I've mentioned this briefly on the orbit-python IRC channel, and I believe I
have now discovered the source of the problem.
What I'm seeing is double-precision floating-point numbers (CORBA "double")
are being crammed into single-precision floating-point containers on their way
out of orbit-python servers. I have short demo code, in case anyone wants to
see it in action. I first discovered it when trying to pass back a Python
"time.time ()" value, which takes a bit more than "float" precision to
represent accurately.
After building C versions of my test client and server (which made me glad I'm
using Python for my real app!), I determined that the problem was in the o-p
*server* code, which allowed me to look closer at the marshal_double function
in marshal.c and finally see what I think is the error.
Sorry this isn't in the form of a patch against the latest CVS source, but
it's short and simple enough that it shouldn't be any trouble for a person
with CVS access to apply.
In marshal.c, version 0.3.0, line 465 (in marshal_double) looks like this:
v = (CORBA_float)PyFloat_AsDouble(arg);
I believe it should look more like this, using CORBA_double as a cast:
v = (CORBA_double)PyFloat_AsDouble(arg);
If I am mistaken about this, I apologize for wasting everyone's time.
--
Chip
|
