *     GIT_REPOSITORY https://github.com/biojppm/rapidyaml.git
 *     GIT_TAG v0.10.0
 *     GIT_SHALLOW FALSE  # ensure submodules are checked out
 * )
 * FetchContent_MakeAvailable(ryml)
 * add_executable(ryml-quickstart ${ryml_SOURCE_DIR}/samples/quickstart.cpp)
 * target_link_libraries(ryml-quickstart ryml::ryml)
 * add_custom_target(run ryml-quickstart
 *     COMMAND $<TARGET_FILE:ryml-quickstart>
 *     DEPENDS ryml-quickstart
 *     COMMENT "running: $<TARGET_FILE:ryml-quickstart>")
 * ```
 * Now run the following commands in the same folder:
 * ```bash
 * # configure the project
 * cmake -S . -B build
 * # build and run
 * cmake --build build --target ryml-quickstart -j
 * # optionally, open in your IDE
 * cmake --open build
 * ```
 *
 * @{ */

/** @cond dev */
int report_checks();
/** @endcond */


/** @defgroup doc_sample_helpers Sample helpers
 *
 * Helper utilities used in the sample.
 */

//-----------------------------------------------------------------------------

// ryml can be used as a single header, or as a simple library:
#if defined(RYML_SINGLE_HEADER) // using the single header directly in the executable
    #define RYML_SINGLE_HDR_DEFINE_NOW
    #include <ryml_all.hpp>
#elif defined(RYML_SINGLE_HEADER_LIB) // using the single header from a library
    #include <ryml_all.hpp>
#else
    #include <ryml.hpp>
    // <ryml_std.hpp> is needed if interop with std containers is
    // desired; ryml itself does not use any STL container.
    // For this sample, we will be using std interop, so...
    #include <ryml_std.hpp> // optional header, provided for std:: interop
    #include <c4/format.hpp> // needed for the examples below
    // optional header, definitions for error utilities to implement
    // error callbacks:
    #include <c4/yml/error.def.hpp>
#endif

// these are needed for the examples below
#include <iostream>
#include <sstream>
#include <vector>
#include <map>
#ifdef C4_EXCEPTIONS
#include <stdexcept>
#else
#include <csetjmp>
#endif


//-----------------------------------------------------------------------------

/** @cond dev */
// CONTENTS:
//
// (Each function addresses a topic and is fully self-contained. Jump
// to the function to find out about its topic.)
void sample_lightning_overview();   ///< lightning overview of most common features
void sample_quick_overview();       ///< quick overview of most common features
void sample_substr();               ///< about ryml's string views (from c4core)
void sample_parse_file();           ///< ready-to-go example of parsing a file from disk
void sample_parse_in_place();       ///< parse a mutable YAML source buffer
void sample_parse_in_arena();       ///< parse a read-only YAML source buffer
void sample_parse_reuse_tree();     ///< parse into an existing tree, maybe into a node
void sample_parse_reuse_parser();   ///< reuse an existing parser
void sample_parse_reuse_tree_and_parser(); ///< how to reuse existing trees and parsers
void sample_iterate_trees();        ///< visit individual nodes and iterate through trees
void sample_create_trees();         ///< programatically create trees
void sample_tree_arena();           ///< interact with the tree's serialization arena
void sample_fundamental_types();    ///< serialize/deserialize fundamental types
void sample_empty_null_values();    ///< serialize/deserialize/query empty or null values
void sample_formatting();           ///< control formatting when serializing/deserializing
void sample_base64();               ///< encode/decode base64
void sample_user_scalar_types();    ///< serialize/deserialize scalar (leaf/string) types
void sample_user_container_types(); ///< serialize/deserialize container (map or seq) types
void sample_std_types();            ///< serialize/deserialize STL containers
void sample_float_precision();      ///< control precision of serialized floats
void sample_emit_to_container();    ///< emit to memory, eg a string or vector-like container
void sample_emit_to_stream();       ///< emit to a stream, eg std::ostream
void sample_emit_to_file();         ///< emit to a FILE*
void sample_emit_nested_node();     ///< pick a nested node as the root when emitting
void sample_style();                ///< query/set node styles
void sample_style_flow_ml_indent(); ///< control indentation of FLOW_ML containers
void sample_style_flow_ml_filter(); ///< set the parser to pick FLOW_SL even if the container is multiline
void sample_json();                 ///< JSON parsing and emitting
void sample_anchors_and_aliases();  ///< deal with YAML anchors and aliases
void sample_anchors_and_aliases_create(); ///< how to create YAML anchors and aliases
void sample_tags();                 ///< deal with YAML type tags
void sample_tag_directives();       ///< deal with YAML tag namespace directives
void sample_docs();                 ///< deal with YAML docs
void sample_error_handler();        ///< set custom error handlers
void sample_error_basic();          ///< handler for basic errors, and obtain a full error message with basic context
void sample_error_parse();          ///< handler for parse errors, and obtain a full error message with parse context
void sample_error_visit();          ///< handler for visit errors, and obtain a full error message with visit context
void sample_error_visit_location(); ///< obtaining the YAML location from a visit error
void sample_global_allocator();     ///< set a global allocator for ryml
void sample_per_tree_allocator();   ///< set per-tree allocators
void sample_static_trees();         ///< how to use static trees in ryml
void sample_location_tracking();    ///< track node YAML source locations in the parsed tree

int main()
{
    sample_lightning_overview();
    sample_quick_overview();
    sample_substr();
    sample_parse_file();
    sample_parse_in_place();
    sample_parse_in_arena();
    sample_parse_reuse_tree();
    sample_parse_reuse_parser();
    sample_parse_reuse_tree_and_parser();
    sample_iterate_trees();
    sample_create_trees();
    sample_tree_arena();
    sample_fundamental_types();
    sample_empty_null_values();
    sample_formatting();
    sample_base64();
    sample_user_scalar_types();
    sample_user_container_types();
    sample_float_precision();
    sample_std_types();
    sample_emit_to_container();
    sample_emit_to_stream();
    sample_emit_to_file();
    sample_emit_nested_node();
    sample_style();
    sample_style_flow_ml_indent();
    sample_style_flow_ml_filter();
    sample_json();
    sample_anchors_and_aliases();
    sample_tags();
    sample_tag_directives();
    sample_docs();
    sample_error_handler();
    sample_error_basic();
    sample_error_parse();
    sample_error_visit();
    sample_error_visit_location();
    sample_global_allocator();
    sample_per_tree_allocator();
    sample_static_trees();
    sample_location_tracking();
    return report_checks();
}

/** @endcond */


//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------

C4_SUPPRESS_WARNING_GCC_CLANG_PUSH
C4_SUPPRESS_WARNING_GCC_CLANG("-Wcast-qual")
C4_SUPPRESS_WARNING_GCC_CLANG("-Wold-style-cast")
C4_SUPPRESS_WARNING_GCC("-Wuseless-cast")

//-----------------------------------------------------------------------------
// first, some helpers used in this quickstart

/** @addtogroup doc_sample_helpers
 *
 * @{ */


/** an example error handler, required for some of the quickstart
 * examples.
 * @ingroup doc_sample_helpers */
struct ErrorHandlerExample
{
    ErrorHandlerExample() : defaults(ryml::get_callbacks()) {}
    ryml::Callbacks defaults;
public:
    // utilities used below
    template<class Fn> bool check_error_occurs(Fn &&fn);
    template<class Fn> bool check_assertion_occurs(Fn &&fn);
    void check_enabled() const;
    void check_disabled() const;
    ryml::Callbacks callbacks();
public:
    // these are the functions that we want to execute on error
    [[noreturn]] void on_error_basic(ryml::csubstr msg, ryml::ErrorDataBasic const& errdata);
    [[noreturn]] void on_error_parse(ryml::csubstr msg, ryml::ErrorDataParse const& errdata);
    [[noreturn]] void on_error_visit(ryml::csubstr msg, ryml::ErrorDataVisit const& errdata);
public:
    // these are the functions that we set ryml to call
    [[noreturn]] static void s_error_basic(ryml::csubstr msg, ryml::ErrorDataBasic const& errdata, void *this_);
    [[noreturn]] static void s_error_parse(ryml::csubstr msg, ryml::ErrorDataParse const& errdata, void *this_);
    [[noreturn]] static void s_error_visit(ryml::csubstr msg, ryml::ErrorDataVisit const& errdata, void *this_);
public:
    // the handlers save these fields to be able to check on them
    // after the error is triggered. They are only valid after an
    // error, and before the next error.
    std::string       saved_msg_short;
    std::string       saved_msg_full;
    std::string       saved_msg_full_with_context;
    ryml::Location    saved_basic_loc;
    ryml::Location    saved_parse_loc;
    ryml::Tree const* saved_visit_tree;
    ryml::id_type     saved_visit_id;
};

/** Shows how to create a scoped error handler.
 * @ingroup doc_sample_helpers */
struct ScopedErrorHandlerExample : public ErrorHandlerExample
{
    ScopedErrorHandlerExample() : ErrorHandlerExample() { ryml::set_callbacks(callbacks()); check_enabled(); }
    ~ScopedErrorHandlerExample() { ryml::set_callbacks(this->defaults); check_disabled(); }
};


// helper functions for sample_parse_file()
template<class CharContainer> CharContainer file_get_contents(const char *filename);
template<class CharContainer> size_t        file_get_contents(const char *filename, CharContainer *v);
template<class CharContainer> void          file_put_contents(const char *filename, CharContainer const& v, const char* access="wb");
void                                        file_put_contents(const char *filename, const char *buf, size_t sz, const char* access);


bool report_check(int line, const char *predicate, bool result);


#if defined(__DOXYGEN__) || defined(_DOXYGEN_)
/// a quick'n'dirty assertion to verify a predicate
#   define CHECK(predicate) assert(predicate)  // enable doxygen to link to the functions called inside CHECK()
#else
#   if !(defined(__GNUC__) && (__GNUC__ == 4 && __GNUC_MINOR__ >= 8))
    /// a quick'n'dirty assertion to verify a predicate
#       define CHECK(predicate) do { if(!report_check(__LINE__, #predicate, (predicate))) { RYML_DEBUG_BREAK(); } } while(0)
#   else // GCC 4.8 has a problem with the CHECK() macro
/// a quick'n'dirty assertion to verify a predicate
#       define CHECK CheckPredicate{__FILE__, __LINE__}
        struct CheckPredicate
        {
            const char *file;
            const int line;
            void operator() (bool result) const
            {
                if (!report_check(line, nullptr, result))
                {
                    RYML_DEBUG_BREAK();
                }
            }
        };
#   endif // GCC 4.8
#endif // doxygen

/** @} */ // doc_sample_helpers


//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------

/** a lightning tour over most features
 * see @ref sample_quick_overview */
void sample_lightning_overview()
{
    // Parse YAML code in place, potentially mutating the buffer:
    char yml_buf[] = "{foo: 1, bar: [2, 3], john: doe}";
    ryml::Tree tree = ryml::parse_in_place(yml_buf);

    // read from the tree:
    ryml::NodeRef bar = tree["bar"];
    CHECK(bar[0].val() == "2");
    CHECK(bar[1].val() == "3");
    CHECK(bar[0].val().str == yml_buf + 15); // points at the source buffer
    CHECK(bar[1].val().str == yml_buf + 18);

    // deserializing:
    int bar0 = 0, bar1 = 0;
    bar[0] >> bar0;
    bar[1] >> bar1;
    CHECK(bar0 == 2);
    CHECK(bar1 == 3);

    // serializing:
    bar[0] << 10; // creates a string in the tree's arena
    bar[1] << 11;
    CHECK(bar[0].val() == "10");
    CHECK(bar[1].val() == "11");

    // add nodes
    bar.append_child() << 12; // see also operator= (explanation below)
    CHECK(bar[2].val() == "12");

    // emit tree
    // to std::string
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"({foo: 1,bar: [10,11,12],john: doe})");
    std::cout << tree; // emit to ostream
    ryml::emit_yaml(tree, stdout); // emit to FILE*

    // emit node
    ryml::ConstNodeRef foo = tree["foo"];
    // to std::string
    CHECK(ryml::emitrs_yaml<std::string>(foo) == "foo: 1\n");
    std::cout << foo; // emit node to ostream
    ryml::emit_yaml(foo, stdout); // emit node to FILE*
}


//-----------------------------------------------------------------------------

/** a brief tour over most features */
void sample_quick_overview()
{
    // Parse YAML code in place, potentially mutating the buffer:
    char yml_buf[] = R"(
foo: 1
bar: [2, 3]
john: doe)";
    ryml::Tree tree = ryml::parse_in_place(yml_buf);

    // The resulting tree contains only views to the parsed string. If
    // the string was parsed in place, then the string must outlive
    // the tree! This works in this case because both `yml_buf` and `tree`
    // live on the same scope, so have the same lifetime.

    // It is also possible to:
    //
    //   - parse a read-only buffer using parse_in_arena(). This
    //     copies the YAML buffer to the tree's arena, and spares the
    //     headache of the string's lifetime.
    //
    //   - reuse an existing tree (advised)
    //
    //   - reuse an existing parser (advised)
    //
    //   - parse into an existing node deep in a tree
    //
    // Note: it will always be significantly faster to parse in place
    // and reuse tree+parser.
    //
    // Below you will find samples that show how to achieve reuse; but
    // please note that for brevity and clarity, many of the examples
    // here are parsing in the arena, and not reusing tree or parser.


    //------------------------------------------------------------------
    // API overview

    // ryml has a two-level API:
    //
    // The lower level index API is based on the indices of nodes,
    // where the node's id is the node's position in the tree's data
    // array. This API is very efficient, but somewhat difficult to use:
    ryml::id_type root_id = tree.root_id();
    ryml::id_type bar_id = tree.find_child(root_id, "bar"); // need to get the index right
    CHECK(tree.is_map(root_id)); // all of the index methods are in the tree
    CHECK(tree.is_seq(bar_id));  // ... and receive the subject index

    // The node API is a lightweight abstraction sitting on top of the
    // index API, but offering a much more convenient interaction:
    ryml::ConstNodeRef root = tree.rootref();  // a const node reference
    ryml::ConstNodeRef bar = tree["bar"];
    CHECK(root.is_map());
    CHECK(bar.is_seq());

    // A node ref is a lightweight handle to the tree and associated id:
    CHECK(root.tree() == &tree); // a node ref points at its tree, WITHOUT refcount
    CHECK(root.id() == root_id); // a node ref's id is the index of the node
    CHECK(bar.id() == bar_id);   // a node ref's id is the index of the node

    // The node API translates very cleanly to the index API, so most
    // of the code examples below are using the node API.

    // WARNING. A node ref holds a raw pointer to the tree. Care must
    // be taken to ensure the lifetimes match, so that a node will
    // never access the tree after the goes out of scope.


    //------------------------------------------------------------------
    // To read the parsed tree

    // ConstNodeRef::operator[] does a lookup, is O(num_children[node]).
    CHECK(tree["foo"].is_keyval());
    CHECK(tree["foo"].val() == "1");   // get the val of a node (must be leaf node, otherwise it is a container and has no val)
    CHECK(tree["foo"].key() == "foo"); // get the key of a node (must be child of a map, otherwise it has no key)
    CHECK(tree["bar"].is_seq());
    CHECK(tree["bar"].has_key());
    CHECK(tree["bar"].key() == "bar");
    // maps use string keys, seqs use index keys:
    CHECK(tree["bar"][0].val() == "2");
    CHECK(tree["bar"][1].val() == "3");
    CHECK(tree["john"].val() == "doe");
    // An index key is the position of the child within its parent,
    // so even maps can also use int keys, if the key position is
    // known.
    CHECK(tree[0].id() == tree["foo"].id());
    CHECK(tree[1].id() == tree["bar"].id());
    CHECK(tree[2].id() == tree["john"].id());
    // Tree::operator[](int) searches a ***root*** child by its position.
    CHECK(tree[0].id() == tree["foo"].id());  // 0: first child of root
    CHECK(tree[1].id() == tree["bar"].id());  // 1: second child of root
    CHECK(tree[2].id() == tree["john"].id()); // 2: third child of root
    // NodeRef::operator[](int) searches a ***node*** child by its position:
    CHECK(bar[0].val() == "2"); // 0 means first child of bar
    CHECK(bar[1].val() == "3"); // 1 means second child of bar
    // NodeRef::operator[](string):
    // A string key is the key of the node: lookup is by name. So it
    // is only available for maps, and it is NOT available for seqs,
    // since seq members do not have keys.
    CHECK(tree["foo"].key() == "foo");
    CHECK(tree["bar"].key() == "bar");
    CHECK(tree["john"].key() == "john");
    CHECK(bar.is_seq());
    // CHECK(bar["BOOM!"].is_seed()); // error, seqs do not have key lookup

    // Note that maps can also use index keys as well as string keys:
    CHECK(root["foo"].id() == root[0].id());
    CHECK(root["bar"].id() == root[1].id());
    CHECK(root["john"].id() == root[2].id());

    // IMPORTANT. The ryml tree uses an index-based linked list for
    // storing children, so the complexity of
    // `Tree::operator[csubstr]` and `Tree::operator[id_type]` is O(n),
    // linear on the number of root children. If you use
    // `Tree::operator[]` with a large tree where the root has many
    // children, you will see a performance hit.
    //
    // To avoid this hit, you can create your own accelerator
    // structure. For example, before doing a lookup, do a single
    // traverse at the root level to fill an `map<csubstr,id_type>`
    // mapping key names to node indices; with a node index, a lookup
    // (via `Tree::get()`) is O(1), so this way you can get O(log n)
    // lookup from a key. (But please do not use `std::map` if you
    // care about performance; use something else like a flat map or
    // sorted vector).
    //
    // As for node refs, the difference from `NodeRef::operator[]` and
    // `ConstNodeRef::operator[]` to `Tree::operator[]` is that the
    // latter refers to the root node, whereas the former are invoked
    // on their target node. But the lookup process works the same for
    // both and their algorithmic complexity is the same: they are
    // both linear in the number of direct children. But of course,
    // depending on the data, that number may be very different from
    // one to another.


    //------------------------------------------------------------------
    // Hierarchy:

    {
        ryml::ConstNodeRef foo = root.first_child();
        ryml::ConstNodeRef john = root.last_child();
        CHECK(tree.size() == 6); // O(1) number of nodes in the tree
        CHECK(root.num_children() == 3); // O(num_children[root])
        CHECK(foo.num_siblings() == 3); // O(num_children[parent(foo)])
        CHECK(foo.parent().id() == root.id()); // parent() is O(1)
        CHECK(root.first_child().id() == root["foo"].id()); // first_child() is O(1)
        CHECK(root.last_child().id() == root["john"].id()); // last_child() is O(1)
        CHECK(john.first_sibling().id() == foo.id());
        CHECK(foo.last_sibling().id() == john.id());
        // prev_sibling(), next_sibling(): (both are O(1))
        CHECK(foo.num_siblings() == root.num_children());
        CHECK(foo.prev_sibling().id() == ryml::NONE); // foo is the first_child()
        CHECK(foo.next_sibling().key() == "bar");
        CHECK(foo.next_sibling().next_sibling().key() == "john");
        CHECK(foo.next_sibling().next_sibling().next_sibling().id() == ryml::NONE); // john is the last_child()
    }


    //------------------------------------------------------------------
    // Iterating:
    {
        ryml::csubstr expected_keys[] = {"foo", "bar", "john"};
        // iterate children using the high-level node API:
        {
            ryml::id_type count = 0;
            for(ryml::ConstNodeRef const& child : root.children())
                CHECK(child.key() == expected_keys[count++]);
        }
        // iterate siblings using the high-level node API:
        {
            ryml::id_type count = 0;
            for(ryml::ConstNodeRef const& child : root["foo"].siblings())
                CHECK(child.key() == expected_keys[count++]);
        }
        // iterate children using the lower-level tree index API:
        {
            ryml::id_type count = 0;
            for(ryml::id_type child_id = tree.first_child(root_id); child_id != ryml::NONE; child_id = tree.next_sibling(child_id))
                CHECK(tree.key(child_id) == expected_keys[count++]);
        }
        // iterate siblings using the lower-level tree index API:
        // (notice the only difference from above is in the loop
        // preamble, which calls tree.first_sibling(bar_id) instead of
        // tree.first_child(root_id))
        {
            ryml::id_type count = 0;
            for(ryml::id_type child_id = tree.first_sibling(bar_id); child_id != ryml::NONE; child_id = tree.next_sibling(child_id))
                CHECK(tree.key(child_id) == expected_keys[count++]);
        }
    }


    //------------------------------------------------------------------
    // Gotchas:

    // ryml uses assertions to prevent you from trying to obtain
    // things that do not exist. For example:

    {
        ryml::ConstNodeRef seq_node = tree["bar"];
        ryml::ConstNodeRef val_node = seq_node[0];

        CHECK(seq_node.is_seq());          // seq is a container
        CHECK(!seq_node.has_val());        // ... so it has no val
        //CHECK(seq_node.val() == BOOM!);  // ... so attempting to get a val is undefined behavior

        CHECK(val_node.parent() == seq_node); // belongs to a seq
        CHECK(!val_node.has_key());           // ... so it has no key
        //CHECK(val_node.key() == BOOM!);     // ... so attempting to get a key is undefined behavior

        CHECK(val_node.is_val());         // this node is a val
        //CHECK(val_node.first_child() == BOOM!); // ... so attempting to get a child is undefined behavior

        // assertions are also present in methods that /may/ read the val:
        CHECK(seq_node.is_seq());              // seq is a container
        //CHECK(seq_node.val_is_null() BOOM!); // so cannot get the val to check
    }

    // By default, assertions are enabled unless the NDEBUG macro is
    // defined (which happens in release builds).
    //
    // This adheres to the pay-only-for-what-you-use philosophy: if
    // you are sure that your intent is correct, why would you need to
    // pay the runtime cost for the assertions?
    //
    // The downside, of course, is that you may be unsure, or your
    // code may be wrong; and then release builds may end up doing
    // something wrong.
    //
    // So in that case, you can use the appropriate ryml predicates to
    // check your intent (as in the examples above), or you can
    // explicitly enable/disable assertions, by defining the macro
    // RYML_USE_ASSERT to a proper value (see c4/yml/common.hpp). This
    // will make problematic code trigger a call to the appropriate
    // error callback.
    //
    // Also, to be clear, this does not apply to parse errors
    // occurring when the YAML is parsed. Parse errors are always
    // detected by the parser: this is not done through assertions,
    // but through the appropriate error checking mechanism. So
    // checking for these errors is always enabled and cannot be
    // switched off.


    //------------------------------------------------------------------
    // Deserializing: use operator>>
    {
        int foo = 0, bar0 = 0, bar1 = 0;
        std::string john_str;
        std::string bar_str;
        root["foo"] >> foo;
        root["bar"][0] >> bar0;
        root["bar"][1] >> bar1;
        root["john"] >> john_str; // requires from_chars(std::string). see serialization samples below.
        root["bar"] >> ryml::key(bar_str); // to deserialize the key, use the tag function ryml::key()
        CHECK(foo == 1);
        CHECK(bar0 == 2);
        CHECK(bar1 == 3);
        CHECK(john_str == "doe");
        CHECK(bar_str == "bar");
    }


    //------------------------------------------------------------------
    // Modifying existing nodes: operator= vs operator<<

    // As implied by its name, ConstNodeRef is a reference to a const
    // node. It can be used to read from the node, but not write to it
    // or modify the hierarchy of the node. If any modification is
    // desired then a NodeRef must be used instead:
    ryml::NodeRef wroot = tree.rootref(); // writeable root

    // operator= assigns an existing string to the receiving node.
    // The contents are NOT copied, and the string pointer will be in
    // effect until the tree goes out of scope! So BEWARE to only
    // assign from strings outliving the tree.
    wroot["foo"] = "says you";
    wroot["bar"][0] = "-2";
    wroot["bar"][1] = "-3";
    wroot["john"] = "ron";
    // Now the tree is _pointing_ at the memory of the strings above.
    // In this case it is OK because those are static strings, located
    // in the executable's static section, and will outlive the tree.
    CHECK(root["foo"].val() == "says you");
    CHECK(root["bar"][0].val() == "-2");
    CHECK(root["bar"][1].val() == "-3");
    CHECK(root["john"].val() == "ron");
    // But WATCHOUT: do not assign from temporary objects:
    // {
    //     std::string crash("will dangle");
    //     root["john"] = ryml::to_csubstr(crash);
    // }
    // CHECK(root["john"] == "dangling"); // CRASH! the string was deallocated

    // operator<<: for cases where the lifetime of the string is
    // problematic WRT the tree, you can create and save a string in
    // the tree using operator<<. It first serializes values to a
    // string arena owned by the tree, then assigns the serialized
    // string to the receiving node. This avoids constraints with the
    // lifetime, since the arena lives with the tree.
    CHECK(tree.arena().empty());
    wroot["foo"] << "says who";  // requires to_chars(). see serialization samples below.
    wroot["bar"][0] << 20;
    wroot["bar"][1] << 30;
    wroot["john"] << "deere";
    CHECK(root["foo"].val() == "says who");
    CHECK(root["bar"][0].val() == "20");
    CHECK(root["bar"][1].val() == "30");
    CHECK(root["john"].val() == "deere");
    CHECK(tree.arena() == "says who2030deere"); // the result of serializations to the tree arena
    // using operator<< instead of operator=, the crash above is avoided:
    {
        std::string ok("in_scope");
        // root["john"] = ryml::to_csubstr(ok); // don't, will dangle
        wroot["john"] << ryml::to_csubstr(ok); // OK, copy to the tree's arena
    }
    CHECK(root["john"].val() == "in_scope"); // OK! val is now in the tree's arena
    // serializing floating points:
    wroot["float"] << 2.4f;
    // to force a particular precision or float format:
    // (see sample_float_precision() and sample_formatting())
    wroot["digits"] << ryml::fmt::real(2.4, /*num_digits*/6, ryml::FTOA_FLOAT);
    CHECK(tree.arena() == "says who2030deerein_scope2.42.400000"); // the result of serializations to the tree arena


    //------------------------------------------------------------------
    // Adding new nodes:

    // adding a keyval node to a map:
    CHECK(root.num_children() == 5);
    wroot["newkeyval"] = "shiny and new"; // using these strings
    wroot.append_child() << ryml::key("newkeyval (serialized)") << "shiny and new (serialized)"; // serializes and assigns the serialization
    CHECK(root.num_children() == 7);
    CHECK(root["newkeyval"].key() == "newkeyval");
    CHECK(root["newkeyval"].val() == "shiny and new");
    CHECK(root["newkeyval (serialized)"].key() == "newkeyval (serialized)");
    CHECK(root["newkeyval (serialized)"].val() == "shiny and new (serialized)");
    CHECK( ! tree.in_arena(root["newkeyval"].key())); // it's using directly the static string above
    CHECK( ! tree.in_arena(root["newkeyval"].val())); // it's using directly the static string above
    CHECK(   tree.in_arena(root["newkeyval (serialized)"].key())); // it's using a serialization of the string above
    CHECK(   tree.in_arena(root["newkeyval (serialized)"].val())); // it's using a serialization of the string above
    // adding a val node to a seq:
    CHECK(root["bar"].num_children() == 2);
    wroot["bar"][2] = "oh so nice";
    wroot["bar"][3] << "oh so nice (serialized)";
    CHECK(root["bar"].num_children() == 4);
    CHECK(root["bar"][2].val() == "oh so nice");
    CHECK(root["bar"][3].val() == "oh so nice (serialized)");
    // adding a seq node:
    CHECK(root.num_children() == 7);
    wroot["newseq"] |= ryml::SEQ;
    wroot.append_child() << ryml::key("newseq (serialized)") |= ryml::SEQ;
    CHECK(root.num_children() == 9);
    CHECK(root["newseq"].num_children() == 0);
    CHECK(root["newseq"].is_seq());
    CHECK(root["newseq (serialized)"].num_children() == 0);
    CHECK(root["newseq (serialized)"].is_seq());
    // adding a map node:
    CHECK(root.num_children() == 9);
    wroot["newmap"] |= ryml::MAP;
    wroot.append_child() << ryml::key("newmap (serialized)") |= ryml::MAP;
    CHECK(root.num_children() == 11);
    CHECK(root["newmap"].num_children() == 0);
    CHECK(root["newmap"].is_map());
    CHECK(root["newmap (serialized)"].num_children() == 0);
    CHECK(root["newmap (serialized)"].is_map());
    //
    // When the tree is mutable, operator[] first searches the tree
    // for the does not mutate the tree until the returned node is
    // written to.
    //
    // Until such time, the NodeRef object keeps in itself the required
    // information to write to the proper place in the tree. This is
    // called being in a "seed" state.
    //
    // This means that passing a key/index which does not exist will
    // not mutate the tree, but will instead store (in the node) the
    // proper place of the tree to be able to do so, if and when it is
    // required. This is why the node is said to be in "seed" state -
    // it allows creating the entry in the tree in the future.
    //
    // This is a significant difference from eg, the behavior of
    // std::map, which mutates the map immediately within the call to
    // operator[].
    //
    // All of the points above apply only if the tree is mutable. If
    // the tree is const, then a NodeRef cannot be obtained from it;
    // only a ConstNodeRef, which can never be used to mutate the
    // tree.
    //
    CHECK(!root.has_child("I am not nothing"));
    ryml::NodeRef nothing;
    CHECK(nothing.invalid());    // invalid because it points at nothing
    nothing = wroot["I am nothing"];
    CHECK(!nothing.invalid());   // points at the tree, and a specific place in the tree
    CHECK(nothing.is_seed()); // ... but nothing is there yet.
    CHECK(!root.has_child("I am nothing")); // same as above
    CHECK(!nothing.readable()); // ... and this node cannot be used to
                                // read anything from the tree
    ryml::NodeRef something = wroot["I am something"];
    ryml::ConstNodeRef constsomething = wroot["I am something"];
    CHECK(!root.has_child("I am something")); // same as above
    CHECK(!something.invalid());
    CHECK(something.is_seed()); // same as above
    CHECK(!something.readable()); // same as above
    CHECK(constsomething.invalid()); // NOTE: because a ConstNodeRef cannot be
                                     // used to mutate a tree, it is only valid()
                                     // if it is pointing at an existing node.
    something = "indeed";  // this will commit the seed to the tree, mutating at the proper place
    CHECK(root.has_child("I am something"));
    CHECK(root["I am something"].val() == "indeed");
    CHECK(!something.invalid()); // it was already valid
    CHECK(!something.is_seed()); // now the tree has this node, so the
                                 // ref is no longer a seed
    CHECK(something.readable()); // and it is now readable
    //
    // now the constref is also valid (but it needs to be reassigned):
    ryml::ConstNodeRef constsomethingnew = wroot["I am something"];
    CHECK(!constsomethingnew.invalid());
    CHECK(constsomethingnew.readable());
    // note that the old constref is now stale, because it only keeps
    // the state at creation:
    CHECK(constsomething.invalid());
    CHECK(!constsomething.readable());
    //
    // -----------------------------------------------------------
    // Remember: a seed node cannot be used to read from the tree!
    // -----------------------------------------------------------
    //
    // The seed node needs to be created and become readable first.
    //
    // Trying to invoke any tree-reading method on a node that is not
    // readable will cause an assertion (see RYML_USE_ASSERT).
    //
    // It is your responsibility to verify that the preconditions are
    // met. If you are not sure about the structure of your data,
    // write your code defensively to signify your full intent:
    //
    ryml::NodeRef wbar = wroot["bar"];
    if(wbar.readable() && wbar.is_seq()) // .is_seq() requires .readable()
    {
        CHECK(wbar[0].readable() && wbar[0].val() == "20");
        CHECK( ! wbar[100].readable());
        CHECK( ! wbar[100].readable() || wbar[100].val() == "100"); // <- no crash because it is not .readable(), so never tries to call .val()
        // this would work as well:
        CHECK( ! wbar[0].is_seed() && wbar[0].val() == "20");
        CHECK(wbar[100].is_seed() || wbar[100].val() == "100");
    }


    //------------------------------------------------------------------
    // .operator[]() vs .at()

    // (Const)NodeRef::operator[] is an analogue to std::vector::operator[].
    // (Const)NodeRef::at() is an analogue to std::vector::at()
    //
    // at() will always check the subject node is .readable().
    //
    // [] is meant for the happy path, and unverified in Release
    // builds.
    {
        // in this example we will be checking errors, so set up a
        // temporary error handler to catch them:
        ScopedErrorHandlerExample errh; // calls ryml::set_callbacks()
        // instantiate the tree after errh
        ryml::Tree err_tree = ryml::parse_in_arena("{foo: bar}");
        // ... so that the tree uses the current callbacks:
        CHECK(err_tree.callbacks() == errh.callbacks());
        // node does not exist, only a seed node
        ryml::NodeRef seed_node = err_tree["this"];
        // ... therefore not .readable()
        CHECK(!seed_node.readable());
        // using .at() reliably produces an error:
        CHECK(errh.check_error_occurs([&]{
            return seed_node.at("is").at("an").at("invalid").at("operation");
            //              ^
            //              error occurs here because it is unreadable
        }));
        // ... but using [] fails only when RYML_USE_ASSERT is
        // defined. otherwise, it's the dreaded Undefined Behavior:
        CHECK(errh.check_assertion_occurs([&]{
            return seed_node["is"]["an"]["invalid"]["operation"];
            //              ^
            //              assertion occurs here because it is unreadable
        }));
    }


    //------------------------------------------------------------------
    // Emitting:

    ryml::csubstr expected_result = R"(foo: says who
bar: [20,30,oh so nice,oh so nice (serialized)]
john: in_scope
float: 2.4
digits: 2.400000
newkeyval: shiny and new
newkeyval (serialized): shiny and new (serialized)
newseq: []
newseq (serialized): []
newmap: {}
newmap (serialized): {}
I am something: indeed
)";

    // emit to a FILE*
    ryml::emit_yaml(tree, stdout);
    // emit to a stream
    std::stringstream ss;
    ss << tree;
    std::string stream_result = ss.str();
    // emit to a buffer:
    std::string str_result = ryml::emitrs_yaml<std::string>(tree);
    // can emit to any given buffer:
    char buf[1024];
    ryml::csubstr buf_result = ryml::emit_yaml(tree, buf);
    // now check
    CHECK(buf_result == expected_result);
    CHECK(str_result == expected_result);
    CHECK(stream_result == expected_result);
    // There are many possibilities to emit to buffer;
    // please look at the emit sample functions below.


    //------------------------------------------------------------------
    // ConstNodeRef vs NodeRef

    ryml::NodeRef noderef = tree["bar"][0];
    ryml::ConstNodeRef constnoderef = tree["bar"][0];

    // ConstNodeRef cannot be used to mutate the tree:
    //constnoderef = "21";  // compile error
    //constnoderef << "22"; // compile error
    // ... but a NodeRef can:
    noderef = "21";         // ok, can assign because it's not const
    CHECK(tree["bar"][0].val() == "21");
    noderef << "22";        // ok, can serialize and assign because it's not const
    CHECK(tree["bar"][0].val() == "22");

    // it is not possible to obtain a NodeRef from a ConstNodeRef:
    // noderef = constnoderef; // compile error

    // it is always possible to obtain a ConstNodeRef from a NodeRef:
    constnoderef = noderef;    // ok can assign const <- nonconst

    // If a tree is const, then only ConstNodeRef's can be
    // obtained from that tree:
    ryml::Tree const& consttree = tree;
    //noderef = consttree["bar"][0];    // compile error
    noderef = tree["bar"][0];           // ok
    constnoderef = consttree["bar"][0]; // ok

    // ConstNodeRef and NodeRef can be compared for equality.
    // Equality means they point at the same node.
    CHECK(constnoderef == noderef);
    CHECK(!(constnoderef != noderef));


    //------------------------------------------------------------------
    // Getting the location of nodes in the source:
    //
    // Location tracking is opt-in:
    ryml::EventHandlerTree evt_handler = {};
    ryml::Parser parser(&evt_handler, ryml::ParserOptions().locations(true));
    // Now the parser will start by building the accelerator structure:
    ryml::Tree tree2 = parse_in_arena(&parser, "expected.yml", expected_result);
    // ... and use it when querying
    ryml::ConstNodeRef subject_node = tree2["bar"][1];
    CHECK(subject_node.val() == "30");
    ryml::Location loc = subject_node.location(parser);
    CHECK(parser.location_contents(loc).begins_with("30"));
    CHECK(loc.line == 1u);
    CHECK(loc.col == 9u);
    // For further details in location tracking,
    // refer to the sample function below.

    //------------------------------------------------------------------
    // Dealing with UTF8
    ryml::Tree langs = ryml::parse_in_arena(R"(
en: Planet (Gas)
fr: Planète (Gazeuse)
ru: Планета (Газ)
ja: 惑星（ガス）
zh: 行星（气体）
# UTF8 decoding only happens in double-quoted strings,
# as per the YAML standard
decode this: "\u263A c\x61f\xE9"
and this as well: "\u2705 \U0001D11E"
not decoded: '\u263A \xE2\x98\xBA'
neither this: '\u2705 \U0001D11E'
)");
    // in-place UTF8 just works:
    CHECK(langs["en"].val() == "Planet (Gas)");
    CHECK(langs["fr"].val() == "Planète (Gazeuse)");
    CHECK(langs["ru"].val() == "Планета (Газ)");
    CHECK(langs["ja"].val() == "惑星（ガス）");
    CHECK(langs["zh"].val() == "行星（气体）");
    // and \x \u \U codepoints are decoded, but only when they appear
    // inside double-quoted strings, as dictated by the YAML
    // standard:
    CHECK(langs["decode this"].val() == "☺ café");
    CHECK(langs["and this as well"].val() == "✅ 𝄞");
    CHECK(langs["not decoded"].val() == "\\u263A \\xE2\\x98\\xBA");
    CHECK(langs["neither this"].val() == "\\u2705 \\U0001D11E");
}


//-----------------------------------------------------------------------------

/** demonstrate usage of ryml::substr and ryml::csubstr
 *
 * These types are imported from the c4core library into the ryml
 * namespace You may have noticed above the use of a `csubstr`
 * class. This class is defined in another library,
 * [c4core](https://github.com/biojppm/c4core), which is imported by
 * ryml. This is a library I use with my projects consisting of
 * multiplatform low-level utilities. One of these is `c4::csubstr`
 * (the name comes from "constant substring") which is a non-owning
 * read-only string view, with many methods that make it practical to
 * use (I would certainly argue more practical than `std::string`). In
 * fact, `c4::csubstr` and its writeable counterpart `c4::substr` are
 * the workhorses of the ryml parsing and serialization code.
 *
 * @see doc_substr */
void sample_substr()
{
    // substr is a mutable view: pointer and length to a string in memory.
    // csubstr is a const-substr (immutable).

    // construct from explicit args
    {
        const char foobar_str[] = "foobar";
        auto s = ryml::csubstr(foobar_str, strlen(foobar_str));
        CHECK(s == "foobar");
        CHECK(s.size() == 6);
        CHECK(s.data() == foobar_str);
        CHECK(s.size() == s.len);
        CHECK(s.data() == s.str);
    }

    // construct from a string array
    {
        const char foobar_str[] = "foobar";
        ryml::csubstr s = foobar_str;
        CHECK(s == "foobar");
        CHECK(s != "foobar0");
        CHECK(s.size() == 6); // does not include the terminating \0
        CHECK(s.data() == foobar_str);
        CHECK(s.size() == s.len);
        CHECK(s.data() == s.str);
    }
    // you can also declare directly in-place from an array:
    {
        ryml::csubstr s = "foobar";
        CHECK(s == "foobar");
        CHECK(s != "foobar0");
        CHECK(s.size() == 6);
        CHECK(s.size() == s.len);
        CHECK(s.data() == s.str);
    }

    // construct from a C-string:
    //
    // Since the input is only a pointer, the string length can only
    // be found with a call to strlen(). To make this cost evident, we
    // require calling to_csubstr():
    {
        const char *foobar_str = "foobar";
        ryml::csubstr s = ryml::to_csubstr(foobar_str);
        CHECK(s == "foobar");
        CHECK(s != "foobar0");
        CHECK(s.size() == 6);
        CHECK(s.size() == s.len);
        CHECK(s.data() == s.str);
    }

    // construct from a std::string: same approach as above.
    // requires inclusion of the <ryml/std/string.hpp> header
    // or of the umbrella header <ryml_std.hpp>.
    //
    // not including <string> in the default header is a deliberate
    // design choice to avoid including the heavy std:: allocation
    // machinery
    {
        std::string foobar_str = "foobar";
        ryml::csubstr s = ryml::to_csubstr(foobar_str); // defined in <ryml/std/string.hpp>
        CHECK(s == "foobar");
        CHECK(s != "foobar0");
        CHECK(s.size() == 6);
        CHECK(s.size() == s.len);
        CHECK(s.data() == s.str);
    }

    // convert substr -> csubstr
    {
        char buf[] = "foo";
        ryml::substr foo = buf;
        CHECK(foo.len == 3);
        CHECK(foo.data() == buf);
        ryml::csubstr cfoo = foo;
        CHECK(cfoo.data() == buf);
    }
    // cannot convert csubstr -> substr:
    {
        // ryml::substr foo2 = cfoo; // compile error: cannot write to csubstr
    }

    // construct from char[]/const char[]: mutable vs immutable memory
    {
        char const foobar_str_ro[] = "foobar"; // ro := read-only
        char       foobar_str_rw[] = "foobar"; // rw := read-write
        static_assert(std::is_array<decltype(foobar_str_ro)>::value, "this is an array");
        static_assert(std::is_array<decltype(foobar_str_rw)>::value, "this is an array");
        // csubstr <- read-only memory
        {
            ryml::csubstr foobar = foobar_str_ro;
            CHECK(foobar.data() == foobar_str_ro);
            CHECK(foobar.size() == strlen(foobar_str_ro));
            CHECK(foobar == "foobar"); // AKA strcmp
        }
        // csubstr <- read-write memory: you can create an immutable csubstr from mutable memory
        {
            ryml::csubstr foobar = foobar_str_rw;
            CHECK(foobar.data() == foobar_str_rw);
            CHECK(foobar.size() == strlen(foobar_str_rw));
            CHECK(foobar == "foobar"); // AKA strcmp
        }
        // substr <- read-write memory.
        {
            ryml::substr foobar = foobar_str_rw;
            CHECK(foobar.data() == foobar_str_rw);
            CHECK(foobar.size() == strlen(foobar_str_rw));
            CHECK(foobar == "foobar"); // AKA strcmp
        }
        // substr <- ro is impossible.
        {
            //ryml::substr foobar = foobar_str_ro; // compile error!
        }
    }

    // construct from char*/const char*: mutable vs immutable memory.
    // use to_substr()/to_csubstr()
    {
        char const* foobar_str_ro = "foobar";       // ro := read-only
        char        foobar_str_rw_[] = "foobar";    // rw := read-write
        char      * foobar_str_rw = foobar_str_rw_; // rw := read-write
        static_assert(!std::is_array<decltype(foobar_str_ro)>::value, "this is a decayed pointer");
        static_assert(!std::is_array<decltype(foobar_str_rw)>::value, "this is a decayed pointer");
        // csubstr <- read-only memory
        {
            //ryml::csubstr foobar = foobar_str_ro; // compile error: length is not known
            ryml::csubstr foobar = ryml::to_csubstr(foobar_str_ro);
            CHECK(foobar.data() == foobar_str_ro);
            CHECK(foobar.size() == strlen(foobar_str_ro));
            CHECK(foobar == "foobar"); // AKA strcmp
        }
        // csubstr <- read-write memory: you can create an immutable csubstr from mutable memory
        {
            ryml::csubstr foobar = ryml::to_csubstr(foobar_str_rw);
            CHECK(foobar.data() == foobar_str_rw);
            CHECK(foobar.size() == strlen(foobar_str_rw));
            CHECK(foobar == "foobar"); // AKA strcmp
        }
        // substr <- read-write memory.
        {
            ryml::substr foobar = ryml::to_substr(foobar_str_rw);
            CHECK(foobar.data() == foobar_str_rw);
            CHECK(foobar.size() == strlen(foobar_str_rw));
            CHECK(foobar == "foobar"); // AKA strcmp
        }
        // substr <- read-only is impossible.
        {
            //ryml::substr foobar = ryml::to_substr(foobar_str_ro); // compile error!
        }
    }

    // substr is mutable, without changing the size:
    {
        char buf[] = "foobar";
        ryml::substr foobar = buf;
        CHECK(foobar == "foobar");
        foobar[0] = 'F';            CHECK(foobar == "Foobar");
        foobar.back() = 'R';        CHECK(foobar == "FoobaR");
        foobar.reverse();           CHECK(foobar == "RabooF");
        foobar.reverse();           CHECK(foobar == "FoobaR");
        foobar.reverse_sub(1, 4);   CHECK(foobar == "FabooR");
        foobar.reverse_sub(1, 4);   CHECK(foobar == "FoobaR");
        foobar.reverse_range(2, 5); CHECK(foobar == "FoaboR");
        foobar.reverse_range(2, 5); CHECK(foobar == "FoobaR");
        foobar.replace('o', '0');   CHECK(foobar == "F00baR");
        foobar.replace('a', '_');   CHECK(foobar == "F00b_R");
        foobar.replace("_0b", 'a'); CHECK(foobar == "FaaaaR");
        foobar.toupper();           CHECK(foobar == "FAAAAR");
        foobar.tolower();           CHECK(foobar == "faaaar");
        foobar.fill('.');           CHECK(foobar == "......");
        // see also:
        // - .erase()
        // - .replace_all()
    }

    // sub-views
    {
        ryml::csubstr s = "fooFOObarBAR";
        CHECK(s.len == 12u);
        // sub(): <- first,[num]
        CHECK(s.sub(0)     == "fooFOObarBAR");
        CHECK(s.sub(0, 12) == "fooFOObarBAR");
        CHECK(s.sub(0,  3) == "foo"         );
        CHECK(s.sub(3)     ==    "FOObarBAR");
        CHECK(s.sub(3,  3) ==    "FOO"      );
        CHECK(s.sub(6)     ==       "barBAR");
        CHECK(s.sub(6,  3) ==       "bar"   );
        CHECK(s.sub(9)     ==          "BAR");
        CHECK(s.sub(9,  3) ==          "BAR");
        // first(): <- length
        CHECK(s.first(0) == ""   );
        CHECK(s.first(1) == "f"  );
        CHECK(s.first(2) != "f"  );
        CHECK(s.first(2) == "fo" );
        CHECK(s.first(3) == "foo");
        // last(): <- length
        CHECK(s.last(0) ==    "");
        CHECK(s.last(1) ==   "R");
        CHECK(s.last(2) ==  "AR");
        CHECK(s.last(3) == "BAR");
        // range(): <- first, last
        CHECK(s.range(0, 12) == "fooFOObarBAR");
        CHECK(s.range(1, 12) ==  "ooFOObarBAR");
        CHECK(s.range(1, 11) ==  "ooFOObarBA" );
        CHECK(s.range(2, 10) ==   "oFOObarB"  );
        CHECK(s.range(3,  9) ==    "FOObar"   );
        // offs(): offset from beginning, end
        CHECK(s.offs(0, 0) == "fooFOObarBAR");
        CHECK(s.offs(1, 0) ==  "ooFOObarBAR");
        CHECK(s.offs(1, 1) ==  "ooFOObarBA" );
        CHECK(s.offs(2, 1) ==   "oFOObarBA" );
        CHECK(s.offs(2, 2) ==   "oFOObarB"  );
        CHECK(s.offs(3, 3) ==    "FOObar"  );
        // right_of(): <- pos, include_pos
        CHECK(s.right_of(0,  true) == "fooFOObarBAR");
        CHECK(s.right_of(0, false) ==  "ooFOObarBAR");
        CHECK(s.right_of(1,  true) ==  "ooFOObarBAR");
        CHECK(s.right_of(1, false) ==   "oFOObarBAR");
        CHECK(s.right_of(2,  true) ==   "oFOObarBAR");
        CHECK(s.right_of(2, false) ==    "FOObarBAR");
        CHECK(s.right_of(3,  true) ==    "FOObarBAR");
        CHECK(s.right_of(3, false) ==     "OObarBAR");
        // left_of() <- pos, include_pos
        CHECK(s.left_of(12, false) == "fooFOObarBAR");
        CHECK(s.left_of(11,  true) == "fooFOObarBAR");
        CHECK(s.left_of(11, false) == "fooFOObarBA" );
        CHECK(s.left_of(10,  true) == "fooFOObarBA" );
        CHECK(s.left_of(10, false) == "fooFOObarB"  );
        CHECK(s.left_of( 9,  true) == "fooFOObarB"  );
        CHECK(s.left_of( 9, false) == "fooFOObar"   );
        // left_of(),right_of() <- substr
        ryml::csubstr FOO = s.sub(3, 3);
        CHECK(s.is_super(FOO)); // required for the following
        CHECK(s.left_of(FOO) == "foo");
        CHECK(s.right_of(FOO) == "barBAR");
    }

    // printing a substr/csubstr using printf-like
    {
        ryml::csubstr s = "some substring";
        ryml::csubstr some = s.first(4);
        CHECK(some == "some");
        CHECK(s == "some substring");
        // To print a csubstr using printf(), use the %.*s format specifier:
        {
            char result[32] = {0};
            std::snprintf(result, sizeof(result), "%.*s", (int)some.len, some.str);
            printf("~~~%s~~~\n", result);
            CHECK(ryml::to_csubstr((const char*)result) == "some");
            CHECK(ryml::to_csubstr((const char*)result) == some);
        }
        // But NOTE: because this is a string view type, in general
        // the C-string is NOT zero terminated.  So NEVER print it
        // directly, or it will overflow past the end of the given
        // substr, with a potential unbounded access.  For example,
        // this is bad:
        {
            char result[32] = {0};
            std::snprintf(result, sizeof(result), "%s", some.str); // ERROR! do not print the c-string directly
            CHECK(ryml::to_csubstr((const char*)result) == "some substring");
            CHECK(ryml::to_csubstr((const char*)result) == s);
        }
    }

    // printing a substr/csubstr using ostreams
    {
        ryml::csubstr s = "some substring";
        ryml::csubstr some = s.first(4);
        CHECK(some == "some");
        CHECK(s == "some substring");
        // simple! just use plain operator<<
        {
            std::stringstream ss;
            ss << s;
            CHECK(ss.str() == "some substring"); // as expected
            CHECK(ss.str() == s); // as expected
        }
        // But NOTE: because this is a string view type, in general
        // the C-string is NOT zero terminated.  So NEVER print it
        // directly, or it will overflow past the end of the given
        // substr, with a potential unbounded access.  For example,
        // this is bad:
        {
            std::stringstream ss;
            ss << some.str; // ERROR! do not print the c-string directly
            CHECK(ss.str() == "some substring"); // NOT "some"
            CHECK(ss.str() == s); // NOT some
        }
        // this is also bad (the same)
        {
            std::stringstream ss;
            ss << some.data(); // ERROR! do not print the c-string directly
            CHECK(ss.str() == "some substring"); // NOT "some"
            CHECK(ss.str() == s); // NOT some
        }
        // this is ok:
        {
            std::stringstream ss;
            ss << some;
            CHECK(ss.str() == "some"); // ok
            CHECK(ss.str() == some); // ok
        }
    }

    // is_sub(),is_super()
    {
        ryml::csubstr foobar = "foobar";
        ryml::csubstr foo = foobar.first(3);
        CHECK(foo.is_sub(foobar));
        CHECK(foo.is_sub(foo));
        CHECK(!foo.is_super(foobar));
        CHECK(!foobar.is_sub(foo));
        // identity comparison is true:
        CHECK(foo.is_super(foo));
        CHECK(foo.is_sub(foo));
        CHECK(foobar.is_sub(foobar));
        CHECK(foobar.is_super(foobar));
    }

    // overlaps()
    {
        ryml::csubstr foobar = "foobar";
        ryml::csubstr foo = foobar.first(3);
        ryml::csubstr oba = foobar.offs(2, 1);
        ryml::csubstr abc = "abc";
        CHECK(foobar.overlaps(foo));
        CHECK(foobar.overlaps(oba));
        CHECK(foo.overlaps(foobar));
        CHECK(foo.overlaps(oba));
        CHECK(!foo.overlaps(abc));
        CHECK(!abc.overlaps(foo));
    }

    // triml(): trim characters from the left
    // trimr(): trim characters from the right
    // trim(): trim characters from left AND right
    {
        CHECK(ryml::csubstr(" \t\n\rcontents without whitespace\t \n\r").trim("\t \n\r") == "contents without whitespace");
        ryml::csubstr aaabbb = "aaabbb";
        ryml::csubstr aaa___bbb = "aaa___bbb";
        // trim a character:
        CHECK(aaabbb.triml('a') == aaabbb.last(3)); // bbb
        CHECK(aaabbb.trimr('a') == aaabbb);
        CHECK(aaabbb.trim ('a') == aaabbb.last(3)); // bbb
        CHECK(aaabbb.triml('b') == aaabbb);
        CHECK(aaabbb.trimr('b') == aaabbb.first(3)); // aaa
        CHECK(aaabbb.trim ('b') == aaabbb.first(3)); // aaa
        CHECK(aaabbb.triml('c') == aaabbb);
        CHECK(aaabbb.trimr('c') == aaabbb);
        CHECK(aaabbb.trim ('c') == aaabbb);
        CHECK(aaa___bbb.triml('a') == aaa___bbb.last(6)); // ___bbb
        CHECK(aaa___bbb.trimr('a') == aaa___bbb);
        CHECK(aaa___bbb.trim ('a') == aaa___bbb.last(6)); // ___bbb
        CHECK(aaa___bbb.triml('b') == aaa___bbb);
        CHECK(aaa___bbb.trimr('b') == aaa___bbb.first(6)); // aaa___
        CHECK(aaa___bbb.trim ('b') == aaa___bbb.first(6)); // aaa___
        CHECK(aaa___bbb.triml('c') == aaa___bbb);
        CHECK(aaa___bbb.trimr('c') == aaa___bbb);
        CHECK(aaa___bbb.trim ('c') == aaa___bbb);
        // trim ANY of the characters:
        CHECK(aaabbb.triml("ab") == "");
        CHECK(aaabbb.trimr("ab") == "");
        CHECK(aaabbb.trim ("ab") == "");
        CHECK(aaabbb.triml("ba") == "");
        CHECK(aaabbb.trimr("ba") == "");
        CHECK(aaabbb.trim ("ba") == "");
        CHECK(aaabbb.triml("cd") == aaabbb);
        CHECK(aaabbb.trimr("cd") == aaabbb);
        CHECK(aaabbb.trim ("cd") == aaabbb);
        CHECK(aaa___bbb.triml("ab") == aaa___bbb.last(6)); // ___bbb
        CHECK(aaa___bbb.triml("ba") == aaa___bbb.last(6)); // ___bbb
        CHECK(aaa___bbb.triml("cd") == aaa___bbb);
        CHECK(aaa___bbb.trimr("ab") == aaa___bbb.first(6)); // aaa___
        CHECK(aaa___bbb.trimr("ba") == aaa___bbb.first(6)); // aaa___
        CHECK(aaa___bbb.trimr("cd") == aaa___bbb);
        CHECK(aaa___bbb.trim ("ab") == aaa___bbb.range(3, 6)); // ___
        CHECK(aaa___bbb.trim ("ba") == aaa___bbb.range(3, 6)); // ___
        CHECK(aaa___bbb.trim ("cd") == aaa___bbb);
    }

    // unquoted():
    {
        CHECK(ryml::csubstr(R"('this is is single quoted')").unquoted() == "this is is single quoted");
        CHECK(ryml::csubstr(R"("this is is double quoted")").unquoted() == "this is is double quoted");
    }

    // stripl(): remove pattern from the left
    // stripr(): remove pattern from the right
    {
        ryml::csubstr abc___cba = "abc___cba";
        ryml::csubstr abc___abc = "abc___abc";
        CHECK(abc___cba.stripl("abc") == abc___cba.last(6)); // ___cba
        CHECK(abc___cba.stripr("abc") == abc___cba);
        CHECK(abc___cba.stripl("ab") == abc___cba.last(7)); // c___cba
        CHECK(abc___cba.stripr("ab") == abc___cba);
        CHECK(abc___cba.stripl("a") == abc___cba.last(8)); // bc___cba, same as triml('a')
        CHECK(abc___cba.stripr("a") == abc___cba.first(8));
        CHECK(abc___abc.stripl("abc") == abc___abc.last(6)); // ___abc
        CHECK(abc___abc.stripr("abc") == abc___abc.first(6)); // abc___
        CHECK(abc___abc.stripl("ab") == abc___abc.last(7)); // c___cba
        CHECK(abc___abc.stripr("ab") == abc___abc);
        CHECK(abc___abc.stripl("a") == abc___abc.last(8)); // bc___cba, same as triml('a')
        CHECK(abc___abc.stripr("a") == abc___abc);
    }

    // begins_with()/ends_with()
    // begins_with_any()/ends_with_any()
    {
        ryml::csubstr s = "foobar123";
        // char overloads
        CHECK(s.begins_with('f'));
        CHECK(s.ends_with('3'));
        CHECK(!s.ends_with('2'));
        CHECK(!s.ends_with('o'));
        // char[] overloads
        CHECK(s.begins_with("foobar"));
        CHECK(s.begins_with("foo"));
        CHECK(s.begins_with_any("foo"));
        CHECK(!s.begins_with("oof"));
        CHECK(s.begins_with_any("oof"));
        CHECK(s.ends_with("23"));
        CHECK(s.ends_with("123"));
        CHECK(s.ends_with_any("123"));
        CHECK(!s.ends_with("321"));
        CHECK(s.ends_with_any("231"));
    }

    // select()
    {
        ryml::csubstr s = "0123456789";
        CHECK(s.select('0') == s.sub(0, 1));
        CHECK(s.select('1') == s.sub(1, 1));
        CHECK(s.select('2') == s.sub(2, 1));
        CHECK(s.select('8') == s.sub(8, 1));
        CHECK(s.select('9') == s.sub(9, 1));
        CHECK(s.select("0123") == s.range(0, 4));
        CHECK(s.select("012" ) == s.range(0, 3));
        CHECK(s.select("01"  ) == s.range(0, 2));
        CHECK(s.select("0"   ) == s.range(0, 1));
        CHECK(s.select( "123") == s.range(1, 4));
        CHECK(s.select(  "23") == s.range(2, 4));
        CHECK(s.select(   "3") == s.range(3, 4));
    }

    // find()
    {
        ryml::csubstr s012345 = "012345";
        // find single characters:
        CHECK(s012345.find('a') == ryml::npos);
        CHECK(s012345.find('0'    ) == 0u);
        CHECK(s012345.find('0', 1u) == ryml::npos);
        CHECK(s012345.find('1'    ) == 1u);
        CHECK(s012345.find('1', 2u) == ryml::npos);
        CHECK(s012345.find('2'    ) == 2u);
        CHECK(s012345.find('2', 3u) == ryml::npos);
        CHECK(s012345.find('3'    ) == 3u);
        CHECK(s012345.find('3', 4u) == ryml::npos);
        // find patterns
        CHECK(s012345.find("ab"    ) == ryml::npos);
        CHECK(s012345.find("01"    ) == 0u);
        CHECK(s012345.find("01", 1u) == ryml::npos);
        CHECK(s012345.find("12"    ) == 1u);
        CHECK(s012345.find("12", 2u) == ryml::npos);
        CHECK(s012345.find("23"    ) == 2u);
        CHECK(s012345.find("23", 3u) == ryml::npos);
    }

    // count(): count the number of occurrences of a character
    {
        ryml::csubstr buf = "00110022003300440055";
        CHECK(buf.count('1'     ) ==  2u);
        CHECK(buf.count('1',  0u) ==  2u);
        CHECK(buf.count('1',  1u) ==  2u);
        CHECK(buf.count('1',  2u) ==  2u);
        CHECK(buf.count('1',  3u) ==  1u);
        CHECK(buf.count('1',  4u) ==  0u);
        CHECK(buf.count('1',  5u) ==  0u);
        CHECK(buf.count('0'     ) == 10u);
        CHECK(buf.count('0',  0u) == 10u);
        CHECK(buf.count('0',  1u) ==  9u);
        CHECK(buf.count('0',  2u) ==  8u);
        CHECK(buf.count('0',  3u) ==  8u);
        CHECK(buf.count('0',  4u) ==  8u);
        CHECK(buf.count('0',  5u) ==  7u);
        CHECK(buf.count('0',  6u) ==  6u);
        CHECK(buf.count('0',  7u) ==  6u);
        CHECK(buf.count('0',  8u) ==  6u);
        CHECK(buf.count('0',  9u) ==  5u);
        CHECK(buf.count('0', 10u) ==  4u);
        CHECK(buf.count('0', 11u) ==  4u);
        CHECK(buf.count('0', 12u) ==  4u);
        CHECK(buf.count('0', 13u) ==  3u);
        CHECK(buf.count('0', 14u) ==  2u);
        CHECK(buf.count('0', 15u) ==  2u);
        CHECK(buf.count('0', 16u) ==  2u);
        CHECK(buf.count('0', 17u) ==  1u);
        CHECK(buf.count('0', 18u) ==  0u);
        CHECK(buf.count('0', 19u) ==  0u);
        CHECK(buf.count('0', 20u) ==  0u);
    }

    // first_of(),last_of()
    {
        ryml::csubstr s012345 = "012345";
        CHECK(s012345.first_of('a') == ryml::npos);
        CHECK(s012345.first_of("ab") == ryml::npos);
        CHECK(s012345.first_of('0') == 0u);
        CHECK(s012345.first_of("0") == 0u);
        CHECK(s012345.first_of("01") == 0u);
        CHECK(s012345.first_of("10") == 0u);
        CHECK(s012345.first_of("012") == 0u);
        CHECK(s012345.first_of("210") == 0u);
        CHECK(s012345.first_of("0123") == 0u);
        CHECK(s012345.first_of("3210") == 0u);
        CHECK(s012345.first_of("01234") == 0u);
        CHECK(s012345.first_of("43210") == 0u);
        CHECK(s012345.first_of("012345") == 0u);
        CHECK(s012345.first_of("543210") == 0u);
        CHECK(s012345.first_of('5') == 5u);
        CHECK(s012345.first_of("5") == 5u);
        CHECK(s012345.first_of("45") == 4u);
        CHECK(s012345.first_of("54") == 4u);
        CHECK(s012345.first_of("345") == 3u);
        CHECK(s012345.first_of("543") == 3u);
        CHECK(s012345.first_of("2345") == 2u);
        CHECK(s012345.first_of("5432") == 2u);
        CHECK(s012345.first_of("12345") == 1u);
        CHECK(s012345.first_of("54321") == 1u);
        CHECK(s012345.first_of("012345") == 0u);
        CHECK(s012345.first_of("543210") == 0u);
        CHECK(s012345.first_of('0', 6u) == ryml::npos);
        CHECK(s012345.first_of('5', 6u) == ryml::npos);
        CHECK(s012345.first_of("012345", 6u) == ryml::npos);
        //
        CHECK(s012345.last_of('a') == ryml::npos);
        CHECK(s012345.last_of("ab") == ryml::npos);
        CHECK(s012345.last_of('0') == 0u);
        CHECK(s012345.last_of("0") == 0u);
        CHECK(s012345.last_of("01") == 1u);
        CHECK(s012345.last_of("10") == 1u);
        CHECK(s012345.last_of("012") == 2u);
        CHECK(s012345.last_of("210") == 2u);
        CHECK(s012345.last_of("0123") == 3u);
        CHECK(s012345.last_of("3210") == 3u);
        CHECK(s012345.last_of("01234") == 4u);
        CHECK(s012345.last_of("43210") == 4u);
        CHECK(s012345.last_of("012345") == 5u);
        CHECK(s012345.last_of("543210") == 5u);
        CHECK(s012345.last_of('5') == 5u);
        CHECK(s012345.last_of("5") == 5u);
        CHECK(s012345.last_of("45") == 5u);
        CHECK(s012345.last_of("54") == 5u);
        CHECK(s012345.last_of("345") == 5u);
        CHECK(s012345.last_of("543") == 5u);
        CHECK(s012345.last_of("2345") == 5u);
        CHECK(s012345.last_of("5432") == 5u);
        CHECK(s012345.last_of("12345") == 5u);
        CHECK(s012345.last_of("54321") == 5u);
        CHECK(s012345.last_of("012345") == 5u);
        CHECK(s012345.last_of("543210") == 5u);
        CHECK(s012345.last_of('0', 6u) == 0u);
        CHECK(s012345.last_of('5', 6u) == 5u);
        CHECK(s012345.last_of("012345", 6u) == 5u);
    }

    // first_not_of(), last_not_of()
    {
        ryml::csubstr s012345 = "012345";
        CHECK(s012345.first_not_of('a') == 0u);
        CHECK(s012345.first_not_of("ab") == 0u);
        CHECK(s012345.first_not_of('0') == 1u);
        CHECK(s012345.first_not_of("0") == 1u);
        CHECK(s012345.first_not_of("01") == 2u);
        CHECK(s012345.first_not_of("10") == 2u);
        CHECK(s012345.first_not_of("012") == 3u);
        CHECK(s012345.first_not_of("210") == 3u);
        CHECK(s012345.first_not_of("0123") == 4u);
        CHECK(s012345.first_not_of("3210") == 4u);
        CHECK(s012345.first_not_of("01234") == 5u);
        CHECK(s012345.first_not_of("43210") == 5u);
        CHECK(s012345.first_not_of("012345") == ryml::npos);
        CHECK(s012345.first_not_of("543210") == ryml::npos);
        CHECK(s012345.first_not_of('5') == 0u);
        CHECK(s012345.first_not_of("5") == 0u);
        CHECK(s012345.first_not_of("45") == 0u);
        CHECK(s012345.first_not_of("54") == 0u);
        CHECK(s012345.first_not_of("345") == 0u);
        CHECK(s012345.first_not_of("543") == 0u);
        CHECK(s012345.first_not_of("2345") == 0u);
        CHECK(s012345.first_not_of("5432") == 0u);
        CHECK(s012345.first_not_of("12345") == 0u);
        CHECK(s012345.first_not_of("54321") == 0u);
        CHECK(s012345.first_not_of("012345") == ryml::npos);
        CHECK(s012345.first_not_of("543210") == ryml::npos);
        CHECK(s012345.last_not_of('a') == 5u);
        CHECK(s012345.last_not_of("ab") == 5u);
        CHECK(s012345.last_not_of('5') == 4u);
        CHECK(s012345.last_not_of("5") == 4u);
        CHECK(s012345.last_not_of("45") == 3u);
        CHECK(s012345.last_not_of("54") == 3u);
        CHECK(s012345.last_not_of("345") == 2u);
        CHECK(s012345.last_not_of("543") == 2u);
        CHECK(s012345.last_not_of("2345") == 1u);
        CHECK(s012345.last_not_of("5432") == 1u);
        CHECK(s012345.last_not_of("12345") == 0u);
        CHECK(s012345.last_not_of("54321") == 0u);
        CHECK(s012345.last_not_of("012345") == ryml::npos);
        CHECK(s012345.last_not_of("543210") == ryml::npos);
        CHECK(s012345.last_not_of('0') == 5u);
        CHECK(s012345.last_not_of("0") == 5u);
        CHECK(s012345.last_not_of("01") == 5u);
        CHECK(s012345.last_not_of("10") == 5u);
        CHECK(s012345.last_not_of("012") == 5u);
        CHECK(s012345.last_not_of("210") == 5u);
        CHECK(s012345.last_not_of("0123") == 5u);
        CHECK(s012345.last_not_of("3210") == 5u);
        CHECK(s012345.last_not_of("01234") == 5u);
        CHECK(s012345.last_not_of("43210") == 5u);
        CHECK(s012345.last_not_of("012345") == ryml::npos);
        CHECK(s012345.last_not_of("543210") == ryml::npos);
    }

    // first_non_empty_span()
    {
        CHECK(ryml::csubstr("foo bar").first_non_empty_span() == "foo");
        CHECK(ryml::csubstr("       foo bar").first_non_empty_span() == "foo");
        CHECK(ryml::csubstr("\n   \r  \t  foo bar").first_non_empty_span() == "foo");
        CHECK(ryml::csubstr("\n   \r  \t  foo\n\r\t bar").first_non_empty_span() == "foo");
        CHECK(ryml::csubstr("\n   \r  \t  foo\n\r\t bar").first_non_empty_span() == "foo");
        CHECK(ryml::csubstr(",\n   \r  \t  foo\n\r\t bar").first_non_empty_span() == ",");
    }
    // first_uint_span()
    {
        CHECK(ryml::csubstr("1234 asdkjh").first_uint_span() == "1234");
        CHECK(ryml::csubstr("1234\rasdkjh").first_uint_span() == "1234");
        CHECK(ryml::csubstr("1234\tasdkjh").first_uint_span() == "1234");
        CHECK(ryml::csubstr("1234\nasdkjh").first_uint_span() == "1234");
        CHECK(ryml::csubstr("1234]asdkjh").first_uint_span() == "1234");
        CHECK(ryml::csubstr("1234)asdkjh").first_uint_span() == "1234");
        CHECK(ryml::csubstr("1234gasdkjh").first_uint_span() == "");
    }
    // first_int_span()
    {
        CHECK(ryml::csubstr("-1234 asdkjh").first_int_span() == "-1234");
        CHECK(ryml::csubstr("-1234\rasdkjh").first_int_span() == "-1234");
        CHECK(ryml::csubstr("-1234\tasdkjh").first_int_span() == "-1234");
        CHECK(ryml::csubstr("-1234\nasdkjh").first_int_span() == "-1234");
        CHECK(ryml::csubstr("-1234]asdkjh").first_int_span() == "-1234");
        CHECK(ryml::csubstr("-1234)asdkjh").first_int_span() == "-1234");
        CHECK(ryml::csubstr("-1234gasdkjh").first_int_span() == "");
    }
    // first_real_span()
    {
        CHECK(ryml::csubstr("-1234 asdkjh").first_real_span() == "-1234");
        CHECK(ryml::csubstr("-1234\rasdkjh").first_real_span() == "-1234");
        CHECK(ryml::csubstr("-1234\tasdkjh").first_real_span() == "-1234");
        CHECK(ryml::csubstr("-1234\nasdkjh").first_real_span() == "-1234");
        CHECK(ryml::csubstr("-1234]asdkjh").first_real_span() == "-1234");
        CHECK(ryml::csubstr("-1234)asdkjh").first_real_span() == "-1234");
        CHECK(ryml::csubstr("-1234gasdkjh").first_real_span() == "");
        CHECK(ryml::csubstr("1.234 asdkjh").first_real_span() == "1.234");
        CHECK(ryml::csubstr("1.234e+5 asdkjh").first_real_span() == "1.234e+5");
        CHECK(ryml::csubstr("1.234e-5 asdkjh").first_real_span() == "1.234e-5");
        CHECK(ryml::csubstr("1.234 asdkjh").first_real_span() == "1.234");
        CHECK(ryml::csubstr("1.234e+5 asdkjh").first_real_span() == "1.234e+5");
        CHECK(ryml::csubstr("1.234e-5 asdkjh").first_real_span() == "1.234e-5");
        CHECK(ryml::csubstr("-1.234 asdkjh").first_real_span() == "-1.234");
        CHECK(ryml::csubstr("-1.234e+5 asdkjh").first_real_span() == "-1.234e+5");
        CHECK(ryml::csubstr("-1.234e-5 asdkjh").first_real_span() == "-1.234e-5");
        // hexadecimal real numbers
        CHECK(ryml::csubstr("0x1.e8480p+19 asdkjh").first_real_span() == "0x1.e8480p+19");
        CHECK(ryml::csubstr("0x1.e8480p-19 asdkjh").first_real_span() == "0x1.e8480p-19");
        CHECK(ryml::csubstr("-0x1.e8480p+19 asdkjh").first_real_span() == "-0x1.e8480p+19");
        CHECK(ryml::csubstr("-0x1.e8480p-19 asdkjh").first_real_span() == "-0x1.e8480p-19");
        CHECK(ryml::csubstr("+0x1.e8480p+19 asdkjh").first_real_span() == "+0x1.e8480p+19");
        CHECK(ryml::csubstr("+0x1.e8480p-19 asdkjh").first_real_span() == "+0x1.e8480p-19");
        // binary real numbers
        CHECK(ryml::csubstr("0b101.011p+19 asdkjh").first_real_span() == "0b101.011p+19");
        CHECK(ryml::csubstr("0b101.011p-19 asdkjh").first_real_span() == "0b101.011p-19");
        CHECK(ryml::csubstr("-0b101.011p+19 asdkjh").first_real_span() == "-0b101.011p+19");
        CHECK(ryml::csubstr("-0b101.011p-19 asdkjh").first_real_span() == "-0b101.011p-19");
        CHECK(ryml::csubstr("+0b101.011p+19 asdkjh").first_real_span() == "+0b101.011p+19");
        CHECK(ryml::csubstr("+0b101.011p-19 asdkjh").first_real_span() == "+0b101.011p-19");
        // octal real numbers
        CHECK(ryml::csubstr("0o173.045p+19 asdkjh").first_real_span() == "0o173.045p+19");
        CHECK(ryml::csubstr("0o173.045p-19 asdkjh").first_real_span() == "0o173.045p-19");
        CHECK(ryml::csubstr("-0o173.045p+19 asdkjh").first_real_span() == "-0o173.045p+19");
        CHECK(ryml::csubstr("-0o173.045p-19 asdkjh").first_real_span() == "-0o173.045p-19");
        CHECK(ryml::csubstr("+0o173.045p+19 asdkjh").first_real_span() == "+0o173.045p+19");
        CHECK(ryml::csubstr("+0o173.045p-19 asdkjh").first_real_span() == "+0o173.045p-19");
    }
    // see also is_number()

    // basename(), dirname(), extshort(), extlong()
    {
        CHECK(ryml::csubstr("/path/to/file.tar.gz").basename() == "file.tar.gz");
        CHECK(ryml::csubstr("/path/to/file.tar.gz").dirname() == "/path/to/");
        CHECK(ryml::csubstr("C:\\path\\to\\file.tar.gz").basename('\\') == "file.tar.gz");
        CHECK(ryml::csubstr("C:\\path\\to\\file.tar.gz").dirname('\\') == "C:\\path\\to\\");
        CHECK(ryml::csubstr("/path/to/file.tar.gz").extshort() == "gz");
        CHECK(ryml::csubstr("/path/to/file.tar.gz").extlong() == "tar.gz");
        CHECK(ryml::csubstr("/path/to/file.tar.gz").name_wo_extshort() == "/path/to/file.tar");
        CHECK(ryml::csubstr("/path/to/file.tar.gz").name_wo_extlong() == "/path/to/file");
    }

    // split()
    {
        using namespace ryml;
        csubstr parts[] = {"aa", "bb", "cc", "dd", "ee", "ff"};
        {
            size_t count = 0;
            for(csubstr part : csubstr("aa/bb/cc/dd/ee/ff").split('/'))
                CHECK(part == parts[count++]);
            CHECK(count == 6u);
        }
        {
            size_t count = 0;
            for(csubstr part : csubstr("aa.bb.cc.dd.ee.ff").split('.'))
                CHECK(part == parts[count++]);
            CHECK(count == 6u);
        }
        {
            size_t count = 0;
            for(csubstr part : csubstr("aa-bb-cc-dd-ee-ff").split('-'))
                CHECK(part == parts[count++]);
            CHECK(count == 6u);
        }
        // see also next_split()
    }

    //  pop_left(),  pop_right() --- non-greedy version
    // gpop_left(), gpop_right() --- greedy version
    {
        const bool skip_empty = true;
        // pop_left(): pop the last element from the left
        CHECK(ryml::csubstr(  "0/1/2"   ). pop_left('/'            ) ==   "0"    );
        CHECK(ryml::csubstr( "/0/1/2"   ). pop_left('/'            ) == ""       );
        CHECK(ryml::csubstr("//0/1/2"   ). pop_left('/'            ) == ""       );
        CHECK(ryml::csubstr(  "0/1/2"   ). pop_left('/', skip_empty) ==   "0"    );
        CHECK(ryml::csubstr( "/0/1/2"   ). pop_left('/', skip_empty) ==  "/0"    );
        CHECK(ryml::csubstr("//0/1/2"   ). pop_left('/', skip_empty) == "//0"    );
        // gpop_left(): pop all but the first element (greedy pop)
        CHECK(ryml::csubstr(  "0/1/2"   ).gpop_left('/'            ) ==   "0/1"  );
        CHECK(ryml::csubstr( "/0/1/2"   ).gpop_left('/'            ) ==  "/0/1"  );
        CHECK(ryml::csubstr("//0/1/2"   ).gpop_left('/'            ) == "//0/1"  );
        CHECK(ryml::csubstr(  "0/1/2/"  ).gpop_left('/'            ) ==   "0/1/2");
        CHECK(ryml::csubstr( "/0/1/2/"  ).gpop_left('/'            ) ==  "/0/1/2");
        CHECK(ryml::csubstr("//0/1/2/"  ).gpop_left('/'            ) == "//0/1/2");
        CHECK(ryml::csubstr(  "0/1/2//" ).gpop_left('/'            ) ==   "0/1/2/");
        CHECK(ryml::csubstr( "/0/1/2//" ).gpop_left('/'            ) ==  "/0/1/2/");
        CHECK(ryml::csubstr("//0/1/2//" ).gpop_left('/'            ) == "//0/1/2/");
        CHECK(ryml::csubstr(  "0/1/2"   ).gpop_left('/', skip_empty) ==   "0/1"  );
        CHECK(ryml::csubstr( "/0/1/2"   ).gpop_left('/', skip_empty) ==  "/0/1"  );
        CHECK(ryml::csubstr("//0/1/2"   ).gpop_left('/', skip_empty) == "//0/1"  );
        CHECK(ryml::csubstr(  "0/1/2/"  ).gpop_left('/', skip_empty) ==   "0/1"  );
        CHECK(ryml::csubstr( "/0/1/2/"  ).gpop_left('/', skip_empty) ==  "/0/1"  );
        CHECK(ryml::csubstr("//0/1/2/"  ).gpop_left('/', skip_empty) == "//0/1"  );
        CHECK(ryml::csubstr(  "0/1/2//" ).gpop_left('/', skip_empty) ==   "0/1"  );
        CHECK(ryml::csubstr( "/0/1/2//" ).gpop_left('/', skip_empty) ==  "/0/1"  );
        CHECK(ryml::csubstr("//0/1/2//" ).gpop_left('/', skip_empty) == "//0/1"  );
        // pop_right(): pop the last element from the right
        CHECK(ryml::csubstr(  "0/1/2"   ). pop_right('/'            ) ==   "2"    );
        CHECK(ryml::csubstr(  "0/1/2/"  ). pop_right('/'            ) == ""       );
        CHECK(ryml::csubstr(  "0/1/2//" ). pop_right('/'            ) == ""       );
        CHECK(ryml::csubstr(  "0/1/2"   ). pop_right('/', skip_empty) ==   "2"    );
        CHECK(ryml::csubstr(  "0/1/2/"  ). pop_right('/', skip_empty) ==   "2/"   );
        CHECK(ryml::csubstr(  "0/1/2//" ). pop_right('/', skip_empty) ==   "2//"  );
        // gpop_right(): pop all but the first element (greedy pop)
        CHECK(ryml::csubstr(  "0/1/2"   ).gpop_right('/'            ) ==     "1/2");
        CHECK(ryml::csubstr(  "0/1/2/"  ).gpop_right('/'            ) ==     "1/2/"  );
        CHECK(ryml::csubstr(  "0/1/2//" ).gpop_right('/'            ) ==     "1/2//"  );
        CHECK(ryml::csubstr( "/0/1/2"   ).gpop_right('/'            ) ==   "0/1/2");
        CHECK(ryml::csubstr( "/0/1/2/"  ).gpop_right('/'            ) ==   "0/1/2/"  );
        CHECK(ryml::csubstr( "/0/1/2//" ).gpop_right('/'            ) ==   "0/1/2//"  );
        CHECK(ryml::csubstr("//0/1/2"   ).gpop_right('/'            ) ==  "/0/1/2");
        CHECK(ryml::csubstr("//0/1/2/"  ).gpop_right('/'            ) ==  "/0/1/2/"  );
        CHECK(ryml::csubstr("//0/1/2//" ).gpop_right('/'            ) ==  "/0/1/2//"  );
        CHECK(ryml::csubstr(  "0/1/2"   ).gpop_right('/', skip_empty) ==     "1/2");
        CHECK(ryml::csubstr(  "0/1/2/"  ).gpop_right('/', skip_empty) ==     "1/2/"  );
        CHECK(ryml::csubstr(  "0/1/2//" ).gpop_right('/', skip_empty) ==     "1/2//"  );
        CHECK(ryml::csubstr( "/0/1/2"   ).gpop_right('/', skip_empty) ==     "1/2");
        CHECK(ryml::csubstr( "/0/1/2/"  ).gpop_right('/', skip_empty) ==     "1/2/"  );
        CHECK(ryml::csubstr( "/0/1/2//" ).gpop_right('/', skip_empty) ==     "1/2//"  );
        CHECK(ryml::csubstr("//0/1/2"   ).gpop_right('/', skip_empty) ==     "1/2");
        CHECK(ryml::csubstr("//0/1/2/"  ).gpop_right('/', skip_empty) ==     "1/2/"  );
        CHECK(ryml::csubstr("//0/1/2//" ).gpop_right('/', skip_empty) ==     "1/2//"  );
    }
}


//-----------------------------------------------------------------------------


/** demonstrate how to load a YAML file from disk to parse with ryml.
 *
 *  ryml offers no overload to directly parse files from disk; it only
 *  parses source buffers (which may be mutable or immutable). It is
 *  up to the caller to load the file contents into a buffer before
 *  parsing with ryml.
 *
 *  But that does not mean that loading a file is unimportant. There
 *  are many ways to achieve this in C++, but for convenience and to
 *  enable you to quickly get up to speed, here is an example
 *  implementation loading a file from disk and then parsing the
 *  resulting buffer with ryml.
 * @see doc_parse  */
void sample_parse_file()
{
    const char filename[] = "ryml_example.yml";

    // because this is a minimal sample, it assumes nothing on the
    // environment/OS (other than that it can read/write files). So we
    // create the file on the fly:
    file_put_contents(filename, ryml::csubstr(R"(
foo: 1
bar:
- 2
- 3
)"));

    // now we can load it into a std::string (for example):
    {
        std::string contents = file_get_contents<std::string>(filename);
        ryml::Tree tree = ryml::parse_in_arena(ryml::to_csubstr(contents)); // immutable (csubstr) overload
        CHECK(tree["foo"].val() == "1");
        CHECK(tree["bar"][0].val() == "2");
        CHECK(tree["bar"][1].val() == "3");
    }

    // or we can use a vector<char> instead:
    {
        std::vector<char> contents = file_get_contents<std::vector<char>>(filename);
        ryml::Tree tree = ryml::parse_in_place(ryml::to_substr(contents)); // mutable (csubstr) overload
        CHECK(tree["foo"].val() == "1");
        CHECK(tree["bar"][0].val() == "2");
        CHECK(tree["bar"][1].val() == "3");
    }

    // generally, any contiguous char container can be used with ryml,
    // provided that the ryml::substr/ryml::csubstr view can be
    // created out of it.
    //
    // ryml provides the overloads above for these two containers, but
    // if you are using another container it should be very easy (only
    // requires pointer and length).
}


//-----------------------------------------------------------------------------

/** demonstrate in-place parsing of a mutable YAML source buffer.
 * @see doc_parse  */
void sample_parse_in_place()
{
    // Like the name suggests, parse_in_place() directly mutates the
    // source buffer in place
    char src[] = "{foo: 1, bar: [2, 3]}"; // ryml can parse in situ
    ryml::substr srcview = src; // a mutable view to the source buffer
    ryml::Tree tree = ryml::parse_in_place(srcview); // you can also reuse the tree and/or parser
    ryml::ConstNodeRef root = tree.crootref(); // get a constant reference to the root

    CHECK(root.is_map());
    CHECK(root["foo"].is_keyval());
    CHECK(root["foo"].key() == "foo");
    CHECK(root["foo"].val() == "1");
    CHECK(root["bar"].is_seq());
    CHECK(root["bar"].has_key());
    CHECK(root["bar"].key() == "bar");
    CHECK(root["bar"][0].val() == "2");
    CHECK(root["bar"][1].val() == "3");

    // deserializing:
    int foo = 0, bar0 = 0, bar1 = 0;
    root["foo"] >> foo;
    root["bar"][0] >> bar0;
    root["bar"][1] >> bar1;
    CHECK(foo == 1);
    CHECK(bar0 == 2);
    CHECK(bar1 == 3);

    // after parsing, the tree holds views to the source buffer:
    CHECK(root["foo"].val().data()     == src + strlen("{foo: "));
    CHECK(root["foo"].val().begin()    == src + strlen("{foo: "));
    CHECK(root["foo"].val().end()      == src + strlen("{foo: 1"));
    CHECK(root["foo"].val().is_sub(srcview)); // equivalent to the previous three assertions
    CHECK(root["bar"][0].val().data()  == src + strlen("{foo: 1, bar: ["));
    CHECK(root["bar"][0].val().begin() == src + strlen("{foo: 1, bar: ["));
    CHECK(root["bar"][0].val().end()   == src + strlen("{foo: 1, bar: [2"));
    CHECK(root["bar"][0].val().is_sub(srcview)); // equivalent to the previous three assertions
    CHECK(root["bar"][1].val().data()  == src + strlen("{foo: 1, bar: [2, "));
    CHECK(root["bar"][1].val().begin() == src + strlen("{foo: 1, bar: [2, "));
    CHECK(root["bar"][1].val().end()   == src + strlen("{foo: 1, bar: [2, 3"));
    CHECK(root["bar"][1].val().is_sub(srcview)); // equivalent to the previous three assertions

    // NOTE. parse_in_place() cannot accept ryml::csubstr
    // so this will cause a /compile/ error:
    ryml::csubstr csrcview = srcview; // ok, can assign from mutable to immutable
    //tree = ryml::parse_in_place(csrcview); // compile error, cannot mutate an immutable view
    (void)csrcview;
}


//-----------------------------------------------------------------------------

/** demonstrate parsing of a read-only YAML source buffer
 * @see doc_parse  */
void sample_parse_in_arena()
{
    // to parse read-only memory, ryml will copy first to the tree's
    // arena, and then parse the copied buffer:
    ryml::Tree tree = ryml::parse_in_arena("{foo: 1, bar: [2, 3]}");
    ryml::ConstNodeRef root = tree.crootref(); // get a const reference to the root

    CHECK(root.is_map());
    CHECK(root["foo"].is_keyval());
    CHECK(root["foo"].key() == "foo");
    CHECK(root["foo"].val() == "1");
    CHECK(root["bar"].is_seq());
    CHECK(root["bar"].has_key());
    CHECK(root["bar"].key() == "bar");
    CHECK(root["bar"][0].val() == "2");
    CHECK(root["bar"][1].val() == "3");

    // deserializing:
    int foo = 0, bar0 = 0, bar1 = 0;
    root["foo"] >> foo;
    root["bar"][0] >> bar0;
    root["bar"][1] >> bar1;
    CHECK(foo == 1);
    CHECK(bar0 == 2);
    CHECK(bar1 == 3);

    // NOTE. parse_in_arena() cannot accept ryml::substr. Overloads
    // receiving substr buffers are declared, but intentionally left
    // undefined, so this will cause a /linker/ error
    char src[] = "{foo: is it really true}";
    ryml::substr srcview = src;
    //tree = ryml::parse_in_place(srcview); // linker error, overload intentionally undefined

    // If you really intend to parse a mutable buffer in the arena,
    // then simply convert it to immutable prior to calling
    // parse_in_arena():
    ryml::csubstr csrcview = srcview; // assigning from src also works
    tree = ryml::parse_in_arena(csrcview); // OK! csrcview is immutable
    CHECK(tree["foo"].val() == "is it really true");
}


//-----------------------------------------------------------------------------

/** demonstrate reuse/modification of tree when parsing
 * @see doc_parse */
void sample_parse_reuse_tree()
{
    ryml::Tree tree;

    // it will always be faster if the tree's size is conveniently reserved:
    tree.reserve(30); // reserve 30 nodes (good enough for this sample)
    // if you are using the tree's arena to serialize data,
    // then reserve also the arena's size:
    tree.reserve_arena(256); // reserve 256 characters (good enough for this sample)

    // now parse into the tree:
    ryml::csubstr yaml = R"(foo: 1
bar: [2, 3]
)";
    ryml::parse_in_arena(yaml, &tree);

    ryml::ConstNodeRef root = tree.crootref();
    CHECK(root.num_children() == 2);
    CHECK(root.is_map());
    CHECK(root["foo"].is_keyval());
    CHECK(root["foo"].key() == "foo");
    CHECK(root["foo"].val() == "1");
    CHECK(root["bar"].is_seq());
    CHECK(root["bar"].has_key());
    CHECK(root["bar"].key() == "bar");
    CHECK(root["bar"][0].val() == "2");
    CHECK(root["bar"][1].val() == "3");
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(foo: 1
bar: [2,3]
)");

    // WATCHOUT: parsing into an existing tree will APPEND to it:
    ryml::parse_in_arena("{foo2: 12, bar2: [22, 32]}", &tree);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(foo: 1
bar: [2,3]
foo2: 12
bar2: [22,32]
)");
    CHECK(root.num_children() == 4);
    CHECK(root["foo2"].is_keyval());
    CHECK(root["foo2"].key() == "foo2");
    CHECK(root["foo2"].val() == "12");
    CHECK(root["bar2"].is_seq());
    CHECK(root["bar2"].has_key());
    CHECK(root["bar2"].key() == "bar2");
    CHECK(root["bar2"][0].val() == "22");
    CHECK(root["bar2"][1].val() == "32");

    // clear first before parsing into an existing tree.
    tree.clear();
    tree.clear_arena();  // you may or may not want to clear the arena
    ryml::parse_in_arena("- a\n- b\n- {x0: 1, x1: 2}", &tree);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == "- a\n- b\n- {x0: 1,x1: 2}\n");
    CHECK(root.is_seq());
    CHECK(root[0].val() == "a");
    CHECK(root[1].val() == "b");
    CHECK(root[2].is_map());
    CHECK(root[2]["x0"].val() == "1");
    CHECK(root[2]["x1"].val() == "2");

    // we can parse directly into a node nested deep in an existing tree:
    ryml::NodeRef mroot = tree.rootref(); // modifiable root
    ryml::parse_in_arena("champagne: Dom Perignon\ncoffee: Arabica", mroot.append_child());
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- a
- b
- {x0: 1,x1: 2}
- champagne: Dom Perignon
  coffee: Arabica
)");
    CHECK(root.is_seq());
    CHECK(root[0].val() == "a");
    CHECK(root[1].val() == "b");
    CHECK(root[2].is_map());
    CHECK(root[2]["x0"].val() == "1");
    CHECK(root[2]["x1"].val() == "2");
    CHECK(root[3].is_map());
    CHECK(root[3]["champagne"].val() == "Dom Perignon");
    CHECK(root[3]["coffee"].val() == "Arabica");

    // watchout: to add to an existing node within a map, the node's
    // key must be separately set first:
    ryml::NodeRef more = mroot[3].append_child({ryml::KEYMAP, "more"});
    ryml::NodeRef beer = mroot[3].append_child({ryml::KEYSEQ, "beer"});
    ryml::NodeRef always = mroot[3].append_child({ryml::KEY, "always"});
    ryml::parse_in_arena("{vinho verde: Soalheiro, vinho tinto: Redoma 2017}", more);
    ryml::parse_in_arena("- Rochefort 10\n- Busch\n- Leffe Rituel", beer);
    ryml::parse_in_arena("lots\nof\nwater", always);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- a
- b
- {x0: 1,x1: 2}
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
  always: lots of water
)");

    // can append at the top:
    ryml::parse_in_arena("- foo\n- bar\n- baz\n- bat", mroot);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- a
- b
- {x0: 1,x1: 2}
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
  always: lots of water
- foo
- bar
- baz
- bat
)");

    // or nested:
    ryml::parse_in_arena("[Kasteel Donker]", beer);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- a
- b
- {x0: 1,x1: 2}
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
    - Kasteel Donker
  always: lots of water
- foo
- bar
- baz
- bat
)");
}


//-----------------------------------------------------------------------------

/** Demonstrates reuse of an existing parser. Doing this is
 * recommended when multiple files are parsed.
 * @see doc_parse */
void sample_parse_reuse_parser()
{
    ryml::EventHandlerTree evt_handler = {};
    ryml::Parser parser(&evt_handler);

    // it is also advised to reserve the parser depth
    // to the expected depth of the data tree:
    parser.reserve_stack(10); // uses small storage optimization
                              // defaulting to 16 depth, so this
                              // instruction is a no-op, and the stack
                              // will located in the parser object.
    parser.reserve_stack(20); // But this will cause an allocation
                              // because it is above 16.

    ryml::Tree champagnes = parse_in_arena(&parser, "champagnes.yml", "[Dom Perignon, Gosset Grande Reserve, Jacquesson 742]");
    CHECK(ryml::emitrs_yaml<std::string>(champagnes) == "[Dom Perignon,Gosset Grande Reserve,Jacquesson 742]");

    ryml::Tree beers = parse_in_arena(&parser, "beers.yml", "[Rochefort 10, Busch, Leffe Rituel, Kasteel Donker]");
    CHECK(ryml::emitrs_yaml<std::string>(beers) == "[Rochefort 10,Busch,Leffe Rituel,Kasteel Donker]");
}


//-----------------------------------------------------------------------------

/** for ultimate speed when parsing multiple times, reuse both the
 * tree and parser
 * @see doc_parse */
void sample_parse_reuse_tree_and_parser()
{
    ryml::Tree tree;
    // it will always be faster if the tree's size is conveniently reserved:
    tree.reserve(30); // reserve 30 nodes (good enough for this sample)
    // if you are using the tree's arena to serialize data,
    // then reserve also the arena's size:
    tree.reserve(256); // reserve 256 characters (good enough for this sample)

    ryml::EventHandlerTree evt_handler;
    ryml::Parser parser(&evt_handler);
    // it is also advised to reserve the parser depth
    // to the expected depth of the data tree:
    parser.reserve_stack(10); // the parser uses small storage
                              // optimization defaulting to 16 depth,
                              // so this instruction is a no-op, and
                              // the stack will be located in the
                              // parser object.
    parser.reserve_stack(20); // But this will cause an allocation
                              // because it is above 16.

    ryml::csubstr champagnes = "- Dom Perignon\n- Gosset Grande Reserve\n- Jacquesson 742";
    ryml::csubstr beers = "- Rochefort 10\n- Busch\n- Leffe Rituel\n- Kasteel Donker";
    ryml::csubstr wines = "- Soalheiro\n- Niepoort Redoma 2017\n- Vina Esmeralda";

    parse_in_arena(&parser, "champagnes.yml", champagnes, &tree);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- Dom Perignon
- Gosset Grande Reserve
- Jacquesson 742
)");

    // watchout: this will APPEND to the given tree:
    parse_in_arena(&parser, "beers.yml", beers, &tree);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- Dom Perignon
- Gosset Grande Reserve
- Jacquesson 742
- Rochefort 10
- Busch
- Leffe Rituel
- Kasteel Donker
)");

    // if you don't wish to append, clear the tree first:
    tree.clear();
    parse_in_arena(&parser, "wines.yml", wines, &tree);
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(- Soalheiro
- Niepoort Redoma 2017
- Vina Esmeralda
)");
}


//-----------------------------------------------------------------------------

/** shows how to programatically iterate through trees
 * @see doc_tree
 * @see doc_node_classes
 */
void sample_iterate_trees()
{
    const ryml::Tree tree = ryml::parse_in_arena(R"(doe: "a deer, a female deer"
ray: "a drop of golden sun"
pi: 3.14159
xmas: true
french-hens: 3
calling-birds:
   - huey
   - dewey
   - louie
   - fred
xmas-fifth-day:
  calling-birds: four
  french-hens: 3
  golden-rings: 5
  partridges:
    count: 1
    location: a pear tree
  turtle-doves: two
cars: GTO
)");
    ryml::ConstNodeRef root = tree.crootref();

    // iterate children
    {
        std::vector<ryml::csubstr> keys, vals; // to store all the root-level keys, vals
        for(ryml::ConstNodeRef n : root.children())
        {
            keys.emplace_back(n.key());
            vals.emplace_back(n.has_val() ? n.val() : ryml::csubstr{});
        }
        CHECK(keys[0] == "doe");
        CHECK(vals[0] == "a deer, a female deer");
        CHECK(keys[1] == "ray");
        CHECK(vals[1] == "a drop of golden sun");
        CHECK(keys[2] == "pi");
        CHECK(vals[2] == "3.14159");
        CHECK(keys[3] == "xmas");
        CHECK(vals[3] == "true");
        CHECK(root[5].has_key());
        CHECK(root[5].is_seq());
        CHECK(root[5].key() == "calling-birds");
        CHECK(!root[5].has_val()); // it is a map, so not a val
        //CHECK(root[5].val() == ""); // ERROR! node does not have a val.
        CHECK(keys[5] == "calling-birds");
        CHECK(vals[5] == "");
    }

    // iterate siblings
    {
        size_t count = 0;
        ryml::csubstr calling_birds[] = {"huey", "dewey", "louie", "fred"};
        for(ryml::ConstNodeRef n : root["calling-birds"][2].siblings())
            CHECK(n.val() == calling_birds[count++]);
        CHECK(count == 4u);
    }
}


//-----------------------------------------------------------------------------

/** shows how to programatically create trees
 * @see doc_tree
 * @see doc_node_classes
 * */
void sample_create_trees()
{
    ryml::NodeRef doe;
    CHECK(doe.invalid()); // it's pointing at nowhere

    ryml::Tree tree;
    ryml::NodeRef root = tree.rootref();
    root |= ryml::MAP; // mark root as a map
    doe = root["doe"];
    CHECK(!doe.invalid()); // it's now pointing at the tree
    CHECK(doe.is_seed()); // but the tree has nothing there, so this is only a seed

    // set the value of the node
    const char a_deer[] = "a deer, a female deer";
    doe = a_deer;
    // now the node really exists in the tree, and this ref is no
    // longer a seed:
    CHECK(!doe.is_seed());
    // WATCHOUT for lifetimes:
    CHECK(doe.val().str == a_deer); // it is pointing at the initial string
    // If you need to avoid lifetime dependency, serialize the data:
    {
        std::string a_drop = "a drop of golden sun";
        // this will copy the string to the tree's arena:
        // (see the serialization samples below)
        root["ray"] << a_drop;
        // and now you can modify the original string without changing
        // the tree:
        a_drop[0] = 'Z';
        a_drop[1] = 'Z';
    }
    CHECK(root["ray"].val() == "a drop of golden sun");

    // etc.
    root["pi"] << ryml::fmt::real(3.141592654, 5);
    root["xmas"] << ryml::fmt::boolalpha(true);
    root["french-hens"] << 3;
    ryml::NodeRef calling_birds = root["calling-birds"];
    calling_birds |= ryml::SEQ;
    calling_birds.append_child() = "huey";
    calling_birds.append_child() = "dewey";
    calling_birds.append_child() = "louie";
    calling_birds.append_child() = "fred";
    ryml::NodeRef xmas5 = root["xmas-fifth-day"];
    xmas5 |= ryml::MAP;
    xmas5["calling-birds"] = "four";
    xmas5["french-hens"] << 3;
    xmas5["golden-rings"] << 5;
    xmas5["partridges"] |= ryml::MAP;
    xmas5["partridges"]["count"] << 1;
    xmas5["partridges"]["location"] = "a pear tree";
    xmas5["turtle-doves"] = "two";
    root["cars"] = "GTO";

    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(doe: 'a deer, a female deer'
ray: a drop of golden sun
pi: 3.14159
xmas: true
french-hens: 3
calling-birds:
  - huey
  - dewey
  - louie
  - fred
xmas-fifth-day:
  calling-birds: four
  french-hens: 3
  golden-rings: 5
  partridges:
    count: 1
    location: a pear tree
  turtle-doves: two
cars: GTO
)");
}


//-----------------------------------------------------------------------------

/** demonstrates explicit and implicit interaction with the tree's string arena.
 * Notice that ryml only holds strings in the tree's nodes. */
void sample_tree_arena()
{
    // mutable buffers are parsed in situ:
    {
        char buf[] = "[a, b, c, d]";
        ryml::substr yml = buf;
        ryml::Tree tree = ryml::parse_in_place(yml);
        // notice the arena is empty:
        CHECK(tree.arena().empty());
        // and the tree is pointing at the original buffer:
        ryml::NodeRef root = tree.rootref();
        CHECK(root[0].val().is_sub(yml));
        CHECK(root[1].val().is_sub(yml));
        CHECK(root[2].val().is_sub(yml));
        CHECK(root[3].val().is_sub(yml));
        CHECK(yml.is_super(root[0].val()));
        CHECK(yml.is_super(root[1].val()));
        CHECK(yml.is_super(root[2].val()));
        CHECK(yml.is_super(root[3].val()));
    }

    // when parsing immutable buffers, the buffer is first copied to the
    // tree's arena; the copy in the arena is then the buffer which is
    // actually parsed
    {
        ryml::csubstr yml = "[a, b, c, d]";
        ryml::Tree tree = ryml::parse_in_arena(yml);
        // notice the buffer was copied to the arena:
        CHECK(tree.arena().data() != yml.data());
        CHECK(tree.arena() == yml);
        // and the tree is pointing at the arena instead of to the
        // original buffer:
        ryml::NodeRef root = tree.rootref();
        ryml::csubstr arena = tree.arena();
        CHECK(root[0].val().is_sub(arena));
        CHECK(root[1].val().is_sub(arena));
        CHECK(root[2].val().is_sub(arena));
        CHECK(root[3].val().is_sub(arena));
        CHECK(arena.is_super(root[0].val()));
        CHECK(arena.is_super(root[1].val()));
        CHECK(arena.is_super(root[2].val()));
        CHECK(arena.is_super(root[3].val()));
    }

    // the arena is also used when the data is serialized to string
    // with NodeRef::operator<<(): mutable buffer
    {
        char buf[] = "[a, b, c, d]"; // mutable
        ryml::substr yml = buf;
        ryml::Tree tree = ryml::parse_in_place(yml);
        // notice the arena is empty:
        CHECK(tree.arena().empty());
        ryml::NodeRef root = tree.rootref();

        // serialize an integer, and mutate the tree
        CHECK(root[2].val() == "c");
        CHECK(root[2].val().is_sub(yml)); // val is first pointing at the buffer
        root[2] << 12345;
        CHECK(root[2].val() == "12345");
        CHECK(root[2].val().is_sub(tree.arena())); // now val is pointing at the arena
        // notice the serialized string was appended to the tree's arena:
        CHECK(tree.arena() == "12345");

        // serialize an integer, and mutate the tree
        CHECK(root[3].val() == "d");
        CHECK(root[3].val().is_sub(yml)); // val is first pointing at the buffer
        root[3] << 67890;
        CHECK(root[3].val() == "67890");
        CHECK(root[3].val().is_sub(tree.arena())); // now val is pointing at the arena
        // notice the serialized string was appended to the tree's arena:
        CHECK(tree.arena() == "1234567890");
    }
    // the arena is also used when the data is serialized to string
    // with NodeRef::operator<<(): immutable buffer
    {
        ryml::csubstr yml = "[a, b, c, d]"; // immutable
        ryml::Tree tree = ryml::parse_in_arena(yml);
        // notice the buffer was copied to the arena:
        CHECK(tree.arena().data() != yml.data());
        CHECK(tree.arena() == yml);
        ryml::NodeRef root = tree.rootref();

        // serialize an integer, and mutate the tree
        CHECK(root[2].val() == "c");
        root[2] << 12345; // serialize an integer
        CHECK(root[2].val() == "12345");
        // notice the serialized string was appended to the tree's arena:
        // notice also the previous values remain there.
        // RYML DOES NOT KEEP TRACK OF REFERENCES TO THE ARENA.
        CHECK(tree.arena() == "[a, b, c, d]12345");
        // old values:  --------------^

        // serialize an integer, and mutate the tree
        root[3] << 67890;
        CHECK(root[3].val() == "67890");
        // notice the serialized string was appended to the tree's arena:
        // notice also the previous values remain there.
        // RYML DOES NOT KEEP TRACK OF REFERENCES TO THE ARENA.
        CHECK(tree.arena() == "[a, b, c, d]1234567890");
        // old values:  --------------^ ---^^^^^
    }

    // to_arena(): directly serialize values to the arena:
    {
        ryml::Tree tree = ryml::parse_in_arena("{a: b}");
        ryml::csubstr c10 = tree.to_arena(10101010);
        CHECK(c10 == "10101010");
        CHECK(c10.is_sub(tree.arena()));
        CHECK(tree.arena() == "{a: b}10101010");
        CHECK(tree.key(1) == "a");
        CHECK(tree.val(1) == "b");
        tree.set_val(1, c10);
        CHECK(tree.val(1) == c10);
        // and you can also do it through a node:
        ryml::NodeRef root = tree.rootref();
        root["a"].set_val_serialized(2222);
        CHECK(root["a"].val() == "2222");
        CHECK(tree.arena() == "{a: b}101010102222");
    }

    // copy_to_arena(): manually copy a string to the arena:
    {
        ryml::Tree tree = ryml::parse_in_arena("{a: b}");
        ryml::csubstr mystr = "Gosset Grande Reserve";
        ryml::csubstr copied = tree.copy_to_arena(mystr);
        CHECK(!copied.overlaps(mystr));
        CHECK(copied == mystr);
        CHECK(tree.arena() == "{a: b}Gosset Grande Reserve");
    }

    // alloc_arena(): allocate a buffer from the arena:
    {
        ryml::Tree tree = ryml::parse_in_arena("{a: b}");
        ryml::csubstr mystr = "Gosset Grande Reserve";
        ryml::substr copied = tree.alloc_arena(mystr.size());
        CHECK(!copied.overlaps(mystr));
        memcpy(copied.str, mystr.str, mystr.len);
        CHECK(copied == mystr);
        CHECK(tree.arena() == "{a: b}Gosset Grande Reserve");
    }

    // reserve_arena(): ensure the arena has a certain size to avoid reallocations
    {
        ryml::Tree tree = ryml::parse_in_arena("{a: b}");
        CHECK(tree.arena().size() == strlen("{a: b}"));
        tree.reserve_arena(100);
        CHECK(tree.arena_capacity() >= 100);
        CHECK(tree.arena().size() == strlen("{a: b}"));
        tree.to_arena(123456);
        CHECK(tree.arena().first(12) == "{a: b}123456");
    }
}


//-----------------------------------------------------------------------------

/** ryml provides facilities for serializing and deserializing the C++
    fundamental types, including boolean and null values; this is
    provided by the several overloads in @ref doc_to_chars and @ref
    doc_from_chars. To add serialization for user scalar types (ie,
    those types that should be serialized as strings in leaf nodes),
    you just need to define the appropriate overloads of to_chars and
    from_chars for those types; see @ref sample_user_scalar_types for
    an example on how to achieve this, and see @ref doc_serialization
    for more information on serialization. */
void sample_fundamental_types()
{
    ryml::Tree tree;
    CHECK(tree.arena().empty());
    CHECK(tree.to_arena('a') == "a");               CHECK(tree.arena() == "a");
    CHECK(tree.to_arena("bcde") == "bcde");         CHECK(tree.arena() == "abcde");
    CHECK(tree.to_arena(unsigned(0)) == "0");       CHECK(tree.arena() == "abcde0");
    CHECK(tree.to_arena(int(1)) == "1");            CHECK(tree.arena() == "abcde01");
    CHECK(tree.to_arena(uint8_t(0)) == "0");        CHECK(tree.arena() == "abcde010");
    CHECK(tree.to_arena(uint16_t(1)) ==  "1");      CHECK(tree.arena() == "abcde0101");
    CHECK(tree.to_arena(uint32_t(2)) ==  "2");      CHECK(tree.arena() == "abcde01012");
    CHECK(tree.to_arena(uint64_t(3)) ==  "3");      CHECK(tree.arena() == "abcde010123");
    CHECK(tree.to_arena(int8_t( 4))  ==  "4");      CHECK(tree.arena() == "abcde0101234");
    CHECK(tree.to_arena(int8_t(-4))  == "-4");      CHECK(tree.arena() == "abcde0101234-4");
    CHECK(tree.to_arena(int16_t( 5)) ==  "5");      CHECK(tree.arena() == "abcde0101234-45");
    CHECK(tree.to_arena(int16_t(-5)) == "-5");      CHECK(tree.arena() == "abcde0101234-45-5");
    CHECK(tree.to_arena(int32_t( 6)) ==  "6");      CHECK(tree.arena() == "abcde0101234-45-56");
    CHECK(tree.to_arena(int32_t(-6)) == "-6");      CHECK(tree.arena() == "abcde0101234-45-56-6");
    CHECK(tree.to_arena(int64_t( 7)) ==  "7");      CHECK(tree.arena() == "abcde0101234-45-56-67");
    CHECK(tree.to_arena(int64_t(-7)) == "-7");      CHECK(tree.arena() == "abcde0101234-45-56-67-7");
    CHECK(tree.to_arena((void*)1) == "0x1");        CHECK(tree.arena() == "abcde0101234-45-56-67-70x1");
    CHECK(tree.to_arena(float(0.124)) == "0.124");  CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.124");
    CHECK(tree.to_arena(double(0.234)) == "0.234"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.234");

    // write boolean values - see also sample_formatting()
    CHECK(tree.to_arena(bool(true)) == "1");                    CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.2341");
    CHECK(tree.to_arena(bool(false)) == "0");                   CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410");
    CHECK(tree.to_arena(c4::fmt::boolalpha(true)) == "true");   CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410true");
    CHECK(tree.to_arena(c4::fmt::boolalpha(false)) == "false"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse");

    // write special float values
    // see also sample_float_precision()
    const float  fnan = std::numeric_limits<float >::quiet_NaN();
    const double dnan = std::numeric_limits<double>::quiet_NaN();
    const float  finf = std::numeric_limits<float >::infinity();
    const double dinf = std::numeric_limits<double>::infinity();
    CHECK(tree.to_arena( finf) ==  ".inf"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse.inf");
    CHECK(tree.to_arena( dinf) ==  ".inf"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse.inf.inf");
    CHECK(tree.to_arena(-finf) == "-.inf"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse.inf.inf-.inf");
    CHECK(tree.to_arena(-dinf) == "-.inf"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse.inf.inf-.inf-.inf");
    CHECK(tree.to_arena( fnan) ==  ".nan"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse.inf.inf-.inf-.inf.nan");
    CHECK(tree.to_arena( dnan) ==  ".nan"); CHECK(tree.arena() == "abcde0101234-45-56-67-70x10.1240.23410truefalse.inf.inf-.inf-.inf.nan.nan");

    // read special float values
    // see also sample_float_precision()
    C4_SUPPRESS_WARNING_GCC_CLANG_WITH_PUSH("-Wfloat-equal");
    tree = ryml::parse_in_arena(R"({ninf: -.inf, pinf: .inf, nan: .nan})");
    float f = 0.f;
    double d = 0.;
    CHECK(f == 0.f);
    CHECK(d == 0.);
    tree["ninf"] >> f; CHECK(f == -finf);
    tree["ninf"] >> d; CHECK(d == -dinf);
    tree["pinf"] >> f; CHECK(f ==  finf);
    tree["pinf"] >> d; CHECK(d ==  dinf);
    tree["nan" ] >> f; CHECK(std::isnan(f));
    tree["nan" ] >> d; CHECK(std::isnan(d));
    C4_SUPPRESS_WARNING_GCC_CLANG_POP

    // value overflow detection:
    // (for integral types only)
    {
        // we will be detecting errors below, so we use this sample helper
        ScopedErrorHandlerExample err = {};
        ryml::Tree t(err.callbacks()); // instantiate with the error-detecting callbacks
        // create a simple tree with an int value
        ryml::parse_in_arena(R"({val: 258})", &t);
        // by default, overflow is not detected:
        uint8_t valu8 = 0;
        int8_t vali8 = 0;
        t["val"] >> valu8; CHECK(valu8 == 2); // not 257; it wrapped around
        t["val"] >> vali8; CHECK(vali8 == 2); // not 257; it wrapped around
        // ...but there are facilities to detect overflow
        CHECK(ryml::overflows<uint8_t>(t["val"].val()));
        CHECK(ryml::overflows<int8_t>(t["val"].val()));
        CHECK( ! ryml::overflows<int16_t>(t["val"].val()));
        // and there is a format helper
        CHECK(err.check_error_occurs([&]{
            auto checku8 = ryml::fmt::overflow_checked(valu8); // need to declare the wrapper type before using it with >>
            t["val"] >> checku8; // this will cause an error
        }));
        CHECK(err.check_error_occurs([&]{
            auto checki8 = ryml::fmt::overflow_checked(vali8); // need to declare the wrapper type before using it with >>
            t["val"] >> checki8; // this will cause an error
        }));
    }
}


//-----------------------------------------------------------------------------

/** Shows how to deal with empty/null values. See also @ref
 * c4::yml::Tree::val_is_null */
void sample_empty_null_values()
{
    // reading empty/null values - see also sample_formatting()
    ryml::Tree tree = ryml::parse_in_arena(R"(
plain:
squoted: ''
dquoted: ""
literal: |
folded: >
all_null: [~, null, Null, NULL]
non_null: [nULL, non_null, non null, null it is not]
)");
    // first, remember that .has_val() is a structural predicate
    // indicating the node is a leaf, and not a container.
    CHECK(tree["plain"].has_val()); // has a val, even if it's empty!
    CHECK(tree["squoted"].has_val());
    CHECK(tree["dquoted"].has_val());
    CHECK(tree["literal"].has_val());
    CHECK(tree["folded"].has_val());
    CHECK( ! tree["all_null"].has_val());
    CHECK( ! tree["non_null"].has_val());
    // In essence, has_val() is the logical opposite of is_container()
    CHECK( ! tree["plain"].is_container());
    CHECK( ! tree["squoted"].is_container());
    CHECK( ! tree["dquoted"].is_container());
    CHECK( ! tree["literal"].is_container());
    CHECK( ! tree["folded"].is_container());
    CHECK(tree["all_null"].is_container());
    CHECK(tree["non_null"].is_container());
    //
    // Right. How about the contents of each val?
    //
    // all of these scalars have zero-length:
    CHECK(tree["plain"].val().len == 0);
    CHECK(tree["squoted"].val().len == 0);
    CHECK(tree["dquoted"].val().len == 0);
    CHECK(tree["literal"].val().len == 0);
    CHECK(tree["folded"].val().len == 0);
    // but only the empty scalar has null string:
    CHECK(tree["plain"].val().str == nullptr);
    CHECK(tree["squoted"].val().str != nullptr);
    CHECK(tree["dquoted"].val().str != nullptr);
    CHECK(tree["literal"].val().str != nullptr);
    CHECK(tree["folded"].val().str != nullptr);
    // likewise, scalar comparison to nullptr has the same results:
    // (remember that .val() gives you the scalar value, node must
    // have a val, ie must be a leaf node, not a container)
    CHECK(tree["plain"].val() == nullptr);
    CHECK(tree["squoted"].val() != nullptr);
    CHECK(tree["dquoted"].val() != nullptr);
    CHECK(tree["literal"].val() != nullptr);
    CHECK(tree["folded"].val() != nullptr);
    // the tree and node classes provide the corresponding predicate
    // functions .key_is_null() and .val_is_null().
    // (note that these functions have the same preconditions as .val(),
    // because they need get the val to look into its contents)
    CHECK(tree["plain"].val_is_null());
    CHECK( ! tree["squoted"].val_is_null());
    CHECK( ! tree["dquoted"].val_is_null());
    CHECK( ! tree["literal"].val_is_null());
    CHECK( ! tree["folded"].val_is_null());
    // matching to null is case-sensitive. only the cases shown here
    // match to null:
    for(ryml::ConstNodeRef child : tree["all_null"].children())
    {
        CHECK(child.val() != nullptr); // it is pointing at a string, so it is not nullptr!
        CHECK(child.val_is_null());
    }
    for(ryml::ConstNodeRef child : tree["non_null"].children())
    {
        CHECK(child.val() != nullptr);
        CHECK( ! child.val_is_null());
    }
    //
    //
    // Because the meaning of null/~/empty will vary from application
    // to application, ryml makes no assumption on what should be
    // serialized as null. It leaves this decision to the user. But
    // it also provides the proper toolbox for the user to implement
    // its intended solution.
    //
    // writing/disambiguating null values:
    ryml::csubstr null = {};
    ryml::csubstr nonnull = "";
    ryml::csubstr strnull = "null";
    ryml::csubstr tilde = "~";
    CHECK(null   .len == 0); CHECK(null   .str == nullptr); CHECK(null    == nullptr);
    CHECK(nonnull.len == 0); CHECK(nonnull.str != nullptr); CHECK(nonnull != nullptr);
    CHECK(strnull.len != 0); CHECK(strnull.str != nullptr); CHECK(strnull != nullptr);
    CHECK(tilde  .len != 0); CHECK(tilde  .str != nullptr); CHECK(tilde   != nullptr);
    tree.clear();
    tree.clear_arena();
    tree.rootref() |= ryml::MAP;
    // serializes as an empty plain scalar:
    tree["empty_null"] << null; CHECK(tree.arena() == "");
    // serializes as an empty quoted scalar:
    tree["empty_nonnull"] << nonnull; CHECK(tree.arena() == "");
    // serializes as the normal 'null' string:
    tree["str_null"] << strnull; CHECK(tree.arena() == "null");
    // serializes as the normal '~' string:
    tree["str_tilde"] << tilde; CHECK(tree.arena() == "null~");
    // this is the resulting yaml:
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(empty_null: 
empty_nonnull: ''
str_null: null
str_tilde: ~
)");
    // To enforce a particular concept of what is a null string, you
    // can use the appropriate condition based on pointer nulity or
    // other appropriate criteria.
    //
    // As an example, proper comparison to nullptr:
    auto null_if_nullptr = [](ryml::csubstr s) {
        return s.str == nullptr ? "null" : s;
    };
    tree["empty_null"] << null_if_nullptr(null);
    tree["empty_nonnull"] << null_if_nullptr(nonnull);
    tree["str_null"] << null_if_nullptr(strnull);
    tree["str_tilde"] << null_if_nullptr(tilde);
    // this is the resulting yaml:
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(empty_null: null
empty_nonnull: ''
str_null: null
str_tilde: ~
)");
    //
    // As another example, nulity check based on the YAML nulity
    // predicate:
    auto null_if_predicate = [](ryml::csubstr s) {
        return ryml::scalar_is_null(s) ? "null" : s;
    };
    tree["empty_null"] << null_if_predicate(null);
    tree["empty_nonnull"] << null_if_predicate(nonnull);
    tree["str_null"] << null_if_predicate(strnull);
    tree["str_tilde"] << null_if_predicate(tilde);
    // this is the resulting yaml:
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(empty_null: null
empty_nonnull: ''
str_null: null
str_tilde: null
)");
    //
    // As another example, nulity check based on the YAML nulity
    // predicate, but returning "~" to simbolize nulity:
    auto tilde_if_predicate = [](ryml::csubstr s) {
        return ryml::scalar_is_null(s) ? "~" : s;
    };
    tree["empty_null"] << tilde_if_predicate(null);
    tree["empty_nonnull"] << tilde_if_predicate(nonnull);
    tree["str_null"] << tilde_if_predicate(strnull);
    tree["str_tilde"] << tilde_if_predicate(tilde);
    // this is the resulting yaml:
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(empty_null: ~
empty_nonnull: ''
str_null: ~
str_tilde: ~
)");
}


//-----------------------------------------------------------------------------

/** ryml provides facilities for formatting/deformatting (imported
 * from c4core into the ryml namespace). See @ref doc_format_utils
 * . These functions are very useful to serialize and deserialize
 * scalar types; see @ref doc_serialization .
 */
void sample_formatting()
{
    // format(), format_sub(), formatrs(): format arguments
    {
        char buf_[256] = {};
        ryml::substr buf = buf_;
        size_t size = ryml::format(buf, "a={} foo {} {} bar {}", 0.1, 10, 11, 12);
        CHECK(size == strlen("a=0.1 foo 10 11 bar 12"));
        CHECK(buf.first(size) == "a=0.1 foo 10 11 bar 12");
        // it is safe to call on an empty buffer:
        // returns the size needed for the result, and no overflow occurs:
        size = ryml::format({} , "a={} foo {} {} bar {}", "this_is_a", 10, 11, 12);
        CHECK(size == ryml::format(buf, "a={} foo {} {} bar {}", "this_is_a", 10, 11, 12));
        CHECK(size == strlen("a=this_is_a foo 10 11 bar 12"));
        // it is also safe to call on an insufficient buffer:
        char smallbuf[8] = {};
        size = ryml::format(smallbuf, "{} is too large {}", "this", "for the buffer");
        CHECK(size == strlen("this is too large for the buffer"));
        // ... and the result is truncated at the buffer size:
        CHECK(ryml::substr(smallbuf, sizeof(smallbuf)) == "this is\0");

        // format_sub() directly returns the written string:
        ryml::csubstr result = ryml::format_sub(buf, "b={}, damn it.", 1);
        CHECK(result == "b=1, damn it.");
        CHECK(result.is_sub(buf));

        // formatrs() means FORMAT & ReSize:
        //
        // Instead of a substr, it receives any owning linear char container
        // for which to_substr() is defined (using ADL).
        // <ryml_std.hpp> has to_substr() definitions for std::string and
        // std::vector<char>.
        //
        // formatrs() starts by calling format(), and if needed, resizes the container
        // and calls format() again.
        //
        // Note that unless the container is previously sized, this
        // may cause an allocation, which will make your code slower.
        // Make sure to call .reserve() on the container for real
        // production code.
        std::string sbuf;
        ryml::formatrs(&sbuf, "and c={} seems about right", 2);
        CHECK(sbuf == "and c=2 seems about right");
        std::vector<char> vbuf; // works with any linear char container
        ryml::formatrs(&vbuf, "and c={} seems about right", 2);
        CHECK(sbuf == "and c=2 seems about right");
        // with formatrs() it is also possible to append:
        ryml::formatrs_append(&sbuf, ", and finally d={} - done", 3);
        CHECK(sbuf == "and c=2 seems about right, and finally d=3 - done");
    }

    // unformat(): read arguments - opposite of format()
    {
        char buf_[256];

        int a = 0, b = 1, c = 2;
        ryml::csubstr result = ryml::format_sub(buf_, "{} and {} and {}", a, b, c);
        CHECK(result == "0 and 1 and 2");
        int aa = -1, bb = -2, cc = -3;
        size_t num_characters = ryml::unformat(result, "{} and {} and {}", aa, bb, cc);
        CHECK(num_characters != ryml::csubstr::npos); // if a conversion fails, returns ryml::csubstr::npos
        CHECK(num_characters == result.size());
        CHECK(aa == a);
        CHECK(bb == b);
        CHECK(cc == c);

        result = ryml::format_sub(buf_, "{} and {} and {}", 10, 20, 30);
        CHECK(result == "10 and 20 and 30");
        num_characters = ryml::unformat(result, "{} and {} and {}", aa, bb, cc);
        CHECK(num_characters != ryml::csubstr::npos); // if a conversion fails, returns ryml::csubstr::npos
        CHECK(num_characters == result.size());
        CHECK(aa == 10);
        CHECK(bb == 20);
        CHECK(cc == 30);
    }

    // cat(), cat_sub(), catrs(): concatenate arguments
    {
        char buf_[256] = {};
        ryml::substr buf = buf_;
        size_t size = ryml::cat(buf, "a=", 0.1, "foo", 10, 11, "bar", 12);
        CHECK(size == strlen("a=0.1foo1011bar12"));
        CHECK(buf.first(size) == "a=0.1foo1011bar12");
        // it is safe to call on an empty buffer:
        // returns the size needed for the result, and no overflow occurs:
        CHECK(ryml::cat({}, "a=", 0) == 3);
        // it is also safe to call on an insufficient buffer:
        char smallbuf[8] = {};
        size = ryml::cat(smallbuf, "this", " is too large ", "for the buffer");
        CHECK(size == strlen("this is too large for the buffer"));
        // ... and the result is truncated at the buffer size:
        CHECK(ryml::substr(smallbuf, sizeof(smallbuf)) == "this is\0");

        // cat_sub() directly returns the written string:
        ryml::csubstr result = ryml::cat_sub(buf, "b=", 1, ", damn it.");
        CHECK(result == "b=1, damn it.");
        CHECK(result.is_sub(buf));

        // catrs() means CAT & ReSize:
        //
        // Instead of a substr, it receives any owning linear char container
        // for which to_substr() is defined (using ADL).
        // <ryml_std.hpp> has to_substr() definitions for std::string and
        // std::vector<char>.
        //
        // catrs() starts by calling cat(), and if needed, resizes the container
        // and calls cat() again.
        //
        // Note that unless the container is previously sized, this
        // may cause an allocation, which will make your code slower.
        // Make sure to call .reserve() on the container for real
        // production code.
        std::string sbuf;
        ryml::catrs(&sbuf, "and c=", 2, " seems about right");
        CHECK(sbuf == "and c=2 seems about right");
        std::vector<char> vbuf; // works with any linear char container
        ryml::catrs(&vbuf, "and c=", 2, " seems about right");
        CHECK(sbuf == "and c=2 seems about right");
        // with catrs() it is also possible to append:
        ryml::catrs_append(&sbuf, ", and finally d=", 3, " - done");
        CHECK(sbuf == "and c=2 seems about right, and finally d=3 - done");
    }

    // uncat(): read arguments - opposite of cat()
    {
        char buf_[256];

        int a = 0, b = 1, c = 2;
        ryml::csubstr result = ryml::cat_sub(buf_, a, ' ', b, ' ', c);
        CHECK(result == "0 1 2");
        int aa = -1, bb = -2, cc = -3;
        char sep1 = 'a', sep2 = 'b';
        size_t num_characters = ryml::uncat(result, aa, sep1, bb, sep2, cc);
        CHECK(num_characters == result.size());
        CHECK(aa == a);
        CHECK(bb == b);
        CHECK(cc == c);
        CHECK(sep1 == ' ');
        CHECK(sep2 == ' ');

        result = ryml::cat_sub(buf_, 10, ' ', 20, ' ', 30);
        CHECK(result == "10 20 30");
        num_characters = ryml::uncat(result, aa, sep1, bb, sep2, cc);
        CHECK(num_characters == result.size());
        CHECK(aa == 10);
        CHECK(bb == 20);
        CHECK(cc == 30);
        CHECK(sep1 == ' ');
        CHECK(sep2 == ' ');
    }

    // catsep(), catsep_sub(), catseprs(): concatenate arguments, with a separator
    {
        char buf_[256] = {};
        ryml::substr buf = buf_;
        // use ' ' as a separator
        size_t size = ryml::catsep(buf, ' ', "a=", 0, "b=", 1, "c=", 2, 45, 67);
        CHECK(buf.first(size) == "a= 0 b= 1 c= 2 45 67");
        // any separator may be used
        // use " and " as a separator
        size = ryml::catsep(buf, " and ", "a=0", "b=1", "c=2", 45, 67);
        CHECK(buf.first(size) == "a=0 and b=1 and c=2 and 45 and 67");
        // use " ... " as a separator
        size = ryml::catsep(buf, " ... ", "a=0", "b=1", "c=2", 45, 67);
        CHECK(buf.first(size) == "a=0 ... b=1 ... c=2 ... 45 ... 67");
        // use '/' as a separator
        size = ryml::catsep(buf, '/', "a=", 0, "b=", 1, "c=", 2, 45, 67);
        CHECK(buf.first(size) == "a=/0/b=/1/c=/2/45/67");
        // use 888 as a separator
        size = ryml::catsep(buf, 888, "a=0", "b=1", "c=2", 45, 67);
        CHECK(buf.first(size) == "a=0888b=1888c=28884588867");

        // it is safe to call on an empty buffer:
        // returns the size needed for the result, and no overflow occurs:
        CHECK(size == ryml::catsep({}, 888, "a=0", "b=1", "c=2", 45, 67));
        // it is also safe to call on an insufficient buffer:
        char smallbuf[8] = {};
        CHECK(size == ryml::catsep(smallbuf, 888, "a=0", "b=1", "c=2", 45, 67));
        CHECK(size == strlen("a=0888b=1888c=28884588867"));
        // ... and the result is truncated:
        CHECK(ryml::substr(smallbuf, sizeof(smallbuf)) == "a=0888b\0");

        // catsep_sub() directly returns the written substr:
        ryml::csubstr result = ryml::catsep_sub(buf, " and ", "a=0", "b=1", "c=2", 45, 67);
        CHECK(result == "a=0 and b=1 and c=2 and 45 and 67");
        CHECK(result.is_sub(buf));

        // catseprs() means CATSEP & ReSize:
        //
        // Instead of a substr, it receives any owning linear char container
        // for which to_substr() is defined (using ADL).
        // <ryml_std.hpp> has to_substr() definitions for std::string and
        // std::vector<char>.
        //
        // catseprs() starts by calling catsep(), and if needed, resizes the container
        // and calls catsep() again.
        //
        // Note that unless the container is previously sized, this
        // may cause an allocation, which will make your code slower.
        // Make sure to call .reserve() on the container for real
        // production code.
        std::string sbuf;
        ryml::catseprs(&sbuf, " and ", "a=0", "b=1", "c=2", 45, 67);
        CHECK(sbuf == "a=0 and b=1 and c=2 and 45 and 67");
        std::vector<char> vbuf; // works with any linear char container
        ryml::catseprs(&vbuf, " and ", "a=0", "b=1", "c=2", 45, 67);
        CHECK(ryml::to_csubstr(vbuf) == "a=0 and b=1 and c=2 and 45 and 67");

        // with catseprs() it is also possible to append:
        ryml::catseprs_append(&sbuf, " well ", " --- a=0", "b=11", "c=12", 145, 167);
        CHECK(sbuf == "a=0 and b=1 and c=2 and 45 and 67 --- a=0 well b=11 well c=12 well 145 well 167");
    }

    // uncatsep(): read arguments with a separator - opposite of catsep()
    {
        char buf_[256] = {};

        int a = 0, b = 1, c = 2;
        ryml::csubstr result = ryml::catsep_sub(buf_, ' ', a, b, c);
        CHECK(result == "0 1 2");
        int aa = -1, bb = -2, cc = -3;
        char sep = 'b';
        size_t num_characters = ryml::uncatsep(result, sep, aa, bb, cc);
        CHECK(num_characters == result.size());
        CHECK(aa == a);
        CHECK(bb == b);
        CHECK(cc == c);
        CHECK(sep == ' ');

        sep = '_';
        result = ryml::catsep_sub(buf_, ' ', 10, 20, 30);
        CHECK(result == "10 20 30");
        num_characters = ryml::uncatsep(result, sep, aa, bb, cc);
        CHECK(num_characters == result.size());
        CHECK(aa == 10);
        CHECK(bb == 20);
        CHECK(cc == 30);
        CHECK(sep == ' ');
    }

    // formatting individual arguments
    {
        using namespace ryml;  // all the symbols below are in the ryml namespace.
        char buf_[256] = {};   // all the results below are written in this buffer
        substr buf = buf_;
        // --------------------------------------
        // fmt::boolalpha(): format as true/false
        // --------------------------------------
        // just as with std streams, printing a bool will output the integer value:
        CHECK("0" == cat_sub(buf, false));
        CHECK("1" == cat_sub(buf, true));
        // to force a "true"/"false", use fmt::boolalpha:
        CHECK("false" == cat_sub(buf, fmt::boolalpha(false)));
        CHECK("true" == cat_sub(buf, fmt::boolalpha(true)));

        // ---------------------------------
        // fmt::hex(): format as hexadecimal
        // ---------------------------------
        CHECK("0xff"  == cat_sub(buf, fmt::hex(255)));
        CHECK("0x100" == cat_sub(buf, fmt::hex(256)));
        CHECK("-0xff"  == cat_sub(buf, fmt::hex(-255)));
        CHECK("-0x100" == cat_sub(buf, fmt::hex(-256)));
        CHECK("3735928559" == cat_sub(buf,          UINT32_C(0xdeadbeef)));
        CHECK("0xdeadbeef" == cat_sub(buf, fmt::hex(UINT32_C(0xdeadbeef))));
        // ----------------------------
        // fmt::bin(): format as binary
        // ----------------------------
        CHECK("0b1000" == cat_sub(buf, fmt::bin(8)));
        CHECK("0b1001" == cat_sub(buf, fmt::bin(9)));
        CHECK("0b10001" == cat_sub(buf, fmt::bin(17)));
        CHECK("0b11001" == cat_sub(buf, fmt::bin(25)));
        CHECK("-0b1000" == cat_sub(buf, fmt::bin(-8)));
        CHECK("-0b1001" == cat_sub(buf, fmt::bin(-9)));
        CHECK("-0b10001" == cat_sub(buf, fmt::bin(-17)));
        CHECK("-0b11001" == cat_sub(buf, fmt::bin(-25)));
        // ---------------------------
        // fmt::bin(): format as octal
        // ---------------------------
        CHECK("0o77"    == cat_sub(buf, fmt::oct(63)));
        CHECK("0o100"   == cat_sub(buf, fmt::oct(64)));
        CHECK("0o377"   == cat_sub(buf, fmt::oct(255)));
        CHECK("0o400"   == cat_sub(buf, fmt::oct(256)));
        CHECK("0o1000"  == cat_sub(buf, fmt::oct(512)));
        CHECK("-0o77"   == cat_sub(buf, fmt::oct(-63)));
        CHECK("-0o100"  == cat_sub(buf, fmt::oct(-64)));
        CHECK("-0o377"  == cat_sub(buf, fmt::oct(-255)));
        CHECK("-0o400"  == cat_sub(buf, fmt::oct(-256)));
        CHECK("-0o1000" == cat_sub(buf, fmt::oct(-512)));
        // ---------------------------
        // fmt::zpad(): pad with zeros
        // ---------------------------
        CHECK("000063" == cat_sub(buf, fmt::zpad(63, 6)));
        CHECK( "00063" == cat_sub(buf, fmt::zpad(63, 5)));
        CHECK(  "0063" == cat_sub(buf, fmt::zpad(63, 4)));
        CHECK(   "063" == cat_sub(buf, fmt::zpad(63, 3)));
        CHECK(    "63" == cat_sub(buf, fmt::zpad(63, 2)));
        CHECK(    "63" == cat_sub(buf, fmt::zpad(63, 1))); // will never trim the result
        CHECK(    "63" == cat_sub(buf, fmt::zpad(63, 0))); // will never trim the result
        CHECK("0x00003f" == cat_sub(buf, fmt::zpad(fmt::hex(63), 6)));
        CHECK("0o000077" == cat_sub(buf, fmt::zpad(fmt::oct(63), 6)));
        CHECK("0b00011001" == cat_sub(buf, fmt::zpad(fmt::bin(25), 8)));
        // ------------------------------------------------
        // fmt::left(): align left with a given field width
        // ------------------------------------------------
        CHECK("63    " == cat_sub(buf, fmt::left(63, 6)));
        CHECK("63   "  == cat_sub(buf, fmt::left(63, 5)));
        CHECK("63  "   == cat_sub(buf, fmt::left(63, 4)));
        CHECK("63 "    == cat_sub(buf, fmt::left(63, 3)));
        CHECK("63"     == cat_sub(buf, fmt::left(63, 2)));
        CHECK("63"     == cat_sub(buf, fmt::left(63, 1))); // will never trim the result
        CHECK("63"     == cat_sub(buf, fmt::left(63, 0))); // will never trim the result
        // the fill character can be specified (defaults to ' '):
        CHECK("63----" == cat_sub(buf, fmt::left(63, 6, '-')));
        CHECK("63++++" == cat_sub(buf, fmt::left(63, 6, '+')));
        CHECK("63////" == cat_sub(buf, fmt::left(63, 6, '/')));
        CHECK("630000" == cat_sub(buf, fmt::left(63, 6, '0')));
        CHECK("63@@@@" == cat_sub(buf, fmt::left(63, 6, '@')));
        CHECK("0x003f    " == cat_sub(buf, fmt::left(fmt::zpad(fmt::hex(63), 4), 10)));
        // --------------------------------------------------
        // fmt::right(): align right with a given field width
        // --------------------------------------------------
        CHECK("    63" == cat_sub(buf, fmt::right(63, 6)));
        CHECK("   63"  == cat_sub(buf, fmt::right(63, 5)));
        CHECK("  63"   == cat_sub(buf, fmt::right(63, 4)));
        CHECK(" 63"    == cat_sub(buf, fmt::right(63, 3)));
        CHECK("63"     == cat_sub(buf, fmt::right(63, 2)));
        CHECK("63"     == cat_sub(buf, fmt::right(63, 1))); // will never trim the result
        CHECK("63"     == cat_sub(buf, fmt::right(63, 0))); // will never trim the result
        // the fill character can be specified (defaults to ' '):
        CHECK("----63" == cat_sub(buf, fmt::right(63, 6, '-')));
        CHECK("++++63" == cat_sub(buf, fmt::right(63, 6, '+')));
        CHECK("////63" == cat_sub(buf, fmt::right(63, 6, '/')));
        CHECK("000063" == cat_sub(buf, fmt::right(63, 6, '0')));
        CHECK("@@@@63" == cat_sub(buf, fmt::right(63, 6, '@')));
        CHECK("    0x003f" == cat_sub(buf, fmt::right(fmt::zpad(fmt::hex(63), 4), 10)));

        // ------------------------------------------
        // fmt::real(): format floating point numbers
        // ------------------------------------------
        // see also sample_float_precision()
        CHECK("0"       == cat_sub(buf, fmt::real(0.01f, 0)));
        CHECK("0.0"     == cat_sub(buf, fmt::real(0.01f, 1)));
        CHECK("0.01"    == cat_sub(buf, fmt::real(0.01f, 2)));
        CHECK("0.010"   == cat_sub(buf, fmt::real(0.01f, 3)));
        CHECK("0.0100"  == cat_sub(buf, fmt::real(0.01f, 4)));
        CHECK("0.01000" == cat_sub(buf, fmt::real(0.01f, 5)));
        CHECK("1"       == cat_sub(buf, fmt::real(1.01f, 0)));
        CHECK("1.0"     == cat_sub(buf, fmt::real(1.01f, 1)));
        CHECK("1.01"    == cat_sub(buf, fmt::real(1.01f, 2)));
        CHECK("1.010"   == cat_sub(buf, fmt::real(1.01f, 3)));
        CHECK("1.0100"  == cat_sub(buf, fmt::real(1.01f, 4)));
        CHECK("1.01000" == cat_sub(buf, fmt::real(1.01f, 5)));
        CHECK("1"       == cat_sub(buf, fmt::real(1.234234234, 0)));
        CHECK("1.2"     == cat_sub(buf, fmt::real(1.234234234, 1)));
        CHECK("1.23"    == cat_sub(buf, fmt::real(1.234234234, 2)));
        CHECK("1.234"   == cat_sub(buf, fmt::real(1.234234234, 3)));
        CHECK("1.2342"  == cat_sub(buf, fmt::real(1.234234234, 4)));
        CHECK("1.23423" == cat_sub(buf, fmt::real(1.234234234, 5)));
        CHECK("1000000.00000" == cat_sub(buf, fmt::real(1000000.000000000, 5)));
        CHECK("1234234.23423" == cat_sub(buf, fmt::real(1234234.234234234, 5)));
        // AKA %f
        CHECK("1000000.00000" == cat_sub(buf, fmt::real(1000000.000000000, 5, FTOA_FLOAT)));  // AKA %f, same as above
        CHECK("1234234.23423" == cat_sub(buf, fmt::real(1234234.234234234, 5, FTOA_FLOAT)));  // AKA %f
        CHECK("1234234.2342"  == cat_sub(buf, fmt::real(1234234.234234234, 4, FTOA_FLOAT)));  // AKA %f
        CHECK("1234234.234"   == cat_sub(buf, fmt::real(1234234.234234234, 3, FTOA_FLOAT)));  // AKA %f
        CHECK("1234234.23"    == cat_sub(buf, fmt::real(1234234.234234234, 2, FTOA_FLOAT)));  // AKA %f
        // AKA %e
        CHECK("1.00000e+06"   == cat_sub(buf, fmt::real(1000000.000000000, 5, FTOA_SCIENT))); // AKA %e
        CHECK("1.23423e+06"   == cat_sub(buf, fmt::real(1234234.234234234, 5, FTOA_SCIENT))); // AKA %e
        CHECK("1.2342e+06"    == cat_sub(buf, fmt::real(1234234.234234234, 4, FTOA_SCIENT))); // AKA %e
        CHECK("1.234e+06"     == cat_sub(buf, fmt::real(1234234.234234234, 3, FTOA_SCIENT))); // AKA %e
        CHECK("1.23e+06"      == cat_sub(buf, fmt::real(1234234.234234234, 2, FTOA_SCIENT))); // AKA %e
        // AKA %g
        CHECK("1e+06"         == cat_sub(buf, fmt::real(1000000.000000000, 5, FTOA_FLEX)));   // AKA %g
        CHECK("1.2342e+06"    == cat_sub(buf, fmt::real(1234234.234234234, 5, FTOA_FLEX)));   // AKA %g
        CHECK("1.234e+06"     == cat_sub(buf, fmt::real(1234234.234234234, 4, FTOA_FLEX)));   // AKA %g
        CHECK("1.23e+06"      == cat_sub(buf, fmt::real(1234234.234234234, 3, FTOA_FLEX)));   // AKA %g
        CHECK("1.2e+06"       == cat_sub(buf, fmt::real(1234234.234234234, 2, FTOA_FLEX)));   // AKA %g
        // FTOA_HEXA: AKA %a (hexadecimal formatting of floats)
        CHECK("0x1.e8480p+19" == cat_sub(buf, fmt::real(1000000.000000000, 5, FTOA_HEXA)));   // AKA %a
        CHECK("0x1.2d53ap+20" == cat_sub(buf, fmt::real(1234234.234234234, 5, FTOA_HEXA)));   // AKA %a

        // --------------------------------------------------------------
        // fmt::raw(): dump data in machine format (respecting alignment)
        // --------------------------------------------------------------
        {
            C4_SUPPRESS_WARNING_GCC_CLANG_WITH_PUSH("-Wcast-align")  // we're casting the values directly, so alignment is strictly respected.
            const uint32_t payload[] = {10, 20, 30, 40, UINT32_C(0xdeadbeef)};
            // (package payload as a substr, for comparison only)
            csubstr expected = csubstr((const char *)payload, sizeof(payload));
            csubstr actual = cat_sub(buf, fmt::raw(payload));
            CHECK(!actual.overlaps(expected));
            CHECK(0 == memcmp(expected.str, actual.str, expected.len));
            // also possible with variables:
            for(const uint32_t value : payload)
            {
                // (package payload as a substr, for comparison only)
                expected = csubstr((const char *)&value, sizeof(value));
                actual = cat_sub(buf, fmt::raw(value));
                CHECK(actual.size() == sizeof(uint32_t));
                CHECK(!actual.overlaps(expected));
                CHECK(0 == memcmp(expected.str, actual.str, expected.len));
                // with non-const data, fmt::craw() may be needed for disambiguation:
                actual = cat_sub(buf, fmt::craw(value));
                CHECK(actual.size() == sizeof(uint32_t));
                CHECK(!actual.overlaps(expected));
                CHECK(0 == memcmp(expected.str, actual.str, expected.len));
                //
                // read back:
                uint32_t result;
                auto reader = fmt::raw(result); // keeps a reference to result
                CHECK(&result == (uint32_t*)reader.buf);
                CHECK(reader.len == sizeof(uint32_t));
                uncat(actual, reader);
                // and compare:
                // (vs2017/release/32bit does not reload result from cache, so force it)
                result = *(uint32_t*)reader.buf;
                CHECK(result == value); // roundtrip completed successfully
            }
            C4_SUPPRESS_WARNING_GCC_CLANG_POP
        }

        // -------------------------
        // fmt::base64(): see below!
        // -------------------------
    }
}


//-----------------------------------------------------------------------------

/** demonstrates how to read and write base64-encoded blobs.
 @see @ref doc_base64
 */
void sample_base64()
{
    ryml::Tree tree;
    tree.rootref() |= ryml::MAP;
    struct text_and_base64 { ryml::csubstr text, base64; };
    text_and_base64 cases[] = {
        {{"Love all, trust a few, do wrong to none."}, {"TG92ZSBhbGwsIHRydXN0IGEgZmV3LCBkbyB3cm9uZyB0byBub25lLg=="}},
        {{"The fool doth think he is wise, but the wise man knows himself to be a fool."}, {"VGhlIGZvb2wgZG90aCB0aGluayBoZSBpcyB3aXNlLCBidXQgdGhlIHdpc2UgbWFuIGtub3dzIGhpbXNlbGYgdG8gYmUgYSBmb29sLg=="}},
        {{"Brevity is the soul of wit."}, {"QnJldml0eSBpcyB0aGUgc291bCBvZiB3aXQu"}},
        {{"All that glitters is not gold."}, {"QWxsIHRoYXQgZ2xpdHRlcnMgaXMgbm90IGdvbGQu"}},
    };
    // to encode base64 and write the result to val:
    for(text_and_base64 c : cases)
    {
        tree[c.text] << ryml::fmt::base64(c.text);
        CHECK(tree[c.text].val() == c.base64);
    }
    // to encode base64 and write the result to key:
    for(text_and_base64 c : cases)
    {
        tree.rootref().append_child() << ryml::key(ryml::fmt::base64(c.text)) << c.text;
        CHECK(tree[c.base64].val() == c.text);
    }
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"('Love all, trust a few, do wrong to none.': TG92ZSBhbGwsIHRydXN0IGEgZmV3LCBkbyB3cm9uZyB0byBub25lLg==
'The fool doth think he is wise, but the wise man knows himself to be a fool.': VGhlIGZvb2wgZG90aCB0aGluayBoZSBpcyB3aXNlLCBidXQgdGhlIHdpc2UgbWFuIGtub3dzIGhpbXNlbGYgdG8gYmUgYSBmb29sLg==
Brevity is the soul of wit.: QnJldml0eSBpcyB0aGUgc291bCBvZiB3aXQu
All that glitters is not gold.: QWxsIHRoYXQgZ2xpdHRlcnMgaXMgbm90IGdvbGQu
TG92ZSBhbGwsIHRydXN0IGEgZmV3LCBkbyB3cm9uZyB0byBub25lLg==: 'Love all, trust a few, do wrong to none.'
VGhlIGZvb2wgZG90aCB0aGluayBoZSBpcyB3aXNlLCBidXQgdGhlIHdpc2UgbWFuIGtub3dzIGhpbXNlbGYgdG8gYmUgYSBmb29sLg==: 'The fool doth think he is wise, but the wise man knows himself to be a fool.'
QnJldml0eSBpcyB0aGUgc291bCBvZiB3aXQu: Brevity is the soul of wit.
QWxsIHRoYXQgZ2xpdHRlcnMgaXMgbm90IGdvbGQu: All that glitters is not gold.
)");
    char buf1_[128], buf2_[128];
    ryml::substr buf1 = buf1_;  // this is where we will write the result (using >>)
    ryml::substr buf2 = buf2_;  // this is where we will write the result (using deserialize_val()/deserialize_key())
    std::string result = {}; // show also how to decode to a std::string
    // to decode the val base64 and write the result to buf:
    for(const text_and_base64 c : cases)
    {
        // write the decoded result into the given buffer
        tree[c.text] >> ryml::fmt::base64(buf1); // cannot know the needed size
        size_t len = tree[c.text].deserialize_val(ryml::fmt::base64(buf2)); // returns the needed size
        CHECK(len <= buf1.len);
        CHECK(len <= buf2.len);
        CHECK(c.text.len == len);
        CHECK(buf1.first(len) == c.text);
        CHECK(buf2.first(len) == c.text);
        //
        // interop with std::string: using substr
        result.clear(); // this is not needed. We do it just to show that the first call can fail.
        len = tree[c.text].deserialize_val(ryml::fmt::base64(ryml::to_substr(result))); // returns the needed size
        if(len > result.size()) // the size was not enough; resize and call again
        {
            result.resize(len);
            len = tree[c.text].deserialize_val(ryml::fmt::base64(ryml::to_substr(result))); // returns the needed size
        }
        result.resize(len); // trim to the length of the decoded buffer
        CHECK(result == c.text);
        //
        // interop with std::string: using blob
        result.clear(); // this is not needed. We do it just to show that the first call can fail.
        ryml::blob strblob(&result[0], result.size());
        CHECK(strblob.buf == result.data());
        CHECK(strblob.len == result.size());
        len = tree[c.text].deserialize_val(ryml::fmt::base64(strblob)); // returns the needed size
        if(len > result.size()) // the size was not enough; resize and call again
        {
            result.resize(len);
            strblob = {&result[0], result.size()};
            CHECK(strblob.buf == result.data());
            CHECK(strblob.len == result.size());
            len = tree[c.text].deserialize_val(ryml::fmt::base64(strblob)); // returns the needed size
        }
        result.resize(len); // trim to the length of the decoded buffer
        CHECK(result == c.text);
        //
        // Note also these are just syntatic wrappers to simplify client code.
        // You can call into the lower level functions without much effort:
        result.clear(); // this is not needed. We do it just to show that the first call can fail.
        ryml::csubstr encoded = tree[c.text].val();
        CHECK(encoded == c.base64);
        len = base64_decode(encoded, ryml::blob{&result[0], result.size()});
        if(len > result.size()) // the size was not enough; resize and call again
        {
            result.resize(len);
            len = base64_decode(encoded, ryml::blob{&result[0], result.size()});
        }
        result.resize(len); // trim to the length of the decoded buffer
        CHECK(result == c.text);
    }
    // to decode the key base64 and write the result to buf:
    for(const text_and_base64 c : cases)
    {
        // write the decoded result into the given buffer
        tree[c.base64] >> ryml::key(ryml::fmt::base64(buf1)); // cannot know the needed size
        size_t len = tree[c.base64].deserialize_key(ryml::fmt::base64(buf2)); // returns the needed size
        CHECK(len <= buf1.len);
        CHECK(len <= buf2.len);
        CHECK(c.text.len == len);
        CHECK(buf1.first(len) == c.text);
        CHECK(buf2.first(len) == c.text);
        // interop with std::string: using substr
        result.clear(); // this is not needed. We do it just to show that the first call can fail.
        len = tree[c.base64].deserialize_key(ryml::fmt::base64(ryml::to_substr(result))); // returns the needed size
        if(len > result.size()) // the size was not enough; resize and call again
        {
            result.resize(len);
            len = tree[c.base64].deserialize_key(ryml::fmt::base64(ryml::to_substr(result))); // returns the needed size
        }
        result.resize(len); // trim to the length of the decoded buffer
        CHECK(result == c.text);
        //
        // interop with std::string: using blob
        result.clear(); // this is not needed. We do it just to show that the first call can fail.
        ryml::blob strblob = {&result[0], result.size()};
        CHECK(strblob.buf == result.data());
        CHECK(strblob.len == result.size());
        len = tree[c.base64].deserialize_key(ryml::fmt::base64(strblob)); // returns the needed size
        if(len > result.size()) // the size was not enough; resize and call again
        {
            result.resize(len);
            strblob = {&result[0], result.size()};
            CHECK(strblob.buf == result.data());
            CHECK(strblob.len == result.size());
            len = tree[c.base64].deserialize_key(ryml::fmt::base64(strblob)); // returns the needed size
        }
        result.resize(len); // trim to the length of the decoded buffer
        CHECK(result == c.text);
        //
        // Note also these are just syntactic wrappers to simplify client code.
        // You can call into the lower level functions without much effort:
        result.clear(); // this is not needed. We do it just to show that the first call can fail.
        ryml::csubstr encoded = tree[c.base64].key();
        CHECK(encoded == c.base64);
        len = base64_decode(encoded, ryml::blob{&result[0], result.size()});
        if(len > result.size()) // the size was not enough; resize and call again
        {
            result.resize(len);
            len = base64_decode(encoded, ryml::blob{&result[0], result.size()});
        }
        result.resize(len); // trim to the length of the decoded buffer
        CHECK(result == c.text);
    }
    // directly encode variables
    {
        const uint64_t valin = UINT64_C(0xdeadbeef);
        uint64_t valout = 0;
        tree["deadbeef"] << c4::fmt::base64(valin); // sometimes cbase64() is needed to avoid ambiguity
        size_t len = tree["deadbeef"].deserialize_val(ryml::fmt::base64(valout));
        CHECK(len <= sizeof(valout));
        CHECK(valout == UINT64_C(0xdeadbeef)); // base64 roundtrip is bit-accurate
    }
    // directly encode memory ranges
    {
        const uint32_t data_in[11] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 0xdeadbeef};
        uint32_t data_out[11] = {};
        CHECK(memcmp(data_in, data_out, sizeof(data_in)) != 0); // before the roundtrip
        tree["int_data"] << c4::fmt::base64(data_in);
        size_t len = tree["int_data"].deserialize_val(ryml::fmt::base64(data_out));
        CHECK(len <= sizeof(data_out));
        CHECK(memcmp(data_in, data_out, sizeof(data_in)) == 0); // after the roundtrip
    }
}


//-----------------------------------------------------------------------------
// Serialization info

/** @} */ // doc_quickstart
/** @addtogroup doc_serialization
 *
 * @{
 *
 * ## Fundamental types
 *
 * ryml provides serialization/deserialization utilities for all
 * fundamental data types in @ref doc_charconv .
 *
 *  - See @ref sample_fundamental_types() for basic examples
 *    of serialization of fundamental types.
 *  - See @ref sample_empty_null_values() for different ways
 *    to serialize and deserialize empty and null values/
 *  - When serializing floating point values in C++ earlier than
 *    17, be aware that there may be a truncation of the precision
 *    with the default float/double implementations of @ref
 *    doc_to_chars. To enforce a particular precision, use for
 *    example @ref c4::fmt::real, or call directly @ref c4::ftoa() or
 *    @ref c4::dtoa(), or any other method (remember that ryml only
 *    stores the final string in the tree, so nothing prevents you from
 *    creating it in whatever way is most suitable). See the relevant
 *    sample: @ref sample_float_precision().
 *  - You can also serialize and deserialize base64: see @ref
 *    doc_base64 and @ref sample_base64
 *
 * To serialize/deserialize any non-fundamental type will require
 * that you instruct ryml on how to achieve this. That will differ
 * based on whether the type is scalar or container.
 *
 *
 * ## User scalar types
 *
 * See @ref doc_sample_scalar_types for serializing user scalar types
 *   (ie leaf nodes in the YAML tree, containing a string
 *   representation):
 *
 *  - See examples on how to @ref doc_sample_to_chars_scalar
 *  - See examples on how to @ref doc_sample_from_chars_scalar
 *  - See the sample @ref sample_user_scalar_types
 *  - See the sample @ref sample_formatting for examples
 *    of functions from @ref doc_format_utils that will be very
 *    helpful in implementing custom `to_chars()`/`from_chars()`
 *    functions.
 *  - See @ref doc_charconv for the implementations of
 *    `to_chars()`/`from_chars()` for the fundamental types.
 *  - See @ref doc_substr and @ref sample_substr() for the
 *    many useful utilities in the substring class.
 *
 *
 * ## User container types
 *
 * - See @ref doc_sample_container_types for when the type is a
 *   container (ie, a node which has children, which may themselves be
 *   containers).
 *
 *   - See the sample @ref sample_user_container_types
 *
 *   - See the sample @ref sample_std_types, and also...
 *
 *
 * ## STL types
 *
 * ryml does not use any STL containers internally, but it can be
 * used to serialize and deserialize these containers. See @ref
 * sample_std_types() for an example. See the header @ref
 * ryml_std.hpp and also the headers it includes:
 *
 *  - scalar types:
 *    - for `std::string`: @ref ext/c4core/src/c4/std/string.hpp
 *    - for `std::string_view`: @ref ext/c4core/src/c4/std/string_view.hpp
 *    - for `std::vector<char>`: @ref ext/c4core/src/c4/std/vector.hpp
 *  - container types:
 *    - for `std::vector<T>`: @ref src/c4/yml/std/vector.hpp
 *    - for `std::map<K,V>`: @ref src/c4/yml/std/map.hpp
 *
 * @}
 *
 * @addtogroup doc_quickstart
 * @{ */


//-----------------------------------------------------------------------------
// user scalar types: implemented in ryml through to_chars() + from_chars()

/** @addtogroup doc_sample_helpers
 * @{ */

/** @defgroup doc_sample_scalar_types Serialize/deserialize scalar types
 * @{ */

template<class T> struct vec2 { T x, y; };  ///< example scalar type, serialized and deserialized
template<class T> struct vec3 { T x, y, z; };  ///< example scalar type, serialized and deserialized
template<class T> struct vec4 { T x, y, z, w; };  ///< example scalar type, serialized and deserialized

template<class T> struct parse_only_vec2 { T x, y; }; ///< example scalar type, deserialized only
template<class T> struct parse_only_vec3 { T x, y, z; }; ///< example scalar type, deserialized only
template<class T> struct parse_only_vec4 { T x, y, z, w; }; ///< example scalar type, deserialized only

template<class T> struct emit_only_vec2 { T x, y; }; ///< example scalar type, serialized only
template<class T> struct emit_only_vec3 { T x, y, z; }; ///< example scalar type, serialized only
template<class T> struct emit_only_vec4 { T x, y, z, w; }; ///< example scalar type, serialized only

/** @defgroup doc_sample_to_chars_scalar Define to_chars to write scalar types
 *
 * @brief To serialize user scalar types, implement the appropriate
 * function to_chars (see also @ref doc_to_chars):
 *
 * ```cpp
 * // any of these can be used:
 * size_t to_chars(substr buf, T const& v);
 * size_t to_chars(substr buf, T v); // this also works, and is good when the type is small
 * ```
 *
 * See the sample @ref sample_user_scalar_types() for an example usage.
 *
 * Your implementation of to_chars must format v to the given string
 * view + return the number of characters written into it. The view
 * size (buf.len) must be strictly respected. Return the number of
 * characters that need to be written for the value to be completely
 * serialized in the string. So if the return value is larger than
 * buf.len, ryml will know that the buffer resize the buffer and call
 * this again with a larger buffer of the correct size.
 *
 * In your implementation, you may be interested in using the
 * formatting facilities in @ref doc_format_utils and @ref doc_charconv;
 * refer to their documentation for further details. But this is not
 * mandatory, and anything can be used, provided that the implemented
 * `to_chars()` fulfills its contract, described above.
 *
 * @warning Because of [C++'s ADL
 * rules](http://en.cppreference.com/w/cpp/language/adl), **it is
 * required to overload these functions in the namespace of the type**
 * you're serializing (or in the c4 namespace, or in the c4::yml
 * namespace). [Here's an example of an issue where failing to do this
 * was causing problems in some
 * platforms](https://github.com/biojppm/rapidyaml/issues/424)
 *
 * @note Please take note of the following pitfall when using
 * serialization functions: you may have to include the header with
 * your `to_chars()` implementation before any other headers that use
 * functions from it. See the include order at the top of this source
 * file. This constraint also applies to the conversion functions for
 * your types; just like with the STL's headers, they should be
 * included prior to ryml's headers. Lately, some effort was directed
 * to provide forward declarations to alleviate this problem, but it
 * may still occur.
 *
 * @see string.hpp
 * @see string_view.hpp
 * @{
 */
template<class T> size_t to_chars(ryml::substr buf, vec2<T> v) { return ryml::format(buf, "({},{})", v.x, v.y); }
template<class T> size_t to_chars(ryml::substr buf, vec3<T> v) { return ryml::format(buf, "({},{},{})", v.x, v.y, v.z); }
template<class T> size_t to_chars(ryml::substr buf, vec4<T> v) { return ryml::format(buf, "({},{},{},{})", v.x, v.y, v.z, v.w); }

template<class T> size_t to_chars(ryml::substr buf, emit_only_vec2<T> v) { return ryml::format(buf, "({},{})", v.x, v.y); }
template<class T> size_t to_chars(ryml::substr buf, emit_only_vec3<T> v) { return ryml::format(buf, "({},{},{})", v.x, v.y, v.z); }
template<class T> size_t to_chars(ryml::substr buf, emit_only_vec4<T> v) { return ryml::format(buf, "({},{},{},{})", v.x, v.y, v.z, v.w); }
/** @} */


/** @defgroup doc_sample_from_chars_scalar Define from_chars to read scalar types
 *
 * @brief To deserialize user scalar types, implement the
 * function `bool from_chars(csubstr buf, T *val)`; see @ref
 * doc_from_chars.
 *
 * The implementation of from_chars must never read beyond the limit
 * of the given buffer, and must return true/false to indicate
 * success/failure in the deserialization. On failure, it is up to you
 * whether the value is left unchanged; ryml itself does not care
 * about the value when the deserialization failed.
 *
 * In your implementation, you may be interested in using the
 * reading facilities in @ref doc_format_utils and @ref doc_charconv;
 * refer to their documentation for further details. But this is not
 * mandatory, and anything can be used, provided that the implemented
 * from_chars fulfills its contract, described above.
 *
 * @warning Because of [C++'s ADL
 * rules](http://en.cppreference.com/w/cpp/language/adl), **it is
 * required to overload these functions in the namespace of the type**
 * you're serializing (or in the c4 namespace, or in the c4::yml
 * namespace). [Here's an example of an issue where failing to do this
 * was causing problems in some
 * platforms](https://github.com/biojppm/rapidyaml/issues/424)
 *
 * @note Please take note of the following pitfall when using
 * serialization functions: you may have to include the header with
 * your `from_chars()` implementation before any other headers that use
 * functions from it. See the include order at the top of this source
 * file. This constraint also applies to the conversion functions for
 * your types; just like with the STL's headers, they should be
 * included prior to ryml's headers. Lately, some effort was directed
 * to provide forward declarations to alleviate this problem, but it
 * may still occur.
 *
 * @{
 */
template<class T> bool from_chars(ryml::csubstr buf, vec2<T> *v) { size_t ret = ryml::unformat(buf, "({},{})", v->x, v->y); return ret != ryml::yml::npos; }
template<class T> bool from_chars(ryml::csubstr buf, vec3<T> *v) { size_t ret = ryml::unformat(buf, "({},{},{})", v->x, v->y, v->z); return ret != ryml::yml::npos; }
template<class T> bool from_chars(ryml::csubstr buf, vec4<T> *v) { size_t ret = ryml::unformat(buf, "({},{},{},{})", v->x, v->y, v->z, v->w); return ret != ryml::yml::npos; }

template<class T> bool from_chars(ryml::csubstr buf, parse_only_vec2<T> *v) { size_t ret = ryml::unformat(buf, "({},{})", v->x, v->y); return ret != ryml::yml::npos; }
template<class T> bool from_chars(ryml::csubstr buf, parse_only_vec3<T> *v) { size_t ret = ryml::unformat(buf, "({},{},{})", v->x, v->y, v->z); return ret != ryml::yml::npos; }
template<class T> bool from_chars(ryml::csubstr buf, parse_only_vec4<T> *v) { size_t ret = ryml::unformat(buf, "({},{},{},{})", v->x, v->y, v->z, v->w); return ret != ryml::yml::npos; }
/** @} */ // doc_sample_from_chars_scalar

/** @} */ // doc_sample_scalar_types
/** @} */ // doc_sample_helpers


/** to add scalar types (ie leaf types converting to/from string),
 * define the functions above for those types. See @ref
 * doc_sample_scalar_types. */
void sample_user_scalar_types()
{
    ryml::Tree t;

    auto r = t.rootref();
    r |= ryml::MAP;

    vec2<int> v2in{10, 11};
    vec2<int> v2out{1, 2};
    r["v2"] << v2in; // serializes to the tree's arena, and then sets the keyval
    r["v2"] >> v2out;
    CHECK(v2in.x == v2out.x);
    CHECK(v2in.y == v2out.y);
    vec3<int> v3in{100, 101, 102};
    vec3<int> v3out{1, 2, 3};
    r["v3"] << v3in; // serializes to the tree's arena, and then sets the keyval
    r["v3"] >> v3out;
    CHECK(v3in.x == v3out.x);
    CHECK(v3in.y == v3out.y);
    CHECK(v3in.z == v3out.z);
    vec4<int> v4in{1000, 1001, 1002, 1003};
    vec4<int> v4out{1, 2, 3, 4};
    r["v4"] << v4in; // serializes to the tree's arena, and then sets the keyval
    r["v4"] >> v4out;
    CHECK(v4in.x == v4out.x);
    CHECK(v4in.y == v4out.y);
    CHECK(v4in.z == v4out.z);
    CHECK(v4in.w == v4out.w);
    CHECK(ryml::emitrs_yaml<std::string>(t) == R"(v2: '(10,11)'
v3: '(100,101,102)'
v4: '(1000,1001,1002,1003)'
)");

    // note that only the used functions are needed:
    //   - if a type is only parsed, then only from_chars() is needed
    //   - if a type is only emitted, then only to_chars() is needed
    emit_only_vec2<int> eov2in{20, 21}; // only has to_chars()
    parse_only_vec2<int> pov2out{1, 2}; // only has from_chars()
    r["v2"] << eov2in; // serializes to the tree's arena, and then sets the keyval
    r["v2"] >> pov2out;
    CHECK(eov2in.x == pov2out.x);
    CHECK(eov2in.y == pov2out.y);
    emit_only_vec3<int> eov3in{30, 31, 32}; // only has to_chars()
    parse_only_vec3<int> pov3out{1, 2, 3}; // only has from_chars()
    r["v3"] << eov3in; // serializes to the tree's arena, and then sets the keyval
    r["v3"] >> pov3out;
    CHECK(eov3in.x == pov3out.x);
    CHECK(eov3in.y == pov3out.y);
    CHECK(eov3in.z == pov3out.z);
    emit_only_vec4<int> eov4in{40, 41, 42, 43}; // only has to_chars()
    parse_only_vec4<int> pov4out{1, 2, 3, 4}; // only has from_chars()
    r["v4"] << eov4in; // serializes to the tree's arena, and then sets the keyval
    r["v4"] >> pov4out;
    CHECK(eov4in.x == pov4out.x);
    CHECK(eov4in.y == pov4out.y);
    CHECK(eov4in.z == pov4out.z);
    CHECK(ryml::emitrs_yaml<std::string>(t) == R"(v2: '(20,21)'
v3: '(30,31,32)'
v4: '(40,41,42,43)'
)");
}


//-----------------------------------------------------------------------------
// user container types: implemented in ryml through write() + read()

/** @addtogroup doc_sample_helpers
 * @{ */

/** @defgroup doc_sample_container_types Serialize/deserialize container types
 *
 * To serialize/deserialize container types to a tree, implement the
 * appropriate functions:
 *
 * ```cpp
 * void write(ryml::NodeRef *n, T const& seq);
 * bool read(ryml::ConstNodeRef const& n, T *seq);
 * ```
 *
 * @warning Because of [C++'s ADL
 * rules](http://en.cppreference.com/w/cpp/language/adl), **it is
 * required to overload these functions in the namespace of the type**
 * you're serializing (or in the c4 namespace, or in the c4::yml
 * namespace). [Here's an example of an issue where failing to do this
 * was causing problems in some
 * platforms](https://github.com/biojppm/rapidyaml/issues/424)
 *
 * @note Please take note of the following pitfall when using
 * serialization functions: you may have to include the header with
 * your `write()` or `read()` implementation before any other headers
 * that use functions from it. See the include order at the top of
 * this source file. This constraint also applies to the conversion
 * functions for your types; just like with the STL's headers, they
 * should be included prior to ryml's headers. Lately, some effort was
 * directed to provide forward declarations to alleviate this problem,
 * but it may still occur.
 *
 * @see sample_container_types
 * @see sample_std_types
 *
 * @{ */


/** example user container type: seq-like */
template<class T>
struct my_seq_type
{
    std::vector<T> seq_member;
};
/** example user container type: map-like */
template<class K, class V>
struct my_map_type
{
    std::map<K, V> map_member;
};
/** example user container type with nested container members.
 * notice all the members have user-defined serialization methods. */
struct my_type
{
    // these are leaf nodes:
    vec2<int> v2;
    vec3<int> v3;
    vec4<int> v4;
    // these are container nodes:
    my_seq_type<int> seq;
    my_map_type<int, int> map;
};

template<class T>
void write(ryml::NodeRef *n, my_seq_type<T> const& seq)
{
    *n |= ryml::SEQ;
    for(auto const& v : seq.seq_member)
        n->append_child() << v;
}
template<class K, class V>
void write(ryml::NodeRef *n, my_map_type<K, V> const& map)
{
    *n |= ryml::MAP;
    for(auto const& v : map.map_member)
        n->append_child() << ryml::key(v.first) << v.second;
}
void write(ryml::NodeRef *n, my_type const& val)
{
    *n |= ryml::MAP;
    // these are leaf nodes:
    n->append_child() << ryml::key("v2") << val.v2;
    n->append_child() << ryml::key("v3") << val.v3;
    n->append_child() << ryml::key("v4") << val.v4;
    // these are container nodes:
    n->append_child() << ryml::key("seq") << val.seq;
    n->append_child() << ryml::key("map") << val.map;
}

template<class T>
bool read(ryml::ConstNodeRef const& n, my_seq_type<T> *seq)
{
    seq->seq_member.resize(static_cast<size_t>(n.num_children())); // num_children() is O(N)
    size_t pos = 0;
    for(auto const ch : n.children())
        ch >> seq->seq_member[pos++];
    return true;
}
template<class K, class V>
bool read(ryml::ConstNodeRef const& n, my_map_type<K, V> *map)
{
    K k{};
    V v{};
    for(auto const ch : n)
    {
        ch >> c4::yml::key(k) >> v;
        map->map_member.emplace(std::make_pair(std::move(k), std::move(v)));
    }
    return true;
}
bool read(ryml::ConstNodeRef const& n, my_type *val)
{
    // these are leaf nodes:
    n["v2"] >> val->v2;
    n["v3"] >> val->v3;
    n["v4"] >> val->v4;
    // these are container nodes:
    n["seq"] >> val->seq;
    n["map"] >> val->map;
    return true;
}

/** @} */ // doc_sample_container_types

/** @} */ // sample_helpers


/** shows how to serialize/deserialize container types.
 * @see doc_sample_container_types
 * @see sample_std_types
 * */
void sample_user_container_types()
{
    my_type mt_in{
        {20, 21},
        {30, 31, 32},
        {40, 41, 42, 43},
        {{101, 102, 103, 104, 105, 106, 107}},
        {{{1001, 2001}, {1002, 2002}, {1003, 2003}}},
    };
    my_type mt_out;

    ryml::Tree t;
    t.rootref() << mt_in;  // read from this
    t.crootref() >> mt_out; // assign here
    CHECK(mt_out.v2.x == mt_in.v2.x);
    CHECK(mt_out.v2.y == mt_in.v2.y);
    CHECK(mt_out.v3.x == mt_in.v3.x);
    CHECK(mt_out.v3.y == mt_in.v3.y);
    CHECK(mt_out.v3.z == mt_in.v3.z);
    CHECK(mt_out.v4.x == mt_in.v4.x);
    CHECK(mt_out.v4.y == mt_in.v4.y);
    CHECK(mt_out.v4.z == mt_in.v4.z);
    CHECK(mt_out.v4.w == mt_in.v4.w);
    CHECK(mt_in.seq.seq_member.size() > 0);
    CHECK(mt_out.seq.seq_member.size() == mt_in.seq.seq_member.size());
    for(size_t i = 0; i < mt_in.seq.seq_member.size(); ++i)
    {
        CHECK(mt_out.seq.seq_member[i] == mt_in.seq.seq_member[i]);
    }
    CHECK(mt_in.map.map_member.size() > 0);
    CHECK(mt_out.map.map_member.size() == mt_in.map.map_member.size());
    for(auto const& kv : mt_in.map.map_member)
    {
        CHECK(mt_out.map.map_member.find(kv.first) != mt_out.map.map_member.end());
        CHECK(mt_out.map.map_member[kv.first] == kv.second);
    }
    CHECK(ryml::emitrs_yaml<std::string>(t) == R"(v2: '(20,21)'
v3: '(30,31,32)'
v4: '(40,41,42,43)'
seq:
  - 101
  - 102
  - 103
  - 104
  - 105
  - 106
  - 107
map:
  1001: 2001
  1002: 2002
  1003: 2003
)");
}


//-----------------------------------------------------------------------------

/** demonstrates usage with the std implementations provided by ryml
    in the ryml_std.hpp header
  @see @ref doc_sample_container_types
  @see also the STL section in @ref doc_serialization */
void sample_std_types()
{
    std::string yml_std_string = R"(- v2: '(20,21)'
  v3: '(30,31,32)'
  v4: '(40,41,42,43)'
  seq:
    - 101
    - 102
    - 103
    - 104
    - 105
    - 106
    - 107
  map:
    1001: 2001
    1002: 2002
    1003: 2003
- v2: '(120,121)'
  v3: '(130,131,132)'
  v4: '(140,141,142,143)'
  seq:
    - 1101
    - 1102
    - 1103
    - 1104
    - 1105
    - 1106
    - 1107
  map:
    11001: 12001
    11002: 12002
    11003: 12003
- v2: '(220,221)'
  v3: '(230,231,232)'
  v4: '(240,241,242,243)'
  seq:
    - 2101
    - 2102
    - 2103
    - 2104
    - 2105
    - 2106
    - 2107
  map:
    21001: 22001
    21002: 22002
    21003: 22003
)";
    // parse in-place using the std::string above
    ryml::Tree tree = ryml::parse_in_place(ryml::to_substr(yml_std_string));
    // my_type is a container-of-containers type. see above its
    // definition implementation for ryml.
    std::vector<my_type> vmt;
    tree.rootref() >> vmt;
    CHECK(vmt.size() == 3);
    ryml::Tree tree_out;
    tree_out.rootref() << vmt;
    CHECK(ryml::emitrs_yaml<std::string>(tree_out) == yml_std_string);
}


//-----------------------------------------------------------------------------

/** control precision of serialized floats */
void sample_float_precision()
{
    std::vector<double> reference{1.23234412342131234, 2.12323123143434237, 3.67847983572591234};
    // A safe precision for comparing doubles. May vary depending on
    // compiler flags. Double goes to about 15 digits, so 14 should be
    // safe enough for this test to succeed.
    const double precision_safe = 1.e-14;
    const size_t num_digits_safe = 14;
    const size_t num_digits_original = 17;
    auto get_num_digits = [](ryml::csubstr number){ return number.sub(2).len; };
    //
    // no significant precision is lost when reading
    // floating point numbers:
    {
        ryml::Tree tree = ryml::parse_in_arena(R"([1.23234412342131234, 2.12323123143434237, 3.67847983572591234])");
        std::vector<double> output;
        tree.rootref() >> output;
        CHECK(output.size() == reference.size());
        for(size_t i = 0; i < reference.size(); ++i)
        {
            CHECK(get_num_digits(tree[(ryml::id_type)i].val()) == num_digits_original);
            CHECK(fabs(output[i] - reference[i]) < precision_safe);
        }
    }
    //
    // However, depending on the compilation settings, there may be a
    // significant precision loss when serializing with the default
    // approach, operator<<(double):
    {
        ryml::Tree serialized;
        serialized.rootref() << reference;
        std::cout << serialized;
        // Without std::to_chars() there is a loss of precision:
        #if (!C4CORE_HAVE_STD_TOCHARS) // This macro is defined when std::to_chars() is available.
        CHECK(ryml::emitrs_yaml<std::string>(serialized) == R"(- 1.23234
- 2.12323
- 3.67848
)" || (bool)"this is indicative; the exact results will vary from platform to platform.");
        C4_UNUSED(num_digits_safe);
        #else  // ... but when using C++17 and above, the results are eminently equal:
        CHECK((ryml::emitrs_yaml<std::string>(serialized) == R"(- 1.2323441234213124
- 2.1232312314343424
- 3.6784798357259123
)") || (bool)"this is indicative; the exact results will vary from platform to platform.");
        size_t pos = 0;
        for(ryml::ConstNodeRef child : serialized.rootref().children())
        {
            CHECK(get_num_digits(child.val()) >= num_digits_safe);
            double out = {};
            child >> out;
            CHECK(fabs(out - reference[pos++]) < precision_safe);
        }
        #endif
    }
    //
    // The difference is explained by the availability of
    // fastfloat::from_chars(), std::from_chars() and std::to_chars().
    //
    // ryml prefers the fastfloat::from_chars() version. Unfortunately
    // fastfloat does not have to_chars() (see
    // https://github.com/fastfloat/fast_float/issues/23).
    //
    // When C++17 is used, ryml uses std::to_chars(), which produces
    // good defaults.
    //
    // However, with earlier standards, or in some library
    // implementations, there's only snprintf() available. Every other
    // std library function will either disrespect the string limits,
    // or more precisely, accept no string size limits. So the
    // implementation of c4core (which ryml uses) falls back to
    // snprintf("%g"), and that picks by default a (low) number of
    // digits.
    //
    // But all is not lost for C++11/C++14 users!
    //
    // To force a particular precision when serializing, you can use
    // c4::fmt::real() (brought into the ryml:: namespace). Or you can
    // serialize the number yourself! The small downside is that you
    // have to build the container.
    //
    // First a function to check the result:
    auto check_precision = [&](ryml::Tree const& serialized){
        std::cout << serialized;
        // now it works!
        CHECK((ryml::emitrs_yaml<std::string>(serialized) == R"(- 1.23234412342131239
- 2.12323123143434245
- 3.67847983572591231
)") || (bool)"this is indicative; the exact results will vary from platform to platform.");
        size_t pos = 0;
        for(ryml::ConstNodeRef child : serialized.rootref().children())
        {
            CHECK(get_num_digits(child.val()) == num_digits_original);
            double out = {};
            child >> out;
            CHECK(fabs(out - reference[pos++]) < precision_safe);
        }
    };
    //
    // Serialization example using fmt::real()
    {
        ryml::Tree serialized;
        ryml::NodeRef root = serialized.rootref();
        root |= ryml::SEQ;
        for(const double v : reference)
            root.append_child() << ryml::fmt::real(v, num_digits_original, ryml::FTOA_FLOAT);
        check_precision(serialized); // OK - now within bounds!
    }
    //
    // Serialization example using snprintf
    {
        ryml::Tree serialized;
        ryml::NodeRef root = serialized.rootref();
        root |= ryml::SEQ;
        char tmp[64];
        for(const double v : reference)
        {
            // reuse a buffer to serialize.
            // add 1 to the significant digits because the %g
            // specifier counts the integral digits.
            (void)snprintf(tmp, sizeof(tmp), "%.18g", v);
            // copy the serialized string to the tree (operator<<
            // copies to the arena, operator= just assigns the string
            // pointer and would be wrong in this case):
            root.append_child() << ryml::to_csubstr((const char*)tmp);
        }
        check_precision(serialized); // OK - now within bounds!
    }
}


//-----------------------------------------------------------------------------

/** demonstrates how to emit to a linear container of char */
void sample_emit_to_container()
{
    // it is possible to emit to any linear container of char.

    ryml::csubstr ymla = "- 1\n- 2\n";
    ryml::csubstr ymlb = R"(- a
- b
- x0: 1
  x1: 2
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
- foo
- bar
- baz
- bat
)";
    const ryml::Tree treea = ryml::parse_in_arena(ymla);
    const ryml::Tree treeb = ryml::parse_in_arena(ymlb);

    // eg, std::vector<char>
    {
        // do a blank call on an empty buffer to find the required size.
        // no overflow will occur, and returns a substr with the size
        // required to output
        ryml::csubstr output = ryml::emit_yaml(treea, treea.root_id(), ryml::substr{}, /*error_on_excess*/false);
        CHECK(output.str == nullptr);
        CHECK(output.len > 0);
        size_t num_needed_chars = output.len;
        std::vector<char> buf(num_needed_chars);
        // now try again with the proper buffer
        output = ryml::emit_yaml(treea, treea.root_id(), ryml::to_substr(buf), /*error_on_excess*/true);
        CHECK(output == ymla);

        // it is possible to reuse the buffer and grow it as needed.
        // first do a blank run to find the size:
        output = ryml::emit_yaml(treeb, treeb.root_id(), ryml::substr{}, /*error_on_excess*/false);
        CHECK(output.str == nullptr);
        CHECK(output.len > 0);
        CHECK(output.len == ymlb.len);
        num_needed_chars = output.len;
        buf.resize(num_needed_chars);
        // now try again with the proper buffer
        output = ryml::emit_yaml(treeb, treeb.root_id(), ryml::to_substr(buf), /*error_on_excess*/true);
        CHECK(output == ymlb);

        // there is a convenience wrapper performing the same as above:
        // provided to_substr() is defined for that container.
        output = ryml::emitrs_yaml(treeb, &buf);
        CHECK(output == ymlb);

        // or you can just output a new container:
        // provided to_substr() is defined for that container.
        std::vector<char> another = ryml::emitrs_yaml<std::vector<char>>(treeb);
        CHECK(ryml::to_csubstr(another) == ymlb);

        // you can also emit nested nodes:
        another = ryml::emitrs_yaml<std::vector<char>>(treeb[3][2]);
        CHECK(ryml::to_csubstr(another) == R"(more:
  vinho verde: Soalheiro
  vinho tinto: Redoma 2017
)");
    }


    // eg, std::string. notice this is the same code as above
    {
        // do a blank call on an empty buffer to find the required size.
        // no overflow will occur, and returns a substr with the size
        // required to output
        ryml::csubstr output = ryml::emit_yaml(treea, treea.root_id(), ryml::substr{}, /*error_on_excess*/false);
        CHECK(output.str == nullptr);
        CHECK(output.len > 0);
        size_t num_needed_chars = output.len;
        std::string buf;
        buf.resize(num_needed_chars);
        // now try again with the proper buffer
        output = ryml::emit_yaml(treea, treea.root_id(), ryml::to_substr(buf), /*error_on_excess*/true);
        CHECK(output == ymla);

        // it is possible to reuse the buffer and grow it as needed.
        // first do a blank run to find the size:
        output = ryml::emit_yaml(treeb, treeb.root_id(), ryml::substr{}, /*error_on_excess*/false);
        CHECK(output.str == nullptr);
        CHECK(output.len > 0);
        CHECK(output.len == ymlb.len);
        num_needed_chars = output.len;
        buf.resize(num_needed_chars);
        // now try again with the proper buffer
        output = ryml::emit_yaml(treeb, treeb.root_id(), ryml::to_substr(buf), /*error_on_excess*/true);
        CHECK(output == ymlb);

        // there is a convenience wrapper performing the above instructions:
        // provided to_substr() is defined for that container
        output = ryml::emitrs_yaml(treeb, &buf);
        CHECK(output == ymlb);

        // or you can just output a new container:
        // provided to_substr() is defined for that container.
        std::string another = ryml::emitrs_yaml<std::string>(treeb);
        CHECK(ryml::to_csubstr(another) == ymlb);

        // you can also emit nested nodes:
        another = ryml::emitrs_yaml<std::string>(treeb[3][2]);
        CHECK(ryml::to_csubstr(another) == R"(more:
  vinho verde: Soalheiro
  vinho tinto: Redoma 2017
)");
    }
}


//-----------------------------------------------------------------------------

/** demonstrates how to emit to a stream-like structure */
void sample_emit_to_stream()
{
    ryml::csubstr ymlb = R"(- a
- b
- x0: 1
  x1: 2
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
- foo
- bar
- baz
- bat
)";
    const ryml::Tree tree = ryml::parse_in_arena(ymlb);

    std::string s;

    // emit a full tree
    {
        std::stringstream ss;
        ss << tree; // works with any stream having .operator<<() and .write()
        s = ss.str();
        CHECK(ryml::to_csubstr(s) == ymlb);
    }

    // emit a full tree as json
    {
        std::stringstream ss;
        ss << ryml::as_json(tree); // works with any stream having .operator<<() and .write()
        s = ss.str();
        CHECK(ryml::to_csubstr(s) == R"([
  "a",
  "b",
  {
    "x0": 1,
    "x1": 2
  },
  {
    "champagne": "Dom Perignon",
    "coffee": "Arabica",
    "more": {
      "vinho verde": "Soalheiro",
      "vinho tinto": "Redoma 2017"
    },
    "beer": [
      "Rochefort 10",
      "Busch",
      "Leffe Rituel"
    ]
  },
  "foo",
  "bar",
  "baz",
  "bat"
]
)");
    }

    // emit a nested node
    {
        std::stringstream ss;
        ss << tree[3][2]; // works with any stream having .operator<<() and .write()
        s = ss.str();
        CHECK(ryml::to_csubstr(s) == R"(more:
  vinho verde: Soalheiro
  vinho tinto: Redoma 2017
)");
    }

    // emit a nested node as json
    {
        std::stringstream ss;
        ss << ryml::as_json(tree[3][2]); // works with any stream having .operator<<() and .write()
        s = ss.str();
        CHECK(ryml::to_csubstr(s) == R"("more": {
  "vinho verde": "Soalheiro",
  "vinho tinto": "Redoma 2017"
}
)");
    }
}


//-----------------------------------------------------------------------------

/** demonstrates how to emit to a FILE* */
void sample_emit_to_file()
{
    ryml::csubstr yml = R"(- a
- b
- x0: 1
  x1: 2
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
- foo
- bar
- baz
- bat
)";
    const ryml::Tree tree = ryml::parse_in_arena(yml);
    // this is emitting to stdout, but of course you can pass in any
    // FILE* obtained from fopen()
    size_t len = ryml::emit_yaml(tree, tree.root_id(), stdout);
    // the return value is the number of characters that were written
    // to the file
    CHECK(len == yml.len);
}


//-----------------------------------------------------------------------------

/** just like parsing into a nested node, you can also emit from a nested node. */
void sample_emit_nested_node()
{
    const ryml::Tree tree = ryml::parse_in_arena(R"(- a
- b
- x0: 1
  x1: 2
- champagne: Dom Perignon
  coffee: Arabica
  more:
    vinho verde: Soalheiro
    vinho tinto: Redoma 2017
  beer:
    - Rochefort 10
    - Busch
    - Leffe Rituel
    - - and so
      - many other
      - wonderful beers
- more
- seq
- members
- here
)");
    CHECK(ryml::emitrs_yaml<std::string>(tree[3]["beer"]) == R"(beer:
  - Rochefort 10
  - Busch
  - Leffe Rituel
  - - and so
    - many other
    - wonderful beers
)");
    CHECK(ryml::emitrs_yaml<std::string>(tree[3]["beer"][0]) == "Rochefort 10");
    CHECK(ryml::emitrs_yaml<std::string>(tree[3]["beer"][3]) == R"(- and so
- many other
- wonderful beers
)");
}


//-----------------------------------------------------------------------------

/** [experimental] query/set/modify node style to control
 * formatting of emitted YAML code. */
void sample_style()
{
    // we will be using this helper throughout this function
    auto tostr = [](ryml::ConstNodeRef n) { return ryml::emitrs_yaml<std::string>(n); };
    // let's parse this yaml:
    ryml::csubstr yaml = R"(block map:
  block key: block val
block seq:
  - block val 1
  - block val 2
  - 'quoted'
flow map, singleline: {flow key: flow val}
flow seq, singleline: [flow val,flow val]
flow map, multiline: {
    flow key: flow val
  }
flow seq, multiline: [
    flow val,
    flow val
  ]
)";
    ryml::Tree tree = ryml::parse_in_arena(yaml);
    // while parsing, ryml marks parsed nodes with their original style:
    CHECK(tree.rootref().is_block());
    CHECK(tree["block map"].is_key_plain());
    CHECK(tree["block seq"].is_key_plain());
    CHECK(tree["flow map, singleline"].is_key_plain());
    CHECK(tree["flow seq, singleline"].is_key_plain());
    CHECK(tree["flow map, multiline"].is_key_plain());
    CHECK(tree["flow seq, multiline"].is_key_plain());
    CHECK(tree["block map"].is_block());
    CHECK(tree["block seq"].is_block());
    // flow is either singleline (FLOW_SL) or multiline (FLOW_ML)
    CHECK(tree["flow map, singleline"].is_flow_sl());
    CHECK(tree["flow seq, singleline"].is_flow_sl());
    CHECK(tree["flow map, multiline"].is_flow_ml());
    CHECK(tree["flow seq, multiline"].is_flow_ml());
    // is_flow() is equivalent to (is_flow_sl() || is_flow_ml())
    CHECK(tree["flow map, singleline"].is_flow());
    CHECK(tree["flow seq, singleline"].is_flow());
    CHECK(tree["flow map, multiline"].is_flow());
    CHECK(tree["flow seq, multiline"].is_flow());
    //
    // since the tree nodes are marked with their original parsed
    // style, emitting the parsed tree will preserve the original
    // style (minus whitespace):
    //
    CHECK(tostr(tree) == yaml); // same as before!
    //
    // you can set/modify the style programatically!
    //
    // here are more examples.
    //
    {
        ryml::NodeRef n = tree["block map"]; // Let's look at one node
        // It looks like this originally:
        CHECK(tostr(n) == "block map:\n  block key: block val\n");
        // let's modify its style:
        n.set_key_style(ryml::KEY_SQUO);      // scalar style: to single-quoted scalar
        n.set_container_style(ryml::FLOW_SL); // container style: to flow singleline
        // now it looks like this:
        CHECK(tostr(n) == "'block map': {block key: block val}\n");
    }
    // next example
    {
        ryml::NodeRef n = tree["block seq"];
        CHECK(tostr(n) == "block seq:\n  - block val 1\n  - block val 2\n  - 'quoted'\n");
        n.set_key_style(ryml::KEY_DQUO);       // scalar style: to double-quoted scalar
        n.set_container_style(ryml::FLOW_ML);  // container style: to flow multiline
        n[2].set_val_style(ryml::VAL_PLAIN);   // scalar style: to plain
        CHECK(tostr(n) == "\"block seq\": [\n    block val 1,\n    block val 2,\n    quoted\n  ]\n");
    }
    // next example
    {
        ryml::NodeRef n = tree["flow map, singleline"];
        CHECK(tostr(n) == "flow map, singleline: {flow key: flow val}\n");
        n.set_container_style(ryml::BLOCK);
        n["flow key"].set_val_style(ryml::VAL_LITERAL);
        CHECK(tostr(n) == "flow map, singleline:\n  flow key: |-\n    flow val\n");
    }
    // next example
    {
        ryml::NodeRef n = tree["flow map, multiline"];
        CHECK(tostr(n) == "flow map, multiline: {\n    flow key: flow val\n  }\n");
        n.set_container_style(ryml::BLOCK);
        CHECK(tostr(n) == "flow map, multiline:\n  flow key: flow val\n");
    }
    // next example
    {
        ryml::NodeRef n = tree["flow seq, singleline"];
        CHECK(tostr(n) == "flow seq, singleline: [flow val,flow val]\n");
        n.set_key_style(ryml::KEY_FOLDED);
        n.set_container_style(ryml::BLOCK);
        n[0].set_val_style(ryml::VAL_SQUO);
        n[1].set_val_style(ryml::VAL_DQUO);
        CHECK(tostr(n) == "? >-\n  flow seq, singleline\n:\n  - 'flow val'\n  - \"flow val\"\n");
    }
    // next example
    {
        ryml::NodeRef n = tree["flow seq, multiline"];
        CHECK(tostr(n) == "flow seq, multiline: [\n    flow val,\n    flow val\n  ]\n");
        n.set_container_style(ryml::FLOW_SL);
        CHECK(tostr(n) == "flow seq, multiline: [flow val,flow val]\n");
    }
    // note the full tree now:
    CHECK(tostr(tree) != yaml);
    CHECK(tostr(tree) ==
          R"('block map': {block key: block val}
"block seq": [
    block val 1,
    block val 2,
    quoted
  ]
flow map, singleline:
  flow key: |-
    flow val
? >-
  flow seq, singleline
:
  - 'flow val'
  - "flow val"
flow map, multiline:
  flow key: flow val
flow seq, multiline: [flow val,flow val]
)");
    // you can clear the style of single nodes:
    tree["block map"].clear_style();
    tree["block seq"].clear_style();
    CHECK(tostr(tree) ==
          R"(block map:
  block key: block val
block seq:
  - block val 1
  - block val 2
  - quoted
flow map, singleline:
  flow key: |-
    flow val
? >-
  flow seq, singleline
:
  - 'flow val'
  - "flow val"
flow map, multiline:
  flow key: flow val
flow seq, multiline: [flow val,flow val]
)");
    // you can clear the style recursively:
    tree.rootref().clear_style(/*recurse*/true);
    // when emitting nodes which have no style set, ryml will default
    // to block format for containers, and call
    // ryml::scalar_style_choose() to pick the style for each scalar
    // (at the cost of a scan over each scalar). Note that ryml picks
    // single-quoted for scalars containing commas:
    CHECK(tostr(tree) ==
          R"(block map:
  block key: block val
block seq:
  - block val 1
  - block val 2
  - quoted
'flow map, singleline':
  flow key: flow val
'flow seq, singleline':
  - flow val
  - flow val
'flow map, multiline':
  flow key: flow val
'flow seq, multiline':
  - flow val
  - flow val
)");
    // you can set the style based on type conditions:
    //
    // eg, set a single key to single-quoted
    tree["block map"].set_style_conditionally(/*type_mask*/ryml::KEY,
                                              /*remflags*/ryml::KEY_STYLE,
                                              /*addflags*/ryml::KEY_SQUO,
                                              /*recurse*/false);
    CHECK(tostr(tree) ==
          R"('block map':
  block key: block val
block seq:
  - block val 1
  - block val 2
  - quoted
'flow map, singleline':
  flow key: flow val
'flow seq, singleline':
  - flow val
  - flow val
'flow map, multiline':
  flow key: flow val
'flow seq, multiline':
  - flow val
  - flow val
)");
    // change all keys to single-quoted:
    tree.rootref().set_style_conditionally(/*type_mask*/ryml::KEY,
                                           /*remflags*/ryml::KEY_STYLE,
                                           /*addflags*/ryml::KEY_SQUO,
                                           /*recurse*/true);
    // change all vals to double-quoted
    tree.rootref().set_style_conditionally(/*type_mask*/ryml::VAL,
                                           /*remflags*/ryml::VAL_STYLE,
                                           /*addflags*/ryml::VAL_DQUO,
                                           /*recurse*/true);
    // change all seqs to flow
    tree.rootref().set_style_conditionally(/*type_mask*/ryml::SEQ,
                                           /*remflags*/ryml::CONTAINER_STYLE,
                                           /*addflags*/ryml::FLOW_SL,
                                           /*recurse*/true);
    // change all maps to flow
    tree.rootref().set_style_conditionally(/*type_mask*/ryml::MAP,
                                           /*remflags*/ryml::CONTAINER_STYLE,
                                           /*addflags*/ryml::BLOCK,
                                           /*recurse*/true);
    // done!
    CHECK(tostr(tree) ==
          R"('block map':
  'block key': "block val"
'block seq': ["block val 1","block val 2","quoted"]
'flow map, singleline':
  'flow key': "flow val"
'flow seq, singleline': ["flow val","flow val"]
'flow map, multiline':
  'flow key': "flow val"
'flow seq, multiline': ["flow val","flow val"]
)");
    // you can also set a conditional style in a single node (or its branch if recurse is true):
    tree["flow seq, singleline"].set_style_conditionally(/*type_mask*/ryml::SEQ,
                                                         /*remflags*/ryml::CONTAINER_STYLE,
                                                         /*addflags*/ryml::BLOCK,
                                                         /*recurse*/false);
    CHECK(tostr(tree) ==
          R"('block map':
  'block key': "block val"
'block seq': ["block val 1","block val 2","quoted"]
'flow map, singleline':
  'flow key': "flow val"
'flow seq, singleline':
  - "flow val"
  - "flow val"
'flow map, multiline':
  'flow key': "flow val"
'flow seq, multiline': ["flow val","flow val"]
)");
    // see also:
    //  - ryml::scalar_style_choose()
    //  - ryml::scalar_style_json_choose()
    //  - ryml::scalar_style_query_squo()
    //  - ryml::scalar_style_query_plain()
}


//-----------------------------------------------------------------------------

/** [experimental] control the indentation of emitted FLOW_ML containers */
void sample_style_flow_ml_indent()
{
    // we will be using this helper throughout this function
    auto tostr = [](ryml::ConstNodeRef n, ryml::EmitOptions opts) {
        return ryml::emitrs_yaml<std::string>(n, opts);
    };
    ryml::csubstr yaml = "{map: {seq: [0, 1, 2, 3, [40, 41]]}}";
    ryml::Tree tree = ryml::parse_in_arena(yaml);
    ryml::EmitOptions defaults = {};
    ryml::EmitOptions noindent = ryml::EmitOptions{}.indent_flow_ml(false);
    CHECK(tostr(tree, defaults) == "{map: {seq: [0,1,2,3,[40,41]]}}");
    // let's now set the style to FLOW_ML (it was FLOW_SL)
    tree.rootref().set_container_style(ryml::FLOW_ML);
    tree["map"].set_container_style(ryml::FLOW_ML);
    tree["map"]["seq"].set_container_style(ryml::FLOW_ML);
    tree["map"]["seq"][4].set_container_style(ryml::FLOW_ML);
    // by default FLOW_ML prints one value per line, indented:
    CHECK(tostr(tree, defaults) ==
          R"({
  map: {
    seq: [
      0,
      1,
      2,
      3,
      [
        40,
        41
      ]
    ]
  }
}
)");
    // if we use the noindent options, then each value is put at the
    // beginning of the line
    CHECK(tostr(tree, noindent) ==
          R"({
map: {
seq: [
0,
1,
2,
3,
[
40,
41
]
]
}
}
)");
    // Note that the noindent option will safely respect any prior
    // indent level from enclosing block containers! For example:
    tree.rootref().set_container_style(ryml::BLOCK);
    CHECK(tostr(tree, noindent) == // notice it is indented at the map level
          R"(map: {
  seq: [
  0,
  1,
  2,
  3,
  [
  40,
  41
  ]
  ]
  }
)");
    // Let's set it one BLOCK level further:
    tree["map"].set_container_style(ryml::BLOCK);
    CHECK(tostr(tree, noindent) == // notice it is indented one more level
          R"(map:
  seq: [
    0,
    1,
    2,
    3,
    [
    40,
    41
    ]
    ]
)");
}


//-----------------------------------------------------------------------------

/** [experimental] set the parser to pick FLOW_SL even if the
 * container being parsed is FLOW_ML */
void sample_style_flow_ml_filter()
{
    ryml::csubstr yaml = R"({
  map: {
    seq: [
      0,
      1,
      2,
      3,
      [
        40,
        41
      ]
    ]
  }
}
)";
    ryml::csubstr yaml_not_indented = R"({
map: {
seq: [
0,
1,
2,
3,
[
40,
41
]
]
}
}
)";
    // note that the parser defaults to detect multiline flow
    // (FLOW_ML) containers:
    {
        const ryml::Tree tree = ryml::parse_in_arena(yaml);
        CHECK(tree["map"].is_flow_ml()); // etc
        // emitted yaml is exactly equal to parsed yaml:
        CHECK(ryml::emitrs_yaml<std::string>(tree) == yaml);
    }
    // if you prefer to shorten the emitted yaml, you can set the
    // parser to set singleline flow (FLOW_SL) on all flow containers:
    {
        const ryml::ParserOptions opts = ryml::ParserOptions{}.detect_flow_ml(false);
        const ryml::Tree tree = ryml::parse_in_arena(yaml, opts);
        CHECK(tree["map"].is_flow_sl()); // etc
        // notice how this is smaller now:
        CHECK(ryml::emitrs_yaml<std::string>(tree) ==
              R"({map: {seq: [0,1,2,3,[40,41]]}})");
    }
    // you can also keep FLOW_ML, but control its indentation:
    // (see more details in @ref sample_style_flow_ml_indent())
    {
        const ryml::EmitOptions noindent = ryml::EmitOptions{}.indent_flow_ml(false);
        const ryml::Tree tree = ryml::parse_in_arena(yaml);
        CHECK(tree["map"].is_flow_ml()); // etc
        CHECK(ryml::emitrs_yaml<std::string>(tree, noindent) == yaml_not_indented);
    }
}


//-----------------------------------------------------------------------------

/** shows how to parse and emit JSON.
 *
 * To emit YAML parsed from JSON, see also @ref sample_style() for
 * info on clearing the style flags (example below). */
void sample_json()
{
    ryml::csubstr json = R"({
  "doe": "a deer, a female deer",
  "ray": "a drop of golden sun",
  "me": "a name, I call myself",
  "far": "a long long way to go"
}
)";
    // Since JSON is a subset of YAML, parsing JSON is just the
    // same as YAML:
    ryml::Tree tree = ryml::parse_in_arena(json);
    // If you are sure the source is valid json, you can use the
    // appropriate parse_json overload, which is faster because json
    // has a smaller grammar:
    ryml::Tree json_tree = ryml::parse_json_in_arena(json);
    // to emit JSON:
    CHECK(ryml::emitrs_json<std::string>(tree) == json);
    CHECK(ryml::emitrs_json<std::string>(json_tree) == json);
    // to emit JSON to a stream:
    std::stringstream ss;
    ss << ryml::as_json(tree);  // <- mark it like this
    CHECK(ss.str() == json);
    // Note the following limitations:
    //
    // - YAML streams cannot be emitted as json, and are not
    //   allowed. But you can work around this by emitting the
    //   individual documents separately; see the sample_docs()
    //   below for such an example.
    //
    // - tags cannot be emitted as json, and are not allowed.
    //
    // - anchors and references cannot be emitted as json and
    //   are not allowed.
    //

    // Note that when parsing JSON, ryml will the style of each node
    // in the JSON. This means that if you emit as YAML it will look
    // mostly the same as the JSON:
    std::cout << ryml::emitrs_yaml<std::string>(json_tree);
    CHECK(ryml::emitrs_yaml<std::string>(json_tree) == json);
    // If you want to avoid this, you will need to clear the style.
    json_tree.rootref().clear_style(); // clear the style of the map, but do not recurse
    // note that this is now block mode. That is because when no
    // style is set, the YAML emit function will default to block mode.
    CHECK(ryml::emitrs_yaml<std::string>(json_tree) ==
          R"("doe": "a deer, a female deer"
"ray": "a drop of golden sun"
"me": "a name, I call myself"
"far": "a long long way to go"
)");
    // if you don't want the double quotes in the scalar, you can
    // recurse:
    json_tree.rootref().clear_style(/*recurse*/true);
    // so now when emitting you will get this:
    // (the scalars with a comma are single-quote)
    CHECK(ryml::emitrs_yaml<std::string>(json_tree) ==
          R"(doe: 'a deer, a female deer'
ray: a drop of golden sun
me: 'a name, I call myself'
far: a long long way to go
)");
    // you can do custom style changes based on a type mask. this
    // will change set the style of all scalar values to single-quoted
    json_tree.rootref().set_style_conditionally(ryml::VAL,
                                                /*remflags*/ryml::VAL_STYLE,
                                                /*addflags*/ryml::VAL_SQUO,
                                                /*recurse*/true);
    CHECK(ryml::emitrs_yaml<std::string>(json_tree) ==
          R"(doe: 'a deer, a female deer'
ray: 'a drop of golden sun'
me: 'a name, I call myself'
far: 'a long long way to go'
)");
    // see in particular sample_style() for more examples
}


//-----------------------------------------------------------------------------

/** demonstrates usage with anchors and alias references.

Note that dereferencing is opt-in; after parsing, you have to call
@ref c4::yml::Tree::resolve() explicitly if you want resolved
references in the tree. This method will resolve all references and
substitute the anchored values in place of the reference.

The @ref c4::yml::Tree::resolve() method first does a full traversal
of the tree to gather all anchors and references in a separate
collection, then it goes through that collection to locate the names,
which it does by obeying the YAML standard diktat that

    > an alias node refers to the most recent node in
    > the serialization having the specified anchor

So, depending on the number of anchor/alias nodes, this is a
potentially expensive operation, with a best-case linear complexity
(from the initial traversal) and a worst-case quadratic complexity (if
every node has an alias/anchor). This potential cost is the reason for
requiring an explicit call to @ref c4::yml::Tree::resolve() */
void sample_anchors_and_aliases()
{
    std::string unresolved = R"(base: &base
  name: Everyone has same name
foo: &foo
  <<: *base
  age: 10
bar: &bar
  <<: *base
  age: 20
bill_to: &id001
  street: |-
    123 Tornado Alley
    Suite 16
  city: East Centerville
  state: KS
ship_to: *id001
&keyref key: &valref val
*valref : *keyref
)";
    std::string resolved = R"(base:
  name: Everyone has same name
foo:
  name: Everyone has same name
  age: 10
bar:
  name: Everyone has same name
  age: 20
bill_to:
  street: |-
    123 Tornado Alley
    Suite 16
  city: East Centerville
  state: KS
ship_to:
  street: |-
    123 Tornado Alley
    Suite 16
  city: East Centerville
  state: KS
key: val
val: key
)";

    ryml::Tree tree = ryml::parse_in_arena(ryml::to_csubstr(unresolved));
    // by default, references are not resolved when parsing:
    CHECK( ! tree["base"].has_key_anchor());
    CHECK(   tree["base"].has_val_anchor());
    CHECK(   tree["base"].val_anchor() == "base");
    CHECK(   tree["key"].key_anchor() == "keyref");
    CHECK(   tree["key"].val_anchor() == "valref");
    CHECK(   tree["*valref"].is_key_ref());
    CHECK(   tree["*valref"].is_val_ref());
    CHECK(   tree["*valref"].key_ref() == "valref");
    CHECK(   tree["*valref"].val_ref() == "keyref");

    // to resolve references, simply call tree.resolve(),
    // which will perform the reference instantiations:
    tree.resolve();

    // all the anchors and references are substistuted and then removed:
    CHECK( ! tree["base"].has_key_anchor());
    CHECK( ! tree["base"].has_val_anchor());
    CHECK( ! tree["base"].has_val_anchor());
    CHECK( ! tree["key"].has_key_anchor());
    CHECK( ! tree["key"].has_val_anchor());
    CHECK( ! tree["val"].is_key_ref()); // notice *valref is now turned to val
    CHECK( ! tree["val"].is_val_ref()); // notice *valref is now turned to val

    CHECK(tree["ship_to"]["city"].val() == "East Centerville");
    CHECK(tree["ship_to"]["state"].val() == "KS");
}

/** demonstrates how to use the API to programatically create anchors
 * and aliases */
void sample_anchors_and_aliases_create()
{
    // part 1: anchor/ref
    {
        ryml::Tree t;
        t.rootref() |= ryml::MAP|ryml::BLOCK;
        t["kanchor"] = "2";
        t["kanchor"].set_key_anchor("kanchor");
        t["vanchor"] = "3";
        t["vanchor"].set_val_anchor("vanchor");
        // to set a reference, need to call .set_val_ref()/.set_key_ref()
        t["kref"].set_val_ref("kanchor");
        t["vref"].set_val_ref("vanchor");
        t["nref"] = "*vanchor";  // NOTE: this is not set as a reference in the tree!
        CHECK(ryml::emitrs_yaml<std::string>(t) == R"(&kanchor kanchor: 2
vanchor: &vanchor 3
kref: *kanchor
vref: *vanchor
nref: '*vanchor'
)"); // note that ryml emits nref with quotes to disambiguate (because no style was set)
        t.resolve();
        CHECK(ryml::emitrs_yaml<std::string>(t) == R"(kanchor: 2
vanchor: 3
kref: kanchor
vref: 3
nref: '*vanchor'
)"); // note that nref was not resolved
    }

    // part 2: simple inheritance (ie, adding `<<: *anchor` nodes)
    {
        ryml::Tree t = ryml::parse_in_arena(R"(
orig: &orig {foo: bar, baz: bat}
copy: {}
notcopy: {}
notref: {}
)");
        t["copy"]["<<"].set_val_ref("orig");
        t["notcopy"]["test"].set_val_ref("orig");
        t["notcopy"]["<<"].set_val_ref("orig");
        t["notref"]["<<"] = "*orig"; // not a reference! .set_val_ref() was not called
        CHECK(ryml::emitrs_yaml<std::string>(t) == R"(orig: &orig {foo: bar,baz: bat}
copy: {<<: *orig}
notcopy: {test: *orig,<<: *orig}
notref: {<<: '*orig'}
)");
        t.resolve();
        CHECK(ryml::emitrs_yaml<std::string>(t) == R"(orig: {foo: bar,baz: bat}
copy: {foo: bar,baz: bat}
notcopy: {test: {foo: bar,baz: bat},foo: bar,baz: bat}
notref: {<<: '*orig'}
)");
    }

    // part 3: multiple inheritance (ie, `<<: [*ref1,*ref2,*etc]`)
    {
        ryml::Tree t = ryml::parse_in_arena(R"(orig1: &orig1 {foo: bar}
orig2: &orig2 {baz: bat}
orig3: &orig3 {and: more}
copy: {}
)");
        ryml::NodeRef seq = t["copy"]["<<"];
        seq |= ryml::SEQ;
        seq.append_child().set_val_ref("orig1");
        seq.append_child().set_val_ref("orig2");
        seq.append_child().set_val_ref("orig3");
        CHECK(ryml::emitrs_yaml<std::string>(t) == R"(orig1: &orig1 {foo: bar}
orig2: &orig2 {baz: bat}
orig3: &orig3 {and: more}
copy: {<<: [*orig1,*orig2,*orig3]}
)");
        t.resolve();
        CHECK(ryml::emitrs_yaml<std::string>(t) == R"(orig1: {foo: bar}
orig2: {baz: bat}
orig3: {and: more}
copy: {foo: bar,baz: bat,and: more}
)");
    }
}


//-----------------------------------------------------------------------------

void sample_tags()
{
    const std::string yaml = R"(--- !!map
a: 0
b: 1
--- !map
a: b
--- !!seq
- a
- b
--- !!str a b
--- !!str 'a: b'
---
!!str a: b
--- !!set
? a
? b
--- !!set
a:
--- !!seq
- !!int 0
- !!str 1
)";
    const ryml::Tree tree = ryml::parse_in_arena(ryml::to_csubstr(yaml));
    const ryml::ConstNodeRef root = tree.rootref();
    CHECK(root.is_stream());
    CHECK(root.num_children() == 9);
    for(ryml::ConstNodeRef doc : root.children())
        CHECK(doc.is_doc());
    // tags are kept verbatim from the source:
    CHECK(root[0].has_val_tag());
    CHECK(root[0].val_tag() == "!!map"); // valid only if the node has a val tag
    CHECK(root[1].val_tag() == "!map");
    CHECK(root[2].val_tag() == "!!seq");
    CHECK(root[3].val_tag() == "!!str");
    CHECK(root[4].val_tag() == "!!str");
    CHECK(root[5]["a"].has_key_tag());
    CHECK(root[5]["a"].key_tag() == "!!str"); // valid only if the node has a key tag
    CHECK(root[6].val_tag() == "!!set");
    CHECK(root[7].val_tag() == "!!set");
    CHECK(root[8].val_tag() == "!!seq");
    CHECK(root[8][0].val_tag() == "!!int");
    CHECK(root[8][1].val_tag() == "!!str");
    // ryml also provides a complete toolbox to deal with tags.
    // there is an enumeration for the standard YAML tags:
    CHECK(ryml::to_tag("!map") == ryml::TAG_NONE);
    CHECK(ryml::to_tag("!!map") == ryml::TAG_MAP);
    CHECK(ryml::to_tag("!!seq") == ryml::TAG_SEQ);
    CHECK(ryml::to_tag("!!str") == ryml::TAG_STR);
    CHECK(ryml::to_tag("!!int") == ryml::TAG_INT);
    CHECK(ryml::to_tag("!!set") == ryml::TAG_SET);
    // given a tag enum, you can fetch the short tag string:
    CHECK(ryml::from_tag(ryml::TAG_NONE) == "");
    CHECK(ryml::from_tag(ryml::TAG_MAP) == "!!map");
    CHECK(ryml::from_tag(ryml::TAG_SEQ) == "!!seq");
    CHECK(ryml::from_tag(ryml::TAG_STR) == "!!str");
    CHECK(ryml::from_tag(ryml::TAG_INT) == "!!int");
    CHECK(ryml::from_tag(ryml::TAG_SET) == "!!set");
    // you can also fetch the long tag string:
    CHECK(ryml::from_tag_long(ryml::TAG_NONE) == "");
    CHECK(ryml::from_tag_long(ryml::TAG_MAP) == "<tag:yaml.org,2002:map>");
    CHECK(ryml::from_tag_long(ryml::TAG_SEQ) == "<tag:yaml.org,2002:seq>");
    CHECK(ryml::from_tag_long(ryml::TAG_STR) == "<tag:yaml.org,2002:str>");
    CHECK(ryml::from_tag_long(ryml::TAG_INT) == "<tag:yaml.org,2002:int>");
    CHECK(ryml::from_tag_long(ryml::TAG_SET) == "<tag:yaml.org,2002:set>");
    // and likewise:
    CHECK(ryml::to_tag("!map") == ryml::TAG_NONE);
    CHECK(ryml::to_tag("<tag:yaml.org,2002:map>") == ryml::TAG_MAP);
    CHECK(ryml::to_tag("<tag:yaml.org,2002:seq>") == ryml::TAG_SEQ);
    CHECK(ryml::to_tag("<tag:yaml.org,2002:str>") == ryml::TAG_STR);
    CHECK(ryml::to_tag("<tag:yaml.org,2002:int>") == ryml::TAG_INT);
    CHECK(ryml::to_tag("<tag:yaml.org,2002:set>") == ryml::TAG_SET);
    // to normalize a tag as much as possible, use normalize_tag():
    CHECK(ryml::normalize_tag("!!map") == "!!map");
    CHECK(ryml::normalize_tag("!<tag:yaml.org,2002:map>") == "!!map");
    CHECK(ryml::normalize_tag("<tag:yaml.org,2002:map>") == "!!map");
    CHECK(ryml::normalize_tag("tag:yaml.org,2002:map") == "!!map");
    CHECK(ryml::normalize_tag("!<!!map>") == "<!!map>");
    CHECK(ryml::normalize_tag("!map") == "!map");
    CHECK(ryml::normalize_tag("!my!foo") == "!my!foo");
    // and also for the long form:
    CHECK(ryml::normalize_tag_long("!!map") == "<tag:yaml.org,2002:map>");
    CHECK(ryml::normalize_tag_long("!<tag:yaml.org,2002:map>") == "<tag:yaml.org,2002:map>");
    CHECK(ryml::normalize_tag_long("<tag:yaml.org,2002:map>") == "<tag:yaml.org,2002:map>");
    CHECK(ryml::normalize_tag_long("tag:yaml.org,2002:map") == "<tag:yaml.org,2002:map>");
    CHECK(ryml::normalize_tag_long("!<!!map>") == "<!!map>");
    CHECK(ryml::normalize_tag_long("!map") == "!map");
    // The tree provides the following methods applying to every node
    // with a key and/or val tag:
    ryml::Tree normalized_tree = tree;
    normalized_tree.normalize_tags(); // normalize all tags in short form
    CHECK(ryml::emitrs_yaml<std::string>(normalized_tree) == R"(--- !!map
a: 0
b: 1
--- !map
a: b
--- !!seq
- a
- b
--- !!str a b
--- !!str 'a: b'
---
!!str a: b
--- !!set
a: 
b: 
--- !!set
a: 
--- !!seq
- !!int 0
- !!str 1
)");
    ryml::Tree normalized_tree_long = tree;
    normalized_tree_long.normalize_tags_long(); // normalize all tags in short form
    CHECK(ryml::emitrs_yaml<std::string>(normalized_tree_long) == R"(--- !<tag:yaml.org,2002:map>
a: 0
b: 1
--- !map
a: b
--- !<tag:yaml.org,2002:seq>
- a
- b
--- !<tag:yaml.org,2002:str> a b
--- !<tag:yaml.org,2002:str> 'a: b'
---
!<tag:yaml.org,2002:str> a: b
--- !<tag:yaml.org,2002:set>
a: 
b: 
--- !<tag:yaml.org,2002:set>
a: 
--- !<tag:yaml.org,2002:seq>
- !<tag:yaml.org,2002:int> 0
- !<tag:yaml.org,2002:str> 1
)");
}


//-----------------------------------------------------------------------------

void sample_tag_directives()
{
    const std::string yaml = R"(
%TAG !m! !my-
--- # Bulb here
!m!light fluorescent
...
%TAG !m! !meta-
--- # Color here
!m!light green
)";
    ryml::Tree tree = ryml::parse_in_arena(ryml::to_csubstr(yaml));
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(%TAG !m! !my-
--- !m!light fluorescent
...
%TAG !m! !meta-
--- !m!light green
)");
    // tags are not resolved by default. Use .resolve_tags() to
    // accomplish this:
    tree.resolve_tags();
    CHECK(ryml::emitrs_yaml<std::string>(tree) == R"(%TAG !m! !my-
--- !<!my-light> fluorescent
...
%TAG !m! !meta-
--- !<!meta-light> green
)");
    // see also tree.normalize_tags()
    // see also tree.normalize_tags_long()
}


//-----------------------------------------------------------------------------

void sample_docs()
{
    std::string yml = R"(---
a: 0
b: 1
---
c: 2
d: 3
---
- 4
- 5
- 6
- 7
)";
    ryml::Tree tree = ryml::parse_in_place(ryml::to_substr(yml));
    CHECK(ryml::emitrs_yaml<std::string>(tree) == yml);

    // iteration through docs
    {
        // using the node API
        const ryml::ConstNodeRef stream = tree.rootref();
        CHECK(stream.is_root());
        CHECK(stream.is_stream());
        CHECK(!stream.is_doc());
        CHECK(stream.num_children() == 3);
        for(const ryml::ConstNodeRef doc : stream.children())
            CHECK(doc.is_doc());
        CHECK(tree.docref(0).id() == stream.child(0).id());
        CHECK(tree.docref(1).id() == stream.child(1).id());
        CHECK(tree.docref(2).id() == stream.child(2).id());
        // equivalent: using the lower level index API
        const ryml::id_type stream_id = tree.root_id();
        CHECK(tree.is_root(stream_id));
        CHECK(tree.is_stream(stream_id));
        CHECK(!tree.is_doc(stream_id));
        CHECK(tree.num_children(stream_id) == 3);
        for(ryml::id_type doc_id = tree.first_child(stream_id); doc_id != ryml::NONE; doc_id = tree.next_sibling(stream_id))
            CHECK(tree.is_doc(doc_id));
        CHECK(tree.doc(0) == tree.child(stream_id, 0));
        CHECK(tree.doc(1) == tree.child(stream_id, 1));
        CHECK(tree.doc(2) == tree.child(stream_id, 2));

        // using the node API
        CHECK(stream[0].is_doc());
        CHECK(stream[0].is_map());
        CHECK(stream[0]["a"].val() == "0");
        CHECK(stream[0]["b"].val() == "1");
        // equivalent: using the index API
        const ryml::id_type doc0_id = tree.first_child(stream_id);
        CHECK(tree.is_doc(doc0_id));
        CHECK(tree.is_map(doc0_id));
        CHECK(tree.val(tree.find_child(doc0_id, "a")) == "0");
        CHECK(tree.val(tree.find_child(doc0_id, "b")) == "1");

        // using the node API
        CHECK(stream[1].is_doc());
        CHECK(stream[1].is_map());
        CHECK(stream[1]["c"].val() == "2");
        CHECK(stream[1]["d"].val() == "3");
        // equivalent: using the index API
        const ryml::id_type doc1_id = tree.next_sibling(doc0_id);
        CHECK(tree.is_doc(doc1_id));
        CHECK(tree.is_map(doc1_id));
        CHECK(tree.val(tree.find_child(doc1_id, "c")) == "2");
        CHECK(tree.val(tree.find_child(doc1_id, "d")) == "3");

        // using the node API
        CHECK(stream[2].is_doc());
        CHECK(stream[2].is_seq());
        CHECK(stream[2][0].val() == "4");
        CHECK(stream[2][1].val() == "5");
        CHECK(stream[2][2].val() == "6");
        CHECK(stream[2][3].val() == "7");
        // equivalent: using the index API
        const ryml::id_type doc2_id = tree.next_sibling(doc1_id);
        CHECK(tree.is_doc(doc2_id));
        CHECK(tree.is_seq(doc2_id));
        CHECK(tree.val(tree.child(doc2_id, 0)) == "4");
        CHECK(tree.val(tree.child(doc2_id, 1)) == "5");
        CHECK(tree.val(tree.child(doc2_id, 2)) == "6");
        CHECK(tree.val(tree.child(doc2_id, 3)) == "7");
    }

    // Note: since json does not have streams, you cannot emit the above
    // tree as json when you start from the root:
    //CHECK(ryml::emitrs_json<std::string>(tree) == yml); // RUNTIME ERROR!

    // but, althouth emitting streams as json is not possible,
    // you can iterate through individual documents and emit
    // them separately:
    {
        const std::string expected_json[] = {
            "{\n  \"a\": 0,\n  \"b\": 1\n}\n",
            "{\n  \"c\": 2,\n  \"d\": 3\n}\n",
            "[\n  4,\n  5,\n  6,\n  7\n]\n",
        };
        // using the node API
        {
            ryml::id_type count = 0;
            const ryml::ConstNodeRef stream = tree.rootref();
            CHECK(stream.num_children() == (ryml::id_type)C4_COUNTOF(expected_json));
            for(ryml::ConstNodeRef doc : stream.children())
            {
                CHECK(ryml::emitrs_json<std::string>(doc) == expected_json[count++]);
            }
        }
        // equivalent: using the index API
        {
            ryml::id_type count = 0;
            const ryml::id_type stream_id = tree.root_id();
            CHECK(tree.num_children(stream_id) == (ryml::id_type)C4_COUNTOF(expected_json));
            for(ryml::id_type doc_id = tree.first_child(stream_id); doc_id != ryml::NONE; doc_id = tree.next_sibling(doc_id))
            {
                CHECK(ryml::emitrs_json<std::string>(tree, doc_id) == expected_json[count++]);
            }
        }
    }
}


//-----------------------------------------------------------------------------

// To avoid imposing a particular type of error handling, ryml uses
// error handler callbacks. This enables users to use exceptions, or
// setjmp()/longjmp(), or plain calls to abort(), as they see fit.
//
// However, it is important to note that the error callbacks must never
// return to the caller! Otherwise, an infinite loop or program crash
// will likely occur.
//
// For this reason, to recover from an error when exceptions are disabled,
// then a non-local jump must be performed using setjmp()/longjmp().
// The code below demonstrates both flows.
//
// ryml provides default error handlers, which call
// std::abort(). You can use the cmake option and the macro
// RYML_DEFAULT_CALLBACK_USES_EXCEPTIONS to have the default error
// handler throw an exception instead.

/** demonstrates how to set a custom error handler for ryml */
void sample_error_handler()
{
    ErrorHandlerExample errh; // browse this class to understand more details
    errh.check_disabled();
    // set the global error handlers. Note the error callbacks must
    // never return: they must either throw an exception, use setjmp()
    // and longjmp(), or abort. Otherwise, the parser will enter into
    // an infinite loop, or the program may crash.
    ryml::set_callbacks(errh.callbacks());
    errh.check_enabled();
    CHECK(errh.check_error_occurs([&]{
        ryml::Tree tree = ryml::parse_in_arena("errorhandler.yml", "[a: b\n}");
    }));
    ryml::set_callbacks(errh.defaults); // restore defaults.
    errh.check_disabled();
}


//-----------------------------------------------------------------------------

void sample_error_basic()
{
    auto cause_basic_error = []{
        ryml::TagDirective tag = {};
        return tag.create_from_str("%%%TAG abc");
    };
    {
        ScopedErrorHandlerExample errh; // set the example callbacks (scoped)
        CHECK(errh.check_error_occurs(cause_basic_error));
    }
#ifdef _RYML_WITH_EXCEPTIONS
    bool gotit = false;
    try
    {
        cause_basic_error();
    }
    catch(ryml::ExceptionBasic const& exc)
    {
        gotit = true;
        ryml::csubstr msg = ryml::to_csubstr(exc.what());
        CHECK(!exc.errdata_basic.location.name.empty());
        CHECK(!msg.empty());
    }
    CHECK(gotit);
#endif
}

void sample_error_parse()
{
    ryml::csubstr ymlsrc = R"({
  a: b
   [
)";
    ryml::csubstr ymlfile = "file.yml";
    auto cause_parse_error = [&]{
        return ryml::parse_in_arena(ymlfile, ymlsrc);
    };
    // the YAML in ymlsrc must cause a parse error while it is being
    // parsed. We use our error handler to catch that error, and save
    // the error info:
    ErrorHandlerExample errh;
    {
        ryml::set_callbacks(errh.callbacks());
        CHECK(errh.check_error_occurs(cause_parse_error));
        // the handler in errh saves the error info in itself. Let's
        // use that to see the messages we get.
        //
        // this message is the short message passed into the parse
        // error handler:
        CHECK(errh.saved_msg_short == "invalid character: '['");
        // this message was created inside the handler, by calling
        // ryml::err_parse_format():
        CHECK(ryml::to_csubstr(errh.saved_msg_full).begins_with("file.yml:3: col=4 (12B): ERROR: [parse] invalid character: '['"));
        // If you keep the YAML source buffer around, you can also use
        // it to create/print a larger error message showing the
        // YAML source code context which causes the error:
        std::string msg_ctx = errh.saved_msg_full + "\n";
        ryml::location_format_with_context([&msg_ctx](ryml::csubstr s){
            msg_ctx.append(s.str, s.len);
        }, errh.saved_parse_loc, ymlsrc, "err");
        CHECK(ryml::to_csubstr(msg_ctx).begins_with("file.yml:3: col=4 (12B): ERROR: [parse] invalid character: '['"));
        CHECK(ryml::to_csubstr(msg_ctx).ends_with(R"(file.yml:3: col=4 (12B): err:
err:
err:        [
err:         |
err:         (here)
err:
err: see region:
err:
err:     {
err:       a: b
err:        [
err:         |
err:         (here)
)"));
        //
        // Let's now check the location (see the message above):
        CHECK(errh.saved_parse_loc.name == ymlfile);
        CHECK(errh.saved_parse_loc.line == 3);
        CHECK(errh.saved_parse_loc.col == 4);
        CHECK(errh.saved_parse_loc.offset == 12);
        CHECK(errh.saved_parse_loc.offset <= ymlsrc.len);
        // ... and this is the location in the ryml source code file where
        // this error was found:
        CHECK(!errh.saved_basic_loc.name.empty());
        CHECK(errh.saved_basic_loc.line > 0);
        CHECK(errh.saved_basic_loc.col > 0);
        CHECK(errh.saved_basic_loc.offset > 0);
        ryml::set_callbacks(errh.defaults);
    }
    // A parse error is also a basic error. If no parse error handler
    // is set, then ryml falls back to a basic error:
    {
        ryml::Callbacks cb = errh.callbacks();
        cb.m_error_parse = nullptr;
        ryml::set_callbacks(cb);
        CHECK(ryml::get_callbacks().m_error_parse == nullptr);
        CHECK(errh.check_error_occurs(cause_parse_error));
        // we got a basic error instead of a parse error:
        CHECK(errh.saved_msg_short == "invalid character: '['");
        // notice that the full message now displays this as a basic
        // error:
        CHECK(errh.saved_msg_full == "file.yml:3: col=4 (12B): ERROR: [basic] invalid character: '['");
        // the yml location is now in the location saved from the basic error
        CHECK(errh.saved_basic_loc.name == ymlfile);
        CHECK(errh.saved_basic_loc.line == 3);
        CHECK(errh.saved_basic_loc.col == 4);
        CHECK(errh.saved_basic_loc.offset == 12);
        CHECK(errh.saved_basic_loc.offset <= ymlsrc.len);
        CHECK(errh.saved_parse_loc.name == ryml::csubstr{});
        CHECK(errh.saved_parse_loc.line == ryml::csubstr::npos);
        CHECK(errh.saved_parse_loc.col == ryml::csubstr::npos);
        CHECK(errh.saved_parse_loc.offset == ryml::csubstr::npos);
        ryml::set_callbacks(errh.defaults);
    }
#ifdef _RYML_WITH_EXCEPTIONS
    bool gotit = false;
    try
    {
        cause_parse_error();
    }
    catch(ryml::ExceptionParse const& exc)
    {
        gotit = true;
        ryml::csubstr msg = ryml::to_csubstr(exc.what());
        CHECK(exc.errdata_parse.ymlloc.name == ymlfile);
        CHECK(exc.errdata_parse.ymlloc.line == 3);
        CHECK(exc.errdata_parse.ymlloc.col == 4);
        CHECK(exc.errdata_parse.ymlloc.offset == 12);
        CHECK(exc.errdata_parse.ymlloc.offset <= ymlsrc.len);
        // the message saved in the exception is just the concrete error description:
        CHECK(msg == "invalid character: '['");
        // to print richer error messages, ryml provides helpers to
        // format that description into a complete error message,
        // containing location and source context indication:
        std::string full;
        auto dumpfn = [&full](ryml::csubstr s) { full.append(s.str, s.len); };
        ryml::err_parse_format(dumpfn, msg, exc.errdata_parse);
        full += '\n';
        ryml::location_format_with_context(dumpfn, exc.errdata_parse.ymlloc, ymlsrc, "err", 3);
        CHECK(ryml::to_csubstr(full).begins_with("file.yml:3: col=4 (12B): ERROR: [parse] invalid character: '['"));
        CHECK(ryml::to_csubstr(full).ends_with(R"(file.yml:3: col=4 (12B): err:
err:
err:        [
err:         |
err:         (here)
err:
err: see region:
err:
err:     {
err:       a: b
err:        [
err:         |
err:         (here)
)"));
    }
    CHECK(gotit);
    gotit = false;
    try
    {
        cause_parse_error();
    }
    catch(ryml::ExceptionBasic const& exc) // use references! don't slice the exception
    {
        gotit = true;
        ryml::csubstr msg = ryml::to_csubstr(exc.what());
        CHECK(!exc.errdata_basic.location.name.empty());
        CHECK(!msg.empty());
    }
    CHECK(gotit);
#endif
}


/** Visit errors happen when an error is triggered while reading from
 * a node. */
void sample_error_visit()
{
    ryml::csubstr ymlfile = "file.yml";
    ryml::csubstr ymlsrc = "float: 123.456";
    ErrorHandlerExample errh;
    {
        ryml::set_callbacks(errh.callbacks());
        ryml::Tree tree = ryml::parse_in_arena(ymlfile, ymlsrc);
        CHECK(errh.check_error_occurs([&]{
            int intval = 0;
            tree["float"] >> intval; // cannot deserialize 123.456 to int
        }));
        // the handler in errh saves the error info in itself. Let's
        // use that to see the messages we get.
        //
        // this message is the short message passed into the visit error
        CHECK(errh.saved_msg_short == "could not deserialize value");
        // this message was created inside the handler, by calling
        // ryml::err_visit_format():
        CHECK(ryml::csubstr::npos != ryml::to_csubstr(errh.saved_msg_full).find("ERROR: [visit] could not deserialize value"));
        // The location of the visit error is of the C++ source file where
        // the error was detected -- NOT of the YAML source file:
        CHECK(errh.saved_basic_loc.name != ymlfile);
        // However, note that the tree and node id are available:
        CHECK(errh.saved_visit_tree == &tree);
        CHECK(errh.saved_visit_id == tree["float"].id());
        // see sample_error_visit_location() for an example on how
        // to extract the location.
    }
    // visit errors also fall back to basic errors when the visit
    // handler is not set (similar to the behavior of ExceptionVisit):
    {
        ryml::Callbacks cb = errh.callbacks();
        cb.m_error_visit = nullptr;
        ryml::set_callbacks(cb);
        CHECK(ryml::get_callbacks().m_error_visit == nullptr);
        ryml::Tree tree = ryml::parse_in_arena(ymlfile, ymlsrc);
        CHECK(errh.check_error_occurs([&]{
            int intval = 0;
            tree["float"] >> intval; // cannot deserialize 123.456 to int
        }));
        // we got a basic error instead of a visit error:
        CHECK(errh.saved_msg_short == "could not deserialize value");
        // notice that the full message now displays this as a basic
        // error:
        CHECK(ryml::csubstr::npos != ryml::to_csubstr(errh.saved_msg_full).find("ERROR: [basic] could not deserialize value"));
        // the tree and id are not set, because this was called as a basic error
        CHECK(errh.saved_visit_tree == nullptr);
        CHECK(errh.saved_visit_id == ryml::NONE);
        ryml::set_callbacks(errh.defaults);
    }
#ifdef _RYML_WITH_EXCEPTIONS
    // when using the default ryml callbacks (see
    // RYML_NO_DEFAULT_CALLBACKS), and
    // RYML_DEFAULT_CALLBACK_USES_EXCEPTIONS is defined, the ryml
    // parse handler throws an exception of type ryml::ExceptionVisit,
    // which is derived from ryml::ExceptionBasic.
    {
        const ryml::Tree tree = ryml::parse_in_arena(ymlfile, ymlsrc);
        bool gotit = false;
        try
        {
            int intval = 0;
            tree["float"] >> intval; // cannot deserialize 123.456 to int
        }
        catch(ryml::ExceptionVisit const& exc)
        {
            gotit = true;
            ryml::csubstr msg = ryml::to_csubstr(exc.what());
            CHECK(!exc.errdata_visit.cpploc.name.empty());
            CHECK(exc.errdata_visit.tree == &tree);
            CHECK(exc.errdata_visit.node == tree["float"].id());
            CHECK(!msg.empty());
        }
        CHECK(gotit);
    }
    // you can also catch the exception as its base,
    // ryml::ExceptionBasic:
    {
        const ryml::Tree tree = ryml::parse_in_arena(ymlfile, ymlsrc);
        bool gotit = false;
        try
        {
            int intval = 0;
            tree["float"] >> intval; // cannot deserialize 123.456 to int
        }
        catch(ryml::ExceptionBasic const& exc) // use references! don't slice the exception
        {
            gotit = true;
            ryml::csubstr msg = ryml::to_csubstr(exc.what());
            CHECK(!exc.errdata_basic.location.name.empty());
            CHECK(!msg.empty());
        }
        CHECK(gotit);
    }
#endif
}


/** It is possible to obtain the YAML location from a visit
 * error: when the tree is obtained from parsing YAML, the messages
 * may be enriched by using a parser set to track the locations. See
 * @ref sample_location_tracking() for more details on how to use
 * locations. */
void sample_error_visit_location()
{
    ScopedErrorHandlerExample errh;
    // we will use locations to show the YAML source context of the
    // node where the visit error was triggered. This is a very
    // convenient feature to show detailed messages when deserializing
    // data read from a file (but do note this is opt-in, and it is
    // not mandatory). See sample_location_tracking() for more details
    // on location tracking.
    ryml::ParserOptions opts = ryml::ParserOptions{}.locations(true);
    ryml::EventHandlerTree evt_handler{};
    ryml::Parser parser(&evt_handler, opts);
    ryml::csubstr ymlfile = "file.yml";
    ryml::csubstr ymlsrc = R"(foo: bar
char: a
int: a
float: 123.456
)";
    const ryml::Tree tree = ryml::parse_in_arena(&parser, ymlfile, ymlsrc);
    // This function will cause a visit error when being called:
    auto cause_visit_error = [&]{
        int intval = 0;
        tree["float"] >> intval; // cannot deserialize 123.456 to int
    };
    // Like with the parse error, we will use our error handler to
    // catch that visit error, and save the error info:
    CHECK(evt_handler.callbacks() == errh.callbacks());
    CHECK(parser.callbacks() == errh.callbacks());
    CHECK(tree.callbacks() == errh.callbacks());
    {
        CHECK(errh.check_error_occurs(cause_visit_error));
        // the handler in errh saves the error info in itself. Let's
        // use that to see the messages we get.
        //
        // this message is the short message passed into the visit error
        CHECK(errh.saved_msg_short == "could not deserialize value");
        // this message was created inside the handler, by calling
        // ryml::err_visit_format():
        CHECK(ryml::csubstr::npos != ryml::to_csubstr(errh.saved_msg_full).find("ERROR: [visit] could not deserialize value"));
        // The location of the visit error is of the C++ source file where
        // the error was detected -- NOT of the YAML source file:
        CHECK(errh.saved_basic_loc.name != ymlfile);
        // However, note that the tree and node id are available:
        CHECK(errh.saved_visit_tree == &tree);
        CHECK(errh.saved_visit_id == tree["float"].id());
        // ... which we can use to get the location in the YAML source
        // from the parser (but see @ref sample_location_tracking()):
        ryml::Location ymlloc = errh.saved_visit_tree->location(parser, errh.saved_visit_id);
        CHECK(ymlloc.name == ymlfile);
        // In turn, we can use format_location_context() to
        // print/create an error message pointing at the YAML source
        // code:
        std::string msg = errh.saved_msg_full;
        ryml::location_format_with_context([&msg](ryml::csubstr s){
            msg.append(s.str, s.len);
        }, ymlloc, ymlsrc, "err", /*number of lines to show before the error*/3);
        CHECK(ryml::to_csubstr(msg).ends_with(R"(file.yml:3: col=0 (24B): err:
err:
err:     float: 123.456
err:     |
err:     (here)
err:
err: see region:
err:
err:     foo: bar
err:     char: a
err:     int: a
err:     float: 123.456
err:     |
err:     (here)
)"));
    }
}


//-----------------------------------------------------------------------------

// Please note the following about the use of custom allocators with
// ryml. Due to [the static initialization order
// fiasco](https://en.cppreference.com/w/cpp/language/siof), if you
// use static ryml trees or parsers, you need to make sure that their
// callbacks have the same lifetime. So you can't use ryml's default
// callbacks structure, as it is declared in a ryml file, and the standard
// provides no guarantee on the relative initialization order, such
// that it is constructed before and destroyed after your
// variables (in fact you are pretty much guaranteed to see this
// fail). So please carefully consider your choices, and ponder
// whether you really need to use ryml static trees and parsers. If
// you do need this, then you will need to declare and use a ryml
// callbacks structure that outlives the tree and/or parser.
//
// See also sample_static_trees() for an example on how to use
// trees with static lifetime.

/** @addtogroup doc_sample_helpers
 * @{ */
struct GlobalAllocatorExample
{
    std::vector<char> memory_pool = std::vector<char>(10u * 1024u); // 10KB
    size_t num_allocs = 0, alloc_size = 0, corr_size = 0;
    size_t num_deallocs = 0, dealloc_size = 0;

    void *allocate(size_t len)
    {
        void *ptr = &memory_pool[alloc_size];
        alloc_size += len;
        ++num_allocs;
        // ensure the ptr is aligned
        uintptr_t uptr = (uintptr_t)ptr;
        const uintptr_t align = alignof(max_align_t);
        if (uptr % align)
        {
            uintptr_t prev = uptr - (uptr % align);
            uintptr_t next = prev + align;
            uintptr_t corr = next - uptr;
            ptr = (void*)(((char*)ptr) + corr);
            corr_size += corr;
        }
        C4_CHECK_MSG(alloc_size + corr_size <= memory_pool.size(),
                     "out of memory! requested=%zu+%zu available=%zu\n",
                     alloc_size, corr_size, memory_pool.size());
        return ptr;
    }

    void free(void *mem, size_t len)
    {
        CHECK((char*)mem     >= &memory_pool.front() && (char*)mem     <  &memory_pool.back());
        CHECK((char*)mem+len >= &memory_pool.front() && (char*)mem+len <= &memory_pool.back());
        dealloc_size += len;
        ++num_deallocs;
        // no need to free here
    }

    // bridge
    ryml::Callbacks callbacks()
    {
        ryml::Callbacks cb = {};
        return cb.set_user_data(this)
            .set_allocate(&GlobalAllocatorExample::s_allocate)
            .set_free(&GlobalAllocatorExample::s_free);
    }
    static void* s_allocate(size_t len, void* /*hint*/, void *this_)
    {
        return ((GlobalAllocatorExample*)this_)->allocate(len);
    }
    static void s_free(void *mem, size_t len, void *this_)
    {
        ((GlobalAllocatorExample*)this_)->free(mem, len);
    }

    // checking
    ~GlobalAllocatorExample()
    {
        check_and_reset();
    }
    void check_and_reset()
    {
        std::cout << "size: alloc=" << alloc_size << " dealloc=" << dealloc_size << std::endl;
        std::cout << "count: #allocs=" << num_allocs << " #deallocs=" << num_deallocs << std::endl;
        CHECK(num_allocs == num_deallocs);
        CHECK(alloc_size >= dealloc_size); // failure here means a double free
        CHECK(alloc_size == dealloc_size); // failure here means a leak
        num_allocs = 0;
        num_deallocs = 0;
        alloc_size = 0;
        dealloc_size = 0;
    }
};
/** @} */


/** demonstrates how to set the global allocator for ryml */
void sample_global_allocator()
{
    GlobalAllocatorExample mem;

    // save the existing callbacks for restoring
    ryml::Callbacks defaults = ryml::get_callbacks();

    // set to our callbacks
    ryml::set_callbacks(mem.callbacks());

    // verify that the allocator is in effect
    ryml::Callbacks const& current = ryml::get_callbacks();
    CHECK(current.m_allocate == &mem.s_allocate);
    CHECK(current.m_free == &mem.s_free);

    // so far nothing was allocated
    CHECK(mem.alloc_size == 0);

    // parse one tree and check
    (void)ryml::parse_in_arena(R"({foo: bar})");
    mem.check_and_reset();

    // parse another tree and check
    (void)ryml::parse_in_arena(R"([a, b, c, d, {foo: bar, money: pennys}])");
    mem.check_and_reset();

    // verify that by reserving we save allocations
    {
        ryml::EventHandlerTree evt_handler;
        ryml::Parser parser(&evt_handler); // reuse a parser
        ryml::Tree tree;          // reuse a tree

        tree.reserve(10);         // reserve the number of nodes
        tree.reserve_arena(100);  // reserve the arena size
        parser.reserve_stack(10); // reserve the parser depth.

        // since the parser stack uses Small Storage Optimization,
        // allocations will only happen with capacities higher than 16.
        CHECK(mem.num_allocs == 2); // tree, tree_arena and NOT the parser

        parser.reserve_stack(20); // reserve the parser depth.
        CHECK(mem.num_allocs == 3); // tree, tree_arena and now the parser as well

        // verify that no other allocations occur when parsing
        size_t size_before = mem.alloc_size;
        parse_in_arena(&parser, "", R"([a, b, c, d, {foo: bar, money: pennys}])", &tree);
        CHECK(mem.alloc_size == size_before);
        CHECK(mem.num_allocs == 3);
    }
    mem.check_and_reset();

    // restore defaults.
    ryml::set_callbacks(defaults);
}


//-----------------------------------------------------------------------------

/** @addtogroup doc_sample_helpers
 * @{ */

/** an example for a per-tree memory allocator */
struct PerTreeMemoryExample
{
    std::vector<char> memory_pool = std::vector<char>(10u * 1024u); // 10KB
    size_t num_allocs = 0, alloc_size = 0;
    size_t num_deallocs = 0, dealloc_size = 0;

    ryml::Callbacks callbacks() const
    {
        // Above we used static functions to bridge to our methods.
        // To show a different approach, we employ lambdas here.
        // Note that there can be no captures in the lambdas
        // because these are C-style function pointers.
        ryml::Callbacks cb;
        cb.m_user_data = (void*) this;
        cb.m_allocate = [](size_t len, void *, void *data){ return ((PerTreeMemoryExample*) data)->allocate(len); };
        cb.m_free = [](void *mem, size_t len, void *data){ return ((PerTreeMemoryExample*) data)->free(mem, len); };
        return cb;
    }

    void *allocate(size_t len)
    {
        void *ptr = &memory_pool[alloc_size];
        alloc_size += len;
        ++num_allocs;
        if(C4_UNLIKELY(alloc_size > memory_pool.size()))
        {
            std::cerr << "out of memory! requested=" << alloc_size << " vs " << memory_pool.size() << " available" << std::endl;
            std::abort();
        }
        return ptr;
    }

    void free(void *mem, size_t len)
    {
        CHECK((char*)mem     >= &memory_pool.front() && (char*)mem     <  &memory_pool.back());
        CHECK((char*)mem+len >= &memory_pool.front() && (char*)mem+len <= &memory_pool.back());
        dealloc_size += len;
        ++num_deallocs;
        // no need to free here
    }

    // checking
    ~PerTreeMemoryExample()
    {
        check_and_reset();
    }
    void check_and_reset()
    {
        std::cout << "size: alloc=" << alloc_size << " dealloc=" << dealloc_size << std::endl;
        std::cout << "count: #allocs=" << num_allocs << " #deallocs=" << num_deallocs << std::endl;
        CHECK(num_allocs == num_deallocs);
        CHECK(alloc_size >= dealloc_size); // failure here means a double free
        CHECK(alloc_size == dealloc_size); // failure here means a leak
        num_allocs = 0;
        num_deallocs = 0;
        alloc_size = 0;
        dealloc_size = 0;
    }
};
/** @} */

void sample_per_tree_allocator()
{
    PerTreeMemoryExample mrp;
    PerTreeMemoryExample mr1;
    PerTreeMemoryExample mr2;

    // the trees will use the memory in the resources above,
    // with each tree using a separate resource
    {
        // Watchout: ensure that the lifetime of the callbacks target
        // exceeds the lifetime of the tree.
        ryml::EventHandlerTree evt_handler(mrp.callbacks());
        ryml::Parser parser(&evt_handler);
        ryml::Tree tree1(mr1.callbacks());
        ryml::Tree tree2(mr2.callbacks());

        ryml::csubstr yml1 = "{a: b}";
        ryml::csubstr yml2 = "{c: d, e: f, g: [h, i, 0, 1, 2, 3]}";

        parse_in_arena(&parser, "file1.yml", yml1, &tree1);
        parse_in_arena(&parser, "file2.yml", yml2, &tree2);
    }

    CHECK(mrp.num_allocs == 0); // YAML depth not large enough to warrant a parser allocation
    CHECK(mr1.alloc_size <= mr2.alloc_size); // because yml2 has more nodes
}


//-----------------------------------------------------------------------------


/** shows how to work around the static initialization order fiasco
 * when using a static-duration ryml tree
 * @see https://en.cppreference.com/w/cpp/language/siof */
void sample_static_trees()
{
    // Static trees may incur a static initialization order
    // problem. This happens because a default-constructed tree will
    // obtain the callbacks from the current global setting, which may
    // not have been initialized due to undefined static
    // initialization order:
    //
    // ERROR! depends on ryml::get_callbacks() which may not have been initialized.
    //static ryml::Tree tree;
    //
    // To work around the issue, declare static callbacks
    // to explicitly initialize the static tree:
    static ryml::Callbacks callbacks = {}; // use default callback members
    static ryml::Tree tree(callbacks); // OK
    // now you can use the tree as normal:
    ryml::parse_in_arena(R"(doe: "a deer, a female deer")", &tree);
    CHECK(tree["doe"].val() == "a deer, a female deer");
}


//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------
/** demonstrates how to obtain the (zero-based) location of a node
 * from a recently parsed tree */
void sample_location_tracking()
{
    // NOTE: locations are zero-based. If you intend to show the
    // location to a human user, you may want to pre-increment the line
    // and column by 1.
    ryml::csubstr yaml = R"({
aa: contents,
foo: [one, [two, three]]
})";
    // A parser is needed to track locations, and it has to be
    // explicitly set to do it. Location tracking is disabled by
    // default.
    ryml::ParserOptions opts = {};
    opts.locations(true); // enable locations, default is false
    ryml::EventHandlerTree evt_handler = {};
    ryml::Parser parser(&evt_handler, opts);
    CHECK(parser.options().locations());
    // When locations are enabled, the first task while parsing will
    // consist of building and caching (in the parser) a
    // source-to-node lookup structure to accelerate location lookups.
    //
    // The cost of building the location accelerator is linear in the
    // size of the source buffer. This increased cost is the reason
    // for the opt-in requirement. When locations are disabled there
    // is no cost.
    //
    // Building the location accelerator may trigger an allocation,
    // but this can and should be avoided by reserving prior to
    // parsing:
    parser.reserve_locations(50u); // reserve for 50 lines
    // Now the structure will be built during parsing:
    ryml::Tree tree = parse_in_arena(&parser, "source.yml", yaml);
    // After this, we are ready to query the location from the parser:
    ryml::Location loc = tree.rootref().location(parser);
    // As for the complexity of the query: for large buffers it is
    // O(log(numlines)). For short source buffers (30 lines and less),
    // it is O(numlines), as a plain linear search is faster in this
    // case.
    CHECK(parser.location_contents(loc).begins_with("{"));
    CHECK(loc.offset == 0u);
    CHECK(loc.line == 0u);
    CHECK(loc.col == 0u);
    // on the next call, we only pay O(log(numlines)) because the
    // rebuild is already available:
    loc = tree["aa"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("aa"));
    CHECK(loc.offset == 2u);
    CHECK(loc.line == 1u);
    CHECK(loc.col == 0u);
    // KEYSEQ in flow style: points at the key
    loc = tree["foo"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("foo"));
    CHECK(loc.offset == 16u);
    CHECK(loc.line == 2u);
    CHECK(loc.col == 0u);
    loc = tree["foo"][0].location(parser);
    CHECK(parser.location_contents(loc).begins_with("one"));
    CHECK(loc.line == 2u);
    CHECK(loc.col == 6u);
    // SEQ in flow style: location points at the opening '[' (there's no key)
    loc = tree["foo"][1].location(parser);
    CHECK(parser.location_contents(loc).begins_with("["));
    CHECK(loc.line == 2u);
    CHECK(loc.col == 11u);
    loc = tree["foo"][1][0].location(parser);
    CHECK(parser.location_contents(loc).begins_with("two"));
    CHECK(loc.line == 2u);
    CHECK(loc.col == 12u);
    loc = tree["foo"][1][1].location(parser);
    CHECK(parser.location_contents(loc).begins_with("three"));
    CHECK(loc.line == 2u);
    CHECK(loc.col == 17u);
    // NOTE. The parser locations always point at the latest buffer to
    // be parsed with the parser object, so they must be queried using
    // the corresponding latest tree to be parsed. This means that if
    // the parser is reused, earlier trees will loose the possibility
    // of querying for location. It is undefined behavior to query the
    // parser for the location of a node from an earlier tree:
    ryml::Tree docval = parse_in_arena(&parser, "docval.yaml", "this is a docval");
    // From now on, none of the locations from the previous tree can
    // be queried:
    //loc = tree.rootref().location(parser); // ERROR, undefined behavior
    loc = docval.rootref().location(parser); // OK. this is the latest tree from this parser
    CHECK(parser.location_contents(loc).begins_with("this is a docval"));
    CHECK(loc.line == 0u);
    CHECK(loc.col == 0u);

    // NOTES ABOUT CONTAINER LOCATIONS
    ryml::Tree tree2 = parse_in_arena(&parser, "containers.yaml", R"(
a new: buffer
to: be parsed
map with key:
  first: value
  second: value
seq with key:
  - first value
  - second value
  -
    - nested first value
    - nested second value
  -
    nested first: value
    nested second: value
)");
    // (Likewise, the docval tree can no longer be used to query.)
    //
    // For key-less block-style maps, the location of the container
    // points at the first child's key. For example, in this case
    // the root does not have a key, so its location is taken
    // to be at the first child:
    loc = tree2.rootref().location(parser);
    CHECK(parser.location_contents(loc).begins_with("a new"));
    CHECK(loc.offset == 1u);
    CHECK(loc.line == 1u);
    CHECK(loc.col == 0u);
    // note the first child points exactly at the same place:
    loc = tree2["a new"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("a new"));
    CHECK(loc.offset == 1u);
    CHECK(loc.line == 1u);
    CHECK(loc.col == 0u);
    loc = tree2["to"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("to"));
    CHECK(loc.line == 2u);
    CHECK(loc.col == 0u);
    // but of course, if the block-style map is a KEYMAP, then the
    // location is the map's key, and not the first child's key:
    loc = tree2["map with key"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("map with key"));
    CHECK(loc.line == 3u);
    CHECK(loc.col == 0u);
    loc = tree2["map with key"]["first"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("first"));
    CHECK(loc.line == 4u);
    CHECK(loc.col == 2u);
    loc = tree2["map with key"]["second"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("second"));
    CHECK(loc.line == 5u);
    CHECK(loc.col == 2u);
    // same thing for KEYSEQ:
    loc = tree2["seq with key"].location(parser);
    CHECK(parser.location_contents(loc).begins_with("seq with key"));
    CHECK(loc.line == 6u);
    CHECK(loc.col == 0u);
    loc = tree2["seq with key"][0].location(parser);
    CHECK(parser.location_contents(loc).begins_with("first value"));
    CHECK(loc.line == 7u);
    CHECK(loc.col == 4u);
    loc = tree2["seq with key"][1].location(parser);
    CHECK(parser.location_contents(loc).begins_with("second value"));
    CHECK(loc.line == 8u);
    CHECK(loc.col == 4u);
    // SEQ nested in SEQ: container location points at the first child's "- " dash
    loc = tree2["seq with key"][2].location(parser);
    CHECK(parser.location_contents(loc).begins_with("- nested first value"));
    CHECK(loc.line == 10u);
    CHECK(loc.col == 4u);
    loc = tree2["seq with key"][2][0].location(parser);
    CHECK(parser.location_contents(loc).begins_with("nested first value"));
    CHECK(loc.line == 10u);
    CHECK(loc.col == 6u);
    // MAP nested in SEQ: same as above: point to key
    loc = tree2["seq with key"][3].location(parser);
    CHECK(parser.location_contents(loc).begins_with("nested first: "));
    CHECK(loc.line == 13u);
    CHECK(loc.col == 4u);
    loc = tree2["seq with key"][3][0].location(parser);
    CHECK(parser.location_contents(loc).begins_with("nested first: "));
    CHECK(loc.line == 13u);
    CHECK(loc.col == 4u);
}


//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------
//-----------------------------------------------------------------------------


/** @addtogroup doc_sample_helpers
 * @{ */

namespace /*anon*/ {
static int num_checks = 0;
static int num_failed_checks = 0;
} // namespace /*anon*/


bool report_check(int line, const char *predicate, bool result)
{
    ++num_checks;
    const char *msg = predicate ? "OK! " : "OK!";
    if(!result)
    {
        ++num_failed_checks;
        msg = predicate ?  "FAIL: " : "FAIL";
    }
    std::cout << __FILE__ << ':' << line << ": " << msg << (predicate ? predicate : "") << std::endl;
    return result;
}


int report_checks()
{
    std::cout << "Completed " << num_checks << " checks." << std::endl;
    if(num_failed_checks)
        std::cout << "ERROR: " << num_failed_checks << '/' << num_checks << " checks failed." << std::endl;
    else
        std::cout << "SUCCESS!" << std::endl;
    return num_failed_checks;
}


// methods for the example error handler

#ifndef C4_EXCEPTIONS
/*environment for setjmp*/
static std::jmp_buf s_jmp_env;
static std::string s_jmp_msg;
#endif

/** checking that an assertion occurs while calling fn. assertions are
 * enabled if @ref RYML_USE_ASSERT is defined.
 * @ingroup doc_sample_helpers */
template<class Fn>
bool ErrorHandlerExample::check_assertion_occurs(Fn &&fn)
{
    #if RYML_USE_ASSERT
    return check_error_occurs(std::forward<Fn>(fn));
    #else
    (void)fn; // do nothing otherwise, as there would be undefined behavior
    return true;
    #endif
}
/** checking that an error occurs while calling fn
 * @ingroup doc_sample_helpers */
template<class Fn>
bool ErrorHandlerExample::check_error_occurs(Fn &&fn)
{
    saved_msg_short.clear();
    saved_msg_full.clear();
    saved_msg_full_with_context.clear();
    saved_basic_loc = {};
    saved_parse_loc = {};
    saved_visit_tree = {};
    saved_visit_id = ryml::NONE;
    bool got_error = false;
    #ifdef C4_EXCEPTIONS
    try
    {
        std::forward<Fn>(fn)();
    }
    catch(std::exception const&)
    {
        got_error = true;
    }
    #else
    if(setjmp(s_jmp_env) == 0)
    {
        std::forward<Fn>(fn)();
    }
    else
    {
        got_error = true;
    }
    #endif
    return got_error;
}

/** interrupt execution
 * @ingroup doc_sample_helpers */
[[noreturn]] void stopexec(std::string const& s)
{
    #ifdef C4_EXCEPTIONS
    throw std::runtime_error(s);
    #else
    s_jmp_msg = s;
    std::longjmp(s_jmp_env, 1);  // jump to the corresponding call to setjmp().
    #endif
}
/** this is where the callback implementation goes. Remember that it must not return.
 * @ingroup doc_sample_helpers
 * */
[[noreturn]] void ErrorHandlerExample::on_error_basic(ryml::csubstr msg, ryml::ErrorDataBasic const& errdata)
{
    saved_msg_short.assign(msg.str, msg.len);
    // build a full error message with location
    ryml::err_basic_format([this](ryml::csubstr s){
        saved_msg_full.append(s.str, s.len);
    }, msg, errdata);
    saved_msg_full_with_context = saved_msg_full;
    saved_basic_loc = errdata.location;
    stopexec(saved_msg_short);
}
/** this is where the callback implementation goes. Remember that it must not return.
 * @ingroup doc_sample_helpers
 * @see ryml::format_location_context
 * */
[[noreturn]] void ErrorHandlerExample::on_error_parse(ryml::csubstr msg, ryml::ErrorDataParse const& errdata)
{
    saved_msg_short.assign(msg.str, msg.len);
    // build a full error message with location
    ryml::err_parse_format([this](ryml::csubstr s){
        saved_msg_full.append(s.str, s.len);
    }, msg, errdata);
    // To add the source context, the source buffer is required. If
    // the caller is interested in enriching the full message with the
    // source buffer context, he can ensure that the source buffer is
    // kept, and then arrange the handler to access it.
    // For now, we assign the full message without context:
    saved_msg_full_with_context = saved_msg_full;
    saved_basic_loc = errdata.cpploc;
    saved_parse_loc = errdata.ymlloc;
    stopexec(saved_msg_full);
}
/** this is where the callback implementation goes. Remember that it must not return.
 * @ingroup doc_sample_helpers
 * */
[[noreturn]] void ErrorHandlerExample::on_error_visit(ryml::csubstr msg, ryml::ErrorDataVisit const& errdata)
{
    saved_msg_short.assign(msg.str, msg.len);
    // build a full error message with location
    ryml::err_visit_format([this](ryml::csubstr s){
        saved_msg_full.append(s.str, s.len);
    }, msg, errdata);
    // To add the source context, the source buffer is required. If
    // the caller is interested in enriching the full message with the
    // source buffer context, he can ensure that the source buffer is
    // kept, and then arrange the handler to access it.
    // For now, we assign the full message without context:
    saved_msg_full_with_context = saved_msg_full;
    saved_basic_loc = errdata.cpploc;
    saved_visit_tree = errdata.tree;
    saved_visit_id = errdata.node;
    stopexec(saved_msg_full);
}

/** trampoline function to call the object's method */
[[noreturn]] void ErrorHandlerExample::s_error_basic(ryml::csubstr msg, ryml::ErrorDataBasic const& errdata, void *this_)
{
    static_cast<ErrorHandlerExample*>(this_)->on_error_basic(msg, errdata);
}
/** trampoline function to call the object's method */
[[noreturn]] void ErrorHandlerExample::s_error_parse(ryml::csubstr msg, ryml::ErrorDataParse const& errdata, void *this_)
{
    static_cast<ErrorHandlerExample*>(this_)->on_error_parse(msg, errdata);
}
/** trampoline function to call the object's method */
[[noreturn]] void ErrorHandlerExample::s_error_visit(ryml::csubstr msg, ryml::ErrorDataVisit const& errdata, void *this_)
{
    static_cast<ErrorHandlerExample*>(this_)->on_error_visit(msg, errdata);
}


/** a helper to create the Callbacks object for the custom error handler
 * @ingroup doc_sample_helpers
 * */
ryml::Callbacks ErrorHandlerExample::callbacks()
{
    return ryml::Callbacks{}
        .set_user_data(this)
        .set_error_basic(&ErrorHandlerExample::s_error_basic)
        .set_error_parse(&ErrorHandlerExample::s_error_parse)
        .set_error_visit(&ErrorHandlerExample::s_error_visit);
}

/** test that this handler is currently set */
void ErrorHandlerExample::check_enabled() const
{
    ryml::Callbacks const& current = ryml::get_callbacks();
    CHECK(current.m_error_basic == &s_error_basic);
    CHECK(current.m_error_parse == &s_error_parse);
    CHECK(current.m_error_visit == &s_error_visit);
    CHECK(current.m_allocate == defaults.m_allocate);
    CHECK(current.m_free == defaults.m_free);
}
/** test that this handler is currently not set */
void ErrorHandlerExample::check_disabled() const
{
    ryml::Callbacks const& current = ryml::get_callbacks();
    CHECK(current.m_error_basic != &s_error_basic);
    CHECK(current.m_error_parse != &s_error_parse);
    CHECK(current.m_error_visit != &s_error_visit);
    CHECK(current.m_allocate == defaults.m_allocate);
    CHECK(current.m_free == defaults.m_free);
}


// helper functions for sample_parse_file()

C4_SUPPRESS_WARNING_MSVC_WITH_PUSH(4996) // fopen: this function may be unsafe
/** load a file from disk into an existing CharContainer */
template<class CharContainer>
size_t file_get_contents(const char *filename, CharContainer *v)
{
    std::FILE *fp = std::fopen(filename, "rb"); // NOLINT
    if(fp == nullptr) _RYML_ERR_BASIC("{}: could not open file", filename);
    std::fseek(fp, 0, SEEK_END); // NOLINT
    long sz = std::ftell(fp); // NOLINT
    v->resize(static_cast<typename CharContainer::size_type>(sz));
    if(sz)
    {
        std::rewind(fp); // NOLINT
        size_t ret = std::fread(&(*v)[0], 1, v->size(), fp);
        if(ret != (size_t)sz) _RYML_ERR_BASIC("{}: failed to read: expect {}B, got {}B", filename, sz, ret);
    }
    std::fclose(fp); // NOLINT
    return v->size();
}

/** load a file from disk and return a newly created CharContainer */
template<class CharContainer>
CharContainer file_get_contents(const char *filename)
{
    CharContainer cc;
    file_get_contents(filename, &cc);
    return cc;
}

/** save a buffer into a file */
template<class CharContainer>
void file_put_contents(const char *filename, CharContainer const& v, const char* access)
{
    file_put_contents(filename, v.empty() ? "" : &v[0], v.size(), access);
}

/** save a buffer into a file */
void file_put_contents(const char *filename, const char *buf, size_t sz, const char* access)
{
    std::FILE *fp = std::fopen(filename, access);
    if(fp == nullptr) _RYML_ERR_BASIC("{}: could not open file", filename);
    std::fwrite(buf, 1, sz, fp); // NOLINT
    std::fclose(fp); // NOLINT
}
C4_SUPPRESS_WARNING_MSVC_POP

/** @} */ // doc_sample_helpers

/** @} */ // doc_quickstart

C4_SUPPRESS_WARNING_GCC_CLANG_POP