bug#16603: 24.3.50; Segfault when viewing a backtrace

[Dmitry Antipov]

On 01/31/2014 06:20 AM, Lars Ingebrigtsen wrote:

> (require 'gnus-group)
> (setq debug-on-error t)
> (gnus-read-ephemeral-emacs-bug-group 16577)
>
> Choose Rotem's article, and my Emacs crashes:

Reproduced.  With the only extra eassert:

=== modified file 'src/eval.c'
--- src/eval.c  2014-01-25 03:48:29 +0000
+++ src/eval.c  2014-01-31 06:49:49 +0000
@@ -3191,6 +3191,7 @@
 void
 record_unwind_protect (void (*function) (Lisp_Object), Lisp_Object arg)
 {
+  eassert (specpdl_ptr < specpdl + specpdl_size);
   specpdl_ptr->unwind.kind = SPECPDL_UNWIND;
   specpdl_ptr->unwind.func = function;
   specpdl_ptr->unwind.arg = arg;

I got the following backtrace:

#14 0x00000000005eafb9 in die (msg=0x70d440 "specpdl_ptr < specpdl + specpdl_size", 
file=0x70c498 "../../trunk/src/eval.c",
    line=3194) at ../../trunk/src/alloc.c:6761
#15 0x000000000060d987 in record_unwind_protect (function=0x605b1a 
<restore_stack_limits>, arg=...) at ../../trunk/src/eval.c:3194
#16 0x0000000000605c1f in call_debugger (arg=...) at ../../trunk/src/eval.c:290
#17 0x0000000000609b3b in maybe_call_debugger (conditions=..., sig=..., 
data=...) at ../../trunk/src/eval.c:1724
#18 0x00000000006093a5 in Fsignal (error_symbol=..., data=...) at 
../../trunk/src/eval.c:1542
#19 0x00000000006094be in xsignal (error_symbol=..., data=...) at 
../../trunk/src/eval.c:1579
#20 0x00000000006096e3 in signal_error (s=0x70d008 "Variable binding depth exceeds 
max-specpdl-size", arg=...)
    at ../../trunk/src/eval.c:1634
#21 0x000000000060a6f6 in grow_specpdl () at ../../trunk/src/eval.c:2023
#22 0x000000000060a7e3 in record_in_backtrace (function=..., 
args=0x7ffffff78020, nargs=1) at ../../trunk/src/eval.c:2042
#23 0x000000000060c383 in Ffuncall (nargs=2, args=0x7ffffff78018) at 
../../trunk/src/eval.c:2754

IIUC this is a kind of chicken-egg problem: when we're running out of specpdl
stack, we want to run a debugger, which, in turn, needs some specpdl space to 
run.

Dmitry