Re: Qt3 breakage since 1.5.19
Marc Toussaint wrote: I found your discussion on the broken QT port to cygwin. http://cygwin.com/acronyms/#PPIOSPE I'm redirecting this to the cygwin-xfree list. Please make sure that your mail client respects the Reply-To: address. I have the same problem with uic: it is producing broken code (it looks as if half of the file is simply missing) See my post to the cygwin list: http://cygwin.com/ml/cygwin/2006-02/msg00472.html Is there any solution yet? I'm desparate. Is there any way to undo the qt3 update with the cygwin installer? AFAIK the problem is with cygwin-1.5.19. You could downgrade cygwin to 1.5.18, but you will also need to downgrade a number of other packages, the current versions of which depend on new features in 1.5.19. Besides that, you could help track down the problem in cygwin itself, so that we have a chance in getting this fixed for 1.5.20. Yaakov -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://x.cygwin.com/docs/ FAQ: http://x.cygwin.com/docs/faq/
Qt3 breakage since 1.5.19
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 René Berber wrote: Running Scribus with the debug enabled cygqt-mt-3.dll the program works fine with no mutex failure message and using the latest Cygwin snapshot. As mentioned in my reply to Brian, under gdb there are many SIGSEGV signals received but continuing results in the same behaviour as above: no problem. So there seems to be no regression, there must be something different btw. the original build environment and my PC. I used the code (build script, patches and original code) downloaded using setup.exe. OK, we need to get back to this. Now that cygwin-1.5.19-4 is out, I'm seeing the same breakage with qt3-3.3.4: $ scribus Mutex init failure: Device or resource busy [splashscreen shows, gets as far as loading plugins, then:] Mutex init failure: Invalid argument [exits silently] $ convertall [this is a PyQt app. It exits silently also.] But not everything is affected apparently: $ xxdiff file1 file2 Mutex init failure: Device or resource busy [works, just like before] Mutex lock failure: Invalid argument Mutex unlock failure: Invalid argument Mutex destroy failure: Invalid argument 1) Something changed between 1.5.18 and the 20051207 snapshot to break qt3 threading. 2) It should not be necessary to have a debug library in order for ordinary programs to work. 3) A threaded Qt is required for many packages, including KDE, so disabling threading is not a viable option. Ideas? Yaakov -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.1 (Cygwin) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFD4nhCpiWmPGlmQSMRAh9lAKCtvF9//J6zC4KWQsdJN9wiHPd95QCgxerD UYICx6yxX23jjUdxJAChwyg= =SImM -END PGP SIGNATURE- -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://x.cygwin.com/docs/ FAQ: http://x.cygwin.com/docs/faq/
Re: Qt3 breakage since 1.5.19
Yaakov S (Cygwin Ports) wrote: Running Scribus with the debug enabled cygqt-mt-3.dll the program works fine with no mutex failure message and using the latest Cygwin snapshot. As mentioned in my reply to Brian, under gdb there are many SIGSEGV signals received but continuing results in the same behaviour as above: no problem. So there seems to be no regression, there must be something different btw. the original build environment and my PC. I used the code (build script, patches and original code) downloaded using setup.exe. OK, we need to get back to this. Now that cygwin-1.5.19-4 is out, I'm seeing the same breakage with qt3-3.3.4: $ scribus Mutex init failure: Device or resource busy [splashscreen shows, gets as far as loading plugins, then:] Mutex init failure: Invalid argument [exits silently] Yep, exactly my original problem. $ convertall [this is a PyQt app. It exits silently also.] But not everything is affected apparently: $ xxdiff file1 file2 Mutex init failure: Device or resource busy [works, just like before] Mutex lock failure: Invalid argument Mutex unlock failure: Invalid argument Mutex destroy failure: Invalid argument 1) Something changed between 1.5.18 and the 20051207 snapshot to break qt3 threading. Or the change exposed a bug in Qt. 2) It should not be necessary to have a debug library in order for ordinary programs to work. The library is as close as it gets to a regular library, I separated the debug info, it's just not optimized. It would be interesting to test a rebuilt regular library, have you done that? 3) A threaded Qt is required for many packages, including KDE, so disabling threading is not a viable option. Ideas? All those mutex lock failure point to the threads implementation. If you want to go from what changed in Cygwin's dll forward, it has to be something in threads. Going the other way, what I tried is using gdb trying to look at what breaks the program. The worst part is that under gdb they run fine many times, then break (with the debug library). Yet another approach could be to do the porting of the newer 3.x/4.x version. -- René Berber -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://x.cygwin.com/docs/ FAQ: http://x.cygwin.com/docs/faq/
Re: Qt3 breakage since 1.5.19
René Berber wrote: The library is as close as it gets to a regular library, I separated the debug info, it's just not optimized. It would be interesting to test a rebuilt regular library, have you done that? Answering my own question: no, it doesn't work. Just tested it, rebuilt Qt with original script, build failed as always (uic produced an invalid file, truncated but this time uic did not print the mutex failed... message so it gets even weirder). Scribus dies as before. At the same time I also tested another idea, all the SIGSEGV signals that you can see under gdb are caused by uninitialized variables passed as parameters, so I added -Wall to CFLAGS to see if something jumps up. I'l see what I can find this way. -- René Berber -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://x.cygwin.com/docs/ FAQ: http://x.cygwin.com/docs/faq/
Re: Qt3 breakage since 1.5.19
-BEGIN PGP SIGNED MESSAGE-
Hash: SHA1
René Berber wrote:
Or the change exposed a bug in Qt.
Either way, it was working in 1.5.18 (didn't I already say I'd be
unhappy if my qt3 was broken with 1.5.19?).
The library is as close as it gets to a regular library, I separated the debug
info, it's just not optimized.
It would be interesting to test a rebuilt regular library, have you done that?
Actually, now that you mention it...
I was in the middle of building kdelibs-3.5.1 when gcc choked on
uic-generated code (this happened with both 3.3.4 and 3.3.5). So I went
back to rebuild 3.3.4, and now it won't finish either.
g++ -c -pipe -I/usr/include/postgresql/server
- -I/usr/include/postgresql/pgsql -I/usr/include/postgresql/pgsql/server
- -fno-exceptions -DPIC -Wall -W -O2 -D_REENTRANT -DDESIGNER
- -DQT_INTERNAL_XML -DQT_INTERNAL_WORKSPACE -DQT_INTERNAL_ICONVIE
W -DQT_INTERNAL_TABLE -DQT_TABLET_SUPPORT -DQT_NO_DEBUG
- -DQT_THREAD_SUPPORT
- -I/home/Yaakov/src/kde/qt/3/qt-x11-free-3.3.4/.build/mkspecs/cygwin-g++
- -I. -I/usr/include/freetype2 -I../shared -I../uilib -I../../../include
- -I/usr/X11R6/include -I.moc/release-shared-mt/ -o ./listboxeditor.o
listboxeditor.cpp
listboxeditor.cpp:32:7: warning: no newline at end of file
listboxeditor.cpp: In constructor
'ListBoxEditorBase::ListBoxEditorBase(QWidget*, const char*, bool, uint)':
listboxeditor.cpp:32: error: expected identifier at end of input
listboxeditor.cpp:32: error: expected `(' at end of input
listboxeditor.cpp:32: error: expected `{' at end of input
listboxeditor.cpp: At global scope:
listboxeditor.cpp:32: warning: unused parameter 'parent'
listboxeditor.cpp:32: warning: unused parameter 'name'
listboxeditor.cpp:32: warning: unused parameter 'modal'
listboxeditor.cpp:32: warning: unused parameter 'fl'
listboxeditor.cpp:32: warning: unused parameter 'parent'
listboxeditor.cpp:32: warning: unused parameter 'name'
listboxeditor.cpp:32: warning: unused parameter 'modal'
listboxeditor.cpp:32: warning: unused parameter 'fl'
make: *** [listboxeditor.o] Error 1
Looking at the generated listboxeditor.cpp (and at the KDE package), it
seems that uic doesn't compile .ui files correctly, resulting in the
above gcc errors.
Running the uic command to generate this code in gdb, I get a bunch of
SIGSEGV in pthread_mutex* functions in cygwin1.dll, then:
Program received signal SIGSEGV, Segmentation fault.
0x713f33a4 in QRealMutexPrivate::QRealMutexPrivate ()
from
/home/Yaakov/src/kde/qt/3/qt-x11-free-3.3.4/.build/bin/cygqt-mt-3.dll
The offending code is in src/tools/qmutex_unix.cpp (which has absolutely
no changes from 3.3.4 to 3.3.5).
All those mutex lock failure point to the threads implementation. If you
want
to go from what changed in Cygwin's dll forward, it has to be something in
threads.
Going the other way, what I tried is using gdb trying to look at what breaks
the
program. The worst part is that under gdb they run fine many times, then
break
(with the debug library).
Yet another approach could be to do the porting of the newer 3.x/4.x version.
Same story with 3.3.5, but it's still hard-masked unstable on Gentoo, so
I'm holding it back (at least until I see how KDE 3.5 works with it).
Qt4 is totally different and incompatible with qt3 (but will be parallel
installable, once I get working on it). Let's get this working first,
though.
Yaakov
-BEGIN PGP SIGNATURE-
Version: GnuPG v1.4.1 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iD8DBQFD4sdZpiWmPGlmQSMRApURAKC4SeiVPZJFe9Fbg3U9MjBvl5jDTgCfX9NI
sIjdGmnQKbGmTS5nwhAklVo=
=NNJm
-END PGP SIGNATURE-
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://x.cygwin.com/docs/
FAQ: http://x.cygwin.com/docs/faq/
Re: Qt3 breakage since 1.5.19
Yaakov S (Cygwin Ports) wrote: [snip] Actually, now that you mention it... I was in the middle of building kdelibs-3.5.1 when gcc choked on uic-generated code (this happened with both 3.3.4 and 3.3.5). So I went back to rebuild 3.3.4, and now it won't finish either. g++ -c -pipe -I/usr/include/postgresql/server -I/usr/include/postgresql/pgsql -I/usr/include/postgresql/pgsql/server -fno-exceptions -DPIC -Wall -W -O2 -D_REENTRANT -DDESIGNER -DQT_INTERNAL_XML -DQT_INTERNAL_WORKSPACE -DQT_INTERNAL_ICONVIE W -DQT_INTERNAL_TABLE -DQT_TABLET_SUPPORT -DQT_NO_DEBUG -DQT_THREAD_SUPPORT -I/home/Yaakov/src/kde/qt/3/qt-x11-free-3.3.4/.build/mkspecs/cygwin-g++ -I. -I/usr/include/freetype2 -I../shared -I../uilib -I../../../include -I/usr/X11R6/include -I.moc/release-shared-mt/ -o ./listboxeditor.o listboxeditor.cpp listboxeditor.cpp:32:7: warning: no newline at end of file [snip] Yes, same problem I've been seeing. Looking at the generated listboxeditor.cpp (and at the KDE package), it seems that uic doesn't compile .ui files correctly, resulting in the above gcc errors. Actually is more complicated than that, uic sometimes runs perfectly, then fails like above (except that with a snapshot, previous to latest release, it deadlocked on a mutex). In fact if you look at the output above the failure, uic ran many times generating .h files and I've seen that it does generate a few .cpp files before it fails. Even more strange, I've seen it run on the same file that it failed before when I ran it by hand, but not always... I'm not sure if that means that it corrupts something in the shared memory set by cygwin1.dll, I really don't know how that works. Running the uic command to generate this code in gdb, I get a bunch of SIGSEGV in pthread_mutex* functions in cygwin1.dll, then: Program received signal SIGSEGV, Segmentation fault. 0x713f33a4 in QRealMutexPrivate::QRealMutexPrivate () from /home/Yaakov/src/kde/qt/3/qt-x11-free-3.3.4/.build/bin/cygqt-mt-3.dll The offending code is in src/tools/qmutex_unix.cpp (which has absolutely no changes from 3.3.4 to 3.3.5). Interesting. [snip] Same story with 3.3.5, but it's still hard-masked unstable on Gentoo, so I'm holding it back (at least until I see how KDE 3.5 works with it). Qt4 is totally different and incompatible with qt3 (but will be parallel installable, once I get working on it). Let's get this working first, though. Looks like the path of least resistance is looking at cygwin1.dll changes. I prefer to use gdb, that's why I installed the cygwin code and made a debug qt library. Any idea why the debug library works? I already tought that the optimization may be doing something wrong, but your notes above point in another direction. Regards. -- René Berber -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Problem reports: http://cygwin.com/problems.html Documentation: http://x.cygwin.com/docs/ FAQ: http://x.cygwin.com/docs/faq/
