is_defined() again [Was: Follow up on profiler development]

[John W. Eaton]

On 10-Aug-2011, Daniel Kraft wrote:

| -----BEGIN PGP SIGNED MESSAGE-----
| Hash: SHA1
| 
| Hi again,
| 
| I just observed an "interesting" segfault while working on hierarchical
| profiling (but it happens as well with the version without my work in
| progress):
| 
| Running "demo profshow" three times in a row makes Octave segfault --
| but GDB shows that the segfault happens not inside some profiler data
| collection routine (which I first thought, since I reworked them for
| hierarchical profiling) but rather with is_defined().  This routine
| becomes more and more mystical to me over time! :D
| 
| Any ideas?  But until the problem occurs in a more "direct" situation,
| I'll defer this to later.  (Of course, maybe it is still a problem with
| my profiler code which corrupts some memory but leads only later to a
| segfault ... or something.  Maybe I'll try to run Octave through
| valgrind to find that out -- any experience with that?  Does it work?)

The demo function uses "eval" to create a temporary function called
__demo__.  I noticed that the crash occurs when looking in the symbol
table for the definition of this function.

You can trigger the crash with a demo block as simple as

  %!demo
  %! function f ()
  %! endfunction
  %! f ()
%!demo
%! function f ()
%! endfunction
%! f ()

so it seems unlikely that the problem is with your code.  I opened a
bug report for this problem here:

  https://savannah.gnu.org/bugs/index.php?33981

jwe