Differences between revisions 50 and 51
Revision 50 as of 2014-01-08 21:22:11
Size: 5963
Comment:
Revision 51 as of 2014-04-08 19:45:40
Size: 5999
Editor: ?Solveig
Comment: # apt-get install gdb at the begining.
Deletions are marked like this. Additions are marked like this.
Line 12: Line 12:

{{{
 # apt-get install gdb
}}}

Translation(s): English - Русский


How To Get a Meaningful Backtrace

This page will attempt to explain how to get a meaningful debugging backtrace from a reproducible program crash. For this example, we will install the package with debugging symbols for "hello" or if this is not available rebuild & install the "hello" package so that we keep debugging information.

For reference, a "#" at the beginning of a line means that what follows is a command that has to be executed as root (or sudo). A "$" at the beginning means that what follows is a command that should be executed as your normal user. A "(gdb)" at the beginning means that you should be typing the rest of the command line at the gdb (GNU Debugger) prompt.

 # apt-get install gdb

Installing the debugging symbols

To see if the package you’re trying to debug includes a -dbg package (e.g. for the hello source package, a hello-dbg package exists), search for it in your package manager. On the command line you can use the following command.

 $ apt-cache search hello | grep dbg
 p   hello-dbg            - hello debug symbols

If nothing is returned you have to rebuild the package as explained in the next section (or, especially if you are the maintainer of the package, modify the package to always build a DebugPackage). If an appropriate package was found you should just be able to install the appropriate -dbg package for example with

 # apt-get install hello-dbg

and skip the rebuilding process in the following instructions and go straight to running gdb.

Rebuilding the package you’re debugging

You may skip this section if you were able to install the necessary -dbg package(s) from the previous section.

  • Install the basic development packages and the build-dependencies for the package we want to rebuild. Note that you can skip the rebuilding part and go straight to running gdb, but it is unlikely that you will get a useful backtrace. (If the build-dep line fails, check you have deb-src lines in /etc/apt/sources.list.)

     # apt-get install build-essential fakeroot gdb
     # apt-get build-dep hello
  • Download & rebuild our package from source, keeping debugging symbols

     $ DEB_BUILD_OPTIONS="nostrip noopt" fakeroot apt-get -b source hello
  • Install our newly built package(s). There may be multiple .deb packages generated, so make sure to install only the ones you want. In this example, the .deb generated was called hello_2.1.1-4_i386.deb.

     # dpkg -i hello_2.1.1-4_i386.deb
  • You can ensure the binaries installed from your .deb have debugging symbols with the 'file' command, or with gdb itself (see below).

     $ file /usr/bin/hello # output should contain "not stripped"

Running gdb

  • Now run your program as follows, replacing "[--args]" with any arguments you want to run the program with:

     $ gdb hello
     ... gdb loads ...
     (gdb) set pagination 0
     (gdb) run [--args]
     ... hello loads...
  • Then try to reproduce your crash. If you’re lucky, a crash will occur and you’ll be dropped back to the gdb prompt.
  • If you are not so lucky to get a crash but instead get a freeze, you can still get gdb prompt by pressing CTRL-C in the terminal running gdb.
  • At that point, you can run:

     (gdb) bt
  • You’ll then get a lot of output, which you can then copy & paste to a bug followup e-mail or other bug reporting tool.

  • When you’re done with gdb, you can just run:

     (gdb) quit

If the problem seems to be in a major library such as libc6, xlibs, or libgtk2.0-0, you’ll want to install the appropriate -dbg package (e.g. libc6-dbg in the case of libc6) and then run the problematic program again under gdb.

Often, you will see a backtrace where one or more of the top lines is in malloc() or g_malloc(). When this happens, chances are your backtrace isn’t very useful. The easiest way to find some useful information is to set the environment variable MALLOC_CHECK_ to a value of 2. You can do this while running gdb by doing this:

 $ MALLOC_CHECK_=2 gdb hello

Advanced gdb commands

  • If the program you’re backtracing is multi-threaded, you might want to get a backtrace for all threads:

     (gdb) thread apply all bt
  • Another thing which is quite helpful to report is what variables were set locally at each point in the stack:

     (gdb) bt full
  • You might want to report the output of the combination of the preceding options:

     (gdb) thread apply all bt full
  • And if this is too much irrelevant output, you might want to keep only a few calls, such as the top 10:

     (gdb) thread apply all bt full 10
  • If you have a large backtrace, you can log gdb output to a file (the default is gdb.txt):

     (gdb) set logging on
  • To check you have debugging symbols in your binary:

     $ gdb
      (gdb) symbol-file /usr/bin/hello
    
     # you should see something like this:
      Reading symbols from /usr/bin/hello ... done
      Using host libthread_db library "/lib/tls/i686/cmov/libthread_db.so.1".
      (gdb)
    
     # NB you should _not_ see
      Reading symbols from /usr/bin/hello...(no debugging symbols found)...done

Debugging X Errors

If a GTK program has received an X error; i.e. you see a message of the form:

The program 'preview1' received an X Window System error.

then you can try running the program with --sync, and break on the gdk_x_error function in order to obtain a backtrace, thus:

 (gdb) break gdk_x_error
 (gdb) run --sync

--AriPollak and ?LoicMinier

CategoryDebugging