Re: [Scigraphica-devel] Crash on Linux Alpha Suse 7.1

[R. L. <la...@us...>]

Mikael Borg wrote:
> 
> I have been able to track down the the problem: it is when calling
> gtk_viewport_get_hadjustment, in sg_plot.c:1253:

Line 1235, that is :).

> adj=gtk_viewport_get_hadjustment (GTK_VIEWPORT(plot->sw));
> 
> A few lines later:
> start=adj->value/adj->upper*fac;
> 
> But adj->upper == 0, so the program dies with a floating point exception.

Confirmed.
It's a bug, which is not only in 0.8, but also in CVS.


For example: start SG, do Help->SG Examples->3d

The following values are in sg_plot.c:update_ruler_expose_x()

p_size = 583.40919599999995
s_width = 1
fac = 8.5
adj = {data = {object = {klass = 0x84ad380, flags = 12, ref_count = 2, 
      object_data = 0x845f778}}, lower = 0, upper = 0, value = 0, 
  step_increment = 0, page_increment = 0, page_size = 0}


And thus adj->upper as zero is used in the denominator at line 1248/1249,
letting the value of the variables "start" and "end" explode!
gtk_ruler_set_range() then gets these "NaN" values as parameters!

Same problem occurs in sg_plot.c:update_ruler_expose_y().
In fact I wonder if this miscalculation in sg_plot.c:update_ruler_expose_x/y()
always happens ?


The problem is that not all systems segfault on devision by zero;
in my case it happily continues with a value of "NaN" (I wonder whether
I can change that behaviour).

Anyone has an idea how to tackle this bug?

Regards,
Rob.