Re: COMEDI wrapper (was: Re: Data acquisition in Octave)

[Olaf Till]

On Thu, Nov 20, 2008 at 12:04:49PM -0500, John W. Eaton wrote:
> I have the beginnings of a package here:
> 
>   ftp://velveeta.che.wisc.edu/pub/octave-comedi/comedi-0.1.tar.gz
> 
Some first remarks:

The crash:

octave-3.0.2:1> dev = comedi_open ("/dev/comedi0")
dev = <comedi_t object>
octave-3.0.2:2> comedi_close (dev)
ans = 0
octave-3.0.2:3> comedi_close (dev)
panic: Segmentation fault -- stopping myself...
attempting to save variables to `octave-core'...
error: octave_base_value::save_binary(): wrong type argument `comedi_t'
save to `octave-core' complete
Segmentation fault
address@hidden:~/devel/src/octave-comedi-0.1/src$ 

can probably be cured by something like this patch:

--- orig-comedi_open.cc 2008-11-20 17:16:50.000000000 +0100
+++ comedi_open.cc      2008-11-21 10:45:39.000000000 +0100
@@ -163,6 +163,8 @@
 
   bool print_as_scalar (void) const { return true; }
 
+  void comedi_t_set_null (void) {dev = NULL;}
+
 private:
 
   comedi_t *dev;
@@ -222,6 +224,25 @@
   return retval;
 }
 
+static void
+octave_comedi_t_set_null (octave_value& arg)
+{
+
+  try
+    {
+      octave_base_value& rep = (octave_base_value&) arg.get_rep ();
+
+      octave_comedi_t& val = dynamic_cast<octave_comedi_t&> (rep);
+
+      val.comedi_t_set_null ();
+    }
+  catch (std::bad_cast)
+    {
+      error ("invalid conversion to comedi_t_value");
+    }
+
+}
+
 class
 octave_comedi_range : public octave_base_value
 {
@@ -361,8 +382,11 @@
     {
       comedi_t *it = octave_get_comedi_t_object (args(0));
 
-      if (! error_state)
+      if (! error_state && it) {
         retval = comedi_close (it);
+        octave_value arg0 = args(0);
+        octave_comedi_t_set_null (arg0);
+      }
       else
         error ("comedi_close: expecting comedi_t object argument");
     }

-------------- end of patch -------------------

result of patch:

octave-3.0.2:1> dev = comedi_open ("/dev/comedi0")
dev = <comedi_t object>
octave-3.0.2:2> comedi_close (dev)
ans = 0
octave-3.0.2:3> comedi_close (dev)
error: comedi_close: expecting comedi_t object argument
octave-3.0.2:3> 


comedilib allocates comedi_range's for fields within  comedi_t, frees
the ranges in comedi_close and luckily sets the pointers to NULL after
freeing, so it's probably not problematic to wrap comedi_range into an
octave_value as you did.

However, one usually allocates some more with or for comedilib. I
don't know if wrapping those into octave_value is feasible, since I do
not know the Octave internals good enough:

- comedi_polynomial_t

for comedi_to/from_physical (comedi_to/from_phys is deprecated);
allocated by user, filled in by comedilib, one potentially for each
range in each channel in each device;

could probably have values instead of pointers in the class, so
probably no problems due to (de)allocation

- comedi_calibration_t

one instance (necessary) per device, allocated and freed with
comedilib functions, independently of comedi_t structure; user can
create arbitrarily many (equal) instances --- so it should
automatically be freed if an Octave user variable containing it gets
overwritten or out of scope, or if the user clears it; probably you
have means to do this in Octave ...

- some allocations for filling in the comedi_cmd and comedi_insn
  structures

similar problem as the last


Olaf