README

+++++++++++++
Fault handler
+++++++++++++

Fault handler for SIGSEGV, SIGFPE, SIGABRT, SIGBUS and SIGILL signals: display
the Python traceback and restore the previous handler. Allocate an alternate
stack for this handler, if sigaltstack() is available, to be able to allocate
memory on the stack, even on stack overflow (not available on Windows).

Import the module and call faulthandler.enable() to enable the fault handler.

The fault handler is called on catastrophic cases and so it can only use
signal-safe functions (eg. it doesn't allocate memory on the heap). That's why
the traceback is limited: it only supports ASCII encoding (use the
backslashreplace error handler for non-ASCII characters) and limits each string
to 100 characters, doesn't print the source code in the traceback (only the
filename, the function name and the line number), is limited to 100 frames and
100 threads.

By default, the Python traceback is written to the standard error stream. Start
your graphical applications in a terminal and run your server in foreground to
see the traceback, or pass a file to faulthandler.enable().

faulthandler is implemented in C using signal handlers to be able to dump a
traceback on a crash or when Python is blocked (eg. deadlock).

Website:
https://github.com/haypo/faulthandler/wiki/

faulthandler is part of Python since Python 3.3:
http://docs.python.org/dev/library/faulthandler.html


Example
=======

Example of a segmentation fault on Linux: ::

    $ python
    >>> import faulthandler
    >>> faulthandler.enable()
    >>> faulthandler._sigsegv()
    Fatal Python error: Segmentation fault

    Traceback (most recent call first):
      File "<stdin>", line 1 in <module>
    Segmentation fault


Installation
============

To install faulthandler module, type the following command: ::

    python setup.py install

Then you can test your setup using the following command: ::

    python tests.py

You need a C compiler (eg. gcc) and Python headers to build the faulthandler
module. Eg. on Fedora, you have to install python-devel package (sudo yum
install python-devel).


faulthandler module API
=======================

There are 4 different ways to display the Python traceback:

 * enable(): on a crash
 * dump_tracebacks_later(): after a timeout (useful if your program hangs)
 * register(): by sending a signal (eg. SIGUSR1). It doesn't work on Windows.
 * dump_traceback(): explicitly

Fault handler state (disabled by default):

 * enable(file=sys.stderr, all_threads=False): enable the fault handler
 * disable(): disable the fault handler
 * is_enabled(): get the status of the fault handler

Dump the current traceback:

 * dump_traceback(file=sys.stderr, all_threads=False): dump traceback of the
   current thread, or of all threads if all_threads is True, into file
 * dump_tracebacks_later(timeout, repeat=False, file=sys.stderr,
   exit=False): dump the traceback of all threads in timeout seconds, or each
   timeout seconds if repeat is True. If the function is called twice, the new
   call replaces previous parameters. Exit immediatly if exit is True.
 * cancel_dump_tracebacks_later(): cancel the previous call to
   dump_tracebacks_later()

dump_tracebacks_later() is implemented using the SIGALRM signal and the alarm()
function: if the signal handler is called during a system call, the system call
is interrupted (return EINTR). It it not available on Windows.

enable() and dump_tracebacks_later() keep an internal reference to the output
file. Use disable() and cancel_dump_tracebacks_later() to clear this reference.

Dump the traceback on an user signal:

 * register(signum, file=sys.stderr, all_threads=False): register an handler
   for the signal 'signum': dump the traceback of the current thread, or of all
   threads if all_threads is True, into file". Not available on Windows.
 * unregister(signum): unregister the handler of the signal 'signum' registered
   by register(). Not available on Windows.

Functions to test the fault handler:

 * _fatal_error(message): Exit Python with a fatal error, call Py_FatalError()
   with message.
 * _read_null(): read from the NULL pointer (raise SIGSEGV or SIGBUS depending
   on the platform)
 * _sigabrt(): raise a SIGABRT signal (Aborted)
 * _sigbus(): raise a SIGBUS signal (Bus error)
 * _sigfpe(): raise a SIGFPE signal (Floating point exception), do a division by
   zero
 * _sigill(): raise a SIGILL signal (Illegal instruction)
 * _sigsegv(): raise a SIGSEGV signal (Segmentation fault), read memory from
   NULL (address 0)
 * _stack_overflow(): raise a stack overflow error. Not available on all
   platforms.

register(), unregister(), sigbus() and sigill() are not available on all
operation systems.

faulthandler.version_info is the module version as a tuple: (major, minor),
faulthandler.__version__ is the module version as a string (e.g. "2.0").


Changelog
=========

Version 2.1 (under development)
-------------------------------

Major changes:

 * Add an optional chain argument to faulthandler.register()

Version 2.0 (2011-05-10)
------------------------

Major changes:

 * faulthandler is now part of Python 3.3
 * enable() handles also the SIGABRT signal
 * Add exit option to dump_tracebacks_later(): if True, exit the program
   on timeout after dumping the traceback

Other changes:

 * Change default value of the all_threads argument: dump all threads by
   default because under some rare conditions, it is not possible to get
   the current thread
 * Save/restore errno in signal handlers
 * dump_tracebacks_later() always dump all threads: remove all_threads option
 * Add faulthandler.__version__ attribute (module version as a string)
 * faulthandler.version is now a tuple
 * Rename:

   * dump_traceback_later() to dump_tracebacks_later()
   * cancel_dump_traceback_later() to cancel_dump_tracebacks_later()
   * sigsegv() to _sigsegv()
   * sigfpe() to _sigfpe()
   * sigbus() to _sigbus()
   * sigill() to _sigill()

 * register() and unregister() are no more available on Windows. They were
   useless: only SIGSEGV, SIGABRT and SIGILL can be handled by the application,
   and these signals can only be handled by enable().
 * Add _fatal_error(), _read_null(), _sigabrt() and _stack_overflow() test
   functions
 * register() uses sigaction() SA_RESTART flag to try to not interrupt the
   current system call
 * The fault handler calls the previous signal handler, using sigaction()
   SA_NODEFER flag to call it immediatly
 * enable() raises an OSError if it was not possible to register a signal
   handler
 * Set module size to 0, instead of -1, to be able to unload the module with
   Python 3
 * Fix a reference leak in dump_tracebacks_later()
 * Fix register() if it called twice with the same signal
 * Implement m_traverse for Python 3 to help the garbage collector
 * Move code from faulthandler/\*.c to faulthandler.c and traceback.c: the code
   is simpler and it was easier to integrate faulthandler into Python 3.3 using
   one file (traceback.c already existed in Python)
 * register() uses a static list for all signals instead of reallocating memory
   each time a new signal is registered, because the list is shared with the
   signal handler which may be called anytime.

Version 1.5 (2011-03-24)
------------------------

 * Conform to the PEP 8:

   * Rename isenabled() to is_enabled()
   * Rename dumpbacktrace() to dump_traceback()
   * Rename dumpbacktrace_later() to dump_traceback_later()
   * Rename cancel_dumpbacktrace_later() to cancel_dump_traceback_later()

 * Limit strings to 100 characters
 * dump_traceback_later() signal handler doesn't clear its reference to the
   file, because Py_CLEAR() is not signal safe: you have to call explicitly
   cancel_dump_traceback_later()

Version 1.4 (2011-02-14)
------------------------

 * Add register() and unregister() functions
 * Add optional all_threads argument to enable()
 * Limit the backtrace to 100 threads
 * Allocate an alternative stack for the fatal signal handler to be able to
   display a backtrace on a stack overflow (define HAVE_SIGALTSTACK). Not
   available on Windows.

Version 1.3 (2011-01-31)
------------------------

 * Don't compile dumpbacktrace_later() and cancel_dumpbacktrace_later() on
   Windows because alarm() is missing

Version 1.2 (2011-01-31)
------------------------

 * Add dumpbacktrace_later() and cancel_dumpbacktrace_later() function
 * enable() and dumpbacktrace() get an optional file argument
 * Replace dumpbacktrace_threads() function by a new dumpbacktrace() argument:
   dumpbacktrace(all_threads=True)
 * enable() gets the file descriptor of sys.stderr instead of using the file
   descriptor 2

Version 1.1 (2011-01-03)
------------------------

 * Disable the handler by default, because pkgutil may load the module and so
   enable the handler which is unexpected
 * Add dumpbacktrace() and dumpbacktrace_threads() functions
 * sigill() is available on Windows thanks to Martin's patch
 * Fix dump_ascii() for signed char type (eg. on FreeBSD)
 * Fix tests.py for Python 2.5

Version 1.0 (2010-12-24)
------------------------

 * First public release


Status
======

 * 2011-01-31: Version 1.2 tested with Python 2.5, 2.6, 2.7, 3.1 and 3.2 on
   Debian Sid
 * 2010-12-24: Tested with Python 2.6, 3.1 and 3.2 on Debian Sid
 * 2010-12-24: Tested with Python 2.6 and 3.1 on Windows XP


Similar projects
================

Python debuggers:

 * minidumper is a C extension for writing "minidumps" for post-mortem analysis
   of crashes in Python or its extensions:
   https://bitbucket.org/briancurtin/minidumper/
 * tipper: write the traceback of the current thread into a file on SIGUSR1
   signal: http://pypi.python.org/pypi/tipper/
 * crier: write the traceback of the current thread into a file
   (eg. /tmp/dump-<pid>) if a "request" file is created (eg. /tmp/crier-<pid>).
   Implemented using a thread. https://gist.github.com/737056
 * Python WAD (Wrapped Application Debugger), not update since 2001:
   http://www.dabeaz.com/papers/Python2001/python.html

Application fault handlers:

 * The GNU libc has a fault handler in debug/segfault.c
 * XEmacs has a fault handler displaying the Lisp traceback
 * RPy has a fault handler

System-wide fault handlers:

 * Ubuntu uses Apport: https://wiki.ubuntu.com/Apport
 * The Linux kernel logs also segfaults into /var/log/kern.log (and
   /var/log/syslog). /proc/sys/kernel/core_pattern contols how coredumps are
   created.
 * Windows opens a popup on a fatal error asking if the error should be
   reported to Microsoft


See also
========

 * http://bugs.python.org/issue8863 (may 2010):
   Display Python backtrace on SIGSEGV, SIGFPE and fatal error
 * http://bugs.python.org/issue3999 (sept. 2008):
   Real segmentation fault handler