Development/Tutorials/Debugging/How to create useful crash reports (fr)

Development/Tutorials/Debugging/How to create useful crash reports

Languages: عربي | Asturianu | Català | Česky | Kaszëbsczi | Dansk | Deutsch | English | Esperanto | Español | Eesti | فارسی | Suomi | Français | Galego | Italiano | 日本語 | 한국어 | Norwegian | Polski | Português Brasileiro | Română | Русский | Svenska | Slovenčina | Slovenščina | српски | Türkçe | Tiếng Việt | Українська | 简体中文 | 繁體中文

Introduction

Ce document décrit comment obtenir une "backtrace" utile du plantage des applications KDE. Nous allons commencer par donner quelques informations générales. Ensuite, nous décrirons pour quelques distributions comment préparer vos paquetages KDE pour obtenir la backtrace. Ceci devrait suffire pour la plupart des utilisateurs. Des sections supplémentaires expliquent comment créer des backtraces avec le débogueur GNU et Valgrind, qui sont parfois utiles.

Comment créer des rapports d'erreur utiles

Un bon rapport d'erreur de Bugzilla se compose de deux parties: une description de la façon de reproduire le crash et une backtrace du crash. Si l'un de ces éléments est manquant, il est beaucoup plus difficile (voire impossible) pour les développeurs de s'attaquer au problème.

Une description devrait donner plus d'informations que seulement "ça a buggé". Essayez de décrire tout ce que vous avez fait avant le crash. Avez-vous cliqué sur un bouton, ouvert un site Web particulier ou est-ce l'ouverture d'un fichier qui a causé des problèmes ? Ce petit détail qui peut vous sembler inutile peut être utile pour le développeur, alors il vaut mieux le signaler.

Un article plus précis sur la façon de bien rédiger les descriptions de bogues est disponible ici (en anglais).

Ne mettez pas la backtrace en pièce jointe au rapport de bogue. Copiez/collez-le simplement dedans. De cette manière, il est beaucoup plus facile pour les développeurs de chercher les rapports en double puisque les pièces jointes ne sont pas examinées.

Si vous collez une backtrace dans un rapport de bogue, merci de supprimer toutes les lignes

(no debugging symbols found)

de la backtrace sauf une ou deux car elles rendent le rapport plus difficile à lire.

Même s'il vaut mieux copier/coller les backtraces directement dans le rapport de bogue plutôt que les mettre en pièces jointes, merci de ne pas copier/coller d'autres informations telles que les logs (valgrind, strace ou sortie du terminal) ou des exemples (e-mails, fichiers HTML etc.). Veuillez utiliser les pièces jointes pour ces informations.

Backtraces

Les backtraces sont très importantes. Elles peuvent vous paraître incompréhensibles mais elles contiennent des tonnes d'informations utiles. Une backtrace permet de savoir quelles sont les fonctions qui ont été appelées avant le plantage pour que les développeurs puissent savoir quelle fontion est la cause du problème. Cependant, obtenir de bonnes backtraces présente un inconvénient : les bibliothèques et exécutables requis occupent beaucoup plus d'espace disque que ceux qui sont optimisés. C'est pourquoi beaucoup de distributions choisissent d'installer les versions optimisées, ce qui donne des backtraces inutiles :

(no debugging symbols found)
Using host libthread_db library "/lib/tls/i686/cmov/libthread_db.so.1".
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
[Thread debugging using libthread_db enabled]
[New Thread -1233848624 (LWP 12212)]
[New Thread -1255081072 (LWP 12820)]
[New Thread -1240921200 (LWP 12819)]
[New Thread -1266680944 (LWP 12818)]
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
(no debugging symbols found)
0xffffe410 in __kernel_vsyscall ()
#0  0xffffe410 in __kernel_vsyscall ()
#1  0xb6a1210b in ?? () from /lib/tls/i686/cmov/libpthread.so.0
#2  0xb6a85afe in ?? () from /usr/lib/libX11.so.6
#3  0x00000003 in ?? ()
#4  0x082149c0 in ?? ()
#5  0x00003ffc in ?? ()
#6  0x00000000 in ?? ()

Mais pas d'inquiétude, avec quelques modifications, vous pourrez créer des backtraces complètes pour les applications KDE.

Preparing your KDE packages

If your distribution has debugging-enabled packages, install them.

It is easy to see which debug packages you are missing from looking at the backtrace. For example, take the following line from a backtrace:

#6  0xb7975bdc in ?? () from /usr/lib/libkmailprivate.so.4

The ?? indicates that the library libkmailprivate.so.4 does not have debug information, which might be available in separate debug packages. In this case, it is pretty easy to guess that you need to install debug packages for KMail to get a better backtrace.

Sometimes, you need to install more than one debug package to get a good backtrace. This depends on how the distribution splits up the packages. For example, for some distributions it is enough to install the debug package for kdepim to get enough debugging information for a crash in KMail, for other distributions there is an additional debug package just for KMail.

Here's a list of how to obtain debug packages for some distributions:

• Debian - Debian offers -dbg packages to easy create useful backtraces. Just install the corresponding -dbg package. e.g. kdepim-dbg for KMail crashes. The dependencies of -dbg makes sure to pull in the other right packages (kdelibs-dbg, gdb, and so on).
• FreeBSD ports - Please refer to the KDE on FreeBSD FAQ.
• Gentoo - Gentoo has its own document describing how to proceed.
• Mandriva - Mandriva 2007.0 and up has additional debugging packages for all of KDE (in fact, for all of its packages). Just install the corresponding -debug package, like kdebase-debug and kdemultimedia-debug. You probably want to install kdelibs-debug anyways.
  • Note: the -debug packages are in separate repositories. For instance, for all packages in main, you'll find the debugging package in repository debug_main.
• Kubuntu/Ubuntu - The Ubuntu family makes things quite easy. Every official KDE module has an additional package in the repository, suffixed with -dbg. Always install kdelibs5-dbg, because all KDE applications use kdelibs (kdelibs-dbg for KDE 3 applications). Then you should install a -dbg package for the application which crashed. For example if KOrganizer crashed you should install kdepim-dbg as well. If the program is not from an official KDE module and has no -dbg package, you can install the -dbgsym package from the repository listed on this Debugging Program Crashes page.

  During the Ubuntu development cycle the Apport crash handler is turned on which will report crashes to launchpad.net and do the backtrace for you, if you would rather use the KDE crash handler turn Apport off in /etc/defaults/apport

• openSUSE - You should only install the -debuginfo packages, for example: kdepimlibs4-debuginfo. You can find these packages in KDE repositories. There is also a dedicated openSUSE debugging page.
• Fedora - Fedora has its own document describing how to proceed. (A debuginfo repository has to be enabled.)

If your distribution doesn't have debugging-enabled packages for KDE, you'll have to compile KDE from sources:

• If you're using KDE 3, then at the configure stage, you should supply the parameter --enable-debug=full in order to build debug symbols in the resulting files.
• If you're using KDE 4, then at the cmake stage, you should supply the parameter -DCMAKE_BUILD_TYPE=debugfull. If you want to specify your own CXXFLAGS, then use -DCMAKE_BUILD_TYPE=None CMAKE_CXX_FLAGS="-O0 -g". You can change the CMAKE_CXX_FLAGS as appropriate for your needs.

Then it's just make and make install as you're used to.

Crash!

Now it's time to crash your application. The KDE Crash Dialog should appear right after the crash, which shows the Backtrace tab.

KDE Crash Dialog

Click that tab and wait for a minute. This process may take quite some memory, so things may go sluggish all of a sudden. But the result should look much better. For example:

Using host libthread_db library "/lib/libthread_db.so.1". 
[Thread debugging using libthread_db enabled] 
[New Thread -1232783168 (LWP 7604)] 
[KCrash handler] 
#6  0x0806be76 in TreeMapItem::parent (this=0x0) 
    at /home/bram/KDE/kde3/kdeaddons/konq-plugins/fsview/treemap.h:285 
#7  0x08065fea in TreeMapItemList::compareItems (this=0xbfec04a8, item1=0x0, 
    item2=0x0) 
    at /home/bram/KDE/kde3/kdeaddons/konq-plugins/fsview/treemap.cpp:720 
#8  0xb7281619 in QGList::operator== () from /usr/qt/3/lib/libqt-mt.so.3 
#9  0x0806d498 in QPtrList<TreeMapItem>::operator== (this=0xbfec04a8, 
    list=@0xbfec0468) at /usr/qt/3/include/qptrlist.h:74 
#10 0x08062e18 in TreeMapWidget::mousePressEvent (this=0xbfec03ac, 
    e=0xbfebff1c) 
    at /home/bram/KDE/kde3/kdeaddons/konq-plugins/fsview/treemap.cpp:1840 
#11 0xb7004a63 in QWidget::event () from /usr/qt/3/lib/libqt-mt.so.3 
#12 0xb6f6bca7 in QApplication::internalNotify () 
   from /usr/qt/3/lib/libqt-mt.so.3 
#13 0xb6f6ca88 in QApplication::notify () from /usr/qt/3/lib/libqt-mt.so.3 
#14 0xb7725a84 in KApplication::notify (this=0xbfec055c, receiver=0xbfec03ac, 
    event=0xbfebff1c) 
    at /home/bram/KDE/kde3/kdelibs/kdecore/kapplication.cpp:550 
#15 0xb6f0bfd2 in QETWidget::translateMouseEvent () 
   from /usr/qt/3/lib/libqt-mt.so.3 
#16 0xb6f0b8b0 in QApplication::x11ProcessEvent () 
   from /usr/qt/3/lib/libqt-mt.so.3 
#17 0xb6f1b761 in QEventLoop::processEvents () from /usr/qt/3/lib/libqt-mt.so.3 
#18 0xb6f82831 in QEventLoop::enterLoop () from /usr/qt/3/lib/libqt-mt.so.3 
#19 0xb6f826b6 in QEventLoop::exec () from /usr/qt/3/lib/libqt-mt.so.3 
#20 0xb6f6b72f in QApplication::exec () from /usr/qt/3/lib/libqt-mt.so.3 
#21 0x0805181e in main (argc=134673960, argv=0xffffffff) 
    at /home/bram/KDE/kde3/kdeaddons/konq-plugins/fsview/main.cpp:55

This looks better, right? It shows memory addresses, the source files and line numbers and the parameters passed to functions. Which make it more helpful to the developer where to look for the problem.

Note

You need GDB installed to get the backtrace of a crash. Please read the next section to know what GDB is, and how to install it.

Retrieving a backtrace with GDB

In some cases, it is not possible to create a backtrace with the KDE Crash Dialog. This may be caused by an application which entered an infinite loop, or the crash dialog did not appear at all for some reason. You can try to grab a backtrace with gdb, the GNU Debugger. GDB is widely available through distribution packages.

Invoking GDB differs from the situation. You can run an application from inside gdb, or attach gdb to an already running process. The latter may be useful when an application already has entered an infinite loop. But we will first start with running an application inside gdb. From the shell, run:

$ gdb someKDEapp

The GDB prompt will appear. Note that this does not start the application itself, you should run it by invoking the run command:

(gdb) run

This will run the application like you are used to, and you can work with it like normal (it only consumes far more memory and may feel sluggish). Now it's time to reproduce your crash. When you succeed, the application just closes and you should return to your GDB prompt. Now it's time to run the 'backtrace' command:

Note

Some KDE applications (such as JuK and KTorrent) have special code to ensure that there is only one running instance of the application at a time. For these applications you should type in "run --nofork" at the (gdb) prompt instead of "run" because otherwise gdb will try to debug the wrong process. If you are unsure as to whether to use --nofork just try it. If the application says it's an unknown option you can remove --nofork.

(gdb) thread apply all backtrace

This should give a good backtrace which can be posted at the KDE Bugzilla.

In case you want to attach to an existing process, run the following command in the shell:

$ gdb someKDEapp pid

where pid is the process ID of the process you want to attach to. Once attached, and the process is in an infinite loop, after using the 'backtrace' command again a useful backtrace will appear. You can use 'continue' command to let the application run again and press Ctrl+C in gdb to be able to again enter commands.

Retrieving a backtrace with Valgrind

When it comes to crashes, Valgrind is also a useful tool to create a backtrace. It's not a substitution for GDB, but rather a supplement.

When you run an application in valgrind, every piece of memory read or written by the application is being checked. Valgrind will report erroneous memory operations in the standard output or in a log file. Since most crashes are due to an invalid memory read, valgrind can be useful to track down where the problem occurs.

Note

Valgrind consists of several tools in order to check or profile an application. For this article, we only use memcheck, the default tool when valgrind is being invoked.

Like GDB, Valgrind makes running an application much slower, while consuming a lot more resources.

Start the application within valgrind:

$ valgrind --log-file=someKDEapp someKDEapp

Now reproduce the crash. As soon as this happens, the application and valgrind will terminate. What's left is a file named someKDEapp.pid where pid is replaced by the process ID of the valgrind process. The file may list more errors than the one causing the crash. Here's the bit causing the crash which corresponds to the GDB backtrace above:

==23292== Invalid read of size 4
==23292==    at 0x806BD9E: TreeMapItem::parent() const (treemap.h:285)
==23292==    by 0x8065FB9: TreeMapItemList::compareItems(void*, void*) (treemap.cpp:720)
==23292==    by 0x50AC618: QGList::operator==(QGList const&) const (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==    by 0x806D3BF: QPtrList<TreeMapItem>::operator==(QPtrList<TreeMapItem> const&) const (qptrlist.h:74)
==23292==    by 0x8062DE7: TreeMapWidget::mousePressEvent(QMouseEvent*) (treemap.cpp:1840)
==23292==    by 0x4E2FA62: QWidget::event(QEvent*) (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==    by 0x4D96CA6: QApplication::internalNotify(QObject*, QEvent*) (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==    by 0x4D97A87: QApplication::notify(QObject*, QEvent*) (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==    by 0x4809AC3: KApplication::notify(QObject*, QEvent*) (kapplication.cpp:550)
==23292==    by 0x4D36FD1: QETWidget::translateMouseEvent(_XEvent const*) (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==    by 0x4D368AF: QApplication::x11ProcessEvent(_XEvent*) (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==    by 0x4D46760: QEventLoop::processEvents(unsigned) (in /usr/qt/3/lib/libqt-mt.so.3.3.8)
==23292==  Address 0x2C is not stack'd, malloc'd or (recently) free'd

But to be sure, just attach the whole log file to the crash report.