codecompletionmodel.h

/*
    SPDX-FileCopyrightText: 2007-2008 David Nolden <[email protected]>
    SPDX-FileCopyrightText: 2005-2006 Hamish Rodda <[email protected]>
 
    SPDX-License-Identifier: LGPL-2.0-or-later
*/
 
#ifndef KTEXTEDITOR_CODECOMPLETIONMODEL_H
#define KTEXTEDITOR_CODECOMPLETIONMODEL_H
 
#include <QModelIndex>
#include <ktexteditor/range.h>
#include <ktexteditor_export.h>
 
namespace KTextEditor
{
class Document;
class View;
 
/**
 * \class CodeCompletionModel codecompletionmodel.h <KTextEditor/CodeCompletionModel>
 *
 * \short An item model for providing code completion, and meta information for
 *        enhanced presentation.
 *
 * \section compmodel_intro Introduction
 *
 * The CodeCompletionModel is the actual workhorse to provide code completions
 * in a KTextEditor::View. It is meant to be used in conjunction with the
 * CodeCompletionInterface. The CodeCompletionModel is not meant to be used as
 * is. Rather you need to implement a subclass of CodeCompletionModel to actually
 * generate completions appropriate for your type of Document.
 *
 * \section compmodel_implementing Implementing a CodeCompletionModel
 *
 * The CodeCompletionModel is a QAbstractItemModel, and can be subclassed in the
 * same way. It provides default implementations of several members, however, so
 * in most cases (if your completions are essentially a non-hierarchical, flat list
 * of matches) you will only need to overload few virtual functions.
 *
 * \section compmodel_flatlist Implementing a CodeCompletionModel for a flat list
 *
 * For the simple case of a flat list of completions, you will need to:
 *  - Implement completionInvoked() to actually generate/update the list of completion
 * matches
 *  - implement itemData() (or QAbstractItemModel::data()) to return the information that
 * should be displayed for each match.
 *  - use setRowCount() to reflect the number of matches.
 *
 * \section compmodel_roles_columns Columns and roles
 *
 * \todo document the meaning and usage of the columns and roles used by the
 * CodeCompletionInterface
 *
 * \section compmodel_usage Using the new CodeCompletionModel
 *
 * To start using your CodeCompletionModel, refer to CodeCompletionInterface.
 *
 * \section compmodel_controller ControllerInterface to get more control
 *
 * To have more control over code completion implement
 * CodeCompletionModelControllerInterface in your CodeCompletionModel.
 *
 * \see CodeCompletionInterface, CodeCompletionModelControllerInterface
 * @author Hamish Rodda <[email protected]>
 */
class KTEXTEDITOR_EXPORT CodeCompletionModel : public QAbstractItemModel
{
    Q_OBJECT
 
public:
    explicit CodeCompletionModel(QObject *parent);
    ~CodeCompletionModel() override;
 
    enum Columns {
        Prefix = 0,
        /// Icon representing the type of completion. We have a separate icon field
        /// so that names remain aligned where only some completions have icons,
        /// and so that they can be rearranged by the user.
        Icon,
        Scope,
        Name,
        Arguments,
        Postfix
    };
    static const int ColumnCount = Postfix + 1;
 
    /// @see CompletionProperties
    enum CompletionProperty {
        NoProperty = 0x0,
        FirstProperty = 0x1,
 
        // Access specifiers - no more than 1 per item
        Public = 0x1,
        Protected = 0x2,
        Private = 0x4,
 
        // Extra access specifiers - any number per item
        Static = 0x8,
        Const = 0x10,
 
        // Type - no more than 1 per item (except for Template)
        Namespace = 0x20,
        Class = 0x40,
        Struct = 0x80,
        Union = 0x100,
        Function = 0x200,
        Variable = 0x400,
        Enum = 0x800,
        Template = 0x1000,
        TypeAlias = 0x2000,
 
        // Special attributes - any number per item
        Virtual = 0x4000,
        Override = 0x8000,
        Inline = 0x10000,
        Friend = 0x20000,
        Signal = 0x40000,
        Slot = 0x80000,
 
        // Scope - no more than 1 per item
        LocalScope = 0x100000,
        NamespaceScope = 0x200000,
        GlobalScope = 0x400000,
 
        // Keep this in sync so the code knows when to stop
        LastProperty = GlobalScope
    };
    /// Stores a combination of #CompletionProperty values.
    Q_DECLARE_FLAGS(CompletionProperties, CompletionProperty)
 
    /// @see HighlightMethods
    enum HighlightMethod { NoHighlighting = 0x0, InternalHighlighting = 0x1, CustomHighlighting = 0x2 };
    /// Stores a combination of #HighlightMethod values.
    Q_DECLARE_FLAGS(HighlightMethods, HighlightMethod)
 
    /// Meta information is passed through extra {Qt::ItemDataRole}s.
    /// This information should be returned when requested on the Name column.
    enum ExtraItemDataRoles {
        /// The model should return a set of CompletionProperties.
        CompletionRole = Qt::UserRole,
 
        /// The model should return an index to the scope
        /// -1 represents no scope
        /// \todo how to sort scope?
        ScopeIndex,
 
        /**
         * If requested, your model should try to determine whether the
         * completion in question is a suitable match for the context (ie.
         * is accessible, exported, + returns the data type required).
         *
         * The returned data should ideally be matched against the argument-hint context
         * set earlier by SetMatchContext.
         *
         * Return an integer value that should be positive if the completion is suitable,
         * and zero if the completion is not suitable. The value should be between 0 an 10, where
         * 10 means perfect match.
         *
         * Return QVariant::Invalid if you are unable to determine this.
         */
        MatchQuality,
 
        /**
         * Is requested before MatchQuality(..) is requested. The item on which this is requested
         * is an argument-hint item(@see ArgumentHintDepth). When this role is requested, the item should
         * be noted, and whenever MatchQuality is requested, it should be computed by matching the item given
         * with MatchQuality into the context chosen by SetMatchContext.
         *
         * Feel free to ignore this, but ideally you should return QVariant::Invalid to make clear that your model does not support this.
         * */
        SetMatchContext,
 
        /**
         * Define which highlighting method will be used:
         * - QVariant::Invalid - allows the editor to choose (usually internal highlighting)
         * - QVariant::Integer - highlight as specified by HighlightMethods.
         */
        HighlightingMethod,
 
        /**
         * Allows an item to provide custom highlighting.  Return a
         * QList<QVariant> in the following format:
         * - int startColumn (where 0 = start of the completion entry)
         * - int endColumn (note: not length)
         * - QTextFormat attribute (note: attribute may be a KTextEditor::Attribute, as it is a child class)
         *   If the attribute is invalid, and the item is an argument-hint, the text will be drawn with
         *   a background-color depending on match-quality, or yellow.
         *   You can use that to mark the actual arguments that are matched in an argument-hint.
         *
         * Repeat this triplet as many times as required, however each column must be >= the previous,
         * and startColumn != endColumn.
         */
        CustomHighlight,
 
        /**
         * Returns the inheritance depth of the completion.  For example, a completion
         * which comes from the base class would have depth 0, one from a parent class
         * would have depth 1, one from that class' parent 2, etc. you can use this to
         * symbolize the general distance of a completion-item from a user. It will be used
         * for sorting.
         */
        InheritanceDepth,
 
        /**
         * This allows items in the completion-list to be expandable. If a model returns an QVariant bool value
         * that evaluates to true, the completion-widget will draw a handle to expand the item, and will also make
         * that action accessible through keyboard.
         */
        IsExpandable,
        /**
         * After a model returned true for a row on IsExpandable, the row may be expanded by the user.
         * When this happens, ExpandingWidget is requested.
         *
         * The model may return two types of values:
         * QWidget*:
         *  If the model returns a QVariant of type QWidget*, the code-completion takes over the given widget
         *  and embeds it into the completion-list under the completion-item. The widget will be automatically deleted at some point.
         *  The completion-widget will use the height of the widget as a hint for its preferred size, but it will
         *  resize the widget at will.
         * QString:
         *  If the mode returns a QVariant of type QString, it will create a small html-widget showing the given html-code,
         *  and embed it into the completion-list under the completion-item.
         *
         * @warning
         * @code
         *   QWidget* widget;
         *   return QVariant(widget);
         * @endcode
         * Will not work correctly!
         * Use the following instead.:
         * @code
         *   QVariant v;
         *   v.setValue<QWidget*>(widget);
         *   return v;
         * @endcode
         *
         * */
        ExpandingWidget,
        /**
         * Whenever an item is selected, this will be requested from the underlying model.
         * It may be used as a simple notification that the item was selected.
         *
         * Above that, the model may return a QString, which then should then contain html-code. A html-widget
         * will then be displayed as a one- or two-liner under the currently selected item(it will be partially expanded)
         * */
        ItemSelected,
 
        /**Is this completion-item an argument-hint?
         * The model should return an integral positive number if the item is an argument-hint, and QVariant() or 0 if it is not one.
         *
         * The returned depth-integer is important for sorting and matching.
         * Example:
         * "otherFunction(function1(function2("
         * all functions named function2 should have ArgumentHintDepth 1, all functions found for function1 should have ArgumentHintDepth 2,
         * and all functions named otherFunction should have ArgumentHintDepth 3
         *
         * Later, a completed item may then be matched with the first argument of function2, the return-type of function2 with the first
         * argument-type of function1, and the return-type of function1 with the argument-type of otherFunction.
         *
         * If the model returns a positive value on this role for a row, the content will be treated specially:
         * - It will be shown in a separate argument-hint list
         * - It will be sorted by Argument-hint depth
         * - Match-qualities will be illustrated by differently highlighting the matched argument if possible
         * The argument-hint list strings will be built from all source-model, with a little special behavior:
         * Prefix - Should be all text of the function-signature up to left of the matched argument of the function
         * Name - Should be the type and name of the function's matched argument. This part will be highlighted differently depending on the match-quality
         * Suffix - Should be all the text of the function-signature behind the matched argument
         *
         * Example: You are matching a function with signature "void test(int param1, int param2)", and you are matching the first argument.
         * The model should then return:
         * Prefix: "void test("
         * Name: "int param1"
         * Suffix: ", int param2)"
         *
         * If you don't use the highlighting, matching, etc. you can also return the columns in the usual way.
         * */
        ArgumentHintDepth,
 
        /**
         * This will be requested for each item to ask whether it should be included in computing a best-matches list.
         * If you return a valid positive integer n here, the n best matches will be listed at top of the completion-list separately.
         *
         * This is expensive because all items of the whole completion-list will be tested for their matching-quality, with each of the level 1
         * argument-hints.
         *
         * For that reason the end-user should be able to disable this feature.
         */
        BestMatchesCount,
 
        /**
         * The following three enumeration-values are only used on expanded completion-list items that contain an expanding-widget(@see ExpandingWidget)
         *
         * You can use them to allow the user to interact with the widget by keyboard.
         *
         * AccessibilityNext will be requested on an item if it is expanded, contains an expanding-widget, and the user triggers a special navigation
         * short-cut to go to navigate to the next position within the expanding-widget(if applicable).
         *
         * Return QVariant(true) if the input was used.
         * */
        AccessibilityNext,
        /**
         * AccessibilityPrevious will be requested on an item if it is expanded, contains an expanding-widget, and the user triggers a special navigation
         * short-cut to go to navigate to the previous position within the expanding-widget(if applicable).
         *
         * Return QVariant(true) if the input was used.
         * */
        AccessibilityPrevious,
        /**
         * AccessibilityAccept will be requested on an item if it is expanded, contains an expanding-widget, and the user triggers a special
         * shortcut to trigger the action associated with the position within the expanding-widget the user has navigated to using AccessibilityNext and
         * AccessibilityPrevious.
         *
         * This should return QVariant(true) if an action was triggered, else QVariant(false) or QVariant().
         * */
        AccessibilityAccept,
 
        /**
         * Using this Role, it is possible to greatly optimize the time needed to process very long completion-lists.
         *
         * In the completion-list, the items are usually ordered by some properties like argument-hint depth,
         * inheritance-depth and attributes. Kate usually has to query the completion-models for these values
         * for each item in the completion-list in order to extract the argument-hints and correctly sort the
         * completion-list. However, with a very long completion-list, only a very small fraction of the items is actually
         * visible.
         *
         * By using a tree structure you can give the items in a grouped order to kate, so it does not need to look at each
         * item and query data in order to initially show the completion-list.
         *
         * This is how it works:
         * - You create a tree-structure for your items
         * - Every inner node of the tree defines one attribute value that all sub-nodes have in common.
         *   - When the inner node is queried for GroupRole, it should return the "ExtraItemDataRoles" that all sub-nodes have in common
         *   - When the inner node is then queried for that exact role, it should return that value.
         *   - No other queries will be done to inner nodes.
         * - Every leaf node stands for an actual item in the completion list.
         *
         * The recommended grouping order is: Argument-hint depth, inheritance depth, attributes.
         *
         * This role can also be used to define completely custom groups, bypassing the editors builtin grouping:
         *  - Return Qt::DisplayRole when GroupRole is requested
         *  - Return the label text of the group when Qt::DisplayRole
         *   - Optional: Return an integer sorting-value when InheritanceDepth is  requested. This number will
         *               be used to determine the order of the groups. The order of the builtin groups is:
         *               1 = Best Matches, 100 = Local Scope, 200 = Public, 300 = Protected, 400 = Private, 500 = Namespace, 600 = Global
         *               You can pick any arbitrary number to position your group relative to these builtin groups.
         * */
        GroupRole,
 
        /**
         * Return a nonzero value here to enforce sorting the item at the end of the list.
         */
        UnimportantItemRole,
 
        LastExtraItemDataRole
    };
 
    void setRowCount(int rowCount);
 
    enum InvocationType { AutomaticInvocation, UserInvocation, ManualInvocation };
 
    /**
     * This function is responsible to generating / updating the list of current
     * completions. The default implementation does nothing.
     *
     * When implementing this function, remember to call setRowCount() (or implement
     * rowCount()), and to generate the appropriate change notifications (for instance
     * by calling QAbstractItemModel::reset()).
     * @param view The view to generate completions for
     * @param range The range of text to generate completions for
     * @param invocationType How the code completion was triggered
     * */
    virtual void completionInvoked(KTextEditor::View *view, const KTextEditor::Range &range, InvocationType invocationType);
 
    /**
     * This function is responsible for inserting a selected completion into the
     * view. The default implementation replaces the text that the completions
     * were based on with the Qt::DisplayRole of the Name column of the given match.
     *
     * @param view the view to insert the completion into
     * @param word the Range that the completions are based on (what the user entered
     * so far)
     * @param index identifies the completion match to insert
     * */
    virtual void executeCompletionItem(KTextEditor::View *view, const Range &word, const QModelIndex &index) const;
 
    // Reimplementations
    /**
     * Reimplemented from QAbstractItemModel::columnCount(). The default implementation
     * returns ColumnCount for all indices.
     * */
    int columnCount(const QModelIndex &parent = QModelIndex()) const override;
    /**
     * Reimplemented from QAbstractItemModel::index(). The default implementation
     * returns a standard QModelIndex as long as the row and column are valid.
     * */
    QModelIndex index(int row, int column, const QModelIndex &parent = QModelIndex()) const override;
    /**
     * Reimplemented from QAbstractItemModel::itemData(). The default implementation
     * returns a map with the QAbstractItemModel::data() for all roles that are used
     * by the CodeCompletionInterface. You will need to reimplement either this
     * function or QAbstractItemModel::data() in your CodeCompletionModel.
     * */
    QMap<int, QVariant> itemData(const QModelIndex &index) const override;
    /**
     * Reimplemented from QAbstractItemModel::parent(). The default implementation
     * returns an invalid QModelIndex for all items. This is appropriate for
     * non-hierarchical / flat lists of completions.
     * */
    QModelIndex parent(const QModelIndex &index) const override;
    /**
     * Reimplemented from QAbstractItemModel::rowCount(). The default implementation
     * returns the value set by setRowCount() for invalid (toplevel) indices, and 0
     * for all other indices. This is appropriate for non-hierarchical / flat lists
     * of completions
     * */
    int rowCount(const QModelIndex &parent = QModelIndex()) const override;
 
    /**
     * This function returns true if the model needs grouping, otherwise false.
     * The default is false if not changed via setHasGroups().
     */
    bool hasGroups() const;
 
Q_SIGNALS:
 
    /**
     * Emit this if the code-completion for this model was invoked, some time is needed in order to get the data,
     * and the model is reset once the data is available.
     *
     * This only has an effect if emitted from within completionInvoked(..).
     *
     * This prevents the code-completion list from showing until this model is reset,
     * so there is no annoying flashing in the user-interface resulting from other models
     * supplying their data earlier.
     *
     * @note The implementation may choose to show the completion-list anyway after some timeout
     *
     * @warning If you emit this, you _must_ also reset the model at some point,
     *                  else the code-completion will be completely broken to the user.
     *                  Consider that there may always be additional completion-models apart from yours.
     *
     * @since 4.3
     */
    void waitForReset();
 
    /**
     * Internal
     */
    void hasGroupsChanged(KTextEditor::CodeCompletionModel *model, bool hasGroups);
 
protected:
    void setHasGroups(bool hasGroups);
 
private:
    class CodeCompletionModelPrivate *const d;
};
 
Q_DECLARE_OPERATORS_FOR_FLAGS(CodeCompletionModel::CompletionProperties)
Q_DECLARE_OPERATORS_FOR_FLAGS(CodeCompletionModel::HighlightMethods)
 
}
 
#endif // KTEXTEDITOR_CODECOMPLETIONMODEL_H