Details
P2: Important Description
When performing a signal connection in a thread, there is a chance something is corrupted, causing a segfault. The attached code runs this process many times, and runs successfully under Qt 6.7. Under 6.8 it produces output on Linux such as (this is a debug build of PySide6)
Qt version 6.8.2
Started
ASSERT: "entry < allocated" in file /usr/include/qt6/QtCore/qhash.h, line 286
Aborted (core dumped)
or
Qt version 6.8.2
Started
ASSERT: "!isUnused()" in file /usr/include/qt6/QtCore/qhash.h, line 794
Aborted (core dumped)
On Ubuntu it's similar, but with different messages (not a debug build)
Qt version 6.8.2
Started
terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted (core dumped)
refeyn@refeyn-linux-build:~$ python kdbg2.py
Qt version 6.8.2
Started
Segmentation fault (core dumped)
refeyn@refeyn-linux-build:~$ python kdbg2.py
Qt version 6.8.2
Started
Segmentation fault (core dumped)
refeyn@refeyn-linux-build:~$ python kdbg2.py
Qt version 6.8.2
Started
malloc(): unaligned tcache chunk detected
Aborted (core dumped)
Or on Windows with faulthandler enabled:
Qt version 6.8.0 Started Windows fatal exception: code 0xc0000374 Current thread 0x0000b410 (most recent call first): File "C:\Users\matthew.joyce\Repos\lw8\kdbg2.py", line 19 in onR Thread 0x00009e70 (most recent call first): File "C:\Users\matthew.joyce\Repos\lw8\kdbg2.py", line 35 in <module>
Bisecting pyside, this commit responsible for the change in behaviour is 33bd61d13d8d9e3794b6049891be62f3351313d9 (libpyside: Reimplement signal connections for Python callables not targeting a QMetaMethod)
Oddly this code does use proper slots, so the commit message would imply no change was expected. Dumping the object info shows:
OBJECT D::unnamed
SIGNALS OUT
<None>
SIGNALS IN
<-- C::unnamed onS()
OBJECT C::unnamed
SIGNALS OUT
signal: destroyed(QObject*)
<functor or function pointer>
signal: signal()
--> D::unnamed onS()
SIGNALS IN
<None>