RE: Bug in g++: Testcase exits with segmentation fault
-Original Message-
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED]]On Behalf Of Chris
January
Sent: Tuesday, January 14, 2003 6:01 PM
To: Boris Schaeling
Cc: [EMAIL PROTECTED]
Subject: RE: Bug in g++: Testcase exits with segmentation fault
[...]
The bug is actually in your testcase.
The line:
Observers.push_back(observer());
creates a temporary observer object. So when the line
observer::observer() { Select.give(this); }
calls give(), it leaves a dangling pointer.
Oh, you are right. I've forgotten to use a smart pointer ... thanks for your
feedback!
Boris
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Bug reporting: http://cygwin.com/bugs.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
Bug in g++: Testcase exits with segmentation fault
The testcase attached to this mail compiled with g++ 2.95.3-10 or g++ 3.2 20020927 under Cygwin 1.3.18 exits with segmentation fault. It seems to be an error in g++ as the testcase doesn't work under Linux either. Can anyone help or do I have to report this error to the GCC website? BTW I found two ways to make the testcase work: 1) Change line 64 (and 70 respectively) and use observer *Observer instead of std::listobserver. I compiled the testcase with STLport-4.5.3 under Linux to see if it's a bug in the Standard Library but got again segmentation fault. 2) Remove line 75 which is a simple poll(NULL, 0, 2000). It works without this line but I don't know why. Boris testcase.cpp Description: Binary data -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
1999 C++ bug reappears after upgrade to 1.3.17
I upgraded Cygwin 1.3.12-2 to 1.3.17-1 and now can't run one of my C++ programs any more: A simple call of a virtual member function crashs Windows 98 (no problems under Linux). I found a bug report dated 03/1999 that reminds me of my C++ program and has a testcase attached that causes a Blue Screen of Death: http://sources.redhat.com/ml/cygwin/1999-03/msg00474.html Does this testcase work for anyone under Cygwin 1.3.17-1 and Windows 98? Boris -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
AW: modified poll() patch
-Ursprüngliche Nachricht- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]Im Auftrag von Corinna Vinschen Gesendet: Dienstag, 19. März 2002 11:55 An: [EMAIL PROTECTED] Betreff: Re: modified poll() patch [...] thanks for the patch but could you please send a diff related to the latest from current cvs? I didn't back out your patch since Hi Corinna, I've attached patch and changelog. Boris change.log2 Description: Binary data poll.patch2 Description: Binary data -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
modified poll() patch
Here is the new poll() patch. Now invalid descriptors are recognized (invalid = positive descriptors including 0 which are not open). If there are invalid descriptors poll() returns immediately with the count of invalid descriptors. The timer value is ignored. That's the way Linux handles this case. As far as I can see there is only one bug left in this poll() function: If a descriptor is closed revents is set to POLLIN and not to POLLHUP. That's the way previous poll() functions worked, too. This shouldn't be a problem as read() will return 0 and tell the application that the socket can be closed. I changed the test case, too (see polltest2.c). Hopefully this poll() doesn't cause any problems any more. Boris newpoll2.cc Description: Binary data poll.patch2 Description: Binary data polltest2.c Description: Binary data -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
poll() bugs and patch
There are several bugs in poll(): - poll() must return 0 if no descriptor is ready and the timer expires. It works correctly if all descriptors are non-negative. Otherwise poll() returns the count of negative descriptors which is wrong. - The implementation of poll() must call cygwin_select() even if all descriptors are invalid. Currently poll() returns immediately - and ignores any timer value. The return value is wrong again because of the bug above. - If the timer expires revents must be 0 and must not be set to anything else like POLLNVAL, POLLHUP or whatever. - If cygwin_select() returns -1 revents must be 0 and must not be set to POLLERR. Eg. a signal interrupting poll() doesn't mean an error has occured for a TCP connection. - If an error is pending revents must be set to POLLERR. - The implementation of poll() has a local variable called open_fds that isn't used. I can't see why this variable exists. I've attached three files: - newpoll.cc has a new implementation of poll() which fixes the bugs. - patch.cc is the patch. - polltest.c is a testcase. I rebuilt cygwin1.dll with my poll() function so there should be no problems when compiling. Boris newpoll.cc Description: Binary data patch.cc Description: Binary data polltest.c Description: Binary data -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
AW: poll() bugs and patch
-Ursprüngliche Nachricht- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]Im Auftrag von Corinna Vinschen Gesendet: Mittwoch, 13. März 2002 15:06 An: [EMAIL PROTECTED] Betreff: Re: poll() bugs and patch [...] Thanks for the patch but that's not quite the way to contribute a patch. I'm missing a ChangeLog entry and, worse, we don't have a copyright assignment from you. Due to licensing issues we need the copyright assignment as described on http://cygwin.com/contrib.html before we can incorporate the patch. It's bigger than what is slipping through the cracks as insignificant. Okay. I thought this is only a small patch and were a little bit lazy. I'll see what I can do. :) It would really be nice if you could send us this assignment (which unfortunately has to be send via snail mail so this takes some time) Can I fax the assignment? Boris -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
AW: poll() bugs and patch
-Ursprüngliche Nachricht- Von: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]Im Auftrag von Corinna Vinschen Gesendet: Mittwoch, 13. März 2002 18:14 An: [EMAIL PROTECTED] Betreff: Re: poll() bugs and patch [...] Actually, I'm inclined to just let this one in. There isn't a lot of new code, it's mainly just rearranged. Ok with me. However, I'll not take that patch w/o a ChangeLog entry. Please create one, Boris. Here they are: - change.log - poll.patch - newpoll.cc (new implementation) - polltest.c (testcase) HTH, Boris change.log Description: Binary data poll.patch Description: Binary data polltest.c Description: Binary data newpoll.cc Description: Binary data -- Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple Bug reporting: http://cygwin.com/bugs.html Documentation: http://cygwin.com/docs.html FAQ: http://cygwin.com/faq/
