Simply running './python -X faulthandler' in the source directory gives me this:

% ./python -X faulthandler
Fatal Python error: Py_Initialize: can't initialize faulthandler
SystemError: null argument to internal routine
[1]    25118 abort (core dumped)  ./python -X faulthandler

% gdb ./python core
Core was generated by `./python -X faulthandler'.
Program terminated with signal 6, Aborted.
#0  0x00007f52d7ff9475 in *__GI_raise (sig=<optimized out>)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
64	../nptl/sysdeps/unix/sysv/linux/raise.c: No such file or directory.
(gdb) bt
#0  0x00007f52d7ff9475 in *__GI_raise (sig=<optimized out>)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
#1  0x00007f52d7ffc6f0 in *__GI_abort () at abort.c:92
#2  0x00000000004bc984 in Py_FatalError (msg=0x5b3750 "Py_Initialize: can't initialize faulthandler")
    at Python/pythonrun.c:2283
#3  0x00000000004b85ed in Py_InitializeEx (install_sigs=1) at Python/pythonrun.c:361
#4  0x00000000004b86ea in Py_Initialize () at Python/pythonrun.c:398
#5  0x00000000004d55a6 in Py_Main (argc=3, argv=0x1b8f010) at Modules/main.c:624
#6  0x000000000041b120 in main (argc=3, argv=0x7fffc1ebb558) at ./Modules/python.c:65
(gdb) 
#0  0x00007f52d7ff9475 in *__GI_raise (sig=<optimized out>)
    at ../nptl/sysdeps/unix/sysv/linux/raise.c:64
#1  0x00007f52d7ffc6f0 in *__GI_abort () at abort.c:92
#2  0x00000000004bc984 in Py_FatalError (msg=0x5b3750 "Py_Initialize: can't initialize faulthandler")
    at Python/pythonrun.c:2283
#3  0x00000000004b85ed in Py_InitializeEx (install_sigs=1) at Python/pythonrun.c:361
#4  0x00000000004b86ea in Py_Initialize () at Python/pythonrun.c:398
#5  0x00000000004d55a6 in Py_Main (argc=3, argv=0x1b8f010) at Modules/main.c:624
#6  0x000000000041b120 in main (argc=3, argv=0x7fffc1ebb558) at ./Modules/python.c:65

Here is a patch:


$ hg di
diff --git a/Python/pythonrun.c b/Python/pythonrun.c
--- a/Python/pythonrun.c
+++ b/Python/pythonrun.c
@@ -356,15 +356,15 @@ Py_InitializeEx(int install_sigs)
 
     _PyImportHooks_Init();
 
-    /* initialize the faulthandler module */
-    if (_PyFaulthandler_Init())
-        Py_FatalError("Py_Initialize: can't initialize faulthandler");
-
     /* Initialize _warnings. */
     _PyWarnings_Init();
 
     import_init(interp, sysmod);
 
+    /* initialize the faulthandler module */
+    if (_PyFaulthandler_Init())
+        Py_FatalError("Py_Initialize: can't initialize faulthandler");
+
     _PyTime_Init();
 
     if (initfsencoding(interp) < 0)

I can confirm that it works with the patch. Thanks!

The patch looks good to me.

Hum, a test may be added to ensure that we will not have the regression anymore.

% PYTHONFAULTHANDLER=1 ./python -E -c 'import faulthandler; faulthandler._sigsegv()'
[3]    14516 segmentation fault (core dumped)

Unless I'm missing something, the env. var. is not working as documented.

Patch with two tests is attached: the first does 'python -X faulthandler ...' and passes after Antoine's patch, the second does 'PYTHONFAULTHANDLER=YesPlease python ...' and does not pass.

> PYTHONFAULTHANDLER=1 ./python -E ...

Documentation of the -E option "Ignore all PYTHON* environment
variables, e.g. PYTHONPATH and PYTHONHOME, that might be set."
http://docs.python.org/dev/using/cmdline.html#cmdoption-E

So the option works as expected.

A new version of the tests: one for 'python -X faulthandler', one for 'PYTHONFAULTHANDLER=1 python'. This one sets the environment properly for the second test, but is slightly more invasive.

Both tests fail without Antoine's patch, and both succeed when it is applied.