Missing generated wrappers for RHEL 9 CI tasks

[mrbean-bremen]

The wrapper generation in CI tasks for oldschool builds (RHEL9) fails:

 Trying to determine Qt version...
Determined Qt version is 5.15.9
Parsing typesystem file [:/trolltech/generator/build_all.txt]
/usr/include/qt5ln/include/QtCore/qconfig.h:19: #error: #error "unexpected value for __WORDSIZE macro"
PreProcessing - Generate [.preprocessed.tmp] using [:/trolltech/generator/qtscript_masterinclude.h] and include-paths []
/usr/include/qt5ln/include/QtCore/qsystemdetection.h:184: #error: #error "Qt has not been ported to this OS - see http://www.qt-project.org/"
/usr/include/qt5ln/include/QtCore/qcompilerdetection.h:504: #error: #error "Qt has not been tested with this compiler - see http://www.qt-project.org/"
/usr/include/qt5ln/include/QtCore/qcompilerdetection.h:590: #error: #error Qt requires a C++11 compiler and yours does not seem to be that.
/usr/include/qt5ln/include/QtCore/qbasicatomic.h:61: #error: #error "Qt requires C++11 support"
/usr/include/qt5ln/include/QtCore/qglobalstatic.h:60: syntax error: failed to evaluate #if condition, division/modulo by zero
Building model using [.preprocessed.tmp]
Classes in typesystem: 0
Generated:
  - header....: 0 (0)
  - impl......: 0 (0)
  - modules...: 0 (0)
  - pri.......: 0 (0)

This has not been noticed before because it currently does not generate a CI failure.

[hjmjohnson]

This may be addressed with the work done in #313.

[jcfr]

After inspecting the output of the associated job¹, it looks like the problem still persists:

# workaround to allow to find the Qt include dirs for installed standard qt packages
mkdir /usr/include/qt5ln; ln -s /usr/include/qt5 /usr/include/qt5ln/include
export QTDIR=/usr/include/qt5ln
cd generator
./pythonqt_generator

[...]
Generated:
  - header....: 0 (0)
  - impl......: 0 (0)
  - modules...: 0 (0)
  - pri.......: 0 (0)

Footnotes

1. https://github.com/MeVisLab/pythonqt/actions/runs/18106650418/job/51522795026?pr=313#step:6:22 ↩

[mrbean-bremen]

Yes. I had a look at this some time ago, and the problem is that for some reason some features (like QT_FEATURE_thread) are not defined in qtscript_masterinclude.h, and the QT_CONFIG macro than barks on the feature by causing a division by zero (it does this intentionally to ensure that the features are defined). Something goes wrong during configuration.

I experimentally added a workaround, basically defining QT_BOOTSTRAPPED and redefining QT_CONFIG, which worked to a point, but it generated less wrappers than expected because some features were not defined (as to be expected).

I didn't dig further into this so far, though.

[jcfr]

I didn't dig further into this so far, though.

Et voila. This should now be fixed 🚀

[mrbean-bremen]

Et voila. This should now be fixed 🚀

Excellent, merci!

So /usr/include was missing, I hadn't seen this...

[jcfr]

So /usr/include was missing, I hadn't seen this...

Technically, we could have addressed this by defining __WORDSIZE to 64 in generator/qtscript_masterinclude.h¹ so that qconfig-multilib.h² ends up including QtCore/qconfig-64.h . That said, I considered this as a hack specific to the packaging of Qt on Rocky Linux.

Instead, I opted to implement two changes ...

1. Addition of __x86_64__ in commit fix(generator): Define __x86_64__ to support Qt SDK on multi-lib distribution (only if Q_PROCESSOR_X86_64 is itself defined)
2. Specifying --include-paths=/usr/include in the follow-up commit to ensure that header bits/wordsize.h included in qconfig-multilib.h² is found on Rocky Linux and can define __WORDSIZE by checking if __x86_64__ is itself defined.

Footnotes

1. https://github.com/MeVisLab/pythonqt/blob/master/generator/qtscript_masterinclude.h ↩

2. https://git.rockylinux.org/staging/rpms/qt5-qtbase/-/blob/r8/SOURCES/qconfig-multilib.h ↩ ↩²

[jcfr]

In the future, we could look into (optionally) forwarding -D defines associated with the host system (or target in the context of cross compilation) so that we address most (if not all #error) current reported by the preprocessor.