Re: help understanding a Windows crash report
On 30 October 2014 00:07, Dirk Hohndel d...@hohndel.org wrote: Please test http://subsurface-divelog.org/downloads/daily/subsurface-4.2-349-g2b8043b82b99.exe if you have a chance to make sure that it's not just my system where this works again... This one also implements the long missing don't install 64bit binaries on a 32bit system feature... quickly tested the installer and there are no crashes. lubomir -- ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Oct 30, 2014, at 3:15 AM, Lubomir I. Ivanov neolit...@gmail.com wrote: On 30 October 2014 00:07, Dirk Hohndel d...@hohndel.org wrote: Please test http://subsurface-divelog.org/downloads/daily/subsurface-4.2-349-g2b8043b82b99.exe if you have a chance to make sure that it's not just my system where this works again... This one also implements the long missing don't install 64bit binaries on a 32bit system feature... quickly tested the installer and there are no crashes. Good. Thanks. /D ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Wed, Oct 29, 2014 at 09:19:22PM -0700, Dirk Hohndel wrote: On Oct 29, 2014, at 6:50 PM, Thiago Macieira thi...@macieira.org wrote: I'm fairly certain it's not a Qt bug now. This is a binary incompatibility issue between QtGui and QtWidgets due either a bug in Dirk's packaging or in Fedora's packaging. Dirk: verify that the DLLs you gave me are exactly the ones that Fedora shipped. If they are, the bug is in Fedora's build, somehow. It looks like a stale Qt5Gui.dll got packaged instead of the right one. I am 99% certain that the package I gave you had the correct DLLs from the Fedora packages. I’ll need to hunt down the md5 values for those DLLs to be 100% sure. I am now certain that I am packaging the Qt DLLs as delivered by Fedora 20. I uninstalled the corresponding packages, reinstalled them from the server and compared md5 fingerprints of all the Qt DLLs with the ones from the DLLs that were in the package that crashed... /D ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Wed, Oct 29, 2014 at 6:38 PM, Dirk Hohndel d...@hohndel.org wrote: Running my latest daily build (not on the server, yet, for obvious reasons) I get the following crash. And I have no clue how to read those... Problem signature: Problem Event Name: APPCRASH Application Name:subsurface.exe Application Version: 4.2.0.349 Application Timestamp: 545147c1 Fault Module Name: Qt5Core.dll Fault Module Version:5.3.2.0 Fault Module Timestamp: 541cda2f Exception Code: c005 Exception Offset:00094d90 OS Version: 6.1.7601.2.1.0.256.4 Locale ID: 1033 Additional Information 1:70c7 Additional Information 2: 70c70372fad2f93bb2b69b6202626cb6 Additional Information 3:dfe1 Additional Information 4: dfe1b677b69a5298073538898968a3dd It looks like it's crashing in Qt5Core.dll - but how do I figure out what the rest means? There's an offset but no function name??? Dirk, Can you build the app in debug mode? Qt libraries have them build in debug mode to help that kind of stuff. Tomaz /D ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Wed, Oct 29, 2014 at 06:46:12PM -0200, Tomaz Canabrava wrote: On Wed, Oct 29, 2014 at 6:38 PM, Dirk Hohndel d...@hohndel.org wrote: Running my latest daily build (not on the server, yet, for obvious reasons) I get the following crash. And I have no clue how to read those... Problem signature: Problem Event Name: APPCRASH Application Name:subsurface.exe Application Version: 4.2.0.349 Application Timestamp: 545147c1 Fault Module Name: Qt5Core.dll Fault Module Version:5.3.2.0 Fault Module Timestamp: 541cda2f Exception Code: c005 Exception Offset:00094d90 OS Version: 6.1.7601.2.1.0.256.4 Locale ID: 1033 Additional Information 1:70c7 Additional Information 2: 70c70372fad2f93bb2b69b6202626cb6 Additional Information 3:dfe1 Additional Information 4: dfe1b677b69a5298073538898968a3dd It looks like it's crashing in Qt5Core.dll - but how do I figure out what the rest means? There's an offset but no function name??? Dirk, Can you build the app in debug mode? Qt libraries have them build in debug mode to help that kind of stuff. I remember this rat hole. My Qt libraries come from Fedora (cross built for Win64). A debug build of Marble alone is several hundred megabytes and takes roughly forever... and I have yet to be able to build a single installer that actually works as debug build. Literally. That's why I gave up trying to get Qt5 work in 32bit. Sooo frustrating. /D ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Wed, Oct 29, 2014 at 01:52:07PM -0700, Dirk Hohndel wrote: I remember this rat hole. My Qt libraries come from Fedora (cross built for Win64). A debug build of Marble alone is several hundred megabytes and takes roughly forever... and I have yet to be able to build a single installer that actually works as debug build. Literally. That's why I gave up trying to get Qt5 work in 32bit. Brilliant. Apparently a Fedora update broke things. If I replace the current Qt DLLs with the ones from a built a few months ago everything works. So now I need to undo those updates to mingw64-qt5-qtbase and friends. /D ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Wed, Oct 29, 2014 at 7:04 PM, Dirk Hohndel d...@hohndel.org wrote: On Wed, Oct 29, 2014 at 01:52:07PM -0700, Dirk Hohndel wrote: I remember this rat hole. My Qt libraries come from Fedora (cross built for Win64). A debug build of Marble alone is several hundred megabytes and takes roughly forever... and I have yet to be able to build a single installer that actually works as debug build. Literally. That's why I gave up trying to get Qt5 work in 32bit. Brilliant. Apparently a Fedora update broke things. If I replace the current Qt DLLs with the ones from a built a few months ago everything works. So now I need to undo those updates to mingw64-qt5-qtbase and friends. And poke fedora guys. /D ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface
Re: help understanding a Windows crash report
On Wednesday 29 October 2014 15:07:35 Dirk Hohndel wrote: I have no debug symbols. Thiago is looking into this and he came to the conclusion that this might be a Qt 5.3.2 bug. The previous Fedora version was 5.3.1 - but I can't get those RPMs anymore. I'm fairly certain it's not a Qt bug now. This is a binary incompatibility issue between QtGui and QtWidgets due either a bug in Dirk's packaging or in Fedora's packaging. Dirk: verify that the DLLs you gave me are exactly the ones that Fedora shipped. If they are, the bug is in Fedora's build, somehow. It looks like a stale Qt5Gui.dll got packaged instead of the right one. Explanation: The backtrace that I managed to obtain with Dirk made little sense, but after realising that the debugger wasn't picking up the names of unexported functions, I reconstructed the following backtrace: #3 QPlatformFontDatabase::resolveFontFamilyAlias(const QString ) const #4 QFontComboBoxPrivate::_q_updateModel() [not shown] QFontComboBox::setWritingSystem(QFontDatabase::WritingSystem) #5 QFontComboBox::QFontComboBox(QWidget *) If you look at _q_updateModel, you'll see that it does not call resolveFontFamilyAlias. But by comparing the assembly with the source code, the crash point is on line: if (pfdb-isPrivateFontFamily(list.at(i))) pfdb is a QPlatformFontDatabase. Here's how I know it's a BIC issue: both resolveFontFamilyAlias and isPrivateFontFamily are virtual functions and isPrivateFontFamily was added between Qt 5.3.1 and 5.3.2, inserted just before resolveFontFamilyAlias. That means the caller in QtWidget placed a call to the correct virtual slot, but the QPlatformFontDatabase's virtual table in QtGui contained the old function at that position. This problem cannot happen in a clean build. The only way for this to happen is if Fedora rebuilt without cleaning up the .o files or if they accidentally packaged a pre-release Qt5Gui.dll. -- Thiago Macieira - thiago (AT) macieira.info - thiago (AT) kde.org Software Architect - Intel Open Source Technology Center PGP/GPG: 0x6EF45358; fingerprint: E067 918B B660 DBD1 105C 966C 33F5 F005 6EF4 5358 ___ subsurface mailing list subsurface@subsurface-divelog.org http://lists.subsurface-divelog.org/cgi-bin/mailman/listinfo/subsurface