>>>>> "Adrian" == Adrian Aichner <aichner(a)ecf.teradyne.com> writes:
>>>>> "Martin" == Martin Buchholz <martin(a)xemacs.org> writes:
>>>>> "APA" == Adrian Aichner <aichner(a)ecf.teradyne.com> writes:
APA> case timeout_event:
-> markobj (event->event.timeout.function);
APA> markobj (event->event.timeout.object);
APA> break;
APA> where event->event.timeout.function is:
APA> - event->event.timeout.function {...}
APA> - gu {...}
APA> type 0
APA> val 0
APA> - s {...}
APA> bits 0
APA> val 0
APA> - u {...}
APA> bits 0
APA> val 0
APA> ui 0
APA> i 0
APA> v 0x00000000
APA> cv 0x00000000
Martin> Ah, yes. Lisp_Objects which are Qnull_pointer have been
Martin> outlawed and will crash in mark_object. This is an
Martin> intentional change. This is to catch uninitialized
Martin> Lisp_Objects (also makes garbage collection faster).
Martin> Now we will crash quickly in the first garbage collection
Martin> instead of crashing rarely and more obscurely later,
Martin> e.g. in print_event. Trust me, this is short term pain
Martin> for long term gain.
Adrian> That sounds promising!
Martin> The event object type is particularly prone to
Martin> non-initialization since it contains a union of structs
Martin> possibly containing Lisp_Objects.
Martin> Please test out this patch. Especially Adrian, I would
Martin> appreciate if you could give me a `thumbs up' by Sunday.
Martin> Untested on NT, as usual.
Adrian> How about this Backtrace then? :-)
Here's another. This time it occured when I tried to bring up the CVS
menu for a cvs-examined directory.
Adrian> This occured after rebuilding with you patches successfully applied.
Adrian> I assume it's correct that only
Adrian> event-msw.c had to be re-compiled (for WidowsNT native build)
Adrian> before link.exe ＠C:\WINNT\Profiles\AichnerA\tmp\nma00176. That's what
Adrian> xemacs.mak did anyway.
Adrian> Here's the VC++ 5.0 CallStack, edited for bandwidth:
Adrian> allocate_lisp_storage(unsigned int 2044) line 482 + 3 bytes
Adrian> Fcons(Lisp_Object {...}, Lisp_Object {...}) line 1143 + 71 bytes
Adrian> list2(Lisp_Object {...}, Lisp_Object {...}) line 1197 + 15 bytes
Adrian> throw_or_bomb_out(Lisp_Object {...}, Lisp_Object {...}, int 0, Lisp_Object
Adrian> {...}, Lisp_Object {...}) line 1432 + 13 bytes
Adrian> Fthrow(Lisp_Object {...}, Lisp_Object {...}) line 1468 + 28 bytes
Adrian> Fsignal(Lisp_Object {...}, Lisp_Object {...}) line 2036 + 18 bytes
Adrian> throw_or_bomb_out(Lisp_Object {...}, Lisp_Object {...}, int 0, Lisp_Object
Adrian> {...}, Lisp_Object {...}) line 1432 + 29 bytes
Adrian> ABOVE THREE LINES REPEATED 326 TIMES
Adrian> Fthrow(Lisp_Object {...}, Lisp_Object {...}) line 1468 + 28 bytes
Adrian> It occured after XEmacs had started, reading my extensive
Adrian> site-start.el and .emacs (crashes usually don't happen with
Adrian> -no-init-file -no-site-file, sorry). Hitting the News toolbar button,
Adrian> which starts Gnus for me, crashed XEmacs-21.0-b5++++++
Adrian> Thanks for looking into this! I am confident that this is the right
Adrian> way to a robust XEmacs, EVEN on WindowsNT ;->
Adrian> I'll try my best to give crash feedback. Let me know what additional
Adrian> information might help.
Adrian> Regards,
Adrian> Adrian
Martin> 1998-12-11 Martin Buchholz <martin(a)xemacs.org>
Martin> * events.h (struct timeout_data):
Martin> * event-tty.c (tty_timeout_to_emacs_event):
Martin> * event-msw.c (mswindows_wm_timer_callback):
Martin> * event-Xt.c (Xt_timeout_to_emacs_event):
Martin> * event-msw.c (mswindows_cancel_dispatch_event):
Martin> Make sure Lisp_Objects inside events are initialized to
Adrian> Qnil, not
Martin> Qnull_pointer, which is now illegal.
