gri3-wm/docs/debugging

Debugging i3: How To
====================
Michael Stapelberg <michael+i3@stapelberg.de>
April 2009

This document describes how to debug i3 suitably for sending us useful bug
reports, even if you have no clue of C programming.

First of all: Thank you for being interested in debugging i3. It really means
something to us to get your bug fixed. If you have any questions about the
debugging and/or need further help, do not hesitate to contact us!

== Enabling logging

i3 spits out much information onto stdout. To have a clearly defined place
where log files will be saved, you should redirect stdout and stderr in
xsession. While you’re at it, putting each run of i3 in a separate log file
with date/time in it is a good idea to not get confused about the different
log files later on.

--------------------------------------------------------------------
exec /usr/bin/i3 >/home/michael/i3/i3log-$(date +'%F-%k-%M-%S') 2>&1
--------------------------------------------------------------------

== Enabling core dumps

When i3 crashes, often you have the chance of getting a 'core dump' (an image
of the memory of the i3 process which can be loaded into a debugger). To get a
core dump, you have to make sure that the user limit for core dump files is set
high enough. Many systems ship with a default value which even forbids core
dumps completely. To disable the limit completely and thus enable core dumps,
use the following command (in your .xsession, before starting i3):

-------------------
ulimit -c unlimited
-------------------

Furthermore, to easily recognize core dumps and allow multiple of them, you
should set a custom core dump filename pattern, using a command like the
following:

---------------------------------------------
sudo sysctl -w kernel.core_pattern=core.%e.%p
---------------------------------------------

This will generate files which have the executable’s file name (%e) and the
process id (%p) in it. You can save this setting across reboots using
+/etc/sysctl.conf+.

== Compiling with debug symbols

To actually get useful core dumps, you should make sure that your version of i3
is compiled with debug symbols, that is, that they are not stripped during the
build process. You can check whether your executable contains symbols by
issuing the following command:

----------------
file $(which i3)
----------------

You should get an output like this:
------------------------------------------------------------------------------
/usr/bin/i3: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically
linked (uses shared libs), for GNU/Linux 2.6.18, not stripped
------------------------------------------------------------------------------

Notice the +not stripped+, which is the important part. If you have a version
which is stripped, please have a look if your distribution provides debug
symbols (package +i3-wm-dbg+ on Debian for example) or if you can turn off
stripping. If nothing helps, please build i3 from source.

== Generating a backtrace

Once you have made sure that your i3 is compiled with debug symbols and that
core dumps are enabled, you can start making sense out of the core dumps.

Because the core dump depends on the original executable (and its debug
symbols), please do this as soon as you encounter the problem. If you
re-compile i3, your core dump might be useless afterwards.

Please install +gdb+, a debugger for C. No worries, you don’t need to learn it
now.  Start gdb using the following command (replacing the actual name of the
core dump of course):

----------------------------
gdb $(which i3) core.i3.3849
----------------------------

Then, generate a backtrace using:

--------------
backtrace full
--------------

== Sending bug reports/debugging on IRC

When sending bug reports, please paste the relevant part of the log (if in
doubt, please send us rather too much information than too less) and the whole
backtrace (if there was a core dump).

When debugging with us in IRC, be prepared to use a so called nopaste service
such as http://nopaste.info or http://pastebin.com because pasting large
amounts of text in IRC sometimes leads to incomplete lines (servers have line
length limitations) or flood kicks.