Python 3.14a7: Segfault in sip_api_init_module / _Py_TYPE

[Florian Bruhin]

On 2025-04-13T11:42:06 , Phil Thompson wrote:
> 
> On 13/04/2025 10:45, Florian Bruhin wrote:
> > On 2025-04-13T09:21:41 , Phil Thompson wrote:
> >> 
> >> On 12/04/2025 21:13, Florian Bruhin wrote:
> >> > On 2025-04-11T20:57:32 , Phil Thompson wrote:
> >> >>
> >> >> On 11/04/2025 16:15, Florian Bruhin wrote:
> >> >> > Hey,
> >> >> >
> >> >> > PyQt6 6.9.0 seemed to work fine with Python 3.14 Alpha 6 so far.
> >> >> > However, after upgrading to Alpha 7, "from PyQt6.QtCore" immediately
> >> >> > segfaults:
> >> >> >
> >> >> >     [...]
> >> >> >
> >> >> > I bisected this to this change in CPython, "Cache the result of
> >> >> > tuple_hash", which adds a new ob_hash field to PyTupleObject:
> >> >> >
> >> >> > https://github.com/python/cpython/pull/131529
> >> >> > https://github.com/python/cpython/commit/8614f86b7163b1c39798b481902dbb511292a537
> >> >> >
> >> >> > I'm not sure what happens there under the hood. Is there anything
> >> >> > special PyQt/sip is doing there, or is this simply some sort of CPython
> >> >> > regression in that commit? Happy to report it there, but I don't know
> >> >> > much about the CPython C API at all.
> >> >> >
> >> >> > Florian
> >> >>
> >> >> Just by visual inspection of the code, I think the sequence of calls
> >> >> is...
> >> >>
> >> >> [...]
> >> >>
> >> >> ...so I don't think I'm doing anything bad.
> >> >>
> >> >> Phil
> >> >
> >> > Thanks, I opened a CPython issue here:
> >> > https://github.com/python/cpython/issues/132454
> >> >
> >> > Florian
> >> 
> >> A quick check - I assume you are building PyQt6.sip from source 
> >> against
> >> the Alpha 7 headers.
> > 
> > I just tried rebuilding PyQt6-sip just to be sure:
> > https://bpa.st/SPHQ
> > 
> > Still seeing the same issue with that though.
> > 
> > I also tried rebuilding PyQt itself from sources, but that fails with:
> > 
> >     g++ -c -pipe -O2 -fno-exceptions -Wall -Wextra
> > -mno-direct-extern-access -D_REENTRANT -fPIC -DSIP_PROTECTED_IS_PUBLIC
> > -Dprotected=public -DQT_KEYPAD_NAVIGATION -DQT_NO_EXCEPTIONS
> > -DQT_NO_DEBUG -DQT_PLUGIN -DQT_CORE_LIB -I. -I. -I..
> > -I/home/florian/tmp/pyqt-src/pyqt6-6.9.0/qpy/QtCore
> > -I/usr/local/include/python3.14d -I/usr/include/qt6
> > -I/usr/include/qt6/QtCore -I. -I/usr/lib/qt6/mkspecs/linux-g++ -o
> > sipQtCorecmodule.o sipQtCorecmodule.cpp
> >     In file included from sipAPIQtCore.h:29,
> >                      from sipQtCorecmodule.cpp:24:
> >     ../sip.h:14:10: fatal error: Python.h: No such file or directory
> > 
> > Can I somehow easily teach it to pass
> > -I/home/florian/proj/cpython/Include/ so I don't need to actually
> > install the Python dev build I'm testing against?
> 
> I'm not sure what you mean. The sip module does not use the limited ABI 
> and so must be built against the appropriate Python version.
> 
> Re-building PyQt6 won't make any difference.
> 
> The bug is what you would get if you were using the PyTuple_GET_ITEM 
> macro from a pre-alpha7 header file against an alpha7 installation.
> 
> Phil

I tried again with a clean CPython clone and a clean virtualenv,
and now suddenly it works. So there was probably some sort of caching
going on *somewhere*... I guess that solves it then?

Thanks!

Florian
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://www.riverbankcomputing.com/pipermail/pyqt/attachments/20250413/95a66c88/attachment.sig>