path: root/src/libs/solutions/tasking/tasktree.cpp

// Copyright (C) 2023 The Qt Company Ltd.
// SPDX-License-Identifier: LicenseRef-Qt-Commercial OR GPL-3.0-only WITH Qt-GPL-exception-1.0

#include "tasktree.h"

#include "barrier.h"

#include <QDebug>
#include <QEventLoop>
#include <QFutureWatcher>
#include <QHash>
#include <QMetaEnum>
#include <QMutex>
#include <QPromise>
#include <QPointer>
#include <QSet>
#include <QTime>
#include <QTimer>

using namespace std::chrono;

namespace Tasking {

// That's cut down qtcassert.{c,h} to avoid the dependency.
#define QT_STRING(cond) qDebug("SOFT ASSERT: \"%s\" in %s: %s", cond,  __FILE__, QT_STRINGIFY(__LINE__))
#define QT_ASSERT(cond, action) if (Q_LIKELY(cond)) {} else { QT_STRING(#cond); action; } do {} while (0)
#define QT_CHECK(cond) if (cond) {} else { QT_STRING(#cond); } do {} while (0)

class Guard
{
    Q_DISABLE_COPY(Guard)
public:
    Guard() = default;
    ~Guard() { QT_CHECK(m_lockCount == 0); }
    bool isLocked() const { return m_lockCount; }
private:
    int m_lockCount = 0;
    friend class GuardLocker;
};

class GuardLocker
{
    Q_DISABLE_COPY(GuardLocker)
public:
    GuardLocker(Guard &guard) : m_guard(guard) { ++m_guard.m_lockCount; }
    ~GuardLocker() { --m_guard.m_lockCount; }
private:
    Guard &m_guard;
};

/*!
    \module TaskingSolution
    \title Tasking Solution
    \ingroup solutions-modules
    \brief Contains a general purpose Tasking solution.

    The Tasking solution depends on Qt only, and doesn't depend on any \QC specific code.
*/

/*!
    \namespace Tasking
    \inmodule TaskingSolution
    \brief The Tasking namespace encloses all classes and global functions of the Tasking solution.
*/

/*!
    \class Tasking::TaskInterface
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief TaskInterface is the abstract base class for implementing custom task adapters.
    \reentrant

    To implement a custom task adapter, derive your adapter from the
    \c TaskAdapter<Task> class template. TaskAdapter automatically creates and destroys
    the custom task instance and associates the adapter with a given \c Task type.
*/

/*!
    \fn virtual void TaskInterface::start()

    This method is called by the running TaskTree for starting the \c Task instance.
    Reimplement this method in \c TaskAdapter<Task>'s subclass in order to start the
    associated task.

    Use TaskAdapter::task() to access the associated \c Task instance.

    \sa done(), TaskAdapter::task()
*/

/*!
    \fn void TaskInterface::done(DoneResult result)

    Emit this signal from the \c TaskAdapter<Task>'s subclass, when the \c Task is finished.
    Pass DoneResult::Success as a \a result argument when the task finishes with success;
    otherwise, when an error occurs, pass DoneResult::Error.
*/

/*!
    \class Tasking::TaskAdapter
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief A class template for implementing custom task adapters.
    \reentrant

    The TaskAdapter class template is responsible for creating a task of the \c Task type,
    starting it, and reporting success or an error when the task is finished.
    It also associates the adapter with a given \c Task type.

    Reimplement this class with the actual \c Task type to adapt the task's interface
    into the general TaskTree's interface for managing the \c Task instances.

    Each subclass needs to provide a public default constructor,
    implement the start() method, and emit the done() signal when the task is finished.
    Use task() to access the associated \c Task instance.

    To use your task adapter inside the task tree, create an alias to the
    Tasking::CustomTask template passing your task adapter as a template parameter:
    \code
        // Defines actual worker
        class Worker {...};

        // Adapts Worker's interface to work with task tree
        class WorkerTaskAdapter : public TaskAdapter<Worker> {...};

        // Defines WorkerTask as a new custom task type to be placed inside Group items
        using WorkerTask = CustomTask<WorkerTaskAdapter>;
    \endcode

    Optionally, you may pass a custom \c Deleter for the associated \c Task
    as a second template parameter of your \c TaskAdapter subclass.
    When the \c Deleter parameter is omitted, the \c std::default_delete<Task> is used by default.
    The custom \c Deleter is useful when the destructor of the running \c Task
    may potentially block the caller thread. Instead of blocking, the custom deleter may move
    the running task into a separate thread and implement the blocking destruction there.
    In this way, the fast destruction (seen from the caller thread) of the running task
    with a blocking destructor may be achieved.

    For more information on implementing the custom task adapters, refer to \l {Task Adapters}.

    \sa start(), done(), task()
*/

/*!
    \fn template <typename Task, typename Deleter = std::default_delete<Task>> TaskAdapter<Task, Deleter>::TaskAdapter<Task, Deleter>()

    Creates a task adapter for the given \c Task type.

    Internally, it creates an instance of \c Task, which is accessible via the task() method.
    The optionally provided \c Deleter is used instead of the \c Task destructor.
    When \c Deleter is omitted, the \c std::default_delete<Task> is used by default.

    \sa task()
*/

/*!
    \fn template <typename Task, typename Deleter = std::default_delete<Task>> Task *TaskAdapter<Task, Deleter>::task()

    Returns the pointer to the associated \c Task instance.
*/

/*!
    \fn template <typename Task, typename Deleter = std::default_delete<Task>> Task *TaskAdapter<Task, Deleter>::task() const
    \overload

    Returns the \c const pointer to the associated \c Task instance.
*/

/*!
    \class Tasking::Storage
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief A class template for custom data exchange in the running task tree.
    \reentrant

    The Storage class template is responsible for dynamically creating and destructing objects
    of the custom \c StorageStruct type. The creation and destruction are managed by the
    running task tree. If a Storage object is placed inside a \l {Tasking::Group} {Group} element,
    the running task tree creates the \c StorageStruct object when the group is started and before
    the group's setup handler is called. Later, whenever any handler inside this group is called,
    the task tree activates the previously created instance of the \c StorageStruct object.
    This includes all tasks' and groups' setup and done handlers inside the group where the
    Storage object was placed, also within the nested groups.
    When a copy of the Storage object is passed to the handler via the lambda capture,
    the handler may access the instance activated by the running task tree via the
    \l {Tasking::Storage::operator->()} {operator->()},
    \l {Tasking::Storage::operator*()} {operator*()}, or activeStorage() method.
    If two handlers capture the same Storage object, one of them may store a custom data there,
    and the other may read it afterwards.
    When the group is finished, the previously created instance of the \c StorageStruct
    object is destroyed after the group's done handler is called.

    An example of data exchange between tasks:

    \code
        const Storage<QString> storage;

        const auto onFirstDone = [storage](const Task &task) {
            // Assings QString, taken from the first task result, to the active QString instance
            // of the Storage object.
            *storage = task.getResultAsString();
        };

        const auto onSecondSetup = [storage](Task &task) {
            // Reads QString from the active QString instance of the Storage object and use it to
            // configure the second task before start.
            task.configureWithString(*storage);
        };

        const Group root {
            // The running task tree creates QString instance when root in entered
            storage,
            // The done handler of the first task stores the QString in the storage
            TaskItem(..., onFirstDone),
            // The setup handler of the second task reads the QString from the storage
            TaskItem(onSecondSetup, ...)
        };
    \endcode

    Since the root group executes its tasks sequentially, the \c onFirstDone handler
    is always called before the \c onSecondSetup handler. This means that the QString data,
    read from the \c storage inside the \c onSecondSetup handler's body,
    has already been set by the \c onFirstDone handler.
    You can always rely on it in \l {Tasking::sequential} {sequential} execution mode.

    The Storage internals are shared between all of its copies. That is why the copies of the
    Storage object inside the handlers' lambda captures still refer to the same Storage instance.
    You may place multiple Storage objects inside one \l {Tasking::Group} {Group} element,
    provided that they do not include copies of the same Storage object.
    Otherwise, an assert is triggered at runtime that includes an error message.
    However, you can place copies of the same Storage object in different
    \l {Tasking::Group} {Group} elements of the same recipe. In this case, the running task
    tree will create multiple instances of the \c StorageStruct objects (one for each copy)
    and storage shadowing will take place. Storage shadowing works in a similar way
    to C++ variable shadowing inside the nested blocks of code:

    \code
        Storage<QString> storage;

        const Group root {
            storage,                             // Top copy, 1st instance of StorageStruct
            onGroupSetup([storage] { ... }),     // Top copy is active
            Group {
                storage,                         // Nested copy, 2nd instance of StorageStruct,
                                                 // shadows Top copy
                onGroupSetup([storage] { ... }), // Nested copy is active
            },
            Group {
                onGroupSetup([storage] { ... }), // Top copy is active
            }
        };
    \endcode

    The Storage objects may also be used for passing the initial data to the executed task tree,
    and for reading the final data out of the task tree before it finishes.
    To do this, use \l {TaskTree::onStorageSetup()} {onStorageSetup()} or
    \l {TaskTree::onStorageDone()} {onStorageDone()}, respectively.

    \note If you use an unreachable Storage object inside the handler,
          because you forgot to place the storage in the recipe,
          or placed it, but not in any handler's ancestor group,
          you may expect a crash, preceded by the following message:
          \e {The referenced storage is not reachable in the running tree.
          A nullptr will be returned which might lead to a crash in the calling code.
          It is possible that no storage was added to the tree,
          or the storage is not reachable from where it is referenced.}
*/

/*!
    \fn template <typename StorageStruct> Storage<StorageStruct>::Storage<StorageStruct>()

    Creates a storage for the given \c StorageStruct type.

    \note All copies of \c this object are considered to be the same Storage instance.
*/

/*!
    \fn template <typename StorageStruct> StorageStruct &Storage<StorageStruct>::operator*() const noexcept

    Returns a \e reference to the active \c StorageStruct object, created by the running task tree.
    Use this function only from inside the handler body of any GroupItem element placed
    in the recipe, otherwise you may expect a crash.
    Make sure that Storage is placed in any group ancestor of the handler's group item.

    \note The returned reference is valid as long as the group that created this instance
          is still running.

    \sa activeStorage(), operator->()
*/

/*!
    \fn template <typename StorageStruct> StorageStruct *Storage<StorageStruct>::operator->() const noexcept

    Returns a \e pointer to the active \c StorageStruct object, created by the running task tree.
    Use this function only from inside the handler body of any GroupItem element placed
    in the recipe, otherwise you may expect a crash.
    Make sure that Storage is placed in any group ancestor of the handler's group item.

    \note The returned pointer is valid as long as the group that created this instance
          is still running.

    \sa activeStorage(), operator*()
*/

/*!
    \fn template <typename StorageStruct> StorageStruct *Storage<StorageStruct>::activeStorage() const

    Returns a \e pointer to the active \c StorageStruct object, created by the running task tree.
    Use this function only from inside the handler body of any GroupItem element placed
    in the recipe, otherwise you may expect a crash.
    Make sure that Storage is placed in any group ancestor of the handler's group item.

    \note The returned pointer is valid as long as the group that created this instance
          is still running.

    \sa operator->(), operator*()
*/

/*!
    \class Tasking::GroupItem
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief GroupItem represents the basic element that may be a part of any Group.
    \reentrant

    GroupItem is a basic element that may be a part of any \l {Tasking::Group} {Group}.
    It encapsulates the functionality provided by any GroupItem's subclass.
    It is a value type and it is safe to copy the GroupItem instance,
    even when it is originally created via the subclass' constructor.

    There are four main kinds of GroupItem:
    \table
    \header
        \li GroupItem Kind
        \li Brief Description
    \row
        \li \l CustomTask
        \li Defines asynchronous task type and task's start, done, and error handlers.
            Aliased with a unique task name, such as, \c ConcurrentCallTask<ResultType>
            or \c NetworkQueryTask. Asynchronous tasks are the main reason for using a task tree.
    \row
        \li \l {Tasking::Group} {Group}
        \li A container for other group items. Since the group is of the GroupItem type,
            it's possible to nest it inside another group. The group is seen by its parent
            as a single asynchronous task.
    \row
        \li GroupItem containing \l {Tasking::Storage} {Storage}
        \li Enables the child tasks of a group to exchange data. When GroupItem containing
            \l {Tasking::Storage} {Storage} is placed inside a group, the task tree instantiates
            the storage's data object just before the group is entered,
            and destroys it just after the group is left.
    \row
        \li Other group control items
        \li The items returned by \l {Tasking::parallelLimit()} {parallelLimit()} or
            \l {Tasking::workflowPolicy()} {workflowPolicy()} influence the group's behavior.
            The items returned by \l {Tasking::onGroupSetup()} {onGroupSetup()} or
            \l {Tasking::onGroupDone()} {onGroupDone()} define custom handlers called when
            the group starts or ends execution.
    \endtable
*/

/*!
    \fn template <typename StorageStruct> GroupItem::GroupItem(const Storage<StorageStruct> &storage)

    Constructs a \c GroupItem element holding the \a storage object.

    When the \l {Tasking::Group} {Group} element containing \e this GroupItem is entered
    by the running task tree, an instance of the \c StorageStruct is created dynamically.

    When that group is about to be left after its execution, the previously instantiated
    \c StorageStruct is deleted.

    The dynamically created instance of \c StorageStruct is accessible from inside any
    handler body of the parent \l {Tasking::Group} {Group} element,
    including nested groups and its tasks, via the
    \l {Tasking::Storage::operator->()} {Storage::operator->()},
    \l {Tasking::Storage::operator*()} {Storage::operator*()}, or Storage::activeStorage() method.

    \sa {Tasking::Storage} {Storage}
*/

/*!
    \fn GroupItem::GroupItem(const QList<GroupItem> &items)

    Constructs a \c GroupItem element with a given list of \a items.

    When this \c GroupItem element is parsed by the TaskTree, it is simply replaced with
    its \a items.

    This constructor is useful when constructing a \l {Tasking::Group} {Group} element with
    lists of \c GroupItem elements:

    \code
        static QList<GroupItems> getItems();

        ...

        const Group root {
            parallel,
            finishAllAndSuccess,
            getItems(), // OK, getItems() list is wrapped into a single GroupItem element
            onGroupSetup(...),
            onGroupDone(...)
        };
    \endcode

    If you want to create a subtree, use \l {Tasking::Group} {Group} instead.

    \note Don't confuse this \c GroupItem with the \l {Tasking::Group} {Group} element, as
          \l {Tasking::Group} {Group} keeps its children nested
          after being parsed by the task tree, while this \c GroupItem does not.

    \sa {Tasking::Group} {Group}
*/

/*!
    \fn GroupItem::GroupItem(std::initializer_list<GroupItem> items)
    \overload
    \sa GroupItem(const QList<Tasking::GroupItem> &items)
*/

/*!
    \class Tasking::Group
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief Group represents the basic element for composing declarative recipes describing
           how to execute and handle a nested tree of asynchronous tasks.
    \reentrant

    Group is a container for other group items. It encloses child tasks into one unit,
    which is seen by the group's parent as a single, asynchronous task.
    Since Group is of the GroupItem type, it may also be a child of Group.

    Insert child tasks into the group by using aliased custom task names, such as,
    \c ConcurrentCallTask<ResultType> or \c NetworkQueryTask:

    \code
        const Group group {
            NetworkQueryTask(...),
            ConcurrentCallTask<int>(...)
        };
    \endcode

    The group's behavior may be customized by inserting the items returned by
    \l {Tasking::parallelLimit()} {parallelLimit()} or
    \l {Tasking::workflowPolicy()} {workflowPolicy()} functions:

    \code
        const Group group {
            parallel,
            continueOnError,
            NetworkQueryTask(...),
            NetworkQueryTask(...)
        };
    \endcode

    The group may contain nested groups:

    \code
        const Group group {
            finishAllAndSuccess,
            NetworkQueryTask(...),
            Group {
                NetworkQueryTask(...),
                Group {
                    parallel,
                    NetworkQueryTask(...),
                    NetworkQueryTask(...),
                }
                ConcurrentCallTask<QString>(...)
            }
        };
    \endcode

    The group may dynamically instantiate a custom storage structure, which may be used for
    inter-task data exchange:

    \code
        struct MyCustomStruct { QByteArray data; };

        Storage<MyCustomStruct> storage;

        const auto onFirstSetup = [](NetworkQuery &task) { ... };
        const auto onFirstDone = [storage](const NetworkQuery &task) {
            // storage-> gives a pointer to MyCustomStruct instance,
            // created dynamically by the running task tree.
            storage->data = task.reply()->readAll();
        };
        const auto onSecondSetup = [storage](ConcurrentCall<QImage> &task) {
            // storage-> gives a pointer to MyCustomStruct. Since the group is sequential,
            // the stored MyCustomStruct was already updated inside the onFirstDone handler.
            const QByteArray storedData = storage->data;
        };

        const Group group {
            // When the group is entered by a running task tree, it creates MyCustomStruct
            // instance dynamically. It is later accessible from all handlers via
            // the *storage or storage-> operators.
            sequential,
            storage,
            NetworkQueryTask(onFirstSetup, onFirstDone, CallDoneIf::Success),
            ConcurrentCallTask<QImage>(onSecondSetup)
        };
    \endcode
*/

/*!
    \fn Group::Group(const QList<GroupItem> &children)

    Constructs a group with a given list of \a children.

    This constructor is useful when the child items of the group are not known at compile time,
    but later, at runtime:

    \code
        const QStringList sourceList = ...;

        QList<GroupItem> groupItems { parallel };

        for (const QString &source : sourceList) {
            const NetworkQueryTask task(...); // use source for setup handler
            groupItems << task;
        }

        const Group group(groupItems);
    \endcode
*/

/*!
    \fn Group::Group(std::initializer_list<GroupItem> children)

    Constructs a group from \c std::initializer_list given by \a children.

    This constructor is useful when all child items of the group are known at compile time:

    \code
        const Group group {
            finishAllAndSuccess,
            NetworkQueryTask(...),
            Group {
                NetworkQueryTask(...),
                Group {
                    parallel,
                    NetworkQueryTask(...),
                    NetworkQueryTask(...),
                }
                ConcurrentCallTask<QString>(...)
            }
        };
    \endcode
*/

/*!
    \class Tasking::Sync
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief Synchronously executes a custom handler between other tasks.
    \reentrant

    \c Sync is useful when you want to execute an additional handler between other tasks.
    \c Sync is seen by its parent \l {Tasking::Group} {Group} as any other task.
    Avoid long-running execution of the \c Sync's handler body, since it is executed
    synchronously from the caller thread. If that is unavoidable, consider using
    \c ConcurrentCallTask instead.
*/

/*!
    \fn template <typename Handler> Sync::Sync(Handler &&handler)

    Constructs an element that executes a passed \a handler synchronously.
    The \c Handler is of the \c std::function<DoneResult()> type.
    The DoneResult value, returned by the \a handler, is considered during parent group's
    \l {workflowPolicy} {workflow policy} resolution.
    Optionally, the shortened form of \c std::function<void()> is also accepted.
    In this case, it's assumed that the return value is DoneResult::Success.

    The passed \a handler executes synchronously from the caller thread, so avoid a long-running
    execution of the handler body. Otherwise, consider using \c ConcurrentCallTask.

    \note The \c Sync element is not counted as a task when reporting task tree progress,
          and is not included in TaskTree::taskCount() or TaskTree::progressMaximum().
*/

/*!
    \class Tasking::CustomTask
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief A class template used for declaring custom task items and defining their setup
           and done handlers.
    \reentrant

    Describes custom task items within task tree recipes.

    Custom task names are aliased with unique names using the \c CustomTask template
    with a given TaskAdapter subclass as a template parameter.
    For example, \c ConcurrentCallTask<T> is an alias to the \c CustomTask that is defined
    to work with \c ConcurrentCall<T> as an associated task class.
    The following table contains example custom tasks and their associated task classes:

    \table
    \header
        \li Aliased Task Name (Tasking Namespace)
        \li Associated Task Class
        \li Brief Description
    \row
        \li ConcurrentCallTask<ReturnType>
        \li ConcurrentCall<ReturnType>
        \li Starts an asynchronous task. Runs in a separate thread.
    \row
        \li NetworkQueryTask
        \li NetworkQuery
        \li Sends a network query.
    \row
        \li TaskTreeTask
        \li TaskTree
        \li Starts a nested task tree.
    \row
        \li TimeoutTask
        \li \c std::chrono::milliseconds
        \li Starts a timer.
    \row
        \li WaitForBarrierTask
        \li MultiBarrier<Limit>
        \li Starts an asynchronous task waiting for the barrier to pass.
    \endtable
*/

/*!
    \typealias CustomTask::Task

    Type alias for the task type associated with the custom task's \c Adapter.
*/

/*!
    \typealias CustomTask::Deleter

    Type alias for the task's type deleter associated with the custom task's \c Adapter.
*/

/*!
    \typealias CustomTask::TaskSetupHandler

    Type alias for \c std::function<SetupResult(Task &)>.

    The \c TaskSetupHandler is an optional argument of a custom task element's constructor.
    Any function with the above signature, when passed as a task setup handler,
    will be called by the running task tree after the task is created and before it is started.

    Inside the body of the handler, you may configure the task according to your needs.
    The additional parameters, including storages, may be passed to the handler
    via the lambda capture.
    You can decide dynamically whether the task should be started or skipped with
    success or an error.

    \note Do not start the task inside the start handler by yourself. Leave it for TaskTree,
    otherwise the behavior is undefined.

    The return value of the handler instructs the running task tree on how to proceed
    after the handler's invocation is finished. The return value of SetupResult::Continue
    instructs the task tree to continue running, that is, to execute the associated \c Task.
    The return value of SetupResult::StopWithSuccess or SetupResult::StopWithError
    instructs the task tree to skip the task's execution and finish it immediately with
    success or an error, respectively.

    When the return type is either SetupResult::StopWithSuccess or SetupResult::StopWithError,
    the task's done handler (if provided) isn't called afterwards.

    The constructor of a custom task accepts also functions in the shortened form of
    \c std::function<void(Task &)>, that is, the return value is \c void.
    In this case, it's assumed that the return value is SetupResult::Continue.

    \sa CustomTask(), TaskDoneHandler, GroupSetupHandler
*/

/*!
    \typealias CustomTask::TaskDoneHandler

    Type alias for \c std::function<DoneResult(const Task &, DoneWith)>.

    The \c TaskDoneHandler is an optional argument of a custom task element's constructor.
    Any function with the above signature, when passed as a task done handler,
    will be called by the running task tree after the task execution finished and before
    the final result of the execution is reported back to the parent group.

    Inside the body of the handler you may retrieve the final data from the finished task.
    The additional parameters, including storages, may be passed to the handler
    via the lambda capture.
    It is also possible to decide dynamically whether the task should finish with its return
    value, or the final result should be tweaked.

    The DoneWith argument is optional and your done handler may omit it.
    When provided, it holds the info about the final result of a task that will be
    reported to its parent.

    If you do not plan to read any data from the finished task,
    you may omit the \c {const Task &} argument.

    The returned DoneResult value is optional and your handler may return \c void instead.
    In this case, the final result of the task will be equal to the value indicated by
    the DoneWith argument. When the handler returns the DoneResult value,
    the task's final result may be tweaked inside the done handler's body by the returned value.

    \sa CustomTask(), TaskSetupHandler, GroupDoneHandler
*/

/*!
    \fn template <typename Adapter> template <typename SetupHandler = TaskSetupHandler, typename DoneHandler = TaskDoneHandler> CustomTask<Adapter>::CustomTask(SetupHandler &&setup = TaskSetupHandler(), DoneHandler &&done = TaskDoneHandler(), CallDoneIf callDoneIf = CallDoneIf::SuccessOrError)

    Constructs a \c CustomTask instance and attaches the \a setup and \a done handlers to the task.
    When the running task tree is about to start the task,
    it instantiates the associated \l Task object, invokes \a setup handler with a \e reference
    to the created task, and starts it. When the running task finishes,
    the task tree invokes a \a done handler, with a \c const \e reference to the created task.

    The passed \a setup handler is of the \l TaskSetupHandler type. For example:

    \code
        static void parseAndLog(const QString &input);

        ...

        const QString input = ...;

        const auto onFirstSetup = [input](ConcurrentCall<void> &task) {
            if (input == "Skip")
                return SetupResult::StopWithSuccess; // This task won't start, the next one will
            if (input == "Error")
                return SetupResult::StopWithError; // This task and the next one won't start
            task.setConcurrentCallData(parseAndLog, input);
            // This task will start, and the next one will start after this one finished with success
            return SetupResult::Continue;
        };

        const auto onSecondSetup = [input](ConcurrentCall<void> &task) {
            task.setConcurrentCallData(parseAndLog, input);
        };

        const Group group {
            ConcurrentCallTask<void>(onFirstSetup),
            ConcurrentCallTask<void>(onSecondSetup)
        };
    \endcode

    The \a done handler is of the \l TaskDoneHandler type.
    By default, the \a done handler is invoked whenever the task finishes.
    Pass a non-default value for the \a callDoneIf argument when you want the handler to be called
    only on a successful or failed execution.

    \sa TaskSetupHandler, TaskDoneHandler
*/

/*!
    \enum Tasking::WorkflowPolicy

    This enum describes the possible behavior of the Group element when any group's child task
    finishes its execution. It's also used when the running Group is canceled.

    \value StopOnError
        Default. Corresponds to the stopOnError global element.
        If any child task finishes with an error, the group stops and finishes with an error.
        If all child tasks finished with success, the group finishes with success.
        If a group is empty, it finishes with success.
    \value ContinueOnError
        Corresponds to the continueOnError global element.
        Similar to stopOnError, but in case any child finishes with an error,
        the execution continues until all tasks finish, and the group reports an error
        afterwards, even when some other tasks in the group finished with success.
        If all child tasks finish successfully, the group finishes with success.
        If a group is empty, it finishes with success.
    \value StopOnSuccess
        Corresponds to the stopOnSuccess global element.
        If any child task finishes with success, the group stops and finishes with success.
        If all child tasks finished with an error, the group finishes with an error.
        If a group is empty, it finishes with an error.
    \value ContinueOnSuccess
        Corresponds to the continueOnSuccess global element.
        Similar to stopOnSuccess, but in case any child finishes successfully,
        the execution continues until all tasks finish, and the group reports success
        afterwards, even when some other tasks in the group finished with an error.
        If all child tasks finish with an error, the group finishes with an error.
        If a group is empty, it finishes with an error.
    \value StopOnSuccessOrError
        Corresponds to the stopOnSuccessOrError global element.
        The group starts as many tasks as it can. When any task finishes,
        the group stops and reports the task's result.
        Useful only in parallel mode.
        In sequential mode, only the first task is started, and when finished,
        the group finishes too, so the other tasks are always skipped.
        If a group is empty, it finishes with an error.
    \value FinishAllAndSuccess
        Corresponds to the finishAllAndSuccess global element.
        The group executes all tasks and ignores their return results. When all
        tasks finished, the group finishes with success.
        If a group is empty, it finishes with success.
    \value FinishAllAndError
        Corresponds to the finishAllAndError global element.
        The group executes all tasks and ignores their return results. When all
        tasks finished, the group finishes with an error.
        If a group is empty, it finishes with an error.

    Whenever a child task's result causes the Group to stop, that is,
    in case of StopOnError, StopOnSuccess, or StopOnSuccessOrError policies,
    the Group cancels the other running child tasks (if any - for example in parallel mode),
    and skips executing tasks it has not started yet (for example, in the sequential mode -
    those, that are placed after the failed task). Both canceling and skipping child tasks
    may happen when parallelLimit() is used.

    The table below summarizes the differences between various workflow policies:

    \table
    \header
        \li \l WorkflowPolicy
        \li Executes all child tasks
        \li Result
        \li Result when the group is empty
    \row
        \li StopOnError
        \li Stops when any child task finished with an error and reports an error
        \li An error when at least one child task failed, success otherwise
        \li Success
    \row
        \li ContinueOnError
        \li Yes
        \li An error when at least one child task failed, success otherwise
        \li Success
    \row
        \li StopOnSuccess
        \li Stops when any child task finished with success and reports success
        \li Success when at least one child task succeeded, an error otherwise
        \li An error
    \row
        \li ContinueOnSuccess
        \li Yes
        \li Success when at least one child task succeeded, an error otherwise
        \li An error
    \row
        \li StopOnSuccessOrError
        \li Stops when any child task finished and reports child task's result
        \li Success or an error, depending on the finished child task's result
        \li An error
    \row
        \li FinishAllAndSuccess
        \li Yes
        \li Success
        \li Success
    \row
        \li FinishAllAndError
        \li Yes
        \li An error
        \li An error
    \endtable

    If a child of a group is also a group, the child group runs its tasks according to its own
    workflow policy. When a parent group stops the running child group because
    of parent group's workflow policy, that is, when the StopOnError, StopOnSuccess,
    or StopOnSuccessOrError policy was used for the parent,
    the child group's result is reported according to the
    \b Result column and to the \b {child group's workflow policy} row in the table above.
*/

/*!
    \variable sequential
    A convenient global group's element describing the sequential execution mode.

    This is the default execution mode of the Group element.

    When a Group has no execution mode, it runs in the sequential mode.
    All the direct child tasks of a group are started in a chain, so that when one task finishes,
    the next one starts. This enables you to pass the results from the previous task
    as input to the next task before it starts. This mode guarantees that the next task
    is started only after the previous task finishes.

    \sa parallel, parallelLimit()
*/

/*!
    \variable parallel
    A convenient global group's element describing the parallel execution mode.

    All the direct child tasks of a group are started after the group is started,
    without waiting for the previous child tasks to finish.
    In this mode, all child tasks run simultaneously.

    \sa sequential, parallelLimit()
*/

/*!
    \variable parallelIdealThreadCountLimit
    A convenient global group's element describing the parallel execution mode with a limited
    number of tasks running simultanously. The limit is equal to the ideal number of threads
    excluding the calling thread.

    This is a shortcut to:
    \code
        parallelLimit(qMax(QThread::idealThreadCount() - 1, 1))
    \endcode

    \sa parallel, parallelLimit()
*/

/*!
    \variable stopOnError
    A convenient global group's element describing the StopOnError workflow policy.

    This is the default workflow policy of the Group element.
*/

/*!
    \variable continueOnError
    A convenient global group's element describing the ContinueOnError workflow policy.
*/

/*!
    \variable stopOnSuccess
    A convenient global group's element describing the StopOnSuccess workflow policy.
*/

/*!
    \variable continueOnSuccess
    A convenient global group's element describing the ContinueOnSuccess workflow policy.
*/

/*!
    \variable stopOnSuccessOrError
    A convenient global group's element describing the StopOnSuccessOrError workflow policy.
*/

/*!
    \variable finishAllAndSuccess
    A convenient global group's element describing the FinishAllAndSuccess workflow policy.
*/

/*!
    \variable finishAllAndError
    A convenient global group's element describing the FinishAllAndError workflow policy.
*/

/*!
    \enum Tasking::SetupResult

    This enum is optionally returned from the group's or task's setup handler function.
    It instructs the running task tree on how to proceed after the setup handler's execution
    finished.
    \value Continue
           Default. The group's or task's execution continues normally.
           When a group's or task's setup handler returns void, it's assumed that
           it returned Continue.
    \value StopWithSuccess
           The group's or task's execution stops immediately with success.
           When returned from the group's setup handler, all child tasks are skipped,
           and the group's onGroupDone() handler is invoked with DoneWith::Success.
           The group reports success to its parent. The group's workflow policy is ignored.
           When returned from the task's setup handler, the task isn't started,
           its done handler isn't invoked, and the task reports success to its parent.
    \value StopWithError
           The group's or task's execution stops immediately with an error.
           When returned from the group's setup handler, all child tasks are skipped,
           and the group's onGroupDone() handler is invoked with DoneWith::Error.
           The group reports an error to its parent. The group's workflow policy is ignored.
           When returned from the task's setup handler, the task isn't started,
           its error handler isn't invoked, and the task reports an error to its parent.
*/

/*!
    \enum Tasking::DoneResult

    This enum is optionally returned from the group's or task's done handler function.
    When the done handler doesn't return any value, that is, its return type is \c void,
    its final return value is automatically deduced by the running task tree and reported
    to its parent group.

    When the done handler returns the DoneResult, you can tweak the final return value
    inside the handler.

    When the DoneResult is returned by the group's done handler, the group's workflow policy
    is ignored.

    This enum is also used inside the TaskInterface::done() signal and it indicates whether
    the task finished with success or an error.

    \value Success
           The group's or task's execution ends with success.
    \value Error
           The group's or task's execution ends with an error.
*/

/*!
    \enum Tasking::DoneWith

    This enum is an optional argument for the group's or task's done handler.
    It indicates whether the group or task finished with success or an error, or it was canceled.

    It is also used as an argument inside the TaskTree::done() signal,
    indicating the final result of the TaskTree execution.

    \value Success
           The group's or task's execution ended with success.
    \value Error
           The group's or task's execution ended with an error.
    \value Cancel
           The group's or task's execution was canceled. This happens when the user calls
           TaskTree::cancel() for the running task tree or when the group's workflow policy
           results in canceling some of its running children.
           Tweaking the done handler's final result by returning Tasking::DoneResult from
           the handler is no-op when the group's or task's execution was canceled.
*/

/*!
    \enum Tasking::CallDoneIf

    This enum is an optional argument for the \l onGroupDone() element or custom task's constructor.
    It instructs the task tree on when the group's or task's done handler should be invoked.

    \value SuccessOrError
           The done handler is always invoked.
    \value Success
           The done handler is invoked only after successful execution,
           that is, when DoneWith::Success.
    \value Error
           The done handler is invoked only after failed execution,
           that is, when DoneWith::Error or when DoneWith::Cancel.
*/

/*!
    \typealias GroupItem::GroupSetupHandler

    Type alias for \c std::function<SetupResult()>.

    The \c GroupSetupHandler is an argument of the onGroupSetup() element.
    Any function with the above signature, when passed as a group setup handler,
    will be called by the running task tree when the group execution starts.

    The return value of the handler instructs the running group on how to proceed
    after the handler's invocation is finished. The default return value of SetupResult::Continue
    instructs the group to continue running, that is, to start executing its child tasks.
    The return value of SetupResult::StopWithSuccess or SetupResult::StopWithError
    instructs the group to skip the child tasks' execution and finish immediately with
    success or an error, respectively.

    When the return type is either SetupResult::StopWithSuccess or SetupResult::StopWithError,
    the group's done handler (if provided) is called synchronously immediately afterwards.

    \note Even if the group setup handler returns StopWithSuccess or StopWithError,
    the group's done handler is invoked. This behavior differs from that of task done handler
    and might change in the future.

    The onGroupSetup() element accepts also functions in the shortened form of
    \c std::function<void()>, that is, the return value is \c void.
    In this case, it's assumed that the return value is SetupResult::Continue.

    \sa onGroupSetup(), GroupDoneHandler, CustomTask::TaskSetupHandler
*/

/*!
    \typealias GroupItem::GroupDoneHandler

    Type alias for \c std::function<DoneResult(DoneWith)>.

    The \c GroupDoneHandler is an argument of the onGroupDone() element.
    Any function with the above signature, when passed as a group done handler,
    will be called by the running task tree when the group execution ends.

    The DoneWith argument is optional and your done handler may omit it.
    When provided, it holds the info about the final result of a group that will be
    reported to its parent.

    The returned DoneResult value is optional and your handler may return \c void instead.
    In this case, the final result of the group will be equal to the value indicated by
    the DoneWith argument. When the handler returns the DoneResult value,
    the group's final result may be tweaked inside the done handler's body by the returned value.

    \sa onGroupDone(), GroupSetupHandler, CustomTask::TaskDoneHandler
*/

/*!
    \fn template <typename Handler> GroupItem onGroupSetup(Handler &&handler)

    Constructs a group's element holding the group setup handler.
    The \a handler is invoked whenever the group starts.

    The passed \a handler is either of the \c std::function<SetupResult()> or the
    \c std::function<void()> type. For more information on a possible handler type, refer to
    \l {GroupItem::GroupSetupHandler}.

    When the \a handler is invoked, none of the group's child tasks are running yet.

    If a group contains the Storage elements, the \a handler is invoked
    after the storages are constructed, so that the \a handler may already
    perform some initial modifications to the active storages.

    \sa GroupItem::GroupSetupHandler, onGroupDone()
*/

/*!
    \fn template <typename Handler> GroupItem onGroupDone(Handler &&handler, CallDoneIf callDoneIf = CallDoneIf::SuccessOrError)

    Constructs a group's element holding the group done handler.
    By default, the \a handler is invoked whenever the group finishes.
    Pass a non-default value for the \a callDoneIf argument when you want the handler to be called
    only on a successful or failed execution.
    Depending on the group's workflow policy, this handler may also be called
    when the running group is canceled (e.g. when stopOnError element was used).

    The passed \a handler is of the \c std::function<DoneResult(DoneWith)> type.
    Optionally, each of the return DoneResult type or the argument DoneWith type may be omitted
    (that is, its return type may be \c void). For more information on a possible handler type,
    refer to \l {GroupItem::GroupDoneHandler}.

    When the \a handler is invoked, all of the group's child tasks are already finished.

    If a group contains the Storage elements, the \a handler is invoked
    before the storages are destructed, so that the \a handler may still
    perform a last read of the active storages' data.

    \sa GroupItem::GroupDoneHandler, onGroupSetup()
*/

/*!
    Constructs a group's element describing the \l{Execution Mode}{execution mode}.

    The execution mode element in a Group specifies how the direct child tasks of
    the Group are started.

    For convenience, when appropriate, the \l sequential or \l parallel global elements
    may be used instead.

    The \a limit defines the maximum number of direct child tasks running in parallel:

    \list
        \li When \a limit equals to 0, there is no limit, and all direct child tasks are started
        together, in the oder in which they appear in a group. This means the fully parallel
        execution, and the \l parallel element may be used instead.

        \li When \a limit equals to 1, it means that only one child task may run at the time.
        This means the sequential execution, and the \l sequential element may be used instead.
        In this case, child tasks run in chain, so the next child task starts after
        the previous child task has finished.

        \li When other positive number is passed as \a limit, the group's child tasks run
        in parallel, but with a limited number of tasks running simultanously.
        The \e limit defines the maximum number of tasks running in parallel in a group.
        When the group is started, the first batch of tasks is started
        (the number of tasks in a batch equals to the passed \a limit, at most),
        while the others are kept waiting. When any running task finishes,
        the group starts the next remaining one, so that the \e limit of simultaneously
        running tasks inside a group isn't exceeded. This repeats on every child task's
        finish until all child tasks are started. This enables you to limit the maximum
        number of tasks that run simultaneously, for example if running too many processes might
        block the machine for a long time.
    \endlist

    In all execution modes, a group starts tasks in the oder in which they appear.

    If a child of a group is also a group, the child group runs its tasks according
    to its own execution mode.

    \sa sequential, parallel
*/
GroupItem parallelLimit(int limit)
{
    return Group::parallelLimit(qMax(limit, 0));
}

/*!
    Constructs a group's \l {Workflow Policy} {workflow policy} element for a given \a policy.

    For convenience, global elements may be used instead.

    \sa stopOnError, continueOnError, stopOnSuccess, continueOnSuccess, stopOnSuccessOrError,
        finishAllAndSuccess, finishAllAndError, WorkflowPolicy
*/
GroupItem workflowPolicy(WorkflowPolicy policy)
{
    return Group::workflowPolicy(policy);
}

const GroupItem nullItem = GroupItem({});

const GroupItem sequential = parallelLimit(1);
const GroupItem parallel = parallelLimit(0);
const GroupItem parallelIdealThreadCountLimit = parallelLimit(qMax(QThread::idealThreadCount() - 1, 1));

const GroupItem stopOnError = workflowPolicy(WorkflowPolicy::StopOnError);
const GroupItem continueOnError = workflowPolicy(WorkflowPolicy::ContinueOnError);
const GroupItem stopOnSuccess = workflowPolicy(WorkflowPolicy::StopOnSuccess);
const GroupItem continueOnSuccess = workflowPolicy(WorkflowPolicy::ContinueOnSuccess);
const GroupItem stopOnSuccessOrError = workflowPolicy(WorkflowPolicy::StopOnSuccessOrError);
const GroupItem finishAllAndSuccess = workflowPolicy(WorkflowPolicy::FinishAllAndSuccess);
const GroupItem finishAllAndError = workflowPolicy(WorkflowPolicy::FinishAllAndError);

// Please note the thread_local keyword below guarantees a separate instance per thread.
// The s_activeTaskTrees is currently used internally only and is not exposed in the public API.
// It serves for withLog() implementation now. Add a note here when a new usage is introduced.
static thread_local QList<TaskTree *> s_activeTaskTrees = {};

static TaskTree *activeTaskTree()
{
    QT_ASSERT(s_activeTaskTrees.size(), return nullptr);
    return s_activeTaskTrees.back();
}

DoneResult toDoneResult(bool success)
{
    return success ? DoneResult::Success : DoneResult::Error;
}

static SetupResult toSetupResult(bool success)
{
    return success ? SetupResult::StopWithSuccess : SetupResult::StopWithError;
}

static DoneResult toDoneResult(DoneWith doneWith)
{
    return doneWith == DoneWith::Success ? DoneResult::Success : DoneResult::Error;
}

static DoneWith toDoneWith(DoneResult result)
{
    return result == DoneResult::Success ? DoneWith::Success : DoneWith::Error;
}

class LoopThreadData
{
    Q_DISABLE_COPY_MOVE(LoopThreadData)

public:
    LoopThreadData() = default;
    void pushIteration(int index)
    {
        m_activeLoopStack.push_back(index);
    }
    void popIteration()
    {
        QT_ASSERT(m_activeLoopStack.size(), return);
        m_activeLoopStack.pop_back();
    }
    int iteration() const
    {
        QT_ASSERT(m_activeLoopStack.size(), qWarning(
            "The referenced loop is not reachable in the running tree. "
            "A -1 will be returned which might lead to a crash in the calling code. "
            "It is possible that no loop was added to the tree, "
            "or the loop is not reachable from where it is referenced."); return -1);
        return m_activeLoopStack.last();
    }

private:
    QList<int> m_activeLoopStack;
};

class LoopData
{
public:
    LoopThreadData &threadData() {
        QMutexLocker lock(&m_threadDataMutex);
        return m_threadDataMap.try_emplace(QThread::currentThread()).first->second;
    }

    const std::optional<int> m_loopCount = {};
    const Loop::ValueGetter m_valueGetter = {};
    const Loop::Condition m_condition = {};
    QMutex m_threadDataMutex = {};
    // Use std::map on purpose, so that it doesn't invalidate references on modifications.
    // Don't optimize it by using std::unordered_map.
    std::map<QThread *, LoopThreadData> m_threadDataMap = {};
};

Loop::Loop()
    : m_loopData(new LoopData)
{}

Loop::Loop(int count, const ValueGetter &valueGetter)
    : m_loopData(new LoopData{count, valueGetter})
{}

Loop::Loop(const Condition &condition)
    : m_loopData(new LoopData{{}, {}, condition})
{}

int Loop::iteration() const
{
    return m_loopData->threadData().iteration();
}

const void *Loop::valuePtr() const
{
    return m_loopData->m_valueGetter(iteration());
}

using StoragePtr = void *;

class StorageThreadData
{
    Q_DISABLE_COPY_MOVE(StorageThreadData)

public:
    StorageThreadData() = default;
    void pushStorage(StoragePtr storagePtr)
    {
        m_activeStorageStack.push_back(storagePtr);
    }
    void popStorage()
    {
        QT_ASSERT(m_activeStorageStack.size(), return);
        m_activeStorageStack.pop_back();
    }
    StoragePtr activeStorage() const
    {
        QT_ASSERT(m_activeStorageStack.size(), qWarning(
            "The referenced storage is not reachable in the running tree. "
            "A nullptr will be returned which might lead to a crash in the calling code. "
            "It is possible that no storage was added to the tree, "
            "or the storage is not reachable from where it is referenced."); return nullptr);
        return m_activeStorageStack.last();
    }

private:
    QList<StoragePtr> m_activeStorageStack;
};

class StorageData
{
public:
    StorageThreadData &threadData() {
        QMutexLocker lock(&m_threadDataMutex);
        return m_threadDataMap.try_emplace(QThread::currentThread()).first->second;
    }

    const StorageBase::StorageConstructor m_constructor = {};
    const StorageBase::StorageDestructor m_destructor = {};
    QMutex m_threadDataMutex = {};
    // Use std::map on purpose, so that it doesn't invalidate references on modifications.
    // Don't optimize it by using std::unordered_map.
    std::map<QThread *, StorageThreadData> m_threadDataMap = {};
};

StorageBase::StorageBase(const StorageConstructor &ctor, const StorageDestructor &dtor)
    : m_storageData(new StorageData{ctor, dtor})
{}

void *StorageBase::activeStorageVoid() const
{
    return m_storageData->threadData().activeStorage();
}

void GroupItem::addChildren(const QList<GroupItem> &children)
{
    QT_ASSERT(m_type == Type::Group || m_type == Type::List,
              qWarning("Only Group or List may have children, skipping..."); return);
    if (m_type == Type::List) {
        m_children.append(children);
        return;
    }
    for (const GroupItem &child : children) {
        switch (child.m_type) {
        case Type::List:
            addChildren(child.m_children);
            break;
        case Type::Group:
            m_children.append(child);
            break;
        case Type::GroupData:
            if (child.m_groupData.m_groupHandler.m_setupHandler) {
                QT_ASSERT(!m_groupData.m_groupHandler.m_setupHandler,
                          qWarning("Group setup handler redefinition, overriding..."));
                m_groupData.m_groupHandler.m_setupHandler
                    = child.m_groupData.m_groupHandler.m_setupHandler;
            }
            if (child.m_groupData.m_groupHandler.m_doneHandler) {
                QT_ASSERT(!m_groupData.m_groupHandler.m_doneHandler,
                          qWarning("Group done handler redefinition, overriding..."));
                m_groupData.m_groupHandler.m_doneHandler
                    = child.m_groupData.m_groupHandler.m_doneHandler;
                m_groupData.m_groupHandler.m_callDoneIf
                    = child.m_groupData.m_groupHandler.m_callDoneIf;
            }
            if (child.m_groupData.m_parallelLimit) {
                QT_ASSERT(!m_groupData.m_parallelLimit,
                          qWarning("Group execution mode redefinition, overriding..."));
                m_groupData.m_parallelLimit = child.m_groupData.m_parallelLimit;
            }
            if (child.m_groupData.m_workflowPolicy) {
                QT_ASSERT(!m_groupData.m_workflowPolicy,
                          qWarning("Group workflow policy redefinition, overriding..."));
                m_groupData.m_workflowPolicy = child.m_groupData.m_workflowPolicy;
            }
            if (child.m_groupData.m_loop) {
                QT_ASSERT(!m_groupData.m_loop,
                          qWarning("Group loop redefinition, overriding..."));
                m_groupData.m_loop = child.m_groupData.m_loop;
            }
            break;
        case Type::TaskHandler:
            QT_ASSERT(child.m_taskHandler.m_createHandler,
                      qWarning("Task create handler can't be null, skipping..."); return);
            m_children.append(child);
            break;
        case Type::Storage:
            // Check for duplicates, as can't have the same storage twice on the same level.
            for (const StorageBase &storage : child.m_storageList) {
                if (m_storageList.contains(storage)) {
                    QT_ASSERT(false, qWarning("Can't add the same storage into one Group twice, "
                                              "skipping..."));
                    continue;
                }
                m_storageList.append(storage);
            }
            break;
        }
    }
}

/*!
    \class Tasking::ExecutableItem
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief Base class for executable task items.
    \reentrant

    \c ExecutableItem provides an additional interface for items containing executable tasks.
    Use withTimeout() to attach a timeout to a task.
    Use withLog() to include debugging information about the task startup and the execution result.
*/

/*!
    Attaches \c TimeoutTask to a copy of \c this ExecutableItem, elapsing after \a timeout
    in milliseconds, with an optionally provided timeout \a handler, and returns the coupled item.

    When the ExecutableItem finishes before \a timeout passes, the returned item finishes
    immediately with the task's result. Otherwise, \a handler is invoked (if provided),
    the task is canceled, and the returned item finishes with an error.
*/
ExecutableItem ExecutableItem::withTimeout(milliseconds timeout,
                                           const std::function<void()> &handler) const
{
    const auto onSetup = [timeout](milliseconds &timeoutData) { timeoutData = timeout; };
    return Group {
        parallel,
        stopOnSuccessOrError,
        Group {
            finishAllAndError,
            handler ? TimeoutTask(onSetup, [handler] { handler(); }, CallDoneIf::Success)
                    : TimeoutTask(onSetup)
        },
        *this
    };
}

static QString currentTime() { return QTime::currentTime().toString(Qt::ISODateWithMs); }

/*!
    Attaches a custom debug printout to a copy of \c this ExecutableItem,
    issued on task startup and after the task is finished, and returns the coupled item.

    The debug printout includes a timestamp of the event (start or finish)
    and \a logName to identify the specific task in the debug log.

    The finish printout contains the additional information whether the execution was
    synchronous or asynchronous, its result (the value described by the DoneWith enum),
    and the total execution time in milliseconds.
*/
ExecutableItem ExecutableItem::withLog(const QString &logName) const
{
    const auto header = [logName] {
        return QString("TASK TREE LOG [%1] \"%2\"").arg(currentTime(), logName);
    };
    struct LogStorage
    {
        time_point<system_clock, nanoseconds> start;
        int asyncCount = 0;
    };
    const Storage<LogStorage> storage;
    return Group {
        storage,
        onGroupSetup([storage, header] {
            storage->start = system_clock::now();
            storage->asyncCount = activeTaskTree()->asyncCount();
            qDebug().noquote() << header() << "started.";
        }),
        *this,
        onGroupDone([storage, header](DoneWith result) {
            const auto elapsed = duration_cast<milliseconds>(system_clock::now() - storage->start);
            const int asyncCountDiff = activeTaskTree()->asyncCount() - storage->asyncCount;
            QT_CHECK(asyncCountDiff >= 0);
            const QMetaEnum doneWithEnum = QMetaEnum::fromType<DoneWith>();
            const QString syncType = asyncCountDiff ? QString("asynchronously")
                                                    : QString("synchronously");
            qDebug().noquote().nospace() << header() << " finished " << syncType << " with "
                << doneWithEnum.valueToKey(int(result)) << " within " << elapsed.count() << "ms.";
        })
    };
}

ExecutableItem ExecutableItem::withCancelImpl(
    const std::function<void(QObject *, const std::function<void()> &)> &connectWrapper) const
{
    const auto onSetup = [connectWrapper](Barrier &barrier) {
        connectWrapper(&barrier, [barrierPtr = &barrier] { barrierPtr->advance(); });
    };
    return Group {
        parallel,
        stopOnSuccessOrError,
        Group {
            finishAllAndError,
            BarrierTask(onSetup)
        },
        *this
    };
}

class TaskTreePrivate;
class TaskNode;
class RuntimeContainer;
class RuntimeIteration;
class RuntimeTask;

class ExecutionContextActivator
{
public:
    ExecutionContextActivator(RuntimeIteration *iteration) {
        activateTaskTree(iteration);
        activateContext(iteration);
    }
    ExecutionContextActivator(RuntimeContainer *container) {
        activateTaskTree(container);
        activateContext(container);
    }
    ~ExecutionContextActivator() {
        for (int i = m_activeStorages.size() - 1; i >= 0; --i) // iterate in reverse order
            m_activeStorages[i].m_storageData->threadData().popStorage();
        for (int i = m_activeLoops.size() - 1; i >= 0; --i) // iterate in reverse order
            m_activeLoops[i].m_loopData->threadData().popIteration();
        QT_ASSERT(s_activeTaskTrees.size(), return);
        s_activeTaskTrees.pop_back();
    }

private:
    void activateTaskTree(RuntimeIteration *iteration);
    void activateTaskTree(RuntimeContainer *container);
    void activateContext(RuntimeIteration *iteration);
    void activateContext(RuntimeContainer *container);
    QList<Loop> m_activeLoops;
    QList<StorageBase> m_activeStorages;
};

class ContainerNode
{
    Q_DISABLE_COPY(ContainerNode)

public:
    ContainerNode(ContainerNode &&other) = default;
    ContainerNode(TaskTreePrivate *taskTreePrivate, const GroupItem &task);

    TaskTreePrivate *const m_taskTreePrivate = nullptr;

    const GroupItem::GroupHandler m_groupHandler;
    const int m_parallelLimit = 1;
    const WorkflowPolicy m_workflowPolicy = WorkflowPolicy::StopOnError;
    const std::optional<Loop> m_loop;
    const QList<StorageBase> m_storageList;
    std::vector<TaskNode> m_children;
    const int m_taskCount = 0;
};

class TaskNode
{
    Q_DISABLE_COPY(TaskNode)

public:
    TaskNode(TaskNode &&other) = default;
    TaskNode(TaskTreePrivate *taskTreePrivate, const GroupItem &task)
        : m_taskHandler(task.m_taskHandler)
        , m_container(taskTreePrivate, task)
    {}

    bool isTask() const { return bool(m_taskHandler.m_createHandler); }
    int taskCount() const { return isTask() ? 1 : m_container.m_taskCount; }

    const GroupItem::TaskHandler m_taskHandler;
    ContainerNode m_container;
};

class TaskTreePrivate
{
    Q_DISABLE_COPY_MOVE(TaskTreePrivate)

public:
    TaskTreePrivate(TaskTree *taskTree)
        : q(taskTree) {}

    void start();
    void stop();
    void bumpAsyncCount();
    void advanceProgress(int byValue);
    void emitDone(DoneWith result);
    void callSetupHandler(StorageBase storage, StoragePtr storagePtr) {
        callStorageHandler(storage, storagePtr, &StorageHandler::m_setupHandler);
    }
    void callDoneHandler(StorageBase storage, StoragePtr storagePtr) {
        callStorageHandler(storage, storagePtr, &StorageHandler::m_doneHandler);
    }
    struct StorageHandler {
        StorageBase::StorageHandler m_setupHandler = {};
        StorageBase::StorageHandler m_doneHandler = {};
    };
    typedef StorageBase::StorageHandler StorageHandler::*HandlerPtr; // ptr to class member
    void callStorageHandler(StorageBase storage, StoragePtr storagePtr, HandlerPtr ptr)
    {
        const auto it = m_storageHandlers.constFind(storage);
        if (it == m_storageHandlers.constEnd())
            return;
        const StorageHandler storageHandler = *it;
        if (storageHandler.*ptr) {
            GuardLocker locker(m_guard);
            (storageHandler.*ptr)(storagePtr);
        }
    }

    // Node related methods

    // If returned value != Continue, childDone() needs to be called in parent container (in caller)
    // in order to unwind properly.
    SetupResult start(RuntimeTask *node);
    void stop(RuntimeTask *node);
    bool invokeDoneHandler(RuntimeTask *node, DoneWith doneWith);

    // Container related methods

    SetupResult start(RuntimeContainer *container);
    SetupResult continueStart(RuntimeContainer *container, SetupResult startAction);
    SetupResult startChildren(RuntimeContainer *container);
    SetupResult childDone(RuntimeIteration *iteration, bool success);
    void stop(RuntimeContainer *container);
    bool invokeDoneHandler(RuntimeContainer *container, DoneWith doneWith);
    bool invokeLoopHandler(RuntimeContainer *container);

    template <typename Container, typename Handler, typename ...Args,
              typename ReturnType = std::invoke_result_t<Handler, Args...>>
    ReturnType invokeHandler(Container *container, Handler &&handler, Args &&...args)
    {
        ExecutionContextActivator activator(container);
        GuardLocker locker(m_guard);
        return std::invoke(std::forward<Handler>(handler), std::forward<Args>(args)...);
    }

    static int effectiveLoopCount(const std::optional<Loop> &loop)
    {
        return loop && loop->m_loopData->m_loopCount ? *loop->m_loopData->m_loopCount : 1;
    }

    TaskTree *q = nullptr;
    Guard m_guard;
    int m_progressValue = 0;
    int m_asyncCount = 0;
    QSet<StorageBase> m_storages;
    QHash<StorageBase, StorageHandler> m_storageHandlers;
    std::optional<TaskNode> m_root;
    std::unique_ptr<RuntimeTask> m_runtimeRoot; // Keep me last in order to destruct first
};

static bool initialSuccessBit(WorkflowPolicy workflowPolicy)
{
    switch (workflowPolicy) {
    case WorkflowPolicy::StopOnError:
    case WorkflowPolicy::ContinueOnError:
    case WorkflowPolicy::FinishAllAndSuccess:
        return true;
    case WorkflowPolicy::StopOnSuccess:
    case WorkflowPolicy::ContinueOnSuccess:
    case WorkflowPolicy::StopOnSuccessOrError:
    case WorkflowPolicy::FinishAllAndError:
        return false;
    }
    QT_CHECK(false);
    return false;
}

static bool isProgressive(RuntimeContainer *container);

class RuntimeIteration
{
    Q_DISABLE_COPY(RuntimeIteration)

public:
    RuntimeIteration(int index, RuntimeContainer *container);
    std::optional<Loop> loop() const;
    void deleteChild(RuntimeTask *node);

    const int m_iterationIndex = 0;
    const bool m_isProgressive = true;
    RuntimeContainer *m_container = nullptr;
    int m_doneCount = 0;
    std::vector<std::unique_ptr<RuntimeTask>> m_children = {}; // Owning.
};

class RuntimeContainer
{
    Q_DISABLE_COPY(RuntimeContainer)

public:
    RuntimeContainer(const ContainerNode &taskContainer, RuntimeTask *parentTask)
        : m_containerNode(taskContainer)
        , m_parentTask(parentTask)
        , m_storages(createStorages(taskContainer))
        , m_successBit(initialSuccessBit(taskContainer.m_workflowPolicy))
        , m_shouldIterate(taskContainer.m_loop)
    {}

    ~RuntimeContainer()
    {
        for (int i = m_containerNode.m_storageList.size() - 1; i >= 0; --i) { // iterate in reverse order
            const StorageBase storage = m_containerNode.m_storageList[i];
            StoragePtr storagePtr = m_storages.value(i);
            if (m_callStorageDoneHandlersOnDestruction)
                m_containerNode.m_taskTreePrivate->callDoneHandler(storage, storagePtr);
            storage.m_storageData->m_destructor(storagePtr);
        }
    }

    static QList<StoragePtr> createStorages(const ContainerNode &container);
    bool isStarting() const { return m_startGuard.isLocked(); }
    RuntimeIteration *parentIteration() const;
    bool updateSuccessBit(bool success);
    void deleteFinishedIterations();
    int progressiveLoopCount() const
    {
        return m_containerNode.m_taskTreePrivate->effectiveLoopCount(m_containerNode.m_loop);
    }

    const ContainerNode &m_containerNode; // Not owning.
    RuntimeTask *m_parentTask = nullptr; // Not owning.
    const QList<StoragePtr> m_storages; // Owning.

    bool m_successBit = true;
    bool m_callStorageDoneHandlersOnDestruction = false;
    Guard m_startGuard;

    int m_iterationCount = 0;
    int m_nextToStart = 0;
    int m_runningChildren = 0;
    bool m_shouldIterate = true;
    std::vector<std::unique_ptr<RuntimeIteration>> m_iterations; // Owning.
};

class RuntimeTask
{
public:
    ~RuntimeTask()
    {
        if (m_task) {
            // Ensures the running task's d'tor doesn't emit done() signal. QTCREATORBUG-30204.
            QObject::disconnect(m_task.get(), &TaskInterface::done,
                                m_taskNode.m_container.m_taskTreePrivate->q, nullptr);
        }
    }

    const TaskNode &m_taskNode; // Not owning.
    RuntimeIteration *m_parentIteration = nullptr; // Not owning.
    std::optional<RuntimeContainer> m_container = {}; // Owning.
    std::unique_ptr<TaskInterface> m_task = {}; // Owning.
};

static bool isProgressive(RuntimeContainer *container)
{
    RuntimeIteration *iteration = container->m_parentTask->m_parentIteration;
    return iteration ? iteration->m_isProgressive : true;
}

void ExecutionContextActivator::activateTaskTree(RuntimeIteration *iteration)
{
    activateTaskTree(iteration->m_container);
}

void ExecutionContextActivator::activateTaskTree(RuntimeContainer *container)
{
    s_activeTaskTrees.push_back(container->m_containerNode.m_taskTreePrivate->q);
}

void ExecutionContextActivator::activateContext(RuntimeIteration *iteration)
{
    std::optional<Loop> loop = iteration->loop();
    if (loop) {
        loop->m_loopData->threadData().pushIteration(iteration->m_iterationIndex);
        m_activeLoops.append(*loop);
    }
    activateContext(iteration->m_container);
}

void ExecutionContextActivator::activateContext(RuntimeContainer *container)
{
    const ContainerNode &containerNode = container->m_containerNode;
    for (int i = 0; i < containerNode.m_storageList.size(); ++i) {
        const StorageBase &storage = containerNode.m_storageList[i];
        if (m_activeStorages.contains(storage))
            continue; // Storage shadowing: The storage is already active, skipping it...
        m_activeStorages.append(storage);
        storage.m_storageData->threadData().pushStorage(container->m_storages.value(i));
    }
    // Go to the parent after activating this storages so that storage shadowing works
    // in the direction from child to parent root.
    if (container->parentIteration())
        activateContext(container->parentIteration());
}

void TaskTreePrivate::start()
{
    QT_ASSERT(m_root, return);
    QT_ASSERT(!m_runtimeRoot, return);
    m_asyncCount = 0;
    m_progressValue = 0;
    {
        GuardLocker locker(m_guard);
        emit q->started();
        emit q->asyncCountChanged(m_asyncCount);
        emit q->progressValueChanged(m_progressValue);
    }
    // TODO: check storage handlers for not existing storages in tree
    for (auto it = m_storageHandlers.cbegin(); it != m_storageHandlers.cend(); ++it) {
        QT_ASSERT(m_storages.contains(it.key()), qWarning("The registered storage doesn't "
                  "exist in task tree. Its handlers will never be called."));
    }
    m_runtimeRoot.reset(new RuntimeTask{*m_root});
    start(m_runtimeRoot.get());
    bumpAsyncCount();
}

void TaskTreePrivate::stop()
{
    QT_ASSERT(m_root, return);
    if (!m_runtimeRoot)
        return;
    stop(m_runtimeRoot.get());
    m_runtimeRoot.reset();
    emitDone(DoneWith::Cancel);
}

void TaskTreePrivate::bumpAsyncCount()
{
    if (!m_runtimeRoot)
        return;
    ++m_asyncCount;
    GuardLocker locker(m_guard);
    emit q->asyncCountChanged(m_asyncCount);
}

void TaskTreePrivate::advanceProgress(int byValue)
{
    if (byValue == 0)
        return;
    QT_CHECK(byValue > 0);
    QT_CHECK(m_progressValue + byValue <= m_root->taskCount());
    m_progressValue += byValue;
    GuardLocker locker(m_guard);
    emit q->progressValueChanged(m_progressValue);
}

void TaskTreePrivate::emitDone(DoneWith result)
{
    QT_CHECK(m_progressValue == m_root->taskCount());
    GuardLocker locker(m_guard);
    emit q->done(result);
}

RuntimeIteration::RuntimeIteration(int index, RuntimeContainer *container)
    : m_iterationIndex(index)
    , m_isProgressive(index < container->progressiveLoopCount() && isProgressive(container))
    , m_container(container)
{}

std::optional<Loop> RuntimeIteration::loop() const
{
    return m_container->m_containerNode.m_loop;
}

void RuntimeIteration::deleteChild(RuntimeTask *task)
{
    const auto it = std::find_if(m_children.cbegin(), m_children.cend(), [task](const auto &ptr) {
        return ptr.get() == task;
    });
    if (it != m_children.cend())
        m_children.erase(it);
}

static std::vector<TaskNode> createChildren(TaskTreePrivate *taskTreePrivate,
                                            const QList<GroupItem> &children)
{
    std::vector<TaskNode> result;
    result.reserve(children.size());
    for (const GroupItem &child : children)
        result.emplace_back(taskTreePrivate, child);
    return result;
}

ContainerNode::ContainerNode(TaskTreePrivate *taskTreePrivate, const GroupItem &task)
    : m_taskTreePrivate(taskTreePrivate)
    , m_groupHandler(task.m_groupData.m_groupHandler)
    , m_parallelLimit(task.m_groupData.m_parallelLimit.value_or(1))
    , m_workflowPolicy(task.m_groupData.m_workflowPolicy.value_or(WorkflowPolicy::StopOnError))
    , m_loop(task.m_groupData.m_loop)
    , m_storageList(task.m_storageList)
    , m_children(createChildren(taskTreePrivate, task.m_children))
    , m_taskCount(std::accumulate(m_children.cbegin(), m_children.cend(), 0,
                                  [](int r, const TaskNode &n) { return r + n.taskCount(); })
                  * taskTreePrivate->effectiveLoopCount(m_loop))
{
    for (const StorageBase &storage : m_storageList)
        m_taskTreePrivate->m_storages << storage;
}

QList<StoragePtr> RuntimeContainer::createStorages(const ContainerNode &container)
{
    QList<StoragePtr> storages;
    for (const StorageBase &storage : container.m_storageList) {
        StoragePtr storagePtr = storage.m_storageData->m_constructor();
        storages.append(storagePtr);
        container.m_taskTreePrivate->callSetupHandler(storage, storagePtr);
    }
    return storages;
}

RuntimeIteration *RuntimeContainer::parentIteration() const
{
    return m_parentTask->m_parentIteration;
}

bool RuntimeContainer::updateSuccessBit(bool success)
{
    if (m_containerNode.m_workflowPolicy == WorkflowPolicy::FinishAllAndSuccess
        || m_containerNode.m_workflowPolicy == WorkflowPolicy::FinishAllAndError
        || m_containerNode.m_workflowPolicy == WorkflowPolicy::StopOnSuccessOrError) {
        if (m_containerNode.m_workflowPolicy == WorkflowPolicy::StopOnSuccessOrError)
            m_successBit = success;
        return m_successBit;
    }

    const bool donePolicy = m_containerNode.m_workflowPolicy == WorkflowPolicy::StopOnSuccess
                         || m_containerNode.m_workflowPolicy == WorkflowPolicy::ContinueOnSuccess;
    m_successBit = donePolicy ? (m_successBit || success) : (m_successBit && success);
    return m_successBit;
}

void RuntimeContainer::deleteFinishedIterations()
{
    for (auto it = m_iterations.cbegin(); it != m_iterations.cend(); ) {
        if (it->get()->m_doneCount == int(m_containerNode.m_children.size()))
            it = m_iterations.erase(it);
        else
            ++it;
    }
}

SetupResult TaskTreePrivate::start(RuntimeContainer *container)
{
    const ContainerNode &containerNode = container->m_containerNode;
    SetupResult startAction = SetupResult::Continue;
    if (containerNode.m_groupHandler.m_setupHandler) {
        startAction = invokeHandler(container, containerNode.m_groupHandler.m_setupHandler);
        if (startAction != SetupResult::Continue) {
            if (isProgressive(container))
                advanceProgress(containerNode.m_taskCount);
            // Non-Continue SetupResult takes precedence over the workflow policy.
            container->m_successBit = startAction == SetupResult::StopWithSuccess;
        }
    }
    if (startAction == SetupResult::Continue
        && (containerNode.m_children.empty()
            || (containerNode.m_loop && !invokeLoopHandler(container)))) {
        if (isProgressive(container))
            advanceProgress(containerNode.m_taskCount);
        startAction = toSetupResult(container->m_successBit);
    }
    return continueStart(container, startAction);
}

SetupResult TaskTreePrivate::continueStart(RuntimeContainer *container, SetupResult startAction)
{
    const SetupResult groupAction = startAction == SetupResult::Continue ? startChildren(container)
                                                                         : startAction;
    if (groupAction != SetupResult::Continue) {
        const bool bit = container->updateSuccessBit(groupAction == SetupResult::StopWithSuccess);
        RuntimeIteration *parentIteration = container->parentIteration();
        RuntimeTask *parentTask = container->m_parentTask;
        QT_CHECK(parentTask);
        const bool result = invokeDoneHandler(container, bit ? DoneWith::Success : DoneWith::Error);
        if (parentIteration) {
            parentIteration->deleteChild(parentTask);
            if (!parentIteration->m_container->isStarting())
                childDone(parentIteration, result);
        } else {
            QT_CHECK(m_runtimeRoot.get() == parentTask);
            m_runtimeRoot.reset();
            emitDone(result ? DoneWith::Success : DoneWith::Error);
        }
    }
    return groupAction;
}

SetupResult TaskTreePrivate::startChildren(RuntimeContainer *container)
{
    const ContainerNode &containerNode = container->m_containerNode;
    const int childCount = int(containerNode.m_children.size());

    if (container->m_iterationCount == 0) {
        container->m_iterations.emplace_back(
            std::make_unique<RuntimeIteration>(container->m_iterationCount, container));
        ++container->m_iterationCount;
    } else if (containerNode.m_parallelLimit == 0) {
        container->deleteFinishedIterations();
        if (container->m_iterations.empty())
            return toSetupResult(container->m_successBit);
        return SetupResult::Continue;
    }

    GuardLocker locker(container->m_startGuard);

    while (containerNode.m_parallelLimit == 0
           || container->m_runningChildren < containerNode.m_parallelLimit) {
        container->deleteFinishedIterations();
        if (container->m_nextToStart == childCount) {
            if (container->m_shouldIterate && invokeLoopHandler(container)) {
                container->m_nextToStart = 0;
                container->m_iterations.emplace_back(
                    std::make_unique<RuntimeIteration>(container->m_iterationCount, container));
                ++container->m_iterationCount;
            } else {
                if (container->m_iterations.empty())
                    return toSetupResult(container->m_successBit);
                return SetupResult::Continue;
            }
        }
        RuntimeIteration *iteration = container->m_iterations.back().get();
        RuntimeTask *newTask = new RuntimeTask{containerNode.m_children.at(container->m_nextToStart),
                                               iteration};
        iteration->m_children.emplace_back(newTask);
        ++container->m_runningChildren;
        ++container->m_nextToStart;

        const SetupResult startAction = start(newTask);
        if (startAction == SetupResult::Continue)
            continue;

        const SetupResult finalizeAction = childDone(iteration,
                                                     startAction == SetupResult::StopWithSuccess);
        if (finalizeAction != SetupResult::Continue)
            return finalizeAction;
    }
    return SetupResult::Continue;
}

SetupResult TaskTreePrivate::childDone(RuntimeIteration *iteration, bool success)
{
    RuntimeContainer *container = iteration->m_container;
    const WorkflowPolicy &workflowPolicy = container->m_containerNode.m_workflowPolicy;
    const bool shouldStop = workflowPolicy == WorkflowPolicy::StopOnSuccessOrError
                        || (workflowPolicy == WorkflowPolicy::StopOnSuccess && success)
                        || (workflowPolicy == WorkflowPolicy::StopOnError && !success);
    ++iteration->m_doneCount;
    --container->m_runningChildren;
    if (shouldStop)
        stop(container);

    const bool updatedSuccess = container->updateSuccessBit(success);
    const SetupResult startAction = shouldStop ? toSetupResult(updatedSuccess)
                                               : SetupResult::Continue;

    if (container->isStarting())
        return startAction;
    return continueStart(container, startAction);
}

void TaskTreePrivate::stop(RuntimeContainer *container)
{
    const ContainerNode &containerNode = container->m_containerNode;
    for (auto &iteration : container->m_iterations) {
        for (auto &child : iteration->m_children) {
            ++iteration->m_doneCount;
            stop(child.get());
        }

        if (iteration->m_isProgressive) {
            int skippedTaskCount = 0;
            for (int i = iteration->m_doneCount; i < int(containerNode.m_children.size()); ++i)
                skippedTaskCount += containerNode.m_children.at(i).taskCount();
            advanceProgress(skippedTaskCount);
        }
    }
    const int skippedIterations = container->progressiveLoopCount() - container->m_iterationCount;
    if (skippedIterations > 0) {
        advanceProgress(container->m_containerNode.m_taskCount / container->progressiveLoopCount()
                        * skippedIterations);
    }
}

static bool shouldCall(CallDoneIf callDoneIf, DoneWith result)
{
    if (result == DoneWith::Success)
        return callDoneIf != CallDoneIf::Error;
    return callDoneIf != CallDoneIf::Success;
}

bool TaskTreePrivate::invokeDoneHandler(RuntimeContainer *container, DoneWith doneWith)
{
    DoneResult result = toDoneResult(doneWith);
    const GroupItem::GroupHandler &groupHandler = container->m_containerNode.m_groupHandler;
    if (groupHandler.m_doneHandler && shouldCall(groupHandler.m_callDoneIf, doneWith))
        result = invokeHandler(container, groupHandler.m_doneHandler, doneWith);
    container->m_callStorageDoneHandlersOnDestruction = true;
    // TODO: is it needed?
    container->m_parentTask->m_container.reset();
    return result == DoneResult::Success;
}

bool TaskTreePrivate::invokeLoopHandler(RuntimeContainer *container)
{
    if (container->m_shouldIterate) {
        const LoopData *loopData = container->m_containerNode.m_loop->m_loopData.get();
        if (loopData->m_loopCount) {
            container->m_shouldIterate = container->m_iterationCount < loopData->m_loopCount;
        } else if (loopData->m_condition) {
            container->m_shouldIterate = invokeHandler(container, loopData->m_condition,
                                                       container->m_iterationCount);
        }
    }
    return container->m_shouldIterate;
}

SetupResult TaskTreePrivate::start(RuntimeTask *node)
{
    if (!node->m_taskNode.isTask()) {
        node->m_container.emplace(node->m_taskNode.m_container, node);
        return start(&*node->m_container);
    }

    const GroupItem::TaskHandler &handler = node->m_taskNode.m_taskHandler;
    node->m_task.reset(handler.m_createHandler());
    const SetupResult startAction = handler.m_setupHandler
        ? invokeHandler(node->m_parentIteration, handler.m_setupHandler, *node->m_task.get())
        : SetupResult::Continue;
    if (startAction != SetupResult::Continue) {
        if (node->m_parentIteration->m_isProgressive)
            advanceProgress(1);
        node->m_parentIteration->deleteChild(node);
        return startAction;
    }
    const std::shared_ptr<SetupResult> unwindAction
        = std::make_shared<SetupResult>(SetupResult::Continue);
    QObject::connect(node->m_task.get(), &TaskInterface::done,
                     q, [this, node, unwindAction](DoneResult doneResult) {
        const bool result = invokeDoneHandler(node, toDoneWith(doneResult));
        QObject::disconnect(node->m_task.get(), &TaskInterface::done, q, nullptr);
        node->m_task.release()->deleteLater();
        RuntimeIteration *parentIteration = node->m_parentIteration;
        parentIteration->deleteChild(node);
        if (parentIteration->m_container->isStarting()) {
            *unwindAction = toSetupResult(result);
        } else {
            childDone(parentIteration, result);
            bumpAsyncCount();
        }
    });

    node->m_task->start();
    return *unwindAction;
}

void TaskTreePrivate::stop(RuntimeTask *node)
{
    if (!node->m_task) {
        if (!node->m_container)
            return;
        stop(&*node->m_container);
        node->m_container->updateSuccessBit(false);
        invokeDoneHandler(&*node->m_container, DoneWith::Cancel);
        return;
    }

    invokeDoneHandler(node, DoneWith::Cancel);
    node->m_task.reset();
}

bool TaskTreePrivate::invokeDoneHandler(RuntimeTask *node, DoneWith doneWith)
{
    DoneResult result = toDoneResult(doneWith);
    const GroupItem::TaskHandler &handler = node->m_taskNode.m_taskHandler;
    if (handler.m_doneHandler && shouldCall(handler.m_callDoneIf, doneWith)) {
        result = invokeHandler(node->m_parentIteration,
                               handler.m_doneHandler, *node->m_task.get(), doneWith);
    }
    if (node->m_parentIteration->m_isProgressive)
        advanceProgress(1);
    return result == DoneResult::Success;
}

/*!
    \class Tasking::TaskTree
    \inheaderfile solutions/tasking/tasktree.h
    \inmodule TaskingSolution
    \brief The TaskTree class runs an async task tree structure defined in a declarative way.
    \reentrant

    Use the Tasking namespace to build extensible, declarative task tree
    structures that contain possibly asynchronous tasks, such as QProcess,
    NetworkQuery, or ConcurrentCall<ReturnType>. TaskTree structures enable you
    to create a sophisticated mixture of a parallel or sequential flow of tasks
    in the form of a tree and to run it any time later.

    \section1 Root Element and Tasks

    The TaskTree has a mandatory Group root element, which may contain
    any number of tasks of various types, such as QProcessTask, NetworkQueryTask,
    or ConcurrentCallTask<ReturnType>:

    \code
        using namespace Tasking;

        const Group root {
            QProcessTask(...),
            NetworkQueryTask(...),
            ConcurrentCallTask<int>(...)
        };

        TaskTree *taskTree = new TaskTree(root);
        connect(taskTree, &TaskTree::done, ...);  // finish handler
        taskTree->start();
    \endcode

    The task tree above has a top level element of the Group type that contains
    tasks of the QProcessTask, NetworkQueryTask, and ConcurrentCallTask<int> type.
    After taskTree->start() is called, the tasks are run in a chain, starting
    with QProcessTask. When the QProcessTask finishes successfully, the NetworkQueryTask
    task is started. Finally, when the network task finishes successfully, the
    ConcurrentCallTask<int> task is started.

    When the last running task finishes with success, the task tree is considered
    to have run successfully and the done() signal is emitted with DoneWith::Success.
    When a task finishes with an error, the execution of the task tree is stopped
    and the remaining tasks are skipped. The task tree finishes with an error and
    sends the TaskTree::done() signal with DoneWith::Error.

    \section1 Groups

    The parent of the Group sees it as a single task. Like other tasks,
    the group can be started and it can finish with success or an error.
    The Group elements can be nested to create a tree structure:

    \code
        const Group root {
            Group {
                parallel,
                QProcessTask(...),
                ConcurrentCallTask<int>(...)
            },
            NetworkQueryTask(...)
        };
    \endcode

    The example above differs from the first example in that the root element has
    a subgroup that contains the QProcessTask and ConcurrentCallTask<int>. The subgroup is a
    sibling element of the NetworkQueryTask in the root. The subgroup contains an
    additional \e parallel element that instructs its Group to execute its tasks
    in parallel.

    So, when the tree above is started, the QProcessTask and ConcurrentCallTask<int> start
    immediately and run in parallel. Since the root group doesn't contain a
    \e parallel element, its direct child tasks are run in sequence. Thus, the
    NetworkQueryTask starts when the whole subgroup finishes. The group is
    considered as finished when all its tasks have finished. The order in which
    the tasks finish is not relevant.

    So, depending on which task lasts longer (QProcessTask or ConcurrentCallTask<int>), the
    following scenarios can take place:

    \table
    \header
        \li Scenario 1
        \li Scenario 2
    \row
        \li Root Group starts
        \li Root Group starts
    \row
        \li Sub Group starts
        \li Sub Group starts
    \row
        \li QProcessTask starts
        \li QProcessTask starts
    \row
        \li ConcurrentCallTask<int> starts
        \li ConcurrentCallTask<int> starts
    \row
        \li ...
        \li ...
    \row
        \li \b {QProcessTask finishes}
        \li \b {ConcurrentCallTask<int> finishes}
    \row
        \li ...
        \li ...
    \row
        \li \b {ConcurrentCallTask<int> finishes}
        \li \b {QProcessTask finishes}
    \row
        \li Sub Group finishes
        \li Sub Group finishes
    \row
        \li NetworkQueryTask starts
        \li NetworkQueryTask starts
    \row
        \li ...
        \li ...
    \row
        \li NetworkQueryTask finishes
        \li NetworkQueryTask finishes
    \row
        \li Root Group finishes
        \li Root Group finishes
    \endtable

    The differences between the scenarios are marked with bold. Three dots mean
    that an unspecified amount of time passes between previous and next events
    (a task or tasks continue to run). No dots between events
    means that they occur synchronously.

    The presented scenarios assume that all tasks run successfully. If a task
    fails during execution, the task tree finishes with an error. In particular,
    when QProcessTask finishes with an error while ConcurrentCallTask<int> is still being executed,
    the ConcurrentCallTask<int> is automatically canceled, the subgroup finishes with an error,
    the NetworkQueryTask is skipped, and the tree finishes with an error.

    \section1 Task Types

    Each task type is associated with its corresponding task class that executes
    the task. For example, a QProcessTask inside a task tree is associated with
    the QProcess class that executes the process. The associated objects are
    automatically created, started, and destructed exclusively by the task tree
    at the appropriate time.

    If a root group consists of five sequential QProcessTask tasks, and the task tree
    executes the group, it creates an instance of QProcess for the first
    QProcessTask and starts it. If the QProcess instance finishes successfully,
    the task tree destructs it and creates a new QProcess instance for the
    second QProcessTask, and so on. If the first task finishes with an error, the task
    tree stops creating QProcess instances, and the root group finishes with an
    error.

    The following table shows examples of task types and their corresponding task
    classes:

    \table
    \header
        \li Task Type (Tasking Namespace)
        \li Associated Task Class
        \li Brief Description
    \row
        \li QProcessTask
        \li QProcess
        \li Starts process.
    \row
        \li ConcurrentCallTask<ReturnType>
        \li Tasking::ConcurrentCall<ReturnType>
        \li Starts asynchronous task, runs in separate thread.
    \row
        \li TaskTreeTask
        \li Tasking::TaskTree
        \li Starts nested task tree.
    \row
        \li NetworkQueryTask
        \li NetworkQuery
        \li Starts network download.
    \endtable

    \section1 Task Handlers

    Use Task handlers to set up a task for execution and to enable reading
    the output data from the task when it finishes with success or an error.

    \section2 Task's Start Handler

    When a corresponding task class object is created and before it's started,
    the task tree invokes an optionally user-provided setup handler. The setup
    handler should always take a \e reference to the associated task class object:

    \code
        const auto onSetup = [](QProcess &process) {
            process.setCommand({"sleep", {"3"}});
        };
        const Group root {
            QProcessTask(onSetup)
        };
    \endcode

    You can modify the passed QProcess in the setup handler, so that the task
    tree can start the process according to your configuration.
    You should not call \c {process.start();} in the setup handler,
    as the task tree calls it when needed. The setup handler is optional. When used,
    it must be the first argument of the task's constructor.

    Optionally, the setup handler may return a SetupResult. The returned
    SetupResult influences the further start behavior of a given task. The
    possible values are:

    \table
    \header
        \li SetupResult Value
        \li Brief Description
    \row
        \li Continue
        \li The task will be started normally. This is the default behavior when the
            setup handler doesn't return SetupResult (that is, its return type is
            void).
    \row
        \li StopWithSuccess
        \li The task won't be started and it will report success to its parent.
    \row
        \li StopWithError
        \li The task won't be started and it will report an error to its parent.
    \endtable

    This is useful for running a task only when a condition is met and the data
    needed to evaluate this condition is not known until previously started tasks
    finish. In this way, the setup handler dynamically decides whether to start the
    corresponding task normally or skip it and report success or an error.
    For more information about inter-task data exchange, see \l Storage.

    \section2 Task's Done Handler

    When a running task finishes, the task tree invokes an optionally provided done handler.
    The handler should always take a \c const \e reference to the associated task class object:

    \code
        const auto onSetup = [](QProcess &process) {
            process.setCommand({"sleep", {"3"}});
        };
        const auto onDone = [](const QProcess &process, DoneWith result) {
            if (result == DoneWith::Success)
                qDebug() << "Success" << process.cleanedStdOut();
            else
                qDebug() << "Failure" << process.cleanedStdErr();
        };
        const Group root {
            QProcessTask(onSetup, onDone)
        };
    \endcode

    The done handler may collect output data from QProcess, and store it
    for further processing or perform additional actions.

    \note If the task setup handler returns StopWithSuccess or StopWithError,
          the done handler is not invoked.

    \section1 Group Handlers

    Similarly to task handlers, group handlers enable you to set up a group to
    execute and to apply more actions when the whole group finishes with
    success or an error.

    \section2 Group's Start Handler

    The task tree invokes the group start handler before it starts the child
    tasks. The group handler doesn't take any arguments:

    \code
        const auto onSetup = [] {
            qDebug() << "Entering the group";
        };
        const Group root {
            onGroupSetup(onSetup),
            QProcessTask(...)
        };
    \endcode

    The group setup handler is optional. To define a group setup handler, add an
    onGroupSetup() element to a group. The argument of onGroupSetup() is a user
    handler. If you add more than one onGroupSetup() element to a group, an assert
    is triggered at runtime that includes an error message.

    Like the task's start handler, the group start handler may return SetupResult.
    The returned SetupResult value affects the start behavior of the
    whole group. If you do not specify a group start handler or its return type
    is void, the default group's action is SetupResult::Continue, so that all
    tasks are started normally. Otherwise, when the start handler returns
    SetupResult::StopWithSuccess or SetupResult::StopWithError, the tasks are not
    started (they are skipped) and the group itself reports success or failure,
    depending on the returned value, respectively.

    \code
        const Group root {
            onGroupSetup([] { qDebug() << "Root setup"; }),
            Group {
                onGroupSetup([] { qDebug() << "Group 1 setup"; return SetupResult::Continue; }),
                QProcessTask(...) // Process 1
            },
            Group {
                onGroupSetup([] { qDebug() << "Group 2 setup"; return SetupResult::StopWithSuccess; }),
                QProcessTask(...) // Process 2
            },
            Group {
                onGroupSetup([] { qDebug() << "Group 3 setup"; return SetupResult::StopWithError; }),
                QProcessTask(...) // Process 3
            },
            QProcessTask(...) // Process 4
        };
    \endcode

    In the above example, all subgroups of a root group define their setup handlers.
    The following scenario assumes that all started processes finish with success:

    \table
    \header
        \li Scenario
        \li Comment
    \row
        \li Root Group starts
        \li Doesn't return SetupResult, so its tasks are executed.
    \row
        \li Group 1 starts
        \li Returns Continue, so its tasks are executed.
    \row
        \li Process 1 starts
        \li
    \row
        \li ...
        \li ...
    \row
        \li Process 1 finishes (success)
        \li
    \row
        \li Group 1 finishes (success)
        \li
    \row
        \li Group 2 starts
        \li Returns StopWithSuccess, so Process 2 is skipped and Group 2 reports
            success.
    \row
        \li Group 2 finishes (success)
        \li
    \row
        \li Group 3 starts
        \li Returns StopWithError, so Process 3 is skipped and Group 3 reports
            an error.
    \row
        \li Group 3 finishes (error)
        \li
    \row
        \li Root Group finishes (error)
        \li Group 3, which is a direct child of the root group, finished with an
            error, so the root group stops executing, skips Process 4, which has
            not started yet, and reports an error.
    \endtable

    \section2 Groups's Done Handler

    A Group's done handler is executed after the successful or failed execution of its tasks.
    The final value reported by the group depends on its \l {Workflow Policy}.
    The handler can apply other necessary actions.
    The done handler is defined inside the onGroupDone() element of a group.
    It may take the optional DoneWith argument, indicating the successful or failed execution:

    \code
        const Group root {
            onGroupSetup([] { qDebug() << "Root setup"; }),
            QProcessTask(...),
            onGroupDone([](DoneWith result) {
                if (result == DoneWith::Success)
                    qDebug() << "Root finished with success";
                else
                    qDebug() << "Root finished with an error";
            })
        };
    \endcode

    The group done handler is optional. If you add more than one onGroupDone() to a group,
    an assert is triggered at runtime that includes an error message.

    \note Even if the group setup handler returns StopWithSuccess or StopWithError,
    the group's done handler is invoked. This behavior differs from that of task done handler
    and might change in the future.

    \section1 Other Group Elements

    A group can contain other elements that describe the processing flow, such as
    the execution mode or workflow policy. It can also contain storage elements
    that are responsible for collecting and sharing custom common data gathered
    during group execution.

    \section2 Execution Mode

    The execution mode element in a Group specifies how the direct child tasks of
    the Group are started. The most common execution modes are \l sequential and
    \l parallel. It's also possible to specify the limit of tasks running
    in parallel by using the parallelLimit() function.

    In all execution modes, a group starts tasks in the oder in which they appear.

    If a child of a group is also a group, the child group runs its tasks
    according to its own execution mode.

    \section2 Workflow Policy

    The workflow policy element in a Group specifies how the group should behave
    when any of its \e direct child's tasks finish. For a detailed description of possible
    policies, refer to WorkflowPolicy.

    If a child of a group is also a group, the child group runs its tasks
    according to its own workflow policy.

    \section2 Storage

    Use the \l {Tasking::Storage} {Storage} element to exchange information between tasks.
    Especially, in the sequential execution mode, when a task needs data from another,
    already finished task, before it can start. For example, a task tree that copies data by reading
    it from a source and writing it to a destination might look as follows:

    \code
        static QByteArray load(const QString &fileName) { ... }
        static void save(const QString &fileName, const QByteArray &array) { ... }

        static Group copyRecipe(const QString &source, const QString &destination)
        {
            struct CopyStorage { // [1] custom inter-task struct
                QByteArray content; // [2] custom inter-task data
            };

            // [3] instance of custom inter-task struct manageable by task tree
            const Storage<CopyStorage> storage;

            const auto onLoaderSetup = [source](ConcurrentCall<QByteArray> &async) {
                async.setConcurrentCallData(&load, source);
            };
            // [4] runtime: task tree activates the instance from [7] before invoking handler
            const auto onLoaderDone = [storage](const ConcurrentCall<QByteArray> &async) {
                storage->content = async.result(); // [5] loader stores the result in storage
            };

            // [4] runtime: task tree activates the instance from [7] before invoking handler
            const auto onSaverSetup = [storage, destination](ConcurrentCall<void> &async) {
                const QByteArray content = storage->content; // [6] saver takes data from storage
                async.setConcurrentCallData(&save, destination, content);
            };
            const auto onSaverDone = [](const ConcurrentCall<void> &async) {
                qDebug() << "Save done successfully";
            };

            const Group root {
                // [7] runtime: task tree creates an instance of CopyStorage when root is entered
                storage,
                ConcurrentCallTask<QByteArray>(onLoaderSetup, onLoaderDone, CallDoneIf::Success),
                ConcurrentCallTask<void>(onSaverSetup, onSaverDone, CallDoneIf::Success)
            };
            return root;
        }

        const QString source = ...;
        const QString destination = ...;
        TaskTree taskTree(copyRecipe(source, destination));
        connect(&taskTree, &TaskTree::done,
                &taskTree, [](DoneWith result) {
            if (result == DoneWith::Success)
                qDebug() << "The copying finished successfully.";
        });
        tasktree.start();
    \endcode

    In the example above, the inter-task data consists of a QByteArray content
    variable [2] enclosed in a \c CopyStorage custom struct [1]. If the loader
    finishes successfully, it stores the data in a \c CopyStorage::content
    variable [5]. The saver then uses the variable to configure the saving task [6].

    To enable a task tree to manage the \c CopyStorage struct, an instance of
    \l {Tasking::Storage} {Storage}<\c CopyStorage> is created [3]. If a copy of this object is
    inserted as the group's child item [7], an instance of the \c CopyStorage struct is
    created dynamically when the task tree enters this group. When the task
    tree leaves this group, the existing instance of the \c CopyStorage struct is
    destructed as it's no longer needed.

    If several task trees holding a copy of the common
    \l {Tasking::Storage} {Storage}<\c CopyStorage> instance run simultaneously
    (including the case when the task trees are run in different threads),
    each task tree contains its own copy of the \c CopyStorage struct.

    You can access \c CopyStorage from any handler in the group with a storage object.
    This includes all handlers of all descendant tasks of the group with
    a storage object. To access the custom struct in a handler, pass the
    copy of the \l {Tasking::Storage} {Storage}<\c CopyStorage> object to the handler
    (for example, in a lambda capture) [4].

    When the task tree invokes a handler in a subtree containing the storage [7],
    the task tree activates its own \c CopyStorage instance inside the
    \l {Tasking::Storage} {Storage}<\c CopyStorage> object. Therefore, the \c CopyStorage struct
    may be accessed only from within the handler body. To access the currently active
    \c CopyStorage from within \l {Tasking::Storage} {Storage}<\c CopyStorage>, use the
    \l {Tasking::Storage::operator->()} {Storage::operator->()},
    \l {Tasking::Storage::operator*()} {Storage::operator*()}, or Storage::activeStorage() method.

    The following list summarizes how to employ a Storage object into the task
    tree:
    \list 1
        \li Define the custom structure \c MyStorage with custom data [1], [2]
        \li Create an instance of the \l {Tasking::Storage} {Storage}<\c MyStorage> storage [3]
        \li Pass the \l {Tasking::Storage} {Storage}<\c MyStorage> instance to handlers [4]
        \li Access the \c MyStorage instance in handlers [5], [6]
        \li Insert the \l {Tasking::Storage} {Storage}<\c MyStorage> instance into a group [7]
    \endlist

    \section1 TaskTree class

    TaskTree executes the tree structure of asynchronous tasks according to the
    recipe described by the Group root element.

    As TaskTree is also an asynchronous task, it can be a part of another TaskTree.
    To place a nested TaskTree inside another TaskTree, insert the TaskTreeTask
    element into another Group element.

    TaskTree reports progress of completed tasks when running. The progress value
    is increased when a task finishes or is skipped or canceled.
    When TaskTree is finished and the TaskTree::done() signal is emitted,
    the current value of the progress equals the maximum progress value.
    Maximum progress equals the total number of asynchronous tasks in a tree.
    A nested TaskTree is counted as a single task, and its child tasks are not
    counted in the top level tree. Groups themselves are not counted as tasks,
    but their tasks are counted. \l {Tasking::Sync} {Sync} tasks are not asynchronous,
    so they are not counted as tasks.

    To set additional initial data for the running tree, modify the storage
    instances in a tree when it creates them by installing a storage setup
    handler:

    \code
        Storage<CopyStorage> storage;
        const Group root = ...; // storage placed inside root's group and inside handlers
        TaskTree taskTree(root);
        auto initStorage = [](CopyStorage &storage) {
            storage.content = "initial content";
        };
        taskTree.onStorageSetup(storage, initStorage);
        taskTree.start();
    \endcode

    When the running task tree creates a \c CopyStorage instance, and before any
    handler inside a tree is called, the task tree calls the initStorage handler,
    to enable setting up initial data of the storage, unique to this particular
    run of taskTree.

    Similarly, to collect some additional result data from the running tree,
    read it from storage instances in the tree when they are about to be
    destroyed. To do this, install a storage done handler:

    \code
        Storage<CopyStorage> storage;
        const Group root = ...; // storage placed inside root's group and inside handlers
        TaskTree taskTree(root);
        auto collectStorage = [](const CopyStorage &storage) {
            qDebug() << "final content" << storage.content;
        };
        taskTree.onStorageDone(storage, collectStorage);
        taskTree.start();
    \endcode

    When the running task tree is about to destroy a \c CopyStorage instance, the
    task tree calls the collectStorage handler, to enable reading the final data
    from the storage, unique to this particular run of taskTree.

    \section1 Task Adapters

    To extend a TaskTree with a new task type, implement a simple adapter class
    derived from the TaskAdapter class template. The following class is an
    adapter for a single shot timer, which may be considered as a new asynchronous task:

    \code
        class TimerTaskAdapter : public TaskAdapter<QTimer>
        {
        public:
            TimerTaskAdapter() {
                task()->setSingleShot(true);
                task()->setInterval(1000);
                connect(task(), &QTimer::timeout, this, [this] { emit done(DoneResult::Success); });
            }
        private:
            void start() final { task()->start(); }
        };

        using TimerTask = CustomTask<TimerTaskAdapter>;
    \endcode

    You must derive the custom adapter from the TaskAdapter class template
    instantiated with a template parameter of the class implementing a running
    task. The code above uses QTimer to run the task. This class appears
    later as an argument to the task's handlers. The instance of this class
    parameter automatically becomes a member of the TaskAdapter template, and is
    accessible through the TaskAdapter::task() method. The constructor
    of \c TimerTaskAdapter initially configures the QTimer object and connects
    to the QTimer::timeout() signal. When the signal is triggered, \c TimerTaskAdapter
    emits the TaskInterface::done(DoneResult::Success) signal to inform the task tree that
    the task finished successfully. If it emits TaskInterface::done(DoneResult::Error),
    the task finished with an error.
    The TaskAdapter::start() method starts the timer.

    To make QTimer accessible inside TaskTree under the \c TimerTask name,
    define \c TimerTask to be an alias to the CustomTask<\c TimerTaskAdapter>.
    \c TimerTask becomes a new custom task type, using \c TimerTaskAdapter.

    The new task type is now registered, and you can use it in TaskTree:

    \code
        const auto onSetup = [](QTimer &task) { task.setInterval(2000); };
        const auto onDone = [] { qDebug() << "timer triggered"; };
        const Group root {
            TimerTask(onSetup, onDone)
        };
    \endcode

    When a task tree containing the root from the above example is started, it
    prints a debug message within two seconds and then finishes successfully.

    \note The class implementing the running task should have a default constructor,
    and objects of this class should be freely destructible. It should be allowed
    to destroy a running object, preferably without waiting for the running task
    to finish (that is, safe non-blocking destructor of a running task).
    To achieve a non-blocking destruction of a task that has a blocking destructor,
    consider using the optional \c Deleter template parameter of the TaskAdapter.
*/

/*!
    Constructs an empty task tree. Use setRecipe() to pass a declarative description
    on how the task tree should execute the tasks and how it should handle the finished tasks.

    Starting an empty task tree is no-op and the relevant warning message is issued.

    \sa setRecipe(), start()
*/
TaskTree::TaskTree()
    : d(new TaskTreePrivate(this))
{}

/*!
    \overload

    Constructs a task tree with a given \a recipe. After the task tree is started,
    it executes the tasks contained inside the \a recipe and
    handles finished tasks according to the passed description.

    \sa setRecipe(), start()
*/
TaskTree::TaskTree(const Group &recipe) : TaskTree()
{
    setRecipe(recipe);
}

/*!
    Destroys the task tree.

    When the task tree is running while being destructed, it cancels all the running tasks
    immediately. In this case, no handlers are called, not even the groups' and
    tasks' done handlers or onStorageDone() handlers. The task tree also doesn't emit any
    signals from the destructor, not even done() or progressValueChanged() signals.
    This behavior may always be relied on.
    It is completely safe to destruct the running task tree.

    It's a usual pattern to destruct the running task tree.
    It's guaranteed that the destruction will run quickly, without having to wait for
    the currently running tasks to finish, provided that the used tasks implement
    their destructors in a non-blocking way.

    \note Do not call the destructor directly from any of the running task's handlers
          or task tree's signals. In these cases, use \l deleteLater() instead.

    \sa cancel()
*/
TaskTree::~TaskTree()
{
    QT_ASSERT(!d->m_guard.isLocked(), qWarning("Deleting TaskTree instance directly from "
              "one of its handlers will lead to a crash!"));
    // TODO: delete storages explicitly here?
    delete d;
}

/*!
    Sets a given \a recipe for the task tree. After the task tree is started,
    it executes the tasks contained inside the \a recipe and
    handles finished tasks according to the passed description.

    \note When called for a running task tree, the call is ignored.

    \sa TaskTree(const Tasking::Group &recipe), start()
*/
void TaskTree::setRecipe(const Group &recipe)
{
    QT_ASSERT(!isRunning(), qWarning("The TaskTree is already running, ignoring..."); return);
    QT_ASSERT(!d->m_guard.isLocked(), qWarning("The setRecipe() is called from one of the"
                                               "TaskTree handlers, ignoring..."); return);
    // TODO: Should we clear the m_storageHandlers, too?
    d->m_storages.clear();
    d->m_root.emplace(d, recipe);
}

/*!
    Starts the task tree.

    Use setRecipe() or the constructor to set the declarative description according to which
    the task tree will execute the contained tasks and handle finished tasks.

    When the task tree is empty, that is, constructed with a default constructor,
    a call to \c start() is no-op and the relevant warning message is issued.

    Otherwise, when the task tree is already running, a call to \e start() is ignored and the
    relevant warning message is issued.

    Otherwise, the task tree is started.

    The started task tree may finish synchronously,
    for example when the main group's start handler returns SetupResult::StopWithError.
    For this reason, the connection to the done signal should be established before calling
    \c start(). Use isRunning() in order to detect whether the task tree is still running
    after a call to \c start().

    The task tree implementation relies on the running event loop.
    Make sure you have a QEventLoop or QCoreApplication or one of its
    subclasses running (or about to be run) when calling this method.

    \sa TaskTree(const Tasking::Group &), setRecipe(), isRunning(), cancel()
*/
void TaskTree::start()
{
    QT_ASSERT(!isRunning(), qWarning("The TaskTree is already running, ignoring..."); return);
    QT_ASSERT(!d->m_guard.isLocked(), qWarning("The start() is called from one of the"
                                               "TaskTree handlers, ignoring..."); return);
    d->start();
}

/*!
    \fn void TaskTree::started()

    This signal is emitted when the task tree is started. The emission of this signal is
    followed synchronously by the progressValueChanged() signal with an initial \c 0 value.

    \sa start(), done()
*/

/*!
    \fn void TaskTree::done(DoneWith result)

    This signal is emitted when the task tree finished, passing the final \a result
    of the execution. The task tree neither calls any handler,
    nor emits any signal anymore after this signal was emitted.

    \note Do not delete the task tree directly from this signal's handler.
          Use deleteLater() instead.

    \sa started()
*/

/*!
    Cancels the execution of the running task tree.

    Cancels all the running tasks immediately.
    All running tasks finish with an error, invoking their error handlers.
    All running groups dispatch their handlers according to their workflow policies,
    invoking their done handlers. The storages' onStorageDone() handlers are invoked, too.
    The progressValueChanged() signals are also being sent.
    This behavior may always be relied on.

    The \c cancel() function is executed synchronously, so that after a call to \c cancel()
    all running tasks are finished and the tree is already canceled.
    It's guaranteed that \c cancel() will run quickly, without any blocking wait for
    the currently running tasks to finish, provided the used tasks implement their destructors
    in a non-blocking way.

    When the task tree is empty, that is, constructed with a default constructor,
    a call to \c cancel() is no-op and the relevant warning message is issued.

    Otherwise, when the task tree wasn't started, a call to \c cancel() is ignored.

    \note Do not call this function directly from any of the running task's handlers
          or task tree's signals.

    \sa ~TaskTree()
*/
void TaskTree::cancel()
{
    QT_ASSERT(!d->m_guard.isLocked(), qWarning("The cancel() is called from one of the"
                                               "TaskTree handlers, ignoring..."); return);
    d->stop();
}

/*!
    Returns \c true if the task tree is currently running; otherwise returns \c false.

    \sa start(), cancel()
*/
bool TaskTree::isRunning() const
{
    return bool(d->m_runtimeRoot);
}

/*!
    Executes a local event loop with QEventLoop::ExcludeUserInputEvents and starts the task tree.

    Returns DoneWith::Success if the task tree finished successfully;
    otherwise returns DoneWith::Error.

    \note Avoid using this method from the main thread. Use asynchronous start() instead.
          This method is to be used in non-main threads or in auto tests.

    \sa start()
*/
DoneWith TaskTree::runBlocking()
{
    QPromise<void> dummy;
    dummy.start();
    return runBlocking(dummy.future());
}

/*!
    \overload runBlocking()

    The passed \a future is used for listening to the cancel event.
    When the task tree is canceled, this method cancels the passed \a future.
*/
DoneWith TaskTree::runBlocking(const QFuture<void> &future)
{
    if (future.isCanceled())
        return DoneWith::Cancel;

    DoneWith doneWith = DoneWith::Cancel;
    QEventLoop loop;
    connect(this, &TaskTree::done, &loop, [&loop, &doneWith](DoneWith result) {
        doneWith = result;
        // Otherwise, the tasks from inside the running tree that were deleteLater()
        // will be leaked. Refer to the QObject::deleteLater() docs.
        QMetaObject::invokeMethod(&loop, [&loop] { loop.quit(); }, Qt::QueuedConnection);
    });
    QFutureWatcher<void> watcher;
    connect(&watcher, &QFutureWatcherBase::canceled, this, &TaskTree::cancel);
    watcher.setFuture(future);

    QTimer::singleShot(0, this, &TaskTree::start);

    loop.exec(QEventLoop::ExcludeUserInputEvents);
    if (doneWith == DoneWith::Cancel) {
        auto nonConstFuture = future;
        nonConstFuture.cancel();
    }
    return doneWith;
}

/*!
    Constructs a temporary task tree using the passed \a recipe and runs it blocking.

    The optionally provided \a timeout is used to cancel the tree automatically after
    \a timeout milliseconds have passed.

    Returns DoneWith::Success if the task tree finished successfully;
    otherwise returns DoneWith::Error.

    \note Avoid using this method from the main thread. Use asynchronous start() instead.
          This method is to be used in non-main threads or in auto tests.

    \sa start()
*/
DoneWith TaskTree::runBlocking(const Group &recipe, milliseconds timeout)
{
    QPromise<void> dummy;
    dummy.start();
    return TaskTree::runBlocking(recipe, dummy.future(), timeout);
}

/*!
    \overload runBlocking(const Group &recipe, milliseconds timeout)

    The passed \a future is used for listening to the cancel event.
    When the task tree is canceled, this method cancels the passed \a future.
*/
DoneWith TaskTree::runBlocking(const Group &recipe, const QFuture<void> &future, milliseconds timeout)
{
    const Group root = timeout == milliseconds::max() ? recipe
                                                      : Group { recipe.withTimeout(timeout) };
    TaskTree taskTree(root);
    return taskTree.runBlocking(future);
}

/*!
    Returns the current real count of asynchronous chains of invocations.

    The returned value indicates how many times the control returns to the caller's
    event loop while the task tree is running. Initially, this value is 0.
    If the execution of the task tree finishes fully synchronously, this value remains 0.
    If the task tree contains any asynchronous tasks that are successfully started during
    a call to start(), this value is bumped to 1 just before the call to start() finishes.
    Later, when any asynchronous task finishes and any possible continuations are started,
    this value is bumped again. The bumping continues until the task tree finishes.
    When the task tree emits the done() signal, the bumping stops.
    The asyncCountChanged() signal is emitted on every bump of this value.

    \sa asyncCountChanged()
*/
int TaskTree::asyncCount() const
{
    return d->m_asyncCount;
}

/*!
    \fn void TaskTree::asyncCountChanged(int count)

    This signal is emitted when the running task tree is about to return control to the caller's
    event loop. When the task tree is started, this signal is emitted with \a count value of 0,
    and emitted later on every asyncCount() value bump with an updated \a count value.
    Every signal sent (except the initial one with the value of 0) guarantees that the task tree
    is still running asynchronously after the emission.

    \sa asyncCount()
*/

/*!
    Returns the number of asynchronous tasks contained in the stored recipe.

    \note The returned number doesn't include \l {Tasking::Sync} {Sync} tasks.
    \note Any task or group that was set up using withTimeout() increases the total number of
          tasks by \c 1.

    \sa setRecipe(), progressMaximum()
*/
int TaskTree::taskCount() const
{
    return d->m_root ? d->m_root->taskCount() : 0;
}

/*!
    \fn void TaskTree::progressValueChanged(int value)

    This signal is emitted when the running task tree finished, canceled, or skipped some tasks.
    The \a value gives the current total number of finished, canceled or skipped tasks.
    When the task tree is started, and after the started() signal was emitted,
    this signal is emitted with an initial \a value of \c 0.
    When the task tree is about to finish, and before the done() signal is emitted,
    this signal is emitted with the final \a value of progressMaximum().

    \sa progressValue(), progressMaximum()
*/

/*!
    \fn int TaskTree::progressMaximum() const

    Returns the maximum progressValue().

    \note Currently, it's the same as taskCount(). This might change in the future.

    \sa progressValue()
*/

/*!
    Returns the current progress value, which is between the \c 0 and progressMaximum().

    The returned number indicates how many tasks have been already finished, canceled, or skipped
    while the task tree is running.
    When the task tree is started, this number is set to \c 0.
    When the task tree is finished, this number always equals progressMaximum().

    \sa progressMaximum(), progressValueChanged()
*/
int TaskTree::progressValue() const
{
    return d->m_progressValue;
}

/*!
    \fn template <typename StorageStruct, typename Handler> void TaskTree::onStorageSetup(const Storage<StorageStruct> &storage, Handler &&handler)

    Installs a storage setup \a handler for the \a storage to pass the initial data
    dynamically to the running task tree.

    The \c StorageHandler takes a \e reference to the \c StorageStruct instance:

    \code
        static void save(const QString &fileName, const QByteArray &array) { ... }

        Storage<QByteArray> storage;

        const auto onSaverSetup = [storage](ConcurrentCall<QByteArray> &concurrent) {
            concurrent.setConcurrentCallData(&save, "foo.txt", *storage);
        };

        const Group root {
            storage,
            ConcurrentCallTask(onSaverSetup)
        };

        TaskTree taskTree(root);
        auto initStorage = [](QByteArray &storage){
            storage = "initial content";
        };
        taskTree.onStorageSetup(storage, initStorage);
        taskTree.start();
    \endcode

    When the running task tree enters a Group where the \a storage is placed in,
    it creates a \c StorageStruct instance, ready to be used inside this group.
    Just after the \c StorageStruct instance is created, and before any handler of this group
    is called, the task tree invokes the passed \a handler. This enables setting up
    initial content for the given storage dynamically. Later, when any group's handler is invoked,
    the task tree activates the created and initialized storage, so that it's available inside
    any group's handler.

    \sa onStorageDone()
*/

/*!
    \fn template <typename StorageStruct, typename Handler> void TaskTree::onStorageDone(const Storage<StorageStruct> &storage, Handler &&handler)

    Installs a storage done \a handler for the \a storage to retrieve the final data
    dynamically from the running task tree.

    The \c StorageHandler takes a \c const \e reference to the \c StorageStruct instance:

    \code
        static QByteArray load(const QString &fileName) { ... }

        Storage<QByteArray> storage;

        const auto onLoaderSetup = [](ConcurrentCall<QByteArray> &concurrent) {
            concurrent.setConcurrentCallData(&load, "foo.txt");
        };
        const auto onLoaderDone = [storage](const ConcurrentCall<QByteArray> &concurrent) {
            *storage = concurrent.result();
        };

        const Group root {
            storage,
            ConcurrentCallTask(onLoaderSetup, onLoaderDone, CallDoneIf::Success)
        };

        TaskTree taskTree(root);
        auto collectStorage = [](const QByteArray &storage){
            qDebug() << "final content" << storage;
        };
        taskTree.onStorageDone(storage, collectStorage);
        taskTree.start();
    \endcode

    When the running task tree is about to leave a Group where the \a storage is placed in,
    it destructs a \c StorageStruct instance.
    Just before the \c StorageStruct instance is destructed, and after all possible handlers from
    this group were called, the task tree invokes the passed \a handler. This enables reading
    the final content of the given storage dynamically and processing it further outside of
    the task tree.

    This handler is called also when the running tree is canceled. However, it's not called
    when the running tree is destructed.

    \sa onStorageSetup()
*/

void TaskTree::setupStorageHandler(const StorageBase &storage,
                                   StorageBase::StorageHandler setupHandler,
                                   StorageBase::StorageHandler doneHandler)
{
    auto it = d->m_storageHandlers.find(storage);
    if (it == d->m_storageHandlers.end()) {
        d->m_storageHandlers.insert(storage, {setupHandler, doneHandler});
        return;
    }
    if (setupHandler) {
        QT_ASSERT(!it->m_setupHandler,
                  qWarning("The storage has its setup handler defined, overriding..."));
        it->m_setupHandler = setupHandler;
    }
    if (doneHandler) {
        QT_ASSERT(!it->m_doneHandler,
                  qWarning("The storage has its done handler defined, overriding..."));
        it->m_doneHandler = doneHandler;
    }
}

TaskTreeTaskAdapter::TaskTreeTaskAdapter()
{
    connect(task(), &TaskTree::done, this,
            [this](DoneWith result) { emit done(toDoneResult(result)); });
}

void TaskTreeTaskAdapter::start()
{
    task()->start();
}

using TimeoutCallback = std::function<void()>;

struct TimerData
{
    system_clock::time_point m_deadline;
    QPointer<QObject> m_context;
    TimeoutCallback m_callback;
};

struct TimerThreadData
{
    Q_DISABLE_COPY_MOVE(TimerThreadData)

    TimerThreadData() = default; // defult constructor is required for initializing with {} since C++20 by Mingw 11.20
    QHash<int, TimerData> m_timerIdToTimerData = {};
    QMap<system_clock::time_point, QList<int>> m_deadlineToTimerId = {};
    int m_timerIdCounter = 0;
};

// Please note the thread_local keyword below guarantees a separate instance per thread.
static thread_local TimerThreadData s_threadTimerData = {};

static void removeTimerId(int timerId)
{
    const auto it = s_threadTimerData.m_timerIdToTimerData.constFind(timerId);
    QT_ASSERT(it != s_threadTimerData.m_timerIdToTimerData.cend(),
              qWarning("Removing active timerId failed."); return);

    const system_clock::time_point deadline = it->m_deadline;
    s_threadTimerData.m_timerIdToTimerData.erase(it);

    QList<int> &ids = s_threadTimerData.m_deadlineToTimerId[deadline];
    const int removedCount = ids.removeAll(timerId);
    QT_ASSERT(removedCount == 1, qWarning("Removing active timerId failed."); return);
    if (ids.isEmpty())
        s_threadTimerData.m_deadlineToTimerId.remove(deadline);
}

static void handleTimeout(int timerId)
{
    const auto itData = s_threadTimerData.m_timerIdToTimerData.constFind(timerId);
    if (itData == s_threadTimerData.m_timerIdToTimerData.cend())
        return; // The timer was already activated.

    const auto deadline = itData->m_deadline;
    while (true) {
        auto itMap = s_threadTimerData.m_deadlineToTimerId.begin();
        if (itMap == s_threadTimerData.m_deadlineToTimerId.end())
            return;

        if (itMap.key() > deadline)
            return;

        std::optional<TimerData> timerData;
        QList<int> &idList = *itMap;
        if (!idList.isEmpty()) {
            const int first = idList.first();
            idList.removeFirst();

            const auto it = s_threadTimerData.m_timerIdToTimerData.constFind(first);
            if (it != s_threadTimerData.m_timerIdToTimerData.cend()) {
                timerData = it.value();
                s_threadTimerData.m_timerIdToTimerData.erase(it);
            } else {
                QT_CHECK(false);
            }
        } else {
            QT_CHECK(false);
        }

        if (idList.isEmpty())
            s_threadTimerData.m_deadlineToTimerId.erase(itMap);
        if (timerData && timerData->m_context)
            timerData->m_callback();
    }
}

static int scheduleTimeout(milliseconds timeout, QObject *context, const TimeoutCallback &callback)
{
    const int timerId = ++s_threadTimerData.m_timerIdCounter;
    const system_clock::time_point deadline = system_clock::now() + timeout;
    QTimer::singleShot(timeout, context, [timerId] { handleTimeout(timerId); });
    s_threadTimerData.m_timerIdToTimerData.emplace(timerId, TimerData{deadline, context, callback});
    s_threadTimerData.m_deadlineToTimerId[deadline].append(timerId);
    return timerId;
}

TimeoutTaskAdapter::TimeoutTaskAdapter()
{
    *task() = milliseconds::zero();
}

TimeoutTaskAdapter::~TimeoutTaskAdapter()
{
    if (m_timerId)
        removeTimerId(*m_timerId);
}

void TimeoutTaskAdapter::start()
{
    m_timerId = scheduleTimeout(*task(), this, [this] {
        m_timerId = {};
        emit done(DoneResult::Success);
    });
}

/*!
    \typealias TaskTreeTask

    Type alias for the CustomTask, to be used inside recipes, associated with the TaskTree task.
*/

/*!
    \typealias TimeoutTask

    Type alias for the CustomTask, to be used inside recipes, associated with the
    \c std::chrono::milliseconds type. \c std::chrono::milliseconds is used to set up the
    timeout duration. The default timeout is \c std::chrono::milliseconds::zero(), that is,
    the TimeoutTask finishes as soon as the control returns to the running event loop.

    Example usage:

    \code
        using namespace std::chrono;
        using namespace std::chrono_literals;

        const auto onSetup = [](milliseconds &timeout) { timeout = 1000ms; }
        const auto onDone = [] { qDebug() << "Timed out."; }

        const Group root {
            Timeout(onSetup, onDone)
        };
    \endcode
*/

} // namespace Tasking