The patch from comment #2 was previously shown not to work by Ben.  Did you re-test that failing patch or are your results actually for the second patch attempt in comment #5?

I "fixed" the seg fault reported below by:
change line 5118 of graphics.cc from;

  for (int i = 0; i < n; i++)
to
  for (int i = 1; i < n; i++)

I only guessed at this working, but it does.
A more serious fix is called for.

The problem appears to lie mostly with the fact that OpenGL and the FLTK backend use single precision variables for axis positioning.  When these values overflow the behavior of the plotting functions, such as title, becomes indeterminate.  See bug #32980. 

It is still true that the interpreter shouldn't segfault and perhaps there needs to be some Octave code written to identify when high dynamic range plots are being attempted and stop this behavior.

I've attached a patch which fixes the problem for me.  However, the graphics area of Octave are not my area of expertise so someone with more experience should check this.

(file #23080)

I get a consistently reproducible segmentation fault
with current development version on Mac OS X by doing the following:

>> [X,Y,Z]=sombrero;
>> surf (X, Y, Z*1e303);
>> title ('test')
Segmentation fault

this happens with both graphics_toolkit's but while gnuplot does show the sombrero, I see nothing in the fltk window.

c.