BFCommonKit/Example/Pods/Realm/include/core/realm/group.hpp

/*************************************************************************
 *
 * Copyright 2016 Realm Inc.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 **************************************************************************/

#ifndef REALM_GROUP_HPP
#define REALM_GROUP_HPP

#include <functional>
#include <string>
#include <vector>
#include <map>
#include <stdexcept>
#include <optional>

#include <realm/util/features.h>
#include <realm/exceptions.hpp>
#include <realm/impl/input_stream.hpp>
#include <realm/impl/output_stream.hpp>
#include <realm/impl/cont_transact_hist.hpp>
#include <realm/metrics/metrics.hpp>
#include <realm/table.hpp>
#include <realm/alloc_slab.hpp>

namespace realm {

class DB;
class TableKeys;

namespace _impl {
class GroupFriend;
class TransactLogConvenientEncoder;
class TransactLogParser;
} // namespace _impl


/// A group is a collection of named tables.
///
class Group : public ArrayParent {
public:
    /// Construct a free-standing group. This group instance will be
    /// in the attached state, but neither associated with a file, nor
    /// with an external memory buffer.
    Group();

    enum OpenMode {
        /// Open in read-only mode. Fail if the file does not already exist.
        mode_ReadOnly,
        /// Open in read/write mode. Create the file if it doesn't exist.
        mode_ReadWrite,
        /// Open in read/write mode. Fail if the file does not already exist.
        mode_ReadWriteNoCreate
    };

    /// Equivalent to calling open(const std::string&, const char*, OpenMode)
    /// on an unattached group accessor.
    explicit Group(const std::string& file, const char* encryption_key = nullptr, OpenMode = mode_ReadOnly);

    /// Equivalent to calling open(BinaryData, bool) on an unattached
    /// group accessor. Note that if this constructor throws, the
    /// ownership of the memory buffer will remain with the caller,
    /// regardless of whether \a take_ownership is set to `true` or
    /// `false`.
    explicit Group(BinaryData, bool take_ownership = true);

    struct unattached_tag {
    };

    /// Create a Group instance in its unattached state. It may then
    /// be attached to a database file later by calling one of the
    /// open() methods. You may test whether this instance is
    /// currently in its attached state by calling
    /// is_attached(). Calling any other method (except the
    /// destructor) while in the unattached state has undefined
    /// behavior.
    Group(unattached_tag) noexcept;

    Group(const Group&) = delete;
    Group& operator=(const Group&) = delete;

    ~Group() noexcept override;

    /// Attach this Group instance to the specified database file.
    ///
    /// By default, the specified file is opened in read-only mode
    /// (mode_ReadOnly). This allows opening a file even when the
    /// caller lacks permission to write to that file. The opened
    /// group may still be modified freely, but the changes cannot be
    /// written back to the same file using the commit() function. An
    /// attempt to do that, will cause an exception to be thrown. When
    /// opening in read-only mode, it is an error if the specified
    /// file does not already exist in the file system.
    ///
    /// Alternatively, the file can be opened in read/write mode
    /// (mode_ReadWrite). This allows use of the commit() function,
    /// but, of course, it also requires that the caller has
    /// permission to write to the specified file. When opening in
    /// read-write mode, an attempt to create the specified file will
    /// be made, if it does not already exist in the file system.
    ///
    /// In any case, if the file already exists, it must contain a
    /// valid Realm database. In many cases invalidity will be
    /// detected and cause the InvalidDatabase exception to be thrown,
    /// but you should not rely on it.
    ///
    /// Note that changes made to the database via a Group instance
    /// are not automatically committed to the specified file. You
    /// may, however, at any time, explicitly commit your changes by
    /// calling the commit() method, provided that the specified
    /// open-mode is not mode_ReadOnly. Alternatively, you may call
    /// write() to write the entire database to a new file. Writing
    /// the database to a new file does not end, or in any other way
    /// change the association between the Group instance and the file
    /// that was specified in the call to open().
    ///
    /// A Realm file that contains a history (see Replication::HistoryType) may
    /// be opened via Group::open(), as long as the application can ensure that
    /// there is no concurrent access to the file (see below for more on
    /// concurrency), but if the file is modified via Group::commit() the
    /// history will be discarded. To retain the history, the application must
    /// instead access the file in shared mode, i.e., via DB, and
    /// supply the right kind of replication plugin (see
    /// Replication::get_history_type()).
    ///
    /// A file that is passed to Group::open(), may not be modified by
    /// a third party until after the Group object is
    /// destroyed. Behavior is undefined if a file is modified by a
    /// third party while any Group object is associated with it.
    ///
    /// Calling open() on a Group instance that is already in the
    /// attached state has undefined behavior.
    ///
    /// Accessing a Realm database file through manual construction
    /// of a Group object does not offer any level of thread safety or
    /// transaction safety. When any of those kinds of safety are a
    /// concern, consider using a DB instead. When accessing
    /// a database file in read/write mode through a manually
    /// constructed Group object, it is entirely the responsibility of
    /// the application that the file is not accessed in any way by a
    /// third party during the life-time of that group object. It is,
    /// on the other hand, safe to concurrently access a database file
    /// by multiple manually created Group objects, as long as all of
    /// them are opened in read-only mode, and there is no other party
    /// that modifies the file concurrently.
    ///
    /// Do not call this function on a group instance that is managed
    /// by a shared group. Doing so will result in undefined behavior.
    ///
    /// Even if this function throws, it may have the side-effect of
    /// creating the specified file, and the file may get left behind
    /// in an invalid state. Of course, this can only happen if
    /// read/write mode (mode_ReadWrite) was requested, and the file
    /// did not already exist.
    ///
    /// \param file File system path to a Realm database file.
    ///
    /// \param encryption_key 32-byte key used to encrypt and decrypt
    /// the database file, or nullptr to disable encryption.
    ///
    /// \param mode Specifying a mode that is not mode_ReadOnly
    /// requires that the specified file can be opened in read/write
    /// mode. In general there is no reason to open a group in
    /// read/write mode unless you want to be able to call
    /// Group::commit().
    ///
    /// \throw util::File::AccessError If the file could not be
    /// opened. If the reason corresponds to one of the exception
    /// types that are derived from util::File::AccessError, the
    /// derived exception type is thrown. Note that InvalidDatabase is
    /// among these derived exception types.
    void open(const std::string& file, const char* encryption_key = nullptr, OpenMode mode = mode_ReadOnly);

    /// Attach this Group instance to the specified memory buffer.
    ///
    /// This is similar to constructing a group from a file except
    /// that in this case the database is assumed to be stored in the
    /// specified memory buffer.
    ///
    /// If \a take_ownership is `true`, you pass the ownership of the
    /// specified buffer to the group. In this case the buffer will
    /// eventually be freed using std::free(), so the buffer you pass,
    /// must have been allocated using std::malloc().
    ///
    /// On the other hand, if \a take_ownership is set to `false`, it
    /// is your responsibility to keep the memory buffer alive during
    /// the lifetime of the group, and in case the buffer needs to be
    /// deallocated afterwards, that is your responsibility too.
    ///
    /// If this function throws, the ownership of the memory buffer
    /// will remain with the caller, regardless of whether \a
    /// take_ownership is set to `true` or `false`.
    ///
    /// Calling open() on a Group instance that is already in the
    /// attached state has undefined behavior.
    ///
    /// Do not call this function on a group instance that is managed
    /// by a shared group. Doing so will result in undefined behavior.
    ///
    /// \throw InvalidDatabase If the specified buffer does not appear
    /// to contain a valid database.
    void open(BinaryData, bool take_ownership = true);

    /// A group may be created in the unattached state, and then later
    /// attached to a file with a call to open(). Calling any method
    /// other than open(), and is_attached() on an unattached instance
    /// results in undefined behavior.
    bool is_attached() const noexcept;
    /// A group is frozen only if it is actually a frozen transaction.
    virtual bool is_frozen() const noexcept
    {
        return false;
    }
    /// Returns true if, and only if the number of tables in this
    /// group is zero.
    bool is_empty() const noexcept;

    size_t size() const noexcept;

    int get_history_schema_version() noexcept;

    Replication* get_replication() const
    {
        return *get_repl();
    }

    /// The sync file id is set when a client synchronizes with the server for the
    /// first time. It is used when generating GlobalKeys for tables without a primary
    /// key, where it is used as the "hi" part. This ensures global uniqueness of
    /// GlobalKeys.
    uint64_t get_sync_file_id() const noexcept;
    void set_sync_file_id(uint64_t id);
    void remove_sync_file_id();

    /// Returns the keys for all tables in this group.
    TableKeys get_table_keys() const;

    /// \defgroup group_table_access Table Accessors
    ///
    /// has_table() returns true if, and only if this group contains a table
    /// with the specified name.
    ///
    /// find_table() returns the key of the first table in this group with the
    /// specified name, or `realm::not_found` if this group does not contain a
    /// table with the specified name.
    ///
    /// get_table_name() returns the name of table with the specified key.
    ///
    /// The versions of get_table(), that accepts a \a name argument, return a
    /// table with the specified name, or null if no such table exists.
    ///
    /// add_table() adds a table with the specified name to this group. It
    /// throws TableNameInUse if \a require_unique_name is true and \a name
    /// clashes with the name of an existing table. If \a require_unique_name is
    /// false, it is possible to add more than one table with the same
    /// name. Whenever a table is added the key assigned to it is returned.
    ///
    /// get_or_add_table() checks if a table exists in this group with the specified
    /// name. If it doesn't exist, a table is created.
    ///
    /// remove_table() removes the specified table from this group. A table can
    /// be removed only when it is not the target of a link column of a
    /// different table.
    ///
    /// rename_table() changes the name of a preexisting table. If \a
    /// require_unique_name is false, it becomes possible to have more than one
    /// table with a given name in a single group.
    ///
    /// The template functions work exactly like their non-template namesakes
    /// except as follows: The template versions of get_table() and
    /// get_or_add_table() throw DescriptorMismatch if the dynamic type of the
    /// specified table does not match the statically specified custom table
    /// type. The template versions of add_table() and get_or_add_table() set
    /// the dynamic type (descriptor) to match the statically specified custom
    /// table type.
    ///
    /// \param key Key of table in this group.
    ///
    /// \param name Name of table. All strings are valid table names as long as
    /// they are valid UTF-8 encodings and the number of bytes does not exceed
    /// `max_table_name_length`. A call to add_table() or get_or_add_table()
    /// with a name that is longer than `max_table_name_length` will cause an
    /// exception to be thrown.
    ///
    /// \param new_name New name for preexisting table.
    ///
    /// \param require_unique_name When set to true (the default), it becomes
    /// impossible to add a table with a name that is already in use, or to
    /// rename a table to a name that is already in use.
    ///
    /// \param was_added When specified, the boolean variable is set to true if
    /// the table was added, and to false otherwise. If the function throws, the
    /// boolean variable retains its original value.
    ///
    /// \return get_table(), add_table(), and get_or_add_table() return a table
    /// accessor attached to the requested (or added) table. get_table() may
    /// return null.
    ///
    /// \throw DescriptorMismatch Thrown by get_table() and get_or_add_table()
    /// tf the dynamic table type does not match the statically specified custom
    /// table type (\a T).
    ///
    /// \throw NoSuchTable Thrown by remove_table() and rename_table() if there
    /// is no table with the specified \a name.
    ///
    /// \throw TableNameInUse Thrown by add_table() if \a require_unique_name is
    /// true and \a name clashes with the name of a preexisting table. Thrown by
    /// rename_table() if \a require_unique_name is true and \a new_name clashes
    /// with the name of a preexisting table.
    ///
    /// \throw CrossTableLinkTarget Thrown by remove_table() if the specified
    /// table is the target of a link column of a different table.
    ///
    //@{

    static const size_t max_table_name_length = 63;

    bool has_table(StringData name) const noexcept;
    TableKey find_table(StringData name) const noexcept;
    StringData get_table_name(TableKey key) const;

    TableRef get_table(TableKey key);
    ConstTableRef get_table(TableKey key) const;

    // Catch some implicit conversions
    TableRef get_table(int) = delete;
    ConstTableRef get_table(int) const = delete;

    TableRef get_table(StringData name);
    ConstTableRef get_table(StringData name) const;

    TableRef add_table(StringData name);
    TableRef add_embedded_table(StringData name);
    TableRef add_table_with_primary_key(StringData name, DataType pk_type, StringData pk_name, bool nullable = false);
    TableRef get_or_add_table(StringData name, bool* was_added = nullptr);

    void remove_table(TableKey key);
    void remove_table(StringData name);

    void rename_table(TableKey key, StringData new_name, bool require_unique_name = true);
    void rename_table(StringData name, StringData new_name, bool require_unique_name = true);

    Obj get_object(ObjLink link);
    void validate(ObjLink link) const;

    //@}

    // Serialization

    /// Write this database to the specified output stream.
    ///
    /// \param out The destination output stream to write to.
    ///
    /// \param pad If true, the file is padded to ensure the footer is aligned
    /// to the end of a page
    void write(std::ostream& out, bool pad = false) const;

    /// Write this database to a new file. It is an error to specify a
    /// file that already exists. This is to protect against
    /// overwriting a database file that is currently open, which
    /// would cause undefined behaviour.
    ///
    /// \param file A filesystem path.
    ///
    /// \param encryption_key 32-byte key used to encrypt the database file,
    /// or nullptr to disable encryption.
    ///
    /// \param version If different from 0, the new file will be a full fledged
    /// realm file with free list and history info. The version of the commit
    /// will be set to the value given here.
    ///
    /// \throw util::File::AccessError If the file could not be
    /// opened. If the reason corresponds to one of the exception
    /// types that are derived from util::File::AccessError, the
    /// derived exception type is thrown. In particular,
    /// util::File::Exists will be thrown if the file exists already.
    void write(const std::string& file, const char* encryption_key = nullptr, uint64_t version = 0,
               bool write_history = true) const;

    /// Write this database to a memory buffer.
    ///
    /// Ownership of the returned buffer is transferred to the
    /// caller. The memory will have been allocated using
    /// std::malloc().
    BinaryData write_to_mem() const;

    /// Commit changes to the attached file. This requires that the
    /// attached file is opened in read/write mode.
    ///
    /// Calling this function on an unattached group, a free-standing
    /// group, a group whose attached file is opened in read-only
    /// mode, a group that is attached to a memory buffer, or a group
    /// that is managed by a shared group, is an error and will result
    /// in undefined behavior.
    ///
    /// Table accesors will remain valid across the commit. Note that
    /// this is not the case when working with proper transactions.
    void commit();

    //@{
    /// Some operations on Tables in a Group can cause indirect changes to other
    /// fields, including in other Tables in the same Group. Specifically,
    /// removing a row will set any links to that row to null, and if it had the
    /// last strong links to other rows, will remove those rows. When this
    /// happens, The cascade notification handler will be called with a
    /// CascadeNotification containing information about what indirect changes
    /// will occur, before any changes are made.
    ///
    /// has_cascade_notification_handler() returns true if and only if there is
    /// currently a non-null notification handler registered.
    ///
    /// set_cascade_notification_handler() replaces the current handler (if any)
    /// with the passed in handler. Pass in nullptr to remove the current handler
    /// without registering a new one.
    ///
    /// CascadeNotification contains a vector of rows which will be removed and
    /// a vector of links which will be set to null (or removed, for entries in
    /// LinkLists).
    struct CascadeNotification {
        struct row {
            /// Key identifying a group-level table.
            TableKey table_key;

            /// Key identifying object to be removed.
            ObjKey key;

            row() = default;

            row(TableKey tk, ObjKey k)
                : table_key(tk)
                , key(k)
            {
            }
            bool operator==(const row& r) const noexcept
            {
                return table_key == r.table_key && key == r.key;
            }
            bool operator!=(const row& r) const noexcept
            {
                return !(*this == r);
            }
            /// Trivial lexicographic order
            bool operator<(const row& r) const noexcept
            {
                return table_key < r.table_key || (table_key == r.table_key && key < r.key);
            }
        };

        struct link {
            link() = default;
            link(TableKey tk, ColKey ck, ObjKey k, ObjKey otk)
                : origin_table(tk)
                , origin_col_key(ck)
                , origin_key(k)
                , old_target_key(otk)
            {
            }
            TableKey origin_table; ///< A group-level table.
            ColKey origin_col_key; ///< Link column being nullified.
            ObjKey origin_key;     ///< Row in column being nullified.
            /// The target row index which is being removed. Mostly relevant for
            /// LinkList (to know which entries are being removed), but also
            /// valid for Link.
            ObjKey old_target_key;
        };

        /// A sorted list of rows which will be removed by the current operation.
        std::vector<row> rows;

        /// An unordered list of links which will be nullified by the current
        /// operation.
        std::vector<link> links;
    };

    bool has_cascade_notification_handler() const noexcept;
    void set_cascade_notification_handler(std::function<void(const CascadeNotification&)> new_handler) noexcept;

    //@}

    //@{
    /// During sync operation, schema changes may happen at runtime as connected
    /// clients update their schema as part of an app update. Since this is a
    /// relatively rare event, no attempt is made at limiting the amount of work
    /// the handler is required to do to update its information about table and
    /// column indices (i.e., all table and column indices must be recalculated).
    ///
    /// At the time of writing, only additive schema changes may occur in that
    /// scenario.
    ///
    /// has_schema_change_notification_handler() returns true iff there is currently
    /// a non-null notification handler registered.
    ///
    /// set_schema_change_notification_handler() replaces the current handler (if any)
    /// with the passed in handler. Pass in nullptr to remove the current handler
    /// without registering a new one.

    bool has_schema_change_notification_handler() const noexcept;
    void set_schema_change_notification_handler(std::function<void()> new_handler) noexcept;

    //@}

    // Conversion
    void schema_to_json(std::ostream& out, std::map<std::string, std::string>* renames = nullptr) const;
    void to_json(std::ostream& out, size_t link_depth = 0, std::map<std::string, std::string>* renames = nullptr,
                 JSONOutputMode output_mode = output_mode_json) const;

    /// Compare two groups for equality. Two groups are equal if, and
    /// only if, they contain the same tables in the same order, that
    /// is, for each table T at index I in one of the groups, there is
    /// a table at index I in the other group that is equal to T.
    /// Tables are equal if they have the same content and the same table name.
    bool operator==(const Group&) const;

    /// Compare two groups for inequality. See operator==().
    bool operator!=(const Group& g) const
    {
        return !(*this == g);
    }

    /// Control of what to include when computing memory usage
    enum SizeAggregateControl {
        size_of_state = 1,     ///< size of tables, indexes, toplevel array
        size_of_history = 2,   ///< size of the in-file history compartment
        size_of_freelists = 4, ///< size of the freelists
        size_of_all = 7
    };
    /// Compute the sum of the sizes in number of bytes of all the array nodes
    /// that currently make up this group. When this group represents a snapshot
    /// in a Realm file (such as during a read transaction via a Transaction
    /// instance), this function computes the footprint of that snapshot within
    /// the Realm file.
    ///
    /// If this group accessor is the detached state, this function returns
    /// zero.
    size_t compute_aggregated_byte_size(SizeAggregateControl ctrl = SizeAggregateControl::size_of_all) const noexcept;
    /// Return the size taken up by the current snapshot. This is in contrast to
    /// the number returned by DB::get_stats() which will return the
    /// size of the last snapshot done in that DB. If the snapshots are
    /// identical, the numbers will of course be equal.
    size_t get_used_space() const noexcept;

    /// check that an already attached realm file is valid for read only access.
    /// if not detach the file and throw a FileFormatUpgradeRequired.
    /// return the file format version.
    static int read_only_version_check(SlabAlloc& alloc, ref_type top_ref, const std::string& path);
    void verify() const;
    void validate_primary_columns();
#ifdef REALM_DEBUG
    void print() const;
    void print_free() const;
    MemStats get_stats();
    void enable_mem_diagnostics(bool enable = true)
    {
        m_alloc.enable_debug(enable);
    }
#endif

protected:
    virtual Replication* const* get_repl() const
    {
        return &Table::g_dummy_replication;
    }

private:
    static constexpr char g_class_name_prefix[] = "class_";
    static constexpr size_t g_class_name_prefix_len = 6;

    // nullptr, if we're sharing an allocator provided during initialization
    std::unique_ptr<SlabAlloc> m_local_alloc;
    // in-use allocator. If local, then equal to m_local_alloc.
    SlabAlloc& m_alloc;

    int m_file_format_version;
    /// `m_top` is the root node (or top array) of the Realm, and has the
    /// following layout:
    ///
    /// <pre>
    ///
    ///                                                     Introduced in file
    ///   Slot  Value                                       format version
    ///   ---------------------------------------------------------------------
    ///    1st   m_table_names
    ///    2nd   m_tables
    ///    3rd   Logical file size
    ///    4th   GroupWriter::m_free_positions (optional)
    ///    5th   GroupWriter::m_free_lengths   (optional)
    ///    6th   GroupWriter::m_free_versions  (optional)
    ///    7th   Transaction number / version  (optional)
    ///    8th   History type         (optional)             4
    ///    9th   History ref          (optional)             4
    ///   10th   History version      (optional)             7
    ///   11th   Sync File Id         (optional)            10
    ///
    /// </pre>
    ///
    /// The 'History type' slot stores a value of type
    /// Replication::HistoryType. The 'History version' slot stores a history
    /// schema version as returned by Replication::get_history_schema_version().
    ///
    /// The first three entries are mandatory. In files created by
    /// Group::write(), none of the optional entries are present and the size of
    /// `m_top` is 3. In files updated by Group::commit(), the 4th and 5th entry
    /// are present, and the size of `m_top` is 5. In files updated by way of a
    /// transaction (Transaction::commit()), the 4th, 5th, 6th, and 7th entry
    /// are present, and the size of `m_top` is 7. In files that contain a
    /// changeset history, the 8th, 9th, and 10th entry are present. The 11th entry
    /// will be present if the file is syncked and the client has received a client
    /// file id from the server.
    ///
    /// When a group accessor is attached to a newly created file or an empty
    /// memory buffer where there is no top array yet, `m_top`, `m_tables`, and
    /// `m_table_names` will be left in the detached state until the initiation
    /// of the first write transaction. In particular, they will remain in the
    /// detached state during read transactions that precede the first write
    /// transaction.
    Array m_top;
    Array m_tables;
    ArrayStringShort m_table_names;
    uint64_t m_last_seen_mapping_version = 0;

    typedef std::vector<Table*> TableAccessors;
    mutable TableAccessors m_table_accessors;
    mutable std::mutex m_accessor_mutex;
    mutable int m_num_tables = 0;
    bool m_attached = false;
    bool m_is_writable = true;
    const bool m_is_shared;
    static std::optional<int> fake_target_file_format;

    std::function<void(const CascadeNotification&)> m_notify_handler;
    std::function<void()> m_schema_change_handler;
    std::shared_ptr<metrics::Metrics> m_metrics;
    size_t m_total_rows;

    static constexpr size_t s_table_name_ndx = 0;
    static constexpr size_t s_table_refs_ndx = 1;
    static constexpr size_t s_file_size_ndx = 2;
    static constexpr size_t s_free_pos_ndx = 3;
    static constexpr size_t s_free_size_ndx = 4;
    static constexpr size_t s_free_version_ndx = 5;
    static constexpr size_t s_version_ndx = 6;
    static constexpr size_t s_hist_type_ndx = 7;
    static constexpr size_t s_hist_ref_ndx = 8;
    static constexpr size_t s_hist_version_ndx = 9;
    static constexpr size_t s_sync_file_id_ndx = 10;

    static constexpr size_t s_group_max_size = 11;

    struct shared_tag {
    };
    Group(shared_tag) noexcept;

    Group(SlabAlloc* alloc) noexcept;
    void init_array_parents() noexcept;

    void open(ref_type top_ref, const std::string& file_path);

    // If the underlying memory mappings have been extended, this method is used
    // to update all the tables' allocator wrappers. The allocator wrappers are
    // configure to either allow or deny changes.
    void update_allocator_wrappers(bool writable);

    /// If `top_ref` is not zero, attach this group accessor to the specified
    /// underlying node structure. If `top_ref` is zero and \a
    /// create_group_when_missing is true, create a new node structure that
    /// represents an empty group, and attach this group accessor to it.
    void attach(ref_type top_ref, bool writable, bool create_group_when_missing);

    /// Detach this group accessor from the underlying node structure. If this
    /// group accessors is already in the detached state, this function does
    /// nothing (idempotency).
    void detach() noexcept;

    /// \param writable Must be set to true when, and only when attaching for a
    /// write transaction.
    void attach_shared(ref_type new_top_ref, size_t new_file_size, bool writable);

    void create_empty_group();
    void remove_table(size_t table_ndx, TableKey key);

    void reset_free_space_tracking();

    void remap(size_t new_file_size);
    void remap_and_update_refs(ref_type new_top_ref, size_t new_file_size, bool writable);

    /// Recursively update refs stored in all cached array
    /// accessors. This includes cached array accessors in any
    /// currently attached table accessors. This ensures that the
    /// group instance itself, as well as any attached table accessor
    /// that exists across Group::commit() will remain valid. This
    /// function is not appropriate for use in conjunction with
    /// commits via shared group.
    void update_refs(ref_type top_ref) noexcept;

    // Overriding method in ArrayParent
    void update_child_ref(size_t, ref_type) override;

    // Overriding method in ArrayParent
    ref_type get_child_ref(size_t) const noexcept override;

    class TableWriter;
    class DefaultTableWriter;

    static void write(std::ostream&, int file_format_version, TableWriter&, bool no_top_array,
                      bool pad_for_encryption, uint_fast64_t version_number);

    Table* do_get_table(size_t ndx);
    const Table* do_get_table(size_t ndx) const;
    Table* do_get_table(StringData name);
    const Table* do_get_table(StringData name) const;
    Table* do_add_table(StringData name, bool is_embedded, bool do_repl = true);

    void create_and_insert_table(TableKey key, StringData name);
    Table* create_table_accessor(size_t table_ndx);
    void recycle_table_accessor(Table*);

    void detach_table_accessors() noexcept; // Idempotent

    void mark_all_table_accessors() noexcept;

    void write(util::File& file, const char* encryption_key, uint_fast64_t version_number, bool write_history) const;
    void write(std::ostream&, bool pad, uint_fast64_t version_numer, bool write_history) const;

    std::shared_ptr<metrics::Metrics> get_metrics() const noexcept;
    void set_metrics(std::shared_ptr<metrics::Metrics> other) noexcept;
    void update_num_objects();
    class TransactAdvancer;
    /// Memory mappings must have been updated to reflect any growth in filesize before
    /// calling advance_transact()
    void advance_transact(ref_type new_top_ref, _impl::NoCopyInputStream&, bool writable);
    void refresh_dirty_accessors();
    void flush_accessors_for_commit();

    /// \brief The version of the format of the node structure (in file or in
    /// memory) in use by Realm objects associated with this group.
    ///
    /// Every group contains a file format version field, which is returned
    /// by this function. The file format version field is set to the file format
    /// version specified by the attached file (or attached memory buffer) at the
    /// time of attachment and the value is used to determine if a file format
    /// upgrade is required.
    ///
    /// A value of zero means that the file format is not yet decided. This is
    /// only possible for empty Realms where top-ref is zero. (When group is created
    /// with the unattached_tag). The version number will then be determined in the
    /// subsequent call to Group::open.
    ///
    /// In shared mode (when a Realm file is opened via a DB instance)
    /// it can happen that the file format is upgraded asyncronously (via
    /// another DB instance), and in that case the file format version
    /// field can get out of date, but only for a short while. It is always
    /// guaranteed to be, and remain up to date after the opening process completes
    /// (when DB::do_open() returns).
    ///
    /// An empty Realm file (one whose top-ref is zero) may specify a file
    /// format version of zero to indicate that the format is not yet
    /// decided. In that case the file format version must be changed to a proper
    /// before the opening process completes (Group::open() or DB::open()).
    ///
    /// File format versions:
    ///
    ///   1 Initial file format version
    ///
    ///   2 Various changes.
    ///
    ///   3 Supporting null on string columns broke the file format in following
    ///     way: Index appends an 'X' character to all strings except the null
    ///     string, to be able to distinguish between null and empty
    ///     string. Bumped to 3 because of null support of String columns and
    ///     because of new format of index.
    ///
    ///   4 Introduction of optional in-Realm history of changes (additional
    ///     entries in Group::m_top). Since this change is not forward
    ///     compatible, the file format version had to be bumped. This change is
    ///     implemented in a way that achieves backwards compatibility with
    ///     version 3 (and in turn with version 2).
    ///
    ///   5 Introduced the new Timestamp column type that replaces DateTime.
    ///     When opening an older database file, all DateTime columns will be
    ///     automatically upgraded Timestamp columns.
    ///
    ///   6 Introduced a new structure for the StringIndex. Moved the commit
    ///     logs into the Realm file. Changes to the transaction log format
    ///     including reshuffling instructions. This is the format used in
    ///     milestone 2.0.0.
    ///
    ///   7 Introduced "history schema version" as 10th entry in top array.
    ///
    ///   8 Subtables can now have search index.
    ///
    ///   9 Replication instruction values shuffled, instr_MoveRow added.
    ///
    ///  10 Cluster based table layout. Memory mapping changes which require
    ///     special treatment of large files of preceding versions.
    ///
    ///  11 Same as 10, but version 11 files will have search index added on
    ///     string primary key columns.
    ///
    ///  12 - 19 Room for new file formats in legacy code.
    ///
    ///  20 New data types: Decimal128 and ObjectId. Embedded tables.
    ///
    /// IMPORTANT: When introducing a new file format version, be sure to review
    /// the file validity checks in Group::open() and DB::do_open, the file
    /// format selection logic in
    /// Group::get_target_file_format_version_for_session(), and the file format
    /// upgrade logic in Group::upgrade_file_format(), AND the lists of accepted
    /// file formats and the version deletion list residing in "backup_restore.cpp"

    int get_file_format_version() const noexcept;
    void set_file_format_version(int) noexcept;
    int get_committed_file_format_version() const noexcept;

    /// The specified history type must be a value of Replication::HistoryType.
    static int get_target_file_format_version_for_session(int current_file_format_version, int history_type) noexcept;

    void send_cascade_notification(const CascadeNotification& notification) const;
    void send_schema_change_notification() const;

    static void get_version_and_history_info(const Array& top, _impl::History::version_type& version,
                                             int& history_type, int& history_schema_version) noexcept;
    static ref_type get_history_ref(const Array& top) noexcept;
    void set_history_schema_version(int version);
    template <class Accessor>
    void set_history_parent(Accessor& history_root) noexcept;
    void prepare_top_for_history(int history_type, int history_schema_version, uint64_t file_ident);
    template <class Accessor>
    void prepare_history_parent(Accessor& history_root, int history_type, int history_schema_version,
                                uint64_t file_ident);
    static void validate_top_array(const Array& arr, const SlabAlloc& alloc);

    size_t find_table_index(StringData name) const noexcept;
    TableKey ndx2key(size_t ndx) const;
    size_t key2ndx(TableKey key) const;
    size_t key2ndx_checked(TableKey key) const;
    void set_size() const noexcept;
    std::map<TableRef, ColKey> get_primary_key_columns_from_pk_table(TableRef pk_table);
    void check_table_name_uniqueness(StringData name)
    {
        if (m_table_names.find_first(name) != not_found)
            throw TableNameInUse();
    }

    friend class Table;
    friend class GroupWriter;
    friend class DB;
    friend class _impl::GroupFriend;
    friend class _impl::TransactLogConvenientEncoder;
    friend class _impl::TransactLogParser;
    friend class TrivialReplication;
    friend class metrics::QueryInfo;
    friend class metrics::Metrics;
    friend class Transaction;
    friend class TableKeyIterator;
    friend class CascadeState;
};

class TableKeyIterator {
public:
    bool operator!=(const TableKeyIterator& other)
    {
        return m_pos != other.m_pos;
    }
    TableKeyIterator& operator++();
    TableKey operator*();

private:
    friend class TableKeys;
    const Group* m_group;
    size_t m_pos;
    size_t m_index_in_group = 0;
    TableKey m_table_key;

    TableKeyIterator(const Group* g, size_t p)
        : m_group(g)
        , m_pos(p)
    {
    }
    void load_key();
};

class TableKeys {
public:
    TableKeys(const Group* g)
        : m_iter(g, 0)
    {
    }
    size_t size() const
    {
        return m_iter.m_group->size();
    }
    bool empty() const
    {
        return size() == 0;
    }
    TableKey operator[](size_t p) const;
    TableKeyIterator begin() const
    {
        return TableKeyIterator(m_iter.m_group, 0);
    }
    TableKeyIterator end() const
    {
        return TableKeyIterator(m_iter.m_group, size());
    }

private:
    mutable TableKeyIterator m_iter;
};

// Implementation

inline TableKeys Group::get_table_keys() const
{
    return TableKeys(this);
}

inline bool Group::is_attached() const noexcept
{
    return m_attached;
}

inline bool Group::is_empty() const noexcept
{
    if (!is_attached())
        return false;
    return size() == 0;
}

inline size_t Group::key2ndx(TableKey key) const
{
    size_t idx = key.value & 0xFFFF;
    return idx;
}

inline StringData Group::get_table_name(TableKey key) const
{
    size_t table_ndx = key2ndx_checked(key);
    return m_table_names.get(table_ndx);
}

inline bool Group::has_table(StringData name) const noexcept
{
    size_t ndx = find_table_index(name);
    return ndx != not_found;
}

inline size_t Group::find_table_index(StringData name) const noexcept
{
    if (m_table_names.is_attached())
        return m_table_names.find_first(name);
    return not_found;
}

inline TableKey Group::find_table(StringData name) const noexcept
{
    if (!is_attached())
        return TableKey();
    size_t ndx = find_table_index(name);
    return (ndx != npos) ? ndx2key(ndx) : TableKey{};
}

inline TableRef Group::get_table(TableKey key)
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    auto ndx = key2ndx_checked(key);
    Table* table = do_get_table(ndx); // Throws
    return TableRef(table, table ? table->m_alloc.get_instance_version() : 0);
}

inline ConstTableRef Group::get_table(TableKey key) const
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    auto ndx = key2ndx_checked(key);
    const Table* table = do_get_table(ndx); // Throws
    return ConstTableRef(table, table ? table->m_alloc.get_instance_version() : 0);
}

inline TableRef Group::get_table(StringData name)
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    Table* table = do_get_table(name); // Throws
    return TableRef(table, table ? table->m_alloc.get_instance_version() : 0);
}

inline ConstTableRef Group::get_table(StringData name) const
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    const Table* table = do_get_table(name); // Throws
    return ConstTableRef(table, table ? table->m_alloc.get_instance_version() : 0);
}

inline TableRef Group::add_table(StringData name)
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    check_table_name_uniqueness(name);
    Table* table = do_add_table(name, false); // Throws
    return TableRef(table, table->m_alloc.get_instance_version());
}

inline TableRef Group::add_embedded_table(StringData name)
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    check_table_name_uniqueness(name);
    Table* table = do_add_table(name, true); // Throws
    return TableRef(table, table->m_alloc.get_instance_version());
}

inline TableRef Group::get_or_add_table(StringData name, bool* was_added)
{
    if (!is_attached())
        throw LogicError(LogicError::detached_accessor);
    auto table = do_get_table(name);
    if (was_added)
        *was_added = !table;
    if (!table) {
        table = do_add_table(name, false);
    }
    return TableRef(table, table->m_alloc.get_instance_version());
}

inline void Group::init_array_parents() noexcept
{
    m_table_names.set_parent(&m_top, 0);
    m_tables.set_parent(&m_top, 1);
}

inline void Group::update_child_ref(size_t child_ndx, ref_type new_ref)
{
    m_tables.set(child_ndx, new_ref);
}

inline ref_type Group::get_child_ref(size_t child_ndx) const noexcept
{
    return m_tables.get_as_ref(child_ndx);
}

inline bool Group::has_cascade_notification_handler() const noexcept
{
    return !!m_notify_handler;
}

inline void
Group::set_cascade_notification_handler(std::function<void(const CascadeNotification&)> new_handler) noexcept
{
    m_notify_handler = std::move(new_handler);
}

inline void Group::send_cascade_notification(const CascadeNotification& notification) const
{
    REALM_ASSERT_DEBUG(m_notify_handler);
    m_notify_handler(notification);
}

inline bool Group::has_schema_change_notification_handler() const noexcept
{
    return !!m_schema_change_handler;
}

inline void Group::set_schema_change_notification_handler(std::function<void()> new_handler) noexcept
{
    m_schema_change_handler = std::move(new_handler);
}

inline void Group::send_schema_change_notification() const
{
    if (m_schema_change_handler)
        m_schema_change_handler();
}

inline ref_type Group::get_history_ref(const Array& top) noexcept
{
    bool has_history = (top.is_attached() && top.size() > s_hist_type_ndx);
    if (has_history) {
        // This function is only used is shared mode (from DB)
        REALM_ASSERT(top.size() > s_hist_version_ndx);
        return top.get_as_ref(s_hist_ref_ndx);
    }
    return 0;
}

inline void Group::set_sync_file_id(uint64_t id)
{
    while (m_top.size() < s_sync_file_id_ndx + 1)
        m_top.add(0);
    m_top.set(s_sync_file_id_ndx, RefOrTagged::make_tagged(id));
}

inline void Group::set_history_schema_version(int version)
{
    while (m_top.size() < s_hist_version_ndx + 1)
        m_top.add(0);
    m_top.set(s_hist_version_ndx, RefOrTagged::make_tagged(unsigned(version))); // Throws
}

template <class Accessor>
inline void Group::set_history_parent(Accessor& history_root) noexcept
{
    history_root.set_parent(&m_top, 8);
}

template <class Accessor>
void Group::prepare_history_parent(Accessor& history_root, int history_type, int history_schema_version,
                                   uint64_t file_ident)
{
    prepare_top_for_history(history_type, history_schema_version, file_ident);
    set_history_parent(history_root);
}

class Group::TableWriter {
public:
    struct HistoryInfo {
        ref_type ref = 0;
        int type = 0;
        int version = 0;
        uint64_t sync_file_id = 0;
    };

    virtual ref_type write_names(_impl::OutputStream&) = 0;
    virtual ref_type write_tables(_impl::OutputStream&) = 0;
    virtual HistoryInfo write_history(_impl::OutputStream&) = 0;
    virtual ~TableWriter() noexcept {}
};

inline const Table* Group::do_get_table(size_t ndx) const
{
    return const_cast<Group*>(this)->do_get_table(ndx); // Throws
}

inline const Table* Group::do_get_table(StringData name) const
{
    return const_cast<Group*>(this)->do_get_table(name); // Throws
}

inline void Group::reset_free_space_tracking()
{
    // if used whith a shared allocator, free space should never be reset through
    // Group, but rather through the proper owner of the allocator, which is the DB object.
    REALM_ASSERT(m_local_alloc);
    m_alloc.reset_free_space_tracking(); // Throws
}

inline std::shared_ptr<metrics::Metrics> Group::get_metrics() const noexcept
{
    return m_metrics;
}

inline void Group::set_metrics(std::shared_ptr<metrics::Metrics> shared) noexcept
{
    m_metrics = shared;
}

// The purpose of this class is to give internal access to some, but
// not all of the non-public parts of the Group class.
class _impl::GroupFriend {
public:
    static Allocator& get_alloc(const Group& group) noexcept
    {
        return group.m_alloc;
    }

    static ref_type get_top_ref(const Group& group) noexcept
    {
        return group.m_top.get_ref();
    }

    static ref_type get_history_ref(Allocator& alloc, ref_type top_ref) noexcept
    {
        Array top(alloc);
        if (top_ref != 0)
            top.init_from_ref(top_ref);
        return Group::get_history_ref(top);
    }

    static ref_type get_history_ref(const Group& group) noexcept
    {
        return Group::get_history_ref(group.m_top);
    }

    static int get_file_format_version(const Group& group) noexcept
    {
        return group.get_file_format_version();
    }

    static void get_version_and_history_info(const Allocator& alloc, ref_type top_ref,
                                             _impl::History::version_type& version, int& history_type,
                                             int& history_schema_version) noexcept
    {
        Array top{const_cast<Allocator&>(alloc)};
        if (top_ref != 0)
            top.init_from_ref(top_ref);
        Group::get_version_and_history_info(top, version, history_type, history_schema_version);
    }

    static void set_history_schema_version(Group& group, int version)
    {
        group.set_history_schema_version(version); // Throws
    }

    template <class Accessor>
    static void set_history_parent(Group& group, Accessor& history_root) noexcept
    {
        group.set_history_parent(history_root);
    }

    template <class Accessor>
    static void prepare_history_parent(Group& group, Accessor& history_root, int history_type,
                                       int history_schema_version, uint64_t file_ident = 0)
    {
        group.prepare_history_parent(history_root, history_type, history_schema_version, file_ident); // Throws
    }

    // This is used by upgrade functions in Sync
    static Table* get_table_by_ndx(Group& group, size_t ndx)
    {
        return group.do_get_table(ndx);
    }

    static int get_target_file_format_version_for_session(int current_file_format_version, int history_type) noexcept
    {
        return Group::get_target_file_format_version_for_session(current_file_format_version, history_type);
    }

    static void fake_target_file_format(const std::optional<int> format) noexcept;
};


class CascadeState {
public:
    enum class Mode {
        /// If we remove the last link to an object, delete that object, even if
        /// the link we removed was not a strong link
        All,
        /// If we remove the last link to an object, delete that object only if
        /// the link we removed was a strong link
        Strong,
        /// Never delete objects due to removing links
        None
    };

    struct Link {
        TableKey origin_table; ///< A group-level table.
        ColKey origin_col_key; ///< Link column being nullified.
        ObjKey origin_key;     ///< Row in column being nullified.
        /// The target row index which is being removed. Mostly relevant for
        /// LinkList (to know which entries are being removed), but also
        /// valid for Link.
        ObjLink old_target_link;
    };

    CascadeState(Mode mode = Mode::Strong, Group* g = nullptr) noexcept
        : m_mode(mode)
        , m_group(g)
    {
    }

    /// Indicate which links to take action on. Either all, strong or none.
    Mode m_mode;

    std::vector<std::pair<TableKey, ObjKey>> m_to_be_deleted;
    std::vector<Link> m_to_be_nullified;
    Group* m_group = nullptr;

    bool notification_handler() const noexcept
    {
        return m_group && m_group->has_cascade_notification_handler();
    }

    void send_notifications(Group::CascadeNotification& notifications) const
    {
        REALM_ASSERT_DEBUG(notification_handler());
        m_group->send_cascade_notification(notifications);
    }

    bool enqueue_for_cascade(const Obj& target_obj, bool link_is_strong, bool last_removed)
    {
        // Check if the object should be cascade deleted
        if (m_mode == Mode::None || !last_removed) {
            return false;
        }
        if (m_mode == Mode::All || link_is_strong) {
            bool has_backlinks = target_obj.has_backlinks(m_mode == Mode::Strong);
            if (!has_backlinks) {
                // Object has no more backlinks - add to list for deletion
                m_to_be_deleted.emplace_back(target_obj.get_table()->get_key(), target_obj.get_key());
                return true;
            }
        }
        return false;
    }

    void enqueue_for_nullification(Table& src_table, ColKey src_col_key, ObjKey origin_key, ObjLink target_link)
    {
        // Nullify immediately if we don't need to send cascade notifications
        if (!notification_handler()) {
            Obj obj = src_table.get_object(origin_key);
            obj.nullify_link(src_col_key, target_link);
            return;
        }

        // Otherwise enqueue it
        m_to_be_nullified.push_back({src_table.get_key(), src_col_key, origin_key, target_link});
    }

    void send_notifications()
    {
        if (!notification_handler()) {
            return;
        }
        Group::CascadeNotification notification;
        for (auto& o : m_to_be_deleted)
            notification.rows.emplace_back(o.first, o.second);
        for (auto& l : m_to_be_nullified)
            notification.links.emplace_back(l.origin_table, l.origin_col_key, l.origin_key,
                                            l.old_target_link.get_obj_key());
        send_notifications(notification);
    }
};

} // namespace realm

#endif // REALM_GROUP_HPP