[PySide6] Slot Executed in Signal's Thread Even With Auto & QueuedConnection

[thinkinmachine]

Hi @friedemannkleint ,

Thanks for the reply. Sorry I forgot to add Slot decorator in the example. The result is still the same with the decorator.

import sys

from PySide6.QtCore import QObject, Signal, QThread, Qt, Slot
from PySide6.QtWidgets import QMainWindow, QApplication


class Worker(QObject):
    update = Signal()

    def run(self):
        print(f"worker: {QThread.currentThread().objectName()}")
        self.update.emit()


class Window(QMainWindow):
    def __init__(self, parent=None):
        super().__init__(parent)
        worker = Worker()
        thread = QThread(self)
        thread.setObjectName('test thread')
        worker.moveToThread(thread)
        thread.started.connect(worker.run)
        worker.update.connect(self.update)
        thread.start()

    @Slot()
    def update(self):
        print(f"window: {QThread.currentThread().objectName()}")


if __name__ == '__main__':
    app = QApplication()
    w = Window()
    w.show()
    sys.exit(app.exec())

Subclassing QThread did yield correct result.

import sys

from PySide6.QtCore import QObject, Signal, QThread, Qt, Slot
from PySide6.QtWidgets import QMainWindow, QApplication


class Thread(QThread):
    update = Signal()

    def run(self):
        print(f"thread: {QThread.currentThread().objectName()}")
        self.update.emit()


class Window(QMainWindow):
    def __init__(self, parent=None):
        super().__init__(parent)
        thread = Thread(self)
        thread.setObjectName('test thread')
        thread.update.connect(self.update)
        thread.start()

    @Slot()
    def update(self):
        print(f"window: {QThread.currentThread().objectName()}")


if __name__ == '__main__':
    app = QApplication()
    w = Window()
    w.show()
    sys.exit(app.exec())

However, I remember it's suggested to use a separate worker in some posts I read. Is it a PySide bug that the worker method doesn't work as expected? It's almost identical to the C++ example in the official document.

[Pl45m4]

@friedemannkleint said in [PySide6] Slot Executed in Signal's Thread Even With Auto & QueuedConnection:

The correct way to implement threads is to overwrite QThread.run() to do the work.

Why is there only one correct approach?!
Shouldn't the worker thread approach work the same (as also when using C++)?

Here in this topic the worker approach seems to function correctly (besides the mistakes OP made there).

[thinkinmachine]

@Pl45m4

Thanks for the reply! I did read that post. I’m really confused to see that one works because it’s almost the same as mine. Do you have any idea why this is happening?

[thinkinmachine]

It's reported to Qt Bug Tracker: https://bugreports.qt.io/browse/PYSIDE-2990

Update: tested cpp equivalent version and it worked fine. It should be a PySide issue.

[friedemannkleint]

Please don't use that technique of connecting a worker to the QThread.start() signal. It leaks a thread hanging in its exec() function since that is called by the default implementation of run(), plus the signal execution is blocked by the work.

[thinkinmachine]

@friedemannkleint
Thank you so much. I didn't notice update would be a issue.

However I found another issue. When I use PyCharm's debugger to run the code the thread of worker becomes main thread again. It might have to do with debugger settings.

Update: When the main thread is blocked by breakpoint, the child threads are all blocked as well. This ONLY happens with PyCharm debugging.

[FlowOverNestedThreads]

@thinkinmachine can confirm (dear lord, how have you even pinpointed the reason?), also related

[thinkinmachine]

Also, it doesn't work if lambda is used. Probably because Qt doesn't know which thread the lambda function should be in.

self._worker.update.connect(lambda: self.slotUpdate())

[Pl45m4]

@thinkinmachine said in [PySide6] Slot Executed in Signal's Thread Even With Auto & QueuedConnection:

Probably because Qt doesn't know

It's not a Qt thing I would say, rather than Python-related.
I'm not the best at Python, but AFAIK there is not "context" in Python lambdas (correct me if I'm wrong).
In C++ the Qt Signal-Slot lambda is executed in the "context" (if any) object's thread, or the receiving one.
Like

// ------------------------------------
Worker w;
int result = 42;
emit w.workFinished(result);
// ------------------------------------
connect(&w, &Worker::workFinished, this, [=](int res) {
       int result = res + 123;
       doSomethingWith(result);
});

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

[thinkinmachine]

@FlowOverNestedThreads

I reported this bug to JetBrain PyCharm team (here). Hopefully they can help figure out what's going on. Thanks for providing stackoverflow link.

[thinkinmachine]

@Pl45m4

Thanks for your explanation! I'll keep you posted if I find out something new on this topic.