The Situation
My company has a QML-based application which displays some content using a custom OpenGL-based render plugin (MyGame). This plugin has a few critical needs:
To be able to effect changes in the renderer in response to QML-based signals.
(e.g. change the position of an object rendered by the game)
To only process these changes at a specific spot in MyGame's redraw loop.
(This is very important; MyGame is very sensitive about when changes are allowed.)
To have the plugin redraw at 60Hz (at least).
The Problem
The code we have right now honors (1) and (2), but fails (3); the plugin does not get visually updated consistently. (The updates are erratic, at an estimated 5-10Hz.) I believe that the plugin we have created—based on QQuickFramebufferObject—is not taking proper advantage of how Qt/QML intended the scene graph to be updated.
How can I re-structure my plugin so that I get all three of the above?
The Code
Overview:
The plugin creates a QQuickFramebufferObject (MyPlugin) and a QQuickFramebufferObject::Renderer (MyRenderer).
When MyRenderer::render() is called it calls MyGame::Redraw() itself, and then calls update().
MyGame::Redraw() does what it needs to, and at the right spot where changes can be accepted, emits a timeToMakeChanges QML signal on MyPlugin.
QML listens for the onTimeToMakeChanges signal and invokes methods on the plugin that affect MyGame.
To workaround the problem of low-frequency visual updates, I've found that if I overlay a QML Canvas over my plugin and redraw the canvas frequently using a Timer, my plugin starts to get visually updated at what appears to be around 60Hz. Clearly this is a gross hack.
Following is a summary of the code setup. Please forgive missing/incorrect code; I'm trying to distill thousands of lines of glue code down to the essentials for this question.
MyPlugin.h
#include <QOpenGLFramebufferObject>
#include <QQuickFramebufferObject>
class MyPlugin : public QQuickFramebufferObject {
Q_OBJECT
public:
MyPlugin();
virtual ~MyPlugin();
virtual QQuickFramebufferObject::Renderer* createRenderer() const;
signals:
void timeToMakeChanges();
public slots:
void makeChanges(QVariant inValue);
void HandleWindowChanged(QQuickWindow *inWindow);
private:
MyGame* GetGame() { ... }
};
MyPlugin.cpp
#include "MyPlugin.h"
#include <MyGame.h>
// ******************************************************************
class MyRenderer:
public QObject,
public QQuickFramebufferObject::Renderer,
protected QOpenGLFunctions
{
Q_OBJECT
public:
virtual void render();
private:
static void RequestGameChanges();
};
void MyRenderer::render() {
if ( !m_Initialized ) {
QOpenGLFramebufferObject *theFbo = this->framebufferObject();
InitializeGl( theFbo ); // Not shown
m_MyGame = &MyGame::Create();
m_MyGame->RegisterCallback(
reinterpret_cast<qml_Function>(MyRenderer::RequestGameChanges)
);
m_Initialized = true;
}
m_MyGame->RestoreState();
m_MyGame->Redraw();
m_MyGame->SaveState();
m_PluginItem->window()->resetOpenGLState();
// Tell QML that we want to render again as soon as possible
update();
}
// This gets invoked in the middle of m_MyGame->Redraw()
void MyRenderer::RequestGameChanges() {
emit m_PluginItem->timeToMakeChanges();
}
// ******************************************************************
MyPlugin::MyPlugin() {
setMirrorVertically(true);
connect(
this, SIGNAL(windowChanged(QQuickWindow*)),
this, SLOT(HandleWindowChanged(QQuickWindow*))
);
}
void MyPlugin::HandleWindowChanged(QQuickWindow *inWindow) {
inWindow->setClearBeforeRendering(false);
}
void MyPlugin::makeChanges(QVariant inValue) {
MyGame *theGame = GetGame();
// Send the requested changes to theGame
}
QQuickFramebufferObject::Renderer* MyPlugin::createRenderer() const {
m_Renderer = new MyRenderer( *this );
}
MyApp.qml
import MyPlugin 1.0
Window {
MyPlugin {
property var queuedUpChanges: ([])
onSomeOtherSignal: queueUpChangesToMake();
onTimeToMakeChanges: makeChanges( queuedUpChanges );
}
Canvas { id:hack }
Timer {
interval:10; running:true; repeat:true
onTriggered: hack.changeWhatYouShow();
}
}
Bonus Points
The main question is "How do I modify my code so that I get 60Hz updates?" However, as seen in the QML, the setup above requires me to queue up all changes in QML so that they are able to be applied during the right spot in the MyGame::Render().
Ideally, I'd prefer to write QML without timeToMakeChanges, like:
MyPlugin {
onSomeOtherSignal: makeChanges( ... );
}
If there's a way to accomplish this (other than queuing up the changes in C++ instead)—perhaps something related to synchronize() I'd love to know about it.
I'd make a timer in QML that calls the makeChanges regularly. But store all the state in MyPlugin. Then, in Renderer::synchronize(), copy from MyPlugin to MyRenderer, so it can be used by the MyGame.
(although, I wouldn't do any gamelogic-related calculations in QML ever in the first place)
Related
I am hiding a Qt toolbox window in one of my functions. I need to do some screen capturing stuffs to be executed in function after hiding my toolbox window.
void ProcessData()
{
Toolbox::getInstance()->hide(); //I am trying to hide toolbox here
//screen capture stuffs. - It should be executed once the toolbox window is hidden.
}
But what happens is, Toolbox is hidden only when the ProcessData() returns. This leads to the problem that my captured screen have the toolbox.
Is there any way to wait for the toolbox to be hidden in Qt ?
Apart from using QApplication::processEvents, I would suggest you a more elegant solution:
Reimplement QWidget::hideEvent of the class returned by Toolbox::getInstance to emit a signal, e.g. widgetHidden
protected:
void hideEvent(QHideEvent *event) {
QWidget::hideEvent(event);
Q_EMIT widgetHidden();
}
signals:
void widgetHidden();
Connect widgetHidden to a slot to capture the screen
connect(Toolbox::getInstance(), &MyWidget::widgetHidden, this, &ClassName::onWidgetHidden);
where MyWidget is the class returned by Toolbox::getInstance and ClassName is the name of the class to which ProcessData belongs.
Add a flag to know if it is the right time to capture, as the widget might be hidden in some other way
bool m_captureAllowed = false;
Implement the onWidgetHidden slot
private slots:
void onWidgetHidden() {
if (m_captureAllowed) {
m_captureAllowed = false;
//screen capture stuffs
}
}
Set the flag in ProcessData
void ProcessData()
{
m_captureAllowed = true;
Toolbox::getInstance()->hide(); //I am trying to hide toolbox here
}
I want to write an tool of objective annotation using qt5 MinGw32, which could annotate object in video file and diplay them during playing. So QGraphicsScene is inherited to implementation the function.
Something wrong happens when I change the QGraphicsScene's background frequently(e.g. 30 fps): most of time it works as expected while sometimes the background could not move.
Here is my code:
void MyGraphicsScene::UpdateFrame(QImage image)
{
QPixmap pixmap = QPixmap::fromImage(image);
//fix the view's size and scene's size
views().first()->setFixedSize(pixmap.size());
setSceneRect(0,0, pixmap.width(), pixmap.height());
setBackgroundBrush(QBrush(pixmap));
}
...
//In another thread
void Thread::run()
{
...
myScene.UpdateFrame(newImage);
...
}
I have search through the qt's document and found no answer.
However, there is something strange:
when wrong thing happens, I find the background continues to change, but it didn't show change on the screen unless I move the app to another screen (I have two screen). However, with the app moved, the QGraphicsScene's background just change once and becomes static afterwards.
I guess the background has been changed but doesn't repainted, so I used update(), but it didn't help.
BTW, I couldn't reproduce the occasion, sometiems it happens, somtimes not.
do I need to represented any methods? Or I called the methods in a wrong way? Or is there an alternative approach that would work?
Many thanks for your help in advance.
You should not change QtGui elements from a different thread by calling a method directly.
Use the Qt signal-slot concept.
Calling update() is not necessary.
class MyGraphicsScene{...
....
signals:
void singalFromThread(QImage image);
public:
//CTor
MyGraphicsScene()
{
connect(this, SIGNAL(singalFromThread(QImage)),this,SLOT(UpdateFrame(QImage)));
}
//Call this method from your thread
void updateForwarder(QImage image)
{
//Jump into gui thread
emit singalFromThread(image);
}
public slots:
void UpdateFrame(QImage image)
{
setBackgroundBrush(....);
}
};
In Qt4, we could use the embedInto method of the QX11EmbedWidget class to embed any QWidget into another application. But in Qt5, the QX11EmbedWidget class was dropped. I've been searching Google for many hours, but all I found indicated to use the QWidget::createWindowContainer method instead. However, as far as I understand, this method is rather a replacement for QX11EmbedContainer.
So here's the question: How do we embed a Qt5-Widget into a non-Qt window? -
And in case it metters: The non-Qt window is Gtk3.
I've been asked for the use-case, so let me illustrate briefly: Consider a Gtk-based application, which you are willing to extend - and the UI component, which you need, already exists, but is written in Qt. The goal is to avoid porting the application to Qt or the component to Gtk.
My initial approach was to request the embedding from Qt side. I had no luck with this, as I couldn't find any equivalent for QX11EmbedWidget::embedInto in Qt5. After giving up on this, I decided to give XReparentWindow a try, which was reportedly used with success. However, due to lack of documentation, I wasn't able to reproduce this. My third attempt was to initiate the embedding from server side, which is the Gtk window in my case. I'm glad to report, that it worked :-)
For the Record: How to embed an arbitrary Qt5 Widget into a Gtk window
TL;DR:
Make sure that you write your Gtk code in a unit which is separate from the Qt part. This is necessary in order to avoid name conflicts between Qt5 and Gtk3.
Simply use gtk_socket_add_id to embed any X11 window into a GtkSocket.
Also see the documentation for reference, but note that the example given their doesn't even compile, because they forgot the GTK_SOCKET macro. On the contrary, the following code works.
Details
The QGtkWindow class provides an interface to the Gtk window.
class QGtkWindow : public QObject
{
public:
QGtkWindow();
virtual ~QGtkWindow();
void setCentralWidget( QWidget* const widget );
void show();
private:
GtkWindowAdapter gtkWindow;
QWidget* const container;
}; // QGtkWindow
The GtkWindowAdapter class wraps the Gtk calls and isolates them from the Qt part of the application. An object of this class represents the Gtk window.
class GtkWindowAdapter
{
public:
GtkWindowAdapter();
~GtkWindowAdapter();
void show();
void embed( unsigned long clientWinId );
private:
struct Details;
const std::auto_ptr< Details > pimpl;
}; // GtkWindowAdapter
When instantiated, an GtkWindowAdapter object first initializes Gtk,
static bool gtkInitialized = false;
struct GtkWindowAdapter::Details
{
GtkWidget* widget;
GtkWidget* socket;
void setupUi();
};
GtkWindowAdapter::GtkWindowAdapter()
: pimpl( new Details() )
{
if( !gtkInitialized )
{
int argc = 0;
gtk_init( &argc, NULL );
gtkInitialized = true;
}
pimpl->setupUi();
}
and then setups the Gtk window:
void GtkWindowAdapter::Details::setupUi()
{
widget = gtk_window_new( GTK_WINDOW_TOPLEVEL );
socket = gtk_socket_new();
gtk_widget_show( socket );
gtk_container_add( GTK_CONTAINER ( widget ), socket );
gtk_widget_realize( socket );
}
You might already have noted the embed method which this class provides. This method initiates the embedding of any X11 window. The show method turns the encapsulated Gtk window visible.
void GtkWindowAdapter::embed( unsigned long clientWinId )
{
gtk_socket_add_id( GTK_SOCKET( pimpl->socket ), clientWinId );
}
void GtkWindowAdapter::show()
{
gtk_widget_show( pimpl->widget );
}
Now, the implementation of the QGtkWindow class is fairly simple. When created, it initializes a Gtk window by using the GtkWindowApdater and puts an empty QWidget into that window:
QGtkWindow::QGtkWindow()
: container( new QWidget() )
{
container->setLayout( new QVBoxLayout() );
gtkWindow.embed( container->winId() );
container->show();
}
When the user of the QGtkWindow class decides to put some widget into the window, the setCentralWidget is the way to go. It simply clears the parent widget, which was embedded into the Gtk window originally, then inserts the user's widget instead:
void QGtkWindow::setCentralWidget( QWidget* const widget )
{
qDeleteAll( pimpl->container->layout()->children() );
pimpl->container->layout()->addWidget( widget );
}
void QGtkWindow::show()
{
pimpl->gtkWindow.show();
}
Hope this might spare somebody many hours.
To quote from this link:
https://forum.qt.io/topic/32785/qwindow-qwidget-qt5-x11embedding-how/2
All the x11 stuff has moved to the "extras" department found on
Gitorious. (QX11EmbedWidgets and QX11EmbedContainer and the likes are
not in 5.x)
Try this:
http://qt-project.org/doc/qt-5.1/qtx11extras/qx11info.html
I wanted to do some action when a dialog shows when it opens or when it maximizes from a minimal status or it moves from out of a screen.
Does QT have such a signal?
I am also not sure where to find if QT has a list of signals defined.
Does each QT widget have a 'show' signal?
If you look at Qt source code then you will find QWidget::show to be a slot:
public Q_SLOTS:
// Widget management functions
virtual void setVisible(bool visible);
void setHidden(bool hidden);
void show();
The slot is mainly for us, programmers to make us able to connect with signals for specific purposes like clicking the button we created does something to certain widget. As for Windows or Mac OS, we have the app serving all the events coming from the system via event loop. And QWidget reacts on all the 'signals' in the form of system events coming and yes, may, execute show() or showMaximized() or showMinimized slots then.
But I can assume you want to overload
virtual void showEvent(QShowEvent *);
virtual void hideEvent(QHideEvent *);
Like:
void MyWidget::showEvent(QShowEvent *e)
{
if (isMaximized())
{
if (e->spontaneous())
{
// the author would like to know
// if the event is issued by the system
}
; // the action for maximized
}
else
{
; // the action for normal show
}
}
void MyWidget::hideEvent(QHideEvent *)
{
if (isMinimized())
{
; // the action for minimized
}
else
{
; // the action for hide
}
}
For recognizing cases when the system operates the widget we can use QEvent::spontaneous().
Please also refer to show and hide event doc pages:
http://doc.qt.io/qt-5/qshowevent-members.html
http://doc.qt.io/qt-5/qhideevent.html
I have an application which needs to draw on a pixel by pixel basis at a specified frame rate (simulating an old machine). One caveat is that the main machine engine runs in a background thread in order to ensure that the UI remains responsive and usable during simulation.
Currently, I am toying with using something like this:
class QVideo : public QWidget {
public:
QVideo(QWidget *parent, Qt::WindowFlags f) : QWidget(parent, f), screen_image_(256, 240, QImage::Format_RGB32) {
}
void draw_frame(void *data) {
// render data into screen_image_
}
void start_frame() {
// do any pre-rendering prep work that needs to be done before
// each frame
}
void end_frame() {
update(); // force a paint event
}
void paintEvent(QPaintEvent *) {
QPainter p(this);
p.drawImage(rect(), screen_image_, screen_image_.rect());
}
QImage screen_image_;
};
This is mostly effective, and surprisingly not very slow. However, there is an issue. The update function schedules a paintEvent, it may not hapen right away. In fact, a bunch of paintEvent's may get "combined" according to the Qt documentation.
The negative effect that I am seeing is that after a few minutes of simulation, the screen stops updating (image appears frozen though simulation is still running) until I do something that forces a screen update for example switching the window in and out of maximized.
I have experimented with using QTimer's and other similar mechanism to have the effect of the rendering being in the GUI thread so that I can force immediate updates, but this resulted in unacceptable performance issues.
Is there a better way to draw pixels onto a widget constantly at a fixed interval. Pure Qt solutions are preferred.
EDIT: Since some people choose to have an attitude instead of reading the whole question, I will clarify the issue. I cannot use QWidget::repaint because it has a limitation in that it must be called from the same thread as the event loop. Otherwise, no update occurs and instead I get qDebug messages such as these:
QPixmap: It is not safe to use pixmaps outside the GUI thread
QPixmap: It is not safe to use pixmaps outside the GUI thread
QWidget::repaint: Recursive repaint detected
QPainter::begin: A paint device can only be painted by one painter at a time.
QWidget::repaint: It is dangerous to leave painters active on a widget outside of the PaintEvent
QWidget::repaint: It is dangerous to leave painters active on a widget outside of the PaintEvent
EDIT: to demonstrate the issue I have created this simple example code:
QVideo.h
#include <QWidget>
#include <QPainter>
class QVideo : public QWidget {
Q_OBJECT;
public:
QVideo(QWidget *parent = 0, Qt::WindowFlags f = 0) : QWidget(parent, f), screen_image_(256, 240, QImage::Format_RGB32) {
}
void draw_frame(void *data) {
// render data into screen_image_
// I am using fill here, but in the real thing I am rendering
// on a pixel by pixel basis
screen_image_.fill(rand());
}
void start_frame() {
// do any pre-rendering prep work that needs to be done before
// each frame
}
void end_frame() {
//update(); // force a paint event
repaint();
}
void paintEvent(QPaintEvent *) {
QPainter p(this);
p.drawImage(rect(), screen_image_, screen_image_.rect());
}
QImage screen_image_;
};
main.cc:
#include <QApplication>
#include <QThread>
#include <cstdio>
#include "QVideo.h"
struct Thread : public QThread {
Thread(QVideo *v) : v_(v) {
}
void run() {
while(1) {
v_->start_frame();
v_->draw_frame(0); // contents doesn't matter for this example
v_->end_frame();
QThread::sleep(1);
}
}
QVideo *v_;
};
int main(int argc, char *argv[]) {
QApplication app(argc, argv);
QVideo w;
w.show();
Thread t(&w);
t.start();
return app.exec();
}
I am definitely willing to explore options which don't use a temporary QImage to render. It is just the only class in Qt which seems to have a direct pixel writing interface.
Try emitting a signal from the thread to a slot in the event loop widget that calls repaint(), which will then execute right away. I am doing something like this in my graphing program, which executes the main calculations in one thread, then tells the widget when it is time to repaint() the data.
In similar cases what I did was still using a QTimer, but doing several simulation steps instead of just one. You can even make the program auto-tuning the number of simulation steps to be able to get whatever frames per seconds you like for the screen update.