NumPy frombuffer giving nonsense values when reading C float array on Windows

[sth]

On Tuesday, 26 July 2016 16:36:33 UTC+1, Christian Gollwitzer  wrote:
> Am 26.07.16 um 17:09 schrieb sth:
> > it's difficult to test a .dylib / .so using valgrind
> 
> Why is it difficult? If you have a python script such that
> 
> 	python mytests.py
> 
> loads the .so and runs the tests, then
> 
> 	valgrind --tool=memcheck python mytests.py
> 
> should work. This should immediately spit out an error in case that 
> numpy accesses deleted memory.
> 
> Of course debug information should be enabled when compiling the .so for 
> more useful output, and it could be helpful to have it for numpy/python, 
> too, but that's not a requirement.
> 
> 	Christian

Valgrind isn't giving me any errors; it's reporting possibly-lost memory, but this is constant -- if I call the foreign functions once or 10^6 times, the number of bytes it reports possibly lost (and the number of bytes it reports still reachable) remain the same.