[Python-bugs-list] [ python-Bugs-766669 ] Consistent GPF on exit
SourceForge.net
noreply@sourceforge.net
Sun, 06 Jul 2003 21:21:14 -0700
Bugs item #766669, was opened at 2003-07-06 08:32
Message generated for change (Comment added) made by tim_one
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=766669&group_id=5470
Category: Windows
Group: Python 2.2.3
Status: Open
Resolution: None
Priority: 5
Submitted By: Kurt Grittner (grittkmg)
Assigned to: Nobody/Anonymous (nobody)
Summary: Consistent GPF on exit
Initial Comment:
Using the following script (clt.py):
import os, time, sys
from socket import *
from Tkinter import *
class App:
def __init__(self, master):
self.win = Toplevel(master)
self.button = Button(self.win, text="QUIT",
fg="red", command=self.goaway)
self.button.pack(side=LEFT)
self.hi_there = Button(self.win, text="Hello",
command=self.say_hi)
self.hi_there.pack(side=LEFT)
#self.serverHost = 'localhost'
self.serverHost = '192.168.1.12'
self.serverPort = 50007
self.sockobj = socket(AF_INET, SOCK_STREAM)
self.sockobj.connect((self.serverHost, self.
serverPort))
def say_hi(self):
self.message = ['Hello network world']
for line in self.message:
self.sockobj.send(line)
data = self.sockobj.recv(1024)
print 'Client received:', `data`
def goaway(self):
self.sockobj.shutdown(0)
self.sockobj.close()
self.win.destroy()
def NewClient():
App(root)
def PgmExit():
root.quit()
root = Tk()
Button(root, text='New Client', command=NewClient).
pack()
Button(root, text='Quit All Clients', command=PgmExit).
pack()
root.mainloop()
If you press 'Quit' first, then there are no errors.
If you press 'New Client' even once (whether or not there
is an echo server to talk to) the program runs the other
Toplevel windows seemingly without problems, but then
when you press 'Quit' you die in the following place :
/* Vc98\Crt\Src\Crtexe.c */
#ifdef WPRFLAG
__winitenv = envp;
mainret = wmain(argc, argv, envp);
#else /* WPRFLAG */
__initenv = envp;
mainret = main(argc, argv, envp);
#endif /* WPRFLAG */
#endif /* _WINMAIN_ */
/*
-------------------------------------------------------
----------------- */
/* Error executing following line */
exit(mainret);
/*
-------------------------------------------------------
----------------- */
/* OS error dump
PYTHON_D caused an invalid page fault in
module KERNEL32.DLL at 017f:bff88396.
Registers:
EAX=c00309c4 CS=017f EIP=bff88396 EFLGS=00210216
EBX=0062ffec SS=0187 ESP=0052fecc EBP=00530044
ECX=00000000 DS=0187 ESI=00000000 FS=10e7
EDX=bff76855 ES=0187 EDI=bff79060 GS=0000
Bytes at CS:EIP:
53 56 57 8b 75 10 8b 38 33 db 85 f6 75 2d 8d b5
Stack dump:
/* Vc98\Crt\Src\Crt0dat.c */
/* First debug run */
1020ACE0 jmp doexit+0B6h (1020acf6)
384: }
385:
386:
387: _C_Exit_Done = TRUE;
1020ACE2 mov dword ptr [__C_Exit_Done
(10264728)],1
388:
389: ExitProcess(code);
1020ACEC mov ecx,dword ptr [code]
1020ACEF push ecx
/*
-------------------------------------------------------
----------------- */
/* Error on the following call */
1020ACF0 call dword ptr [__imp__ExitProcess@4
(10256020)]
/*
-------------------------------------------------------
----------------- */
/* Second debug run */
PYTHON_D caused an invalid page fault in
module KERNEL32.DLL at 017f:bff88396.
Registers:
EAX=c00309c4 CS=017f EIP=bff88396 EFLGS=00210216
EBX=0062ffec SS=0187 ESP=0052fecc EBP=00530044
ECX=00000000 DS=0187 ESI=00000000 FS=1b3f
EDX=bff76855 ES=0187 EDI=bff79060 GS=0000
Bytes at CS:EIP:
53 56 57 8b 75 10 8b 38 33 db 85 f6 75 2d 8d b5
Stack dump:
390:
391:
392: }
1020ACF6 mov esp,ebp
1020ACF8 pop ebp
1020ACF9 ret
/* Vc98\Crt\Src\Crtexe.c - Continued */
*/
}
__except ( _XcptFilter(GetExceptionCode(),
GetExceptionInformation()) )
{
/*
* Should never reach here
*/
_exit( GetExceptionCode() );
} /* end of try - except */
}
It seems weird that everything else in the program works
fine right up to ExitProcess. The error occurs on multiple
windows machines. The error does NOT occur on linux
running the same code base. Also, many other
multi-threaded sample programs from other authors
exhibit the same 'crash on exit' syndrome.
Thanks for your attention,
Kurt
----------------------------------------------------------------------
>Comment By: Tim Peters (tim_one)
Date: 2003-07-07 00:21
Message:
Logged In: YES
user_id=31435
You're trying too hard <wink>. I'm not arguing about the
design (for one thing, it's not my design, and I don't know
anything about it), I'm wondering both why I can't reproduce
a problem, and especially why no other report of this has ever
been made. You surely don't believe you're the only
programmer to try sockets in Python on Windows, right?
I also tried it under an up-to-date Win98SE, also with MSVC 6
and all current IE 6 patches. Indeed, that's the machine I've
built the PythonLabs Windows installer *on* for the last two
years. No problem. Everyone who has ever run the Python
test suite on Windows has exercised sockets on Windows,
and so has anyone who has ever run Zope on Windows, etc.
It may indeed be a bit of Rocket Science if you don't have a
theory for why it fails only when you run it!
The reason I'm keen to know is that only someone who can
see the problem can test a putative fix (and so confirm-- or
refute --that the problem goes away).
----------------------------------------------------------------------
Comment By: Kurt Grittner (grittkmg)
Date: 2003-07-07 00:00
Message:
Logged In: YES
user_id=816888
Umm, this is not rocket science. WSACleanup() is being called
in the wrong context. Startup is called in a DLL context. That
puts the info in Thread Local Storage (TLS). The DLL Is
allowed to unload with an atexit() pointer still pending to it.
This is bad design. You should always call startup/cleanup in
pairs and in the same thread context. As I said before this
happens on every win98SE machine that I have tried it on. I
have no firewall or virus scanner. This is a fairly recent
installation of win98 with all of the MS updates for I.E. Apart
from that, the most unusual thing is the presence of the
VisualStudio 6.0 suite of programs.
I can do a new install of win98SE on an old machine, just out
of curiosity, but you only have to read this web site to see
that this design for startup/cleanup calls is a recipie for
disaster:
http://support.microsoft.com/default.aspx?scid=http:
//support.microsoft.com:80/support/kb/articles/Q237/5/72.
ASP&NoWebContent=1
Here is a list of some of the modules running on my machine:
USER32.DLL 4.10.2227 Microsoft Corporation Win32
USER32 core component C:\WINDOWS\SYSTEM\USER32.
DLL 4/21/00 3:33:18 PM GMT Microsoft(R) Windows(R)
Operating System bfc00000 - bfc11000
GDI32.DLL 4.10.1998 Microsoft Corporation Win32
GDI core component C:\WINDOWS\SYSTEM\GDI32.
DLL 4/23/98 4:24:18 AM GMT Microsoft(R) Windows(R)
Operating System bff20000 - bff46000
ADVAPI32.DLL 4.80.1675 Microsoft
Corporation Win32 ADVAPI32 core component C:
\WINDOWS\SYSTEM\ADVAPI32.DLL 4/23/99 4:37:33 PM
GMT Microsoft(R) Windows(R) Operating
System bfe80000 - bfe90000
KERNEL32.DLL 4.10.2222 Microsoft
Corporation Win32 Kernel core component C:
\WINDOWS\SYSTEM\KERNEL32.DLL 4/23/99 12:45:39 AM
GMT Microsoft(R) Windows(R) Operating
System bff70000 - bffe3000
MPR.DLL 4.10.1998 Microsoft Corporation WIN32
Network Interface DLL C:\WINDOWS\SYSTEM\MPR.
DLL 4/23/99 4:37:36 PM GMT Microsoft(R) Windows(R)
Operating System 7fbf0000 - 7fbfe000
MSNP32.DLL 4.10.2222 Microsoft
Corporation Network provider for Microsoft
networks C:\WINDOWS\SYSTEM\MSNP32.DLL 4/23/99
4:37:37 PM GMT Microsoft(R) Windows(R) Operating
System 7fb20000 - 7fb34000
MSNET32.DLL 4.10.2224 Microsoft
Corporation Microsoft 32-bit Network API Library C:
\WINDOWS\SYSTEM\MSNET32.DLL 11/12/99 4:15:15 AM
GMT Microsoft(R) Windows(R) Operating
System 7f300000 - 7f313000
MPREXE.EXE 4.10.1998 Microsoft
Corporation WIN32 Network Interface Service
Process C:\WINDOWS\SYSTEM\MPREXE.EXE 4/23/98
2:19:38 AM GMT Microsoft(R) Windows(R) Operating
System 00500000 - 00507000
MPRSERV.DLL 4.10.2222 Microsoft
Corporation Multinet Router C:
\WINDOWS\SYSTEM\MPRSERV.DLL 4/23/99 4:37:37 PM
GMT Microsoft(R) Windows(R) Operating
System 7fad0000 - 7faf6000
NETAPI32.DLL 4.10.1998 Microsoft
Corporation 32-bit network API DLL C:
\WINDOWS\SYSTEM\NETAPI32.DLL 4/23/99 4:37:37 PM
GMT Microsoft(R) Windows(R) Operating
System 7f990000 - 7f995000
NETBIOS.DLL C:\WINDOWS\SYSTEM\NETBIOS.
DLL 4/23/99 4:37:39 PM GMT 7f840000 - 7f848000
RNR20.DLL 4.10.2222 Microsoft
Corporation Windows Socket2 NameSpace DLL C:
\WINDOWS\SYSTEM\RNR20.DLL 4/23/99 4:38:44 PM
GMT Microsoft(R) Windows(R) Operating
System 783c0000 - 783cf000
MSAFD.DLL 4.10.1998 Microsoft
Corporation Microsoft Windows Sockets 2.0 Service
Provider C:\WINDOWS\SYSTEM\MSAFD.DLL 4/23/99 4:
38:11 PM GMT Microsoft(R) Windows(R) Operating
System 7b410000 - 7b41b000
RPCLTSCM.DLL 4.71.2900 Microsoft
Corporation Remote Process Control LTSCM DLL C:
\WINDOWS\SYSTEM\RPCLTSCM.DLL 4/23/99 4:38:45 PM
GMT Microsoft® Windows® Operating
System 78320000 - 7832a000
WSOCK32.DLL 4.10.1998 Microsoft
Corporation BSD Socket API for Windows C:
\WINDOWS\SYSTEM\WSOCK32.DLL 4/23/99 4:39:14 PM
GMT Microsoft(R) Windows(R) Operating
System 75fa0000 - 75faa000
MSWSOCK.DLL 4.10.2222 Microsoft
Corporation Microsoft WinSock Extension APIs C:
\WINDOWS\SYSTEM\MSWSOCK.DLL 4/23/99 4:38:30 PM
GMT Microsoft(R) Windows(R) Operating
System 794d0000 - 794e5000
WS2_32.DLL 4.10.2222 Microsoft
Corporation Windows Socket 2.0 32-Bit DLL C:
\WINDOWS\SYSTEM\WS2_32.DLL 4/23/99 4:39:13 PM
GMT Microsoft(R) Windows(R) Operating
System 76000000 - 76012000
WS2HELP.DLL 4.10.1998 Microsoft
Corporation Windows Socket 2.0 Helper for Windows
98 C:\WINDOWS\SYSTEM\WS2HELP.DLL 4/23/99 4:
39:13 PM GMT Microsoft(R) Windows(R) Operating
System 75fe0000 - 75fe6000
DIGEST.DLL 6.00.2800.1106 Microsoft
Corporation Digest SSPI Authentication Package C:
\WINDOWS\SYSTEM\DIGEST.DLL 8/29/02 11:21:08 AM
GMT Microsoft® Windows® Operating
System 007a0000 - 007b0000
MSNSSPC.DLL 6.00.7753 Microsoft
Corporation MSN Client for 32 bit platforms C:
\WINDOWS\SYSTEM\MSNSSPC.DLL 4/23/99 4:38:21 PM
GMT Microsoft® Internet Services 7a210000 -
7a22f000
MSAPSSPC.DLL 5.00.7729 Microsoft
Corporation DPA Client for 32 bit platforms C:
\WINDOWS\SYSTEM\MSAPSSPC.DLL 4/23/99 4:38:11 PM
GMT Microsoft® Internet Services 7b3f0000 -
7b401000
MSVCRT40.DLL 4.22.0000 Microsoft
Corporation Microsoft® C Runtime Library C:
\WINDOWS\SYSTEM\MSVCRT40.DLL 4/23/99 4:38:29 PM
GMT Microsoft® Visual C++ 79660000 - 796b5000
SECUR32.DLL 4.10.2222 Microsoft
Corporation Microsoft Win32 Security Services C:
\WINDOWS\SYSTEM\SECUR32.DLL 4/23/99 4:37:39 PM
GMT Microsoft(R) Windows(R) Operating
System 7f870000 - 7f87a000
RPCSS.EXE 4.71.2900 Microsoft
Corporation Distributed COM Services C:
\WINDOWS\SYSTEM\RPCSS.EXE 9/1/98 6:18:39 PM
GMT Microsoft(R) Windows NT(TM) Operating
System 01000000 - 01005000
MSVCRT20.DLL 2.11.000 Microsoft
Corporation Microsoft® C Runtime Library C:
\WINDOWS\SYSTEM\MSVCRT20.DLL 4/23/99 4:37:36 PM
GMT Microsoft® Visual C++ 7fc30000 - 7fc75000
----------------------------------------------------------------------
Comment By: Tim Peters (tim_one)
Date: 2003-07-06 23:13
Message:
Logged In: YES
user_id=31435
Kurt, which versions of Windows and winsock.dll are you
using? I continue to be unable to reproduce any problem
here, and there have been no other reports of this despite
thousands of Python programs using sockets on Windows for
a decade. I have to conclude there's something unique in
your setup. FYI, sometimes virus scanners and/or "personal
firewalls" cause bizarre crashes during startup or shutdown.
Are you running one (or more) of those?
----------------------------------------------------------------------
Comment By: Kurt Grittner (grittkmg)
Date: 2003-07-06 21:29
Message:
Logged In: YES
user_id=816888
I have a working fix for this problem, but the code is a fairly
ugly hack. (A terrible thing to do to such a well organized
codebase).
1. I exported a fini_socket() function from the PYD DLL.
2. I cloned the dynamic lookup function to check for the
optional fini_socket() name as well as the init_socket(). If
found, I remember it in a global.
3. In PyImport_Cleanup() I call the fini_socket() routine (if
present) before these modules start getting freed.
4. In NTInit() I add to a counter instead of using atexit()
5. fini_socket() calls NTcleanup(), which simply calls
WSACleanup 'counter' times to release the winsock DLL.
This works, so now I can go on with my Python study. If
anyone wants my hack code, just email me.
Kurt
----------------------------------------------------------------------
Comment By: Kurt Grittner (grittkmg)
Date: 2003-07-06 16:57
Message:
Logged In: YES
user_id=816888
The minimum required to cause this GPF is the following:
C:\ptest>python
Python 2.3b2 (#43, Jun 29 2003, 16:43:04) [MSC v.1200 32
bit (Intel)] on win32
Type "help", "copyright", "credits" or "license" for more
information.
>>> from socket import *
>>> s=socket(AF_INET, SOCK_STREAM)
>>><Ctrl-Z>
I have found the problem in the source, though I don't know
how to fix it. WSAStartup() is called as a DLL export, but the
WSACleanup() is called as an atexit(). This is too late to call
it. It must be called before the DLL unloads, or the
WSAStartup should be moved to the process level. I don't
know the program well enough to attempt a hack, but I did
verify that it is the WSACleanup line running from the CRT
code that is killing things (every time you even create a
socket).
Notice, this is using the latest version too.
Kurt
----------------------------------------------------------------------
Comment By: Kurt Grittner (grittkmg)
Date: 2003-07-06 14:41
Message:
Logged In: YES
user_id=816888
Hi Tim,
Here is some more info. I stripped out all references to
socket, and saved as clt2.clw to force it to load with
pythonw.exe. Things worked perfectly. Here is the script:
import os, time, sys
#from socket import *
from Tkinter import *
class App:
def __init__(self, master):
self.win = Toplevel(master)
self.button = Button(self.win, text="QUIT", fg="red",
command=self.goaway)
self.button.pack(side=LEFT)
self.hi_there = Button(self.win, text="Hello",
command=self.say_hi)
self.hi_there.pack(side=LEFT)
#self.serverHost = 'localhost'
self.serverHost = '192.168.1.12'
self.serverPort = 50007
#self.sockobj = socket(AF_INET, SOCK_STREAM)
#self.sockobj.connect((self.serverHost, self.serverPort))
def say_hi(self):
self.message = ['Hello network world']
#for line in self.message:
# self.sockobj.send(line)
#data = self.sockobj.recv(1024)
print 'Client received:', `data`
def goaway(self):
#self.sockobj.shutdown(0)
#self.sockobj.close()
self.win.destroy()
def NewClient():
App(root)
def PgmExit():
root.quit()
root = Tk()
Button(root, text='New Client', command=NewClient).pack()
Button(root, text='Quit All Clients', command=PgmExit).pack()
root.mainloop()
So next, I did the reverse... I took out all references to Tkinter
and saved it as clt3.clw (again to use pythonw.exe). It blew
sky high again. Here's the script:
import os, time, sys
from socket import *
class App:
def __init__(self, master):
self.serverHost = '192.168.1.12'
self.serverPort = 50007
self.sockobj = socket(AF_INET, SOCK_STREAM)
self.sockobj.connect((self.serverHost, self.serverPort))
def say_hi(self):
self.message = ['Hello network world']
for line in self.message:
self.sockobj.send(line)
data = self.sockobj.recv(1024)
print 'Client received:', `data`
def goaway(self):
self.sockobj.shutdown(0)
self.sockobj.close()
x = App(0)
x.say_hi()
time.sleep(1)
y = App(1)
y.say_hi()
time.sleep(1)
x.goaway()
y.goaway()
time.sleep(1)
Here's the output from the linux-based echo server:
ns1:/www/python # python select-server.py
select-server loop starting
Connect: ('192.168.1.20', 2382) 135755344
got Hello network world on 135755344
Connect: ('192.168.1.20', 2383) 135744224
got Hello network world on 135744224
got on 135755344
got on 135744224
Here's the (now familiar) GPF output:
PYTHONW caused an invalid page fault in
module KERNEL32.DLL at 017f:bff7b9a6.
Registers:
EAX=00000000 CS=017f EIP=bff7b9a6 EFLGS=00000246
EBX=007961a0 SS=0187 ESP=0095fba8 EBP=0095fbe8
ECX=0095fbf8 DS=0187 ESI=10013168 FS=19d7
EDX=0079d670 ES=0187 EDI=0079e7b0 GS=356e
Bytes at CS:EIP:
ff 76 04 e8 13 89 ff ff 5e c2 04 00 56 8b 74 24
Stack dump:
007961a0 10007eb0 10013168 7800b317 0079eb50 0079d670
00000000 007a2c98 00000000 81dafb90 00794641 760028e8
0079d67c 0079d670 007961a0 0079d67c
And here's the netstat output from right after the crash:
C:\WINDOWS>netstat /a
Active Connections
Proto Local Address Foreign Address State
TCP amd2000:2339 AMD2000:0
LISTENING
TCP amd2000:135 AMD2000:0 LISTENING
TCP amd2000:1243 AMD2000:0
LISTENING
TCP amd2000:1025 AMD2000:0
LISTENING
TCP amd2000:1699 AMD2000:0
LISTENING
TCP amd2000:2339 www.mlgames.org:22
ESTABLISHED
TCP amd2000:2382 www.mlgames.org:50007
TIME_WAIT
TCP amd2000:2383 www.mlgames.org:50007
TIME_WAIT
TCP amd2000:137 AMD2000:0 LISTENING
TCP amd2000:138 AMD2000:0 LISTENING
TCP amd2000:nbsession AMD2000:0
LISTENING
TCP amd2000:1243 www.mlgames.org:22
ESTABLISHED
UDP amd2000:1699 *:*
UDP amd2000:nbname *:*
UDP amd2000:nbdatagram *:*
C:\WINDOWS>
Notice the TIME_WAIT connections to port 50007.
It looks like this is a 'socket' library problem. I also tried
adding del x del y and a long delay to the end of the script to
see if things would change... but no. Same explosion.
So, to answer your question. It happens on both python and
pythonw, and it happens without loading Tkinter. I suppose I
can try the latest development version as you suggested.
Kurt
----------------------------------------------------------------------
Comment By: Tim Peters (tim_one)
Date: 2003-07-06 12:34
Message:
Logged In: YES
user_id=31435
Unassigned. Didn't see any problem on Win98SE, under 2.2.3
or 2.3b2, using either of the self.serverHost assignments.
If you can, please try under 2.3b2. On Windows that ships
with the current version of Tcl/Tk (8.4.3), and some kinds of
Tcl/Tk Windows shutdown races are said to be fixed in 8.4.3.
Question: How do you start this program? With python.exe
or with pythonw.exe? The only known workaround for some
kinds of previous Tcl/Tk shutdown races was to start the
program with pythonw.exe. So if you haven't tried
pythonw.exe, try it and see whether the problem goes away.
If it does, it's almost certainly a bug in Tcl/Tk.
----------------------------------------------------------------------
You can respond by visiting:
https://sourceforge.net/tracker/?func=detail&atid=105470&aid=766669&group_id=5470